2011 219

N° d’ordre : 40673
UNIVERSITÉ DE LILLE I SCIENCES ET TECHNOLOGIES

École Doctorale Sciences Pour l’Ingénieur
HABILITATION À DIRIGER DES RECHERCHES

Spécialité Sciences Mathématiques
présentée par
Philippe LE GROGNEC
Maı̂tre-Assistant à l’École des Mines de Douai
CONTRIBUTION À LA MODÉLISATION DU COMPORTEMENT

MÉCANIQUE DE STRUCTURES MINCES POUR LA PRÉDICTION
DE DIVERS TYPES D’INSTABILITÉ
Applications multi-matériaux
Soutenue le 25 novembre 2011
Membres du jury :
M. KARAMA Moussa Professeur, ENIT Rapporteur

M. ROELANDT Jean-Marc Professeur, UTC Rapporteur
M. SHAHROUR Isam Professeur, Université de Lille I Rapporteur
M. AZARI Zitouni Professeur, ENIM Examinateur
M. NAÏT ABDELAZIZ Moussa Professeur, Université de Lille I Examinateur
M. HARIRI Saı̈d Professeur, École des Mines de Douai Examinateur
Directeur de Recherche :
Professeur Saı̈d HARIRI, École des Mines de Douai
À la mémoire de mon père
Remerciements
Je tiens tout d’abord à remercier Professeur Zitouni Azari de l’École Nationale

d’Ingénieurs de Metz pour m’avoir fait l’honneur de présider le jury de ma soutenance
d’Habilitation à Diriger des Recherches. Je suis également reconnaissant de l’intérêt que
Professeur Moussa Karama de l’École Nationale d’Ingénieurs de Tarbes, Professeur Jean-
Marc Roelandt de l’Université de Technologie de Compiègne et Professeur Isam Shahrour
de l’Université de Lille I ont porté à ce travail en acceptant d’être rapporteurs de ce
mémoire et je remercie Professeur Moussa Naı̈t Abdelaziz de l’Université de Lille I d’avoir
participé au jury de soutenance.
Je remercie Patricia Krawczak, responsable du département Technologie des Polymères

et Composites & Ingénierie Mécanique, ainsi que Christophe Binétruy et Marie-France
Lacrampe, en tant que directeurs des thèses que j’ai encadrées, pour leur confiance et
la liberté qu’ils m’ont accordée dans le cadre de mes activités de recherche. J’adresse
un remerciement tout particulier à Saı̈d Hariri, mon directeur de recherche, pour m’avoir
accueilli au sein de son équipe, m’avoir appris les ficelles du métier tout en me laissant voler
progressivement de mes propres ailes. J’exprime mes plus sincères remerciements à Anh Le
van pour m’avoir initié à la recherche et transmis sa passion pour la mécanique avec toute
la rigueur et le goût du travail bien fait qui le caractérisent. Je lui dois énormément tant sur
le plan scientifique que personnel, pour m’avoir suivi et soutenu pendant de nombreuses
années.
Je tiens à remercier l’ensemble de mes collègues du département qui m’ont apporté leur
aide, à un moment ou à un autre de ces dernières années, ou tout simplement pour leur
présence et leur gentillesse. Je pense en particulier à Stéphane pour son aide technique et
surtout ses conseils avisés, à Laure et Anne, mes deux optionnaires préférées, et à Domi-
nique, toujours là quand il faut. J’ai enfin une pensée toute particulière pour Abdelhadi,
Fanfei, Patrick et Cyril (sans oublier les ”encadrants de l’ombre”, entre autres, Daniel,
Olivier et Hervé), que j’ai eu le plaisir d’encadrer en thèse et qui ont fortement contribué
aux travaux présentés dans ce mémoire.
En dernier lieu, mes pensées se tournent vers Florence et nos quatre enfants Enora,
Malo, Gael et Brendan. Durant ces derniers mois, les petits ont su se faire tout petits et
les grands se sont montrés à la hauteur. Je les remercie tous et en particulier Florence, qui
a fait preuve d’une patience et d’une compréhension infinies.
v
vi
Table des matières
Remerciements v
Table des figures x
Liste des tableaux xii
Nomenclature xiii
Introduction générale 1
Partie I : Présentation générale 5
1 Curriculum Vitæ 7
2 Activités d’enseignement 10
2.1 Statuts et charges d’enseignement . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Nature des enseignements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Interventions diverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Synthèse des activités de recherche 14
4 Activités d’encadrement 16
4.1 Post-Doctorat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2 Thèses de Doctorat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Masters Recherche . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.4 Projets de Découverte de la Recherche . . . . . . . . . . . . . . . . . . . . . 17
4.5 Études Bibliographiques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Production scientifique et technique 18

5.1 Mémoires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Publications dans des revues à comité de lecture . . . . . . . . . . . . . . . 18
5.3 Publications dans des revues sans comité de lecture . . . . . . . . . . . . . . 20
5.4 Participation à des comités de lecture et conseils d’édition . . . . . . . . . . 20
5.5 Communications dans des conférences internationales avec comité de sélection 20
5.6 Communications dans des conférences nationales avec comité de sélection . 23
5.7 Séminaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6 Collaborations académiques et industrielles 24

6.1 Relations nationales et internationales . . . . . . . . . . . . . . . . . . . . . 24
6.2 Implication dans des projets de recherche collaboratifs . . . . . . . . . . . . 24
6.3 Développement industriel et transfert de technologie . . . . . . . . . . . . . 24
6.3.1 Transfert de technologie . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.3.2 Expertise judiciaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Partie II : Présentation des travaux de recherche 27
1 Plasticité en grandes transformations 29

1.1 Intégration locale en plasticité et modélisation numérique . . . . . . . . . . 30
1.1.1 Historique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.1.2 Modélisation de la plasticité . . . . . . . . . . . . . . . . . . . . . . . 31
1.1.3 Intégration locale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
vii
1.1.4 Tenseur tangent consistant . . . . . . . . . . . . . . . . . . . . . . . 34
1.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.2.1 Phénomène de rochet . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.2.1.1 Effet de rochet uniaxial . . . . . . . . . . . . . . . . . . . . 35
1.2.1.2 Effets de rochet multiaxiaux . . . . . . . . . . . . . . . . . 35
1.2.2 Phénomène de striction . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.3 Modélisation de films polymères en grandes déformations plastiques . . . . 38
1.3.1 Problématique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.3.2 Développement d’un modèle élastoplastique à trois phases pour les
polymères semi-cristallins . . . . . . . . . . . . . . . . . . . . . . . . 40
1.3.2.1 Construction du modèle . . . . . . . . . . . . . . . . . . . . 41
1.3.2.1.1 Cas n°1 . . . . . . . . . . . . . . . . . . . . . . . . 41
1.3.2.1.2 Cas n°2 . . . . . . . . . . . . . . . . . . . . . . . . 41
1.3.2.1.3 Cas n°3 . . . . . . . . . . . . . . . . . . . . . . . . 43
1.3.2.2 Calibration du modèle à partir des essais de traction uniaxiale 45
1.3.2.3 Implantation du modèle dans un code de calcul par
éléments finis . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.3.3 Validation du modèle et applications . . . . . . . . . . . . . . . . . . 48
1.3.3.1 Traction uni- et bi-axiale des films mono-couches . . . . . . 48
1.3.3.2 Applications aux films multi-couches . . . . . . . . . . . . . 48
1.3.3.3 Film troué en traction uniaxiale . . . . . . . . . . . . . . . 49
1.3.3.4 Fissuration d’un film (essai DENT) . . . . . . . . . . . . . 51
1.4 Bilan et perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2 Homogénéisation périodique de matériaux composites 57

2.1 Méthodes d’homogénéisation numérique . . . . . . . . . . . . . . . . . . . . 59
2.1.1 Homogénéisation asymptotique d’une structure périodique . . . . . . 59
2.1.2 Modélisation de l’endommagement . . . . . . . . . . . . . . . . . . . 61
2.2 Application au composite interlock . . . . . . . . . . . . . . . . . . . . . . . 62
2.2.1 Modélisation géométrique . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.2 Propriétés matérielles . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.3 Maillage et calcul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.2.4 Résultats numériques . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.2.5 Comparaison 3D/2D . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.6 Comparaison numérique/expérimentale . . . . . . . . . . . . . . . . 67
2.2.7 Endommagement d’un composite interlock . . . . . . . . . . . . . . 68
2.3 Application au sandwich Napcor . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3.1 Modélisation géométrique . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.2 Propriétés matérielles . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.3.3 Résultats numériques . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.3.4 Comparaison 3D/2D . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.3.5 Endommagement du sandwich Napcor . . . . . . . . . . . . . . . . 75
3 Instabilités géométriques : flambement et post-flambement

élastique/plastique des structures minces ou élancées 80
3.1 Méthode générale de résolution analytique d’un problème de flambement
élastique/plastique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.1 Théorie de la bifurcation plastique . . . . . . . . . . . . . . . . . . . 81
3.1.2 Hypothèses sur les états de contraintes . . . . . . . . . . . . . . . . . 82
3.1.3 Poutre de Timoshenko . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.1.4 Plaque sous compression/traction biaxiale . . . . . . . . . . . . . . . 85
viii
3.1.5 Cylindre sous compression axiale . . . . . . . . . . . . . . . . . . . . 87
3.1.6 Poutre mixte acier/béton avec connexion partielle . . . . . . . . . . 89
3.1.7 Flambement du sandwich Napcor en compression suivant l’épaisseur 92
3.1.7.1 Solution analytique pour la valeur critique . . . . . . . . . 94
3.1.7.2 Validations numérique et expérimentale . . . . . . . . . . . 96
3.2 Développements supplémentaires pour une poutre de Bernoulli . . . . . . . 97
3.3 Modélisation numérique du flambement et du post-flambement
élastoplastiques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.1 Élément fini coque élastoplastique en rotations finies . . . . . . . . . 101
3.3.2 Méthodes numériques pour le traitement des fortes non-linéarités . . 102
3.3.2.1 Pilotage par longueur d’arc . . . . . . . . . . . . . . . . . . 102
3.3.2.2 Méthodes de branchement . . . . . . . . . . . . . . . . . . 104
3.3.3 Validation du programme élément fini . . . . . . . . . . . . . . . . . 105
3.3.4 Poutres élastoplastiques . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.3.5 Poutres à parois minces . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.6 Cylindres sous compression axiale . . . . . . . . . . . . . . . . . . . 111
3.3.7 Cylindres sous pression externe . . . . . . . . . . . . . . . . . . . . . 118
Conclusions et perspectives 125
Références 129
Partie III : Recueil de publications 137
ix
Table des figures
1 Courbes de réponse avec rochet multiaxial . . . . . . . . . . . . . . . . . . . 36
2 Courbes de réponse de la striction d’un barreau cylindrique en traction . . 38
3 Évolution des zones plastiques (en noir) durant la striction d’un barreau
cylindrique en traction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Définition de la limite d’allongement et d’un point de la courbe d’équilibre
pour la détermination des paramètres du modèle 8-chaı̂nes [BER, 1998] . . 44
5 Représentation finale du modèle mécanique pour les polymères semi-cristallins 45
6 Détermination des paramètres du modèle à partir de la courbe de traction
uniaxiale pour le PA6 à 90◦ C . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7 Comparaison des résultats de simulation et expérimentaux à 90◦ C dans les
deux cas de chargement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
8 Courbes simulées de traction uni- et bi-axiale du PE avec différentes valeurs
de CR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9 Comparaison des résultats expérimentaux et de l’application de la loi des
mélanges sur différents assemblages multi-couches à 90◦ C . . . . . . . . . . 50
10 Distribution des contraintes de von Mises pour le déplacement imposé maximal 51
11 Comparaison des courbes force-déplacement expérimentale et numériques
(pour le modèle 3-phases avec plusieurs valeurs de CR et pour un modèle
élastoplastique standard) du problème de traction uniaxiale d’un film troué 52
12 Simulation de la propagation de fissure : déformées et contraintes de von
Mises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
13 Comparaison des résultats de simulation et d’essai de fissuration avec
différentes longueurs initiales de ligament . . . . . . . . . . . . . . . . . . . 54
14 Représentation schématique d’une loi de comportement de type traction-
séparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
15 Fils de trame et de chaı̂ne formant la cellule unité du composite interlock 3D 63
16 Maillage de la cellule unité du composite interlock 3D . . . . . . . . . . . . 64
17 Comparaison numérique 3D/analytique 2D des déplacements en traction
d’une, deux et quatre cellules dans la direction trame d’un composite interlock 66
18 Comparaison numérique 3D/analytique 2D des déplacements en flexion
simple de deux cellules dans la direction trame d’un composite interlock . . 67
19 Comparaison expérimentale/numérique du comportement d’un composite
interlock dans les directions trame et chaı̂ne . . . . . . . . . . . . . . . . . . 68
20 Courbes d’endommagement d’un composite interlock en traction uniaxiale
dans la direction trame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
21 Endommagement d’un composite interlock en traction uniaxiale dans la
direction trame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
22 Sandwich Napcor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
23 Cellules unité de la mousse renforcée . . . . . . . . . . . . . . . . . . . . . . 72
24 Validation numérique en flexion simple de l’homogénéisation périodique de
la mousse renforcée du sandwich Napcor . . . . . . . . . . . . . . . . . . . 75
25 Décohésion le long d’un renfort du sandwich Napcor en flexion 3 points . . 75
26 Décohésion de la mousse renforcée du sandwich Napcor en traction uni-
axiale (état stabilisé) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
27 Courbe de décohésion du sandwich Napcor en traction uniaxiale . . . . . . 77
28 Contraintes critiques de flambement plastique d’une plaque rectangulaire
sous compression uniaxiale . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
29 Pentes initiales des branches bifurquées d’un cylindre élastoplastique sous
compression axiale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
30 Poutre bi-couche comprimée avec interaction partielle . . . . . . . . . . . . 91
x
31 Comparaison des charges critiques numériques et analytiques (sous
différentes hypothèses cinématiques) d’une poutre bi-couche avec interac-
tion partielle pour différentes raideurs à l’interface . . . . . . . . . . . . . . 93
32 Modèle géométrique 2D d’une cellule unité de mousse renforcée Napcor . . 94
33 Premier mode de flambement du sandwich Napcor en compression dans
l’épaisseur . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
34 Courbes pression-déplacement expérimentales et numérique de la compres-
sion dans l’épaisseur du sandwich Napcor . . . . . . . . . . . . . . . . . . . 97
35 Zone de décharge élastique durant le flambement plastique d’une poutre
cantilever . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
36 Courbes d’équilibre post-critiques d’une poutre comprimée issues de la va-
leur critique du module tangent : comparaison des différents développements
fractionnaires avec les résultats numériques . . . . . . . . . . . . . . . . . . 100
37 Déformées successives d’un cylindre élastoplastique pincé . . . . . . . . . . 106
38 Panneau cylindrique sous force ponctuelle : relation force(kN)-
déplacement(mm) au centre . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
39 Diagramme de bifurcation élastoplastique d’une poutre comprimée (premier
mode) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
40 Zones de charge plastique/décharge élastique dans une poutre comprimée
sur la branche bifurquée issue de la valeur critique du module tangent (pre-
mier mode) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
41 Zones de charge plastique/décharge élastique dans une poutre comprimée à
la bifurcation au module réduit (modes 1 et 2) . . . . . . . . . . . . . . . . 110
42 Contraintes critiques de bifurcation plastique d’une poutre à section en C . 111
43 Comparaison des déformées modales d’une poutre à section en C pour
différentes longueurs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
44 Contraintes critiques de bifurcation plastique d’une poutre à section rectan-
gulaire creuse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
45 Comparaison des déformées modales d’une poutre à section rectangulaire
creuse pour différentes longueurs . . . . . . . . . . . . . . . . . . . . . . . . 113
46 Courbe d’équilibre d’une poutre élastoplastique à section rectangulaire
creuse comprimée : sensibilité aux imperfections . . . . . . . . . . . . . . . . 114
47 Modes de bifurcation non-axisymétriques d’un cylindre élastoplastique sous
compression axiale (bords libres, forces imposées) . . . . . . . . . . . . . . . 116
48 Modes primaires (”diamant”) d’un cylindre élastoplastique sous compres-
sion axiale (bords encastrés, forces imposées) . . . . . . . . . . . . . . . . . 116
49 Cylindre élastoplastique sous compression axiale (bords encastrés, forces
imposées) : courbes d’équilibre (modes primaires) . . . . . . . . . . . . . . . 117
50 Modes secondaires (”patte d’éléphant”) d’un cylindre élastoplastique sous
compression axiale (bords encastrés, déplacements imposés) . . . . . . . . . 117
51 Cylindre élastoplastique sous compression axiale (bords encastrés,
déplacements imposés) : courbes d’équilibre (modes secondaires) . . . . . . 118
52 Variations des contraintes résiduelles dans l’épaisseur du cylindre sous pres-
sion externe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
53 Pression externe limite d’un cylindre élastoplastique en fonction de l’ampli-
tude des imperfections géométriques, avec ou sans contraintes résiduelles . . 122
54 Déformées post-critiques du tube cylindrique sous pression externe . . . . . 122
xi
Liste des tableaux
1 Propriétés matérielles pour la modélisation du rochet . . . . . . . . . . . . . 35
2 Modules numériques et expérimentaux équivalents d’un composite interlock
en traction et en flexion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3 Propriétés matérielles des constituants du sandwich Napcor droit . . . . . 72
4 Modules effectifs obtenus par homogénéisation périodique de la mousse ren-
forcée du Napcor droit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5 Modules effectifs obtenus par homogénéisation périodique de la mousse ren-
forcée du Napcor X pour différentes inclinaisons des renforts . . . . . . . . 74
6 Paramètres de décohésion pour l’interface âme/renforts du sandwich Napcor 76
7 Charges critiques de flambement plastique d’une poutre à section rectangu-
laire creuse comprimée . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8 Paramètres géométriques et matériels pour la validation numérique du flam-
bement d’une poutre bi-couche avec interaction partielle . . . . . . . . . . . 92
9 Paramètres géométriques considérés pour le flambement plastique d’un cy-
lindre sous compression axiale . . . . . . . . . . . . . . . . . . . . . . . . . . 115
10 Propriétés matérielles considérées pour le flambement plastique d’un cy-
lindre sous compression axiale . . . . . . . . . . . . . . . . . . . . . . . . . . 115
11 Modes de bifurcation et comportements post-critiques d’un cylindre sous
compression axiale en fonction de l’épaisseur et des conditions aux limites . 119
12 Paramètres géométriques du tube cylindrique sous pression externe . . . . . 120
13 Propriétés matérielles du tube cylindrique sous pression externe . . . . . . . 120
xii
Nomenclature
Ωo Domaine de référence
Ωpf (tc ) Domaine en charge plastique sur la branche fondamentale au point critique
∂Y± Frontières inférieure et supérieure
X Position d’un point courant
n Normale extérieure sortante
L Longueur de la poutre
S Section de la poutre
I Moment quadratique de la poutre
a, b Dimensions de la plaque rectangulaire
l Longueur du cylindre
R Rayon du cylindre
e, h, t Épaisseur de la poutre, de la plaque ou du cylindre
T Épaisseur de la plaque équivalente
E Module d’Young
ν Coefficient de Poisson
G Module de cisaillement
k Facteur de correction pour le cisaillement transverse
(λ, µ) Coefficients de Lamé
h, H Module d’écrouissage
ET Module tangent
ER Module réduit
ksc Raideur longitudinale de la connexion à l’interface des deux poutres
u Vecteur des déplacements
Uf Déplacement sur la branche fondamentale (pré-critique)
F Tenseur gradient de la transformation
ε (εe , εp ) Tenseur des déformations de Green-Lagrange linéarisé
(élastiques et plastiques)
E Tenseur des déformations de Green-Lagrange
B Tenseur des dilatations de Cauchy-Green gauche
E, K Tenseurs des macrodéformations de membrane et de courbure
σ Tenseur des contraintes de Cauchy
Π Tenseur des contraintes de Piola-Kirchhoff de première espèce
Σ Tenseur des contraintes de Piola-Kirchhoff de seconde espèce
N, M Résultantes et moments des contraintes moyennes
p Pression hydrostatique
s, S Tenseurs déviateurs des contraintes
(α, a) Variables duales d’écrouissage isotrope
(α, a) Variables duales d’écrouissage cinématique
f Fonction critère de plasticité
λ̇ Multiplicateur plastique
W Énergie potentielle de déformation
whard (wα , wα ) Énergies d’écrouissage (isotrope et cinématique)
D Tenseur d’élasticité
Dp Tenseur tangent matériel élastoplastique
K (K , Kp )
e Tenseur tangent nominal (élastique et plastique)
λc (λT , λR , λE ) Valeur critique (du module tangent, du module réduit et élastique)
X (U, V, W, Θ) Mode de flambement (et ses composantes)
xiii
Knn , Kss , Ktt Modules élastiques dans la zone de cohésion dans les directions normale et
tangentielles
tn , ts , tt Contraintes dans la zone de cohésion dans les directions normale et
tangentielles
δn , δs , δt Déformations dans la zone de cohésion dans les directions normale et
tangentielles
G n , Gs , Gt Énergies dissipées dans la zone de cohésion dans les directions normale et
tangentielles
L Fonction de Langevin
I Tenseur unité du second ordre
I Tenseur unité du quatrième ordre
xiv
Introduction générale
Ce mémoire synthétise l’ensemble de mes travaux de recherche menés à l’École Cen-

trale de Nantes, à l’Université de Nantes puis à l’École des Mines de Douai au sein du
département Technologie des Polymères et Composites & Ingénierie Mécanique (TPCIM).
L’essentiel de ces travaux concerne la modélisation théorique et numérique en mécanique
non-linéaire des solides et s’intéresse de près ou de loin aux instabilités dans les structures.
Ma thèse de doctorat, effectuée au Laboratoire Mécanique et Matériaux (Division
Structures) de l’École Centrale de Nantes de 1998 à 2001, contient déjà la plupart des
ingrédients qui feront le fil conducteur de ma carrière de chercheur. L’objectif de ma
thèse est de contribuer à l’étude du flambement et du post-flambement des structures
élastoplastiques. C’est l’occasion pour moi de mettre en œuvre déjà un certain nombre
d’outils de calcul (méthodes de résolution analytique et programmes de calcul numérique
par éléments finis) qui seront développés au cours des activités ultérieures selon les ap-
plications rencontrées. La notion de mécanique non-linéaire apparaı̂t déjà sous ses deux
formes principales, au travers des non-linéarités matérielles introduites par le comporte-
ment élastoplastique et des non-linéarités géométriques du fait des grandes transformations
et des phénomènes d’instabilité mis en jeu. En 2003, j’ai été recruté comme enseignant-
chercheur dans l’équipe ”Calcul des Structures” du département Mécanique et Compor-
tement des Matériaux de l’École des Mines de Douai. J’ai alors intégré la thématique
générale portant sur la fiabilité des structures et j’ai pris part aux travaux en cours sur la
modélisation de la fissuration de structures sous pression. En 2005 est survenue la fusion
de deux départements à l’École des Mines de Douai : le département Mécanique et Com-
portement des Matériaux et le département Technologie des Polymères et Composites,
spécialisé dans la mise en œuvre des matériaux polymères et composites. De nouvelles
thématiques ont naturellement émergé de ce rapprochement et mes activités se sont diver-
sifiées, ou tout au moins tournées vers des applications polymères et composites, avec tous
les développements et les spécificités que cela sous-entend.
Mes premiers travaux de recherche portent sur la modélisation du comportement non-
linéaire de structures mécaniques en présence d’instabilités. Le flambement (instabilité
géométrique) constitue depuis lors ma principale préoccupation. Il s’agit de la source prin-
cipale de ruine des structures minces sous chargement de compression dominante. De
nombreux travaux théoriques/numériques dans ce domaine sont déjà disponibles dans la
littérature. Mais ceux-ci ne donnent pas toujours, à mon sens, entière satisfaction, dans
la mesure où la plupart des solutions proposées sont semi-analytiques. Dans un sens, les
résultats associés ne sont souvent pas directement exploitables (la résolution n’est pas
menée analytiquement jusqu’à son terme) et, à l’inverse, les méthodes présentées sont
trop restrictives (elles ne concernent qu’un type d’application et ne sont pas facilement
généralisables). Mon objectif est de proposer des outils de modélisation à la fois génériques
(pouvant s’appliquer à diverses géométries, avec différents comportements matériels, sous
des chargements variés), et susceptibles, par l’utilisation d’hypothèses appropriées, de me-
ner à des expressions explicites des chargements critiques.
Pour être tout à fait général, on fait le choix de développer un code de calcul par
éléments finis qui n’impose des restrictions d’aucune sorte. L’intérêt d’un tel programme
”maison” est de pouvoir développer des outils suffisamment robustes pour une description
incrémentale efficace du post-flambement jusqu’à un stade post-critique avancé. Il s’agit
principalement des méthodes de branchement qui permettent de décrire l’ensemble des
courbes triviale et bifurquées, pour les différents modes et à différents ordres, et qui ne
figurent pas dans la plupart des codes commerciaux et des programmes ”maison” d’autres
auteurs. En l’absence de telles techniques, la seule solution pour parvenir au comportement
1
post-bifurqué est de dégénérer le problème de bifurcation en introduisant des imperfections,
ce qui limite la prospection des courbes d’équilibre au seul premier mode primaire. Le code
de calcul développé au cours de ces travaux permet de traduire l’ensemble du squelette
des courbes d’équilibre, dans l’hypothèse d’une structure parfaite (sans aucun défaut).
Ce résultat complet est essentiel dans la mesure où il constitue un support sur lequel
les courbes réelles de comportement (dans des conditions réalistes en présence de défauts
donnés) s’appuieront. En pratique, les structures flambent suivant leur premier mode (dès
le premier point critique rencontré sur la branche triviale), ce qui rend discutable l’intérêt
des modes suivants. En revanche, dans certains cas, la présence d’un mode secondaire (un
nouveau point critique sur la branche bifurquée) ne peut être identifiée et correctement
gérée que par l’utilisation de telles méthodes numériques. Une formulation lagrangienne
totale est adoptée pour la modélisation numérique du problème de mécanique en grandes
transformations. Cette formulation s’avère idéale par sa simplicité de mise en œuvre (le
problème d’objectivité est naturellement résolu et le formalisme de la plasticité est similaire
à celui des petites transformations) et de par la nature des transformations rencontrées pour
les applications de flambement de structures minces envisagées (on se place dans le cadre
de grands déplacements, de rotations finies, avec des déformations modérées). Ce choix est
rarement fait dans la littérature et en particulier dans les codes commerciaux qui visent
également des problèmes avec de très grandes déformations, pour lesquels cette formulation
n’est plus indiquée. C’est du coup l’occasion de valider une telle formulation dans sa
capacité à rendre compte du flambement et du post-flambement avancé des structures.
À l’opposé de cet outil de calcul numérique, on choisit de mettre en œuvre une
méthode de résolution analytique d’un problème de flambement, basée sur la même formu-
lation. Moyennant quelques hypothèses qu’on démontre être peu restrictives, l’approche
développée permet d’aboutir à des solutions analytiques explicites de la charge critique (le
nombre d’applications traitées avec cette méthode témoigne de sa généralité). L’écriture
du problème à partir de l’équation de bifurcation sous sa forme variationnelle permet d’in-
troduire différentes lois de comportement, de réduire les modules 3D aux cas de poutres
ou plaques/coques si nécessaire, de considérer n’importe quelle cinématique et de faire
toutes sortes de simplifications avec la plus grande souplesse. En termes de charge critique,
on retrouve avec cet outil analytique tout un ensemble de résultats connus de manière
systématique, auxquels s’ajoutent certains résultats originaux. L’obtention de la pente
initiale de la branche bifurquée vient compléter la solution dans certains cas, en fournis-
sant un élément de réponse quant à la stabilité de la solution post-critique. Dans le cas
particulier d’une poutre, une méthode alternative permet même d’aller plus loin dans le
développement asymptotique de la branche bifurquée.
Même si certaines applications ont été effectuées dans un cadre élastique linéaire, le
choix d’une loi de comportement élastoplastique s’est globalement imposé dans les outils
développés pour l’analyse du flambement. En effet, dans les applications pratiques (entre
autres, pour des matériaux métalliques), la plasticité apparaı̂t presque systématiquement
durant le post-flambement et fréquemment avant le chargement critique lorsque les struc-
tures deviennent épaisses. D’un point de vue scientifique, la résolution d’un problème de
bifurcation plastique (lorsque la plasticité survient avant le flambement) ne revient pas
simplement à ajouter la composante plasticité à un calcul de flambement élastique, mais
relève de nouvelles questions théoriques et pose de nouvelles difficultés spécifiques. La
nécessité d’implanter une telle loi de comportement dans le programme élément fini ”mai-
son” a entraı̂né une réflexion sur d’autres problèmes en lien avec la plasticité. À partir
d’un socle commun, plusieurs programmes de calcul par éléments finis ont été développés
dans différents cadres (en petites ou grandes transformations) avec différentes formulations
(3D, 2D, coque) pour diverses applications. Outre les instabilités géométriques, plusieurs
autres types d’application ont été abordés dans une moindre mesure. On peut citer les
2
phénomènes de striction (instabilité matérielle) et de rochet (en plasticité cyclique) dont
la prédiction nécessite l’utilisation de types et lois d’écrouissage distincts. Par souci de
généralité et d’approfondissement, une formulation générale de l’intégration locale en plas-
ticité a été mise au point, pour un critère et une loi d’écrouissage arbitraires.
Durant les cinq dernières années, au département TPCIM de l’École des Mines de
Douai, s’est ajoutée une dimension ”matériaux” avec l’opportunité de travailler sur des
applications polymères et composites. Les nouvelles activités entreprises, quel que soit le
matériau considéré, ont en commun l’idée de vouloir mettre en relation le comportement
mécanique avec la structure du matériau issue du procédé de fabrication. Ma participa-
tion dans ces nouvelles activités concerne principalement la modélisation du comporte-
ment mécanique des matériaux considérés et a été l’occasion d’étendre le champ d’appli-
cation des méthodes développées précédemment. D’une part, on s’est intéressé aux très
grandes déformations élastoplastiques de films polymères. L’étude réalisée débute par la
construction d’un modèle de comportement original à trois phases, qui s’appuie sur des
considérations morphologiques, suivie de l’identification des paramètres du modèle sur la
base d’essais de référence, et s’achève par la simulation numérique par éléments finis de cas
tests validés par des résultats expérimentaux (dont la fissuration des films moyennant l’uti-
lisation d’éléments de cohésion). La pierre d’achoppement de ce travail consiste à passer
du modèle théorique au programme numérique, par l’intégration locale des équations de
comportement élastoplastiques qui reprend les concepts généraux utilisés précédemment
avec quelques spécificités dues au comportement hyperélastique d’une des phases.
D’autre part, on s’est intéressé au comportement mécanique de composites à renforts
3D périodiques. La première nécessité pour de tels matériaux consiste à homogénéiser le
comportement (en élasticité linéaire pour commencer) afin de définir un comportement
équivalent de plaque (compte tenu de l’épaisseur mince des structures envisagées). La
notion d’endommagement est abordée par l’étude de la décohésion fibres/matrice. Parmi les
matériaux envisagés, le cas d’un sandwich à renforts dans l’épaisseur soulève des problèmes
de flambement, lorsqu’il est sollicité en compression. Une partie de ce travail s’inscrit donc
dans le prolongement des activités menées depuis ma thèse sur le flambement, et consiste
à adapter certaines des méthodes déjà mises en œuvre au cas des composites. L’analyse
du flambement d’autres structures composites (poutres mixtes avec connexion partielle ou
encore tubes cylindriques composites) est menée en parallèle.
Mon activité de recherche peut se résumer finalement en la conception d’outils de si-
mulation (numérique par éléments finis) à partir de formulations théoriques spécifiques
en mécanique non-linéaire au service du dimensionnement de structures minces (poutres,
plaques, coques cylindriques, films, ...). Parmi les différents modes de ruine rencontrés
en pratique, le flambement (et post-flambement) élastique/plastique représente l’un des
facteurs les plus dimensionnants pour ce type de géométrie et constitue donc mon axe
de recherche principal. L’implantation numérique de modèles de plasticité s’avère natu-
rellement nécessaire pour les applications relatives au flambement plastique, et permet de
s’intéresser à d’autres types d’instabilité (striction) et au phénomène de rochet, ou encore à
la fissuration de films plastiques. Enfin, l’introduction de méthodes d’homogénéisation dans
la démarche de modélisation s’avère indispensable lorsqu’on s’intéresse à des matériaux
composites fortement hétérogènes, et ouvre la voie à de nouvelles perspectives dans la
thématique du flambement qui est la mienne.
Ce mémoire s’articule donc autour de ces trois thématiques que sont le flambement, la
plasticité et l’homogénéisation, et décrit dans chaque cas la problématique, les hypothèses
de travail, les outils développés ou utilisés, les principaux résultats obtenus et les perspec-
tives envisageables. Les travaux présentés ont été réalisés dans le cadre d’encadrements
de thèses et/ou de contrats de recherche, et ont donné lieu à un certain nombre de publi-
3
cations auxquelles il est fait référence et qui complètent le présent rapport sous la forme
d’annexes.
4
Partie I
Présentation générale
6
1 Curriculum Vitæ
Philippe LE GROGNEC 1A rue de la Sensée
Né le 9 janvier 1975 à Vannes (56) 59552 COURCHELETTES
Marié, 4 enfants Tél. : 03-27-96-36-40
Courriel : philippe.le.grognec@mines-douai.fr
Expérience Professionnelle
Depuis décembre 2003 : Enseignant-Chercheur au Département Technologie des

Polymères et Composites & Ingénierie Mécanique à l’École des Mines de Douai
(Maı̂tre-Assistant des Écoles des Mines depuis mai 2007). Activités d’enseignement,
d’encadrement, de développement industriel et de recherche contractualisée.
De septembre 2002 à août 2003 : Attaché Temporaire d’Enseignement et de Re-

cherche (mi-temps) au Département de Physique à l’Université de Nantes. Activités
d’enseignement et de recherche.
D’octobre 2001 à août 2002 : Attaché Temporaire d’Enseignement et de Recherche

(temps plein) au Département Mécanique, Matériaux et Génie Civil à l’École
Centrale de Nantes. Activités d’enseignement et de recherche.
D’octobre 1998 à septembre 2001 : Allocataire de recherche MENRT et Moniteur

au Laboratoire Mécanique et Matériaux (Division Structures) à l’École Centrale de
Nantes. Activités d’enseignement et de recherche.
D’octobre 1997 à juillet 1998 : Scientifique du Contingent à l’École Nationale des

Sous-Officiers d’Active de Saint-Maixent-l’École. Activités d’enseignement et d’en-
cadrement.
De juillet 1996 à octobre 1996 : Ingénieur Stagiaire à Citroën La Janais, Rennes.
Formation
Novembre 2001 : Thèse de Doctorat de l’École Centrale de Nantes et de l’Université

de Nantes : Contribution à la modélisation du flambement et du post-flambement
des structures élastoplastiques. Mention Très Honorable. Laboratoire d’accueil :
Laboratoire Mécanique et Matériaux (Division Structures) à l’École Centrale de
Nantes. Directeur de thèse : Pr. A. Le van (Université de Nantes). Président : Pr.
A. Combescure (INSA de Lyon). Rapporteurs : Q.S. Nguyen (D.R. à l’École Poly-
technique, Palaiseau) et A. Léger (D.R.A. à EDF-R&D, Clamart). Examinateurs :
Pr. G. Rio (Université de Bretagne Sud), Pr. C. Wielgosz (Université de Nantes) et
Pr. D. Le Houedec (École Centrale de Nantes).
Septembre 1997 : Diplôme d’Études Approfondies en Génie Mécanique option

Mécanique des Structures à l’École Centrale de Nantes : Un modèle d’écrouissage
combiné isotrope-cinématique en plasticité. Mention Très Bien. Encadrement : A.
Le van.
7
Juin 1997 : Diplôme d’Ingénieur de l’École Centrale de Nantes option Calcul des
Structures. Mention Bien.
De 1992 à 1994 : Mathématiques Supérieures et Spéciales M’ au Lycée Chateaubriand,

Rennes.
Juin 1992 : Baccalauréat série C. Mention Bien.
Compétences Scientifiques et Pédagogiques
Mécanique non-linéaire des structures :

– Bifurcation, instabilité : flambement plastique de poutres, coques, membranes, et
phénomène de striction ;
– Hyperélasticité et plasticité en grandes transformations, plasticité cyclique : effet
de rochet ;
– Homogénéisation périodique de matériaux composites ;
– Mécanique de la rupture, fatigue : propagation de fissures ;
– Applications multi-matériaux : matériaux métalliques, polymères et composites.
Modélisation analytique et simulation numérique par éléments finis (réalisation de codes
de calcul).
Enseignement de Mécanique des Solides, Mécanique des Vibrations, Mécanique des Mi-
lieux Continus, et de Méthodes Numériques en Mécanique Non-Linéaire (Éléments
Finis).
Références Industrielles et Académiques (collaborations)
– CETIM, IFREMER, DGA, AXS Analyse de Structures, bureau d’études PRODESS,

Faiveley Transport NSF.
– Cabinet d’expertises TECH PLUS.
– Université Technologique de Lisbonne (Portugal), Institut de Mécanique (RWTH-
IAM) d’Aix-la-Chapelle (Allemagne), Université de Galati (Roumanie), Université
de Technologie de Compiègne, Université des Sciences et Technologies de Lille, École
Nationale Supérieure des Arts et Industries Textiles de Roubaix, Institut National
des Sciences Appliquées de Rennes.
Publications et Communications
– 18 publications (dont 16 dans des revues à comité de lecture) + 1 soumise.

– 24 communications dans des conférences avec comité de sélection (dont 18 interna-
tionales).
Connaissances en Informatique
– Systèmes d’exploitation : MS Windows, Unix, Linux.

– Langages de programmation et logiciels scientifiques : Fortran, Python, Maple.
– Logiciels de simulation numérique : Abaqus, Adina, Castem, Samcef, Algor, SDS.
8
Langues Étrangères et Divers
– Anglais : Lu, parlé, écrit.

– Allemand : Niveau scolaire.
– Permis de conduire B.
– Piano, direction chorale, tennis de table.
9
2 Activités d’enseignement
Mes activités d’enseignement ont débuté en 1998 et ont été menées successivement au
sein des trois établissements que sont l’École Centrale de Nantes (ECN), l’Université de
Nantes (UN) et l’École des Mines de Douai (EMD).
2.1 Statuts et charges d’enseignement

Moniteur à l’École Centrale de Nantes (1998-2001) : Durant les trois années de
ma thèse de doctorat, j’ai effectué un service d’enseignement de 64 heures (éq-TD)
par an en tant que moniteur au sein du Département Mécanique, Matériaux et Génie
Civil à l’École Centrale de Nantes. Cet enseignement concernait les élèves-ingénieurs
de l’École Centrale de Nantes en 2ème année dans leur formation de tronc commun
(2A-TC) ainsi que de spécialité pour l’option Simulation en Ingénierie Mécanique
(2A-SIM).
ATER à l’École Centrale de Nantes (2001-2002) : J’ai poursuivi, l’année suivante,

mon enseignement au sein du même département, en tant qu’Attaché Temporaire
d’Enseignement et de Recherche (à temps plein), à raison de 192 heures (éq-TD).
Le contenu de mon service s’est naturellement diversifié, concernant désormais les
élèves-ingénieurs de l’École Centrale de Nantes en 1ère et 2ème année de formation
de tronc commun (1A-TC et 2A-TC) et l’option Simulation en Ingénierie Mécanique
en 2ème et 3ème année de formation de spécialité (2A-SIM et 3A-SIM).
ATER à l’Université de Nantes (2002-2003) : J’ai effectué ensuite un service d’en-

seignement de 96 heures (éq-TD) au sein de la filière Mécanique du Département
de Physique, en tant qu’Attaché Temporaire d’Enseignement et de Recherche (à
mi-temps) à l’Université de Nantes. Les étudiants concernés sont en 2ème année de
DEUG (MIAS), en Licence et Maı̂trise de Mécanique, et en DESS de Modélisation
Numérique en Mécanique.
Enseignant-Chercheur à l’École des Mines de Douai (depuis décembre 2003) :

Je suis actuellement enseignant-chercheur à l’École des Mines de Douai au
Département Technologie des Polymères et Composites & Ingénierie Mécanique. Le
volume des enseignements dispensés est d’environ 150 heures (éq-TD) par an et
concerne les élèves-ingénieurs de l’École des Mines de Douai en 2ème année (tronc
commun) et en 3ème et 4ème année pour les options Ingénierie Mécanique (IM),
Technologie des Polymères et Composites (TPC) et Génie Civil (GC).
2.2 Nature des enseignements
Les disciplines enseignées sont présentées ci-dessous, en décrivant brièvement, pour

chacune d’entre elles, leur contenu, les établissements et le public concernés, ainsi que les
charges horaires associées.
Travaux Dirigés de Mécanique des Solides : Le contenu de cet enseignement, qui

s’intéresse à la dynamique des solides rigides, comprend l’apprentissage des notions
de cinématique, de géométrie des masses et de cinétique, l’utilisation du Principe
Fondamental de la Dynamique, du Principe des Travaux Virtuels et des Équations
10
de Lagrange, et l’étude de différentes applications générales telles que les phénomènes
de frottement et de choc.
– ECN : 50h en 1A-TC (2001-2002)
– UN : 30h en DEUG MIAS 2A (2002-2003)
Travaux Dirigés de Mécanique des Vibrations : Le contenu de ces travaux dirigés

illustre les notions vues en cours, à savoir l’analyse modale de systèmes à un ou
plusieurs degrés de liberté (recherche de fréquences et modes propres), l’obtention
des oscillations libres ou forcées par superposition modale, l’amortissement. Un en-
seignement de spécialité vient compléter ces notions fondamentales avec l’utilisation
des méthodes de condensation numérique et de sous-structuration.
– ECN : 20h en 2A-TC et 5h en 2A-SIM (2001-2002)
Travaux Pratiques de Mécanique des Vibrations : Ces travaux pratiques ont pour
but de familiariser les étudiants avec l’analyse modale expérimentale et les techniques
associées (utilisation de pots vibrants, d’accéléromètres, interprétation de la réponse
en fréquence). Les résultats sont comparés à des solutions de référence analytiques
ou numériques. Dans la continuité des travaux dirigés, des travaux pratiques de
modélisation numérique mettant en œuvre les méthodes de condensation numérique
et de sous-structuration permettent d’illustrer l’efficacité de telles méthodes sur des
problèmes complexes.
– ECN : 100h en 2A-TC et 25h en 2A-SIM (1998-2002)
Cours et Travaux Dirigés de Mécanique des Milieux Continus/Élasticité : Ce

cours constitue une introduction à la mécanique des solides déformables dans le cadre
de l’élasticité linéaire et établit les fondements nécessaires au dimensionnement et
au calcul de structures en général. Il définit donc dans un premier temps les notions
de contraintes et de déformations dans un milieu continu. Les lois de comportement
(et on se restreint au cas de l’élasticité) permettent de relier ces deux grandeurs. Il
s’intéresse ensuite brièvement aux notions d’énergie de déformation et de critère de
limite élastique. Enfin, il présente la mise en équations (et les méthodes de résolution
analytique associées) d’un problème continu à partir du Principe Fondamental de
la Statique ou d’une approche énergétique comme les principes variationnels. Les
travaux dirigés consistent en différents types d’application, essentiellement dans le
cadre de l’élasto-statique, mais aussi en thermo-élasticité ou en élasto-dynamique.
À l’École des Mines de Douai, un polycopié (Élasticité, P. Le Grognec, 60 pages)
est fourni aux étudiants, reprenant les définitions, démonstrations et méthodes
fondamentales.
– ECN : 20h en 2A-TC (2001-2002)
– EMD : 3A-IM, 3A-TPC, 3A-GC (40h/an)
Travaux Pratiques de Résistance des Matériaux : L’objectif est de mettre en pra-

tique les notions abordées en cours à savoir la mesure des déformations au moyen
de jauges extensométriques, et l’étude expérimentale de poutres hyperstatiques, de
concentrations de contraintes dans une plaque trouée et d’une enveloppe mince sous
pression. Une initiation aux éléments finis est également proposée sur le logiciel RDM
Le Mans.
– EMD : 2A-TC (20h/an)
11
Cours et Travaux Dirigés sur la Méthode des Éléments Finis : Ce cours porte
sur les méthodes d’approximation numérique en général et sur la méthode des
éléments finis en particulier, appliquées essentiellement à des problèmes de mécanique
des structures. L’objectif est d’initier les étudiants aux notions de discrétisation
géométrique, d’interpolation élément fini et d’assemblage, ainsi qu’aux outils
mathématiques associés (transformation géométrique, intégration numérique), per-
mettant la résolution approchée d’un problème à géométrie et chargement complexes,
à partir du Principe des Travaux Virtuels en mécanique des solides. Dans le cadre
des travaux dirigés, la construction d’éléments finis classiques (barre, poutre, 2D, ...)
et la résolution ”manuelle” de problèmes simples en élasticité linéaire, en statique
ou dynamique (treillis, portiques, plaques, ...), permettent aux étudiants de mieux
apprécier le fonctionnement d’un code de calcul par éléments finis, dont l’utilisation
est la finalité de cet enseignement.
– ECN : 165h en 2A-TC (1998-2002)
– EMD : 3A-IM (20h/an)
Travaux Pratiques sur la Méthode des Éléments Finis : L’objectif de ces travaux
pratiques est de familiariser les étudiants à l’utilisation d’un code de calcul industriel
(Samcef, Algor, Castem, Adina ou Abaqus, selon les années et les établissements)
basé sur la méthode des éléments finis. Ils consistent en l’apprentissage de la démarche
de modélisation numérique, comprenant la mise en œuvre des données et l’analyse
des résultats et de leur convergence, en s’appuyant sur les acquis théoriques du cours
associé. La réalisation d’un projet permet de valider l’ensemble des acquis théoriques,
la maı̂trise du logiciel et l’esprit d’analyse et de synthèse dans le domaine de la
modélisation numérique.
– ECN : 45h en 2A-TC (1998-2002)
– UN : 45h en Licence de Mécanique (2002-2003)
– EMD : 3A-IM, 3A-TPC (40h/an)
Travaux Pratiques de Méthodes Numériques en Mécanique Non-Linéaire :

Ces travaux pratiques concernent toujours la modélisation numérique par éléments
finis mais cette fois dans le cadre de la mécanique non-linéaire. Ils consistent en
particulier en l’apprentissage des méthodes de calcul incrémentales, nécessaires à
la résolution de problèmes de mécanique non-linéaire. Les domaines d’application
couvrent à la fois l’étude des non-linéarités matérielles et géométriques avec
l’introduction d’un comportement élastoplastique et la prise en compte des grandes
transformations. L’analyse du flambement (et du post-flambement) de structures
minces ou élancées, mettant en jeu la notion de bifurcation et/ou de chargement
limite dans le cadre de grandes transformations (éventuellement élastoplastiques),
constitue la majeure partie des travaux effectués. Des projets sur codes de cal-
cul regroupant ces différents aspects de la mécanique non-linéaire clôturent cet
enseignement en mettant l’accent sur la modélisation et sur l’analyse des résultats.
– UN : 10h en Maı̂trise de Mécanique et 45h en DESS MNM (2002-2003)
– EMD : 4A-IM (40h/an)
Travaux Dirigés de Stabilité : Ces travaux dirigés illustrent les notions de stabilité et
bifurcation de systèmes discrets ou continus sur des exemples de flambement (ou
claquage) de poutres ou plaques minces.
– ECN : 5h en 3A-SIM (2001-2002)
12
2.3 Interventions diverses
Formations professionnelles :
– Contribution à l’action de formation professionnelle ”Calcul des composites”
pour un cadre de la société Stratiforme (2005-2006) : formation théorique en
”Élasticité” (40h), et formation théorique et pratique (Abaqus) en ”Éléments
Finis” (20h).
– Formation professionnelle ”Modélisation numérique par éléments finis” pour un

cadre de la société Faiveley Transport NSF : formation théorique et pratique
(Abaqus) sur la méthode des éléments finis (12-13 mars 2008).
Enseignement dans le cadre d’un accord Erasmus : Missions à l’Université ”Du-

narea de Jos” de Galati (Roumanie), 15-19 mars 2008 et 16-20 mai 2009.
Projets de Découverte de la Recherche : Initiation à la recherche des élèves-

ingénieurs en 3ème année (EMD).
Visites de stage : Élèves-ingénieurs en 1ère , 2ème et 3ème année, Projets de Fin d’Études
(EMD).
Jurys : Participation à des jurys d’admission des élèves-ingénieurs, d’Études Bibliogra-

phiques, de stages (1ère , 2ème et 3ème année), de Projets de Fin d’Études et Projets
Scientifiques et Techniques (EMD).
13
3 Synthèse des activités de recherche
Mes travaux de recherche, initiés en thèse et poursuivis à l’École des Mines de Douai,
évoluent de manière générale autour de la modélisation théorique et numérique du com-
portement non-linéaire des matériaux et des structures. Les thèmes abordés concernent
effectivement à la fois la mécanique des matériaux, au sens où l’on s’intéresse au com-
portement élastique/plastique de matériaux éventuellement hétérogènes (par le biais de
combinaisons ou de méthodes d’homogénéisation à différentes échelles), et la mécanique
des structures, avec l’étude de divers phénomènes de ruine dont principalement le flambe-
ment.
Mon premier axe de recherche s’intéresse à l’identification et à la modélisation

numérique de lois de comportement. L’essentiel de mes travaux concerne l’élastoplasticité
en petites et grandes transformations, avec différentes formulations, lois et critères, selon
le type de matériau considéré. Mon intérêt s’est récemment porté vers la caractérisation
de films plastiques multi-couches (PE/liant/PA) permettant de simuler de manière unifiée
leur réponse en mono- et bi-étirage, en vue de l’optimisation de leur conception. Un modèle
en trois phases combinées a été spécialement développé et calibré, en relation avec la nature
micro-structurale des polymères semi-cristallins considérés. La mise en œuvre numérique
de l’intégration locale en plasticité s’appuie sur les méthodes et outils réalisés auparavant
dans le cadre, entre autres, de travaux sur la striction (instabilité matérielle) et sur l’effet
de rochet sous sollicitation cyclique (avec un écrouissage cinématique non-linéaire), dans
le contexte des matériaux métalliques. À mi-chemin entre les aspects matériau et struc-
ture, des travaux sont actuellement menés dans le domaine des composites renforcés de
fibres. L’objectif est essentiellement d’évaluer les modules élastiques effectifs du matériau
homogène équivalent, par l’utilisation de méthodes numériques ou de solutions analytiques.
La démarche d’homogénéisation asymptotique périodique mise en œuvre, permettant de
remplacer un matériau composite hétérogène par une plaque équivalente et ses raideurs
effectives, a été appliquée successivement à un composite à renfort interlock d’angle couche-
à-couche et à un sandwich renforcé dans l’épaisseur par aiguilletage transverse. Pour ce
dernier, une difficulté majeure réside en la détermination des raideurs en cisaillement trans-
verse.
Mon second axe de recherche concerne la modélisation de la réponse de structures

minces (poutres, plaques, cylindres) à des sollicitations diverses. L’instabilité géométrique
(ou flambement) est le mode de ruine principalement envisagé, car prépondérant pour des
structures minces sous compression dominante. Des travaux ont tout d’abord été menés
sur le flambement plastique, mettant l’accent sur sa nature spécifique (présence de conti-
nua de points de bifurcation) et sur la description non seulement des points critiques mais
aussi des réponses post-critiques, riches en informations sur la stabilité. De nombreuses
applications de claquage ou de bifurcation ont permis de valider les modèles développés,
parmi lesquelles des problèmes de poutres (à parois minces), de plaques ou de cylindres
(avec contraintes résiduelles) sous compression axiale ou pression externe. L’essentiel de ces
activités repose sur le développement de solutions analytiques originales (développements
en séries fractionnaires) et d’outils spécifiques de simulation par éléments finis s’appuyant
sur des méthodes numériques avancées (longueur d’arc, branchement), compte tenu des
très fortes non-linéarités rencontrées. La démarche analytique développée à partir de la
14
résolution de l’équation de bifurcation a été adaptée au cas de structures composites (en
élasticité) : une poutre mixte acier-béton avec connexion partielle et le sandwich à renforts
transverses en compression dans l’épaisseur. En parallèle, la notion de fissuration a été
abordée dans le cadre des matériaux précédemment cités. Les premiers travaux dans ce
domaine concernent les matériaux métalliques et portent sur la nocivité des défauts dans
les équipements sous pression. La caractérisation (semi-analytique) de la nocivité d’une fis-
sure, en élasticité ou plasticité, validée expérimentalement par extensométrie et émission
acoustique, permet d’évaluer la durée de vie résiduelle de la structure considérée, dans le
cadre d’un comportement en fatigue sous chargement cyclique. La fissuration sous char-
gement statique de films plastiques a été simulée numériquement en utilisant le modèle à
trois phases et testée expérimentalement pour validation. Enfin, une réflexion sur l’endom-
magement des matériaux composites à renforts fibreux a été menée moyennant l’utilisation
d’éléments de cohésion dans le cadre de l’homogénéisation.
15
4 Activités d’encadrement
4.1 Post-Doctorat
PD1 - A. El Hakimi, Modélisation numérique de la propagation de fissures dans des
structures sous pression, en collaboration avec le CETIM (Senlis), 2007-2008.
4.2 Thèses de Doctorat

Th1 - A. El Hakimi, Étude numérique et expérimentale de la nocivité des défauts
dans des coques cylindriques et sphériques sous pression, co-dirigée par Pr. S. Hariri
et Pr. A. Laksimi, Université de Technologie de Compiègne, soutenue le 30 juin 2006.
Th2 - F. Zeng, Contribution à la modélisation du comportement mécanique en

grandes déformations élastoplastiques de films plastiques d’emballage, dirigée par
Pr. M.F. Lacrampe, Université des Sciences et Technologies de Lille, soutenue le 7
décembre 2010.
Th3 - P. Lapeyronnie, Mise en œuvre et comportement mécanique de composites

organiques renforcés de structures 3D interlocks, dirigée par Pr. C. Binétruy,
Université des Sciences et Technologies de Lille, soutenue le 14 décembre 2010.
Th4 - C. Lainé, Contribution à la modélisation du comportement mécanique de struc-

tures sandwich 3D par homogénéisation périodique, co-dirigée par Pr. C. Binétruy
et Pr. S. Panier, Université des Sciences et Technologies de Lille, soutenance prévue
en 2012.
Contributions ponctuelles :
Th5 - E. Lagardère, Comportement au fluage de structures sandwich thermoplas-

tiques, dirigée par Pr. P. Krawczak, Université des Sciences et Technologies de Lille,
soutenue le 23 novembre 2006.
Th6 - J. Guilleminot, Modélisation stochastique mésoscopique de milieux aléatoires :

Application à un polymère renforcé de fibres longues, co-dirigée par Pr. C. Binétruy
et Pr. D. Kondo, Université des Sciences et Technologies de Lille, soutenue le 9
décembre 2008.
Th7 - T.T. Dang Huu, Étude de l’infusion et de l’injection de composites dans une
cavité déformable, dirigée par Pr. C. Binétruy, École Nationale Supérieure des
Mines de Paris, soutenue le 4 décembre 2009.
Th8 - F. Zhang, Modélisation en contexte aléatoire des propriétés de transport des

matériaux fibreux, co-dirigée par Pr. C. Binétruy et Pr. C. Robin, Université des
Sciences et Technologies de Lille, soutenance prévue en 2011.
Th9 - O. Syerko, Modélisation du comportement de milieux fibreux 3D secs par ap-

proche d’homogénéisation multi-échelle, co-dirigée par Pr. C. Binétruy, Pr. I. An-
drianov et Pr. S. Comas-Cardona, Université des Sciences et Technologies de Lille,
soutenance prévue en 2012.
16
4.3 Masters Recherche
MR1 - G. Tetsios, Étude et modélisation de propagation automatique de fissures
semi-elliptiques externes dans un cylindre droit, co-encadré par S. Hariri, Université
des Sciences et Technologies de Lille, 2004.
MR2 - E. Lacoste, Contribution à la modélisation mécanique de films polymères en

grandes déformations élastoplastiques, Université de Nantes, 2007.
MR3 - C. Lainé, Endommagement d’un composite sandwich 3D à renforts fibreux

par homogénéisation, Université des Sciences et Technologies de Lille, 2009.
4.4 Projets de Découverte de la Recherche

– R.T. Ma, A. Fakham, Validation numérique du flambement plastique d’une poutre
de Timoshenko, École des Mines de Douai, 2009.
– Z. Meng, A. Verhaeghe, Homogénéisation périodique du comportement mécanique

d’un sandwich 3D Napcor , École des Mines de Douai, 2010.
4.5 Études Bibliographiques

– S. Kowalczyk, D. Vandeneynde, Technologies des pompes hydrauliques, École des
Mines de Douai, 2004.
– A. Vallet, R. Kaczala, Musicalité relative des percussions membranophones selon le

type de conception, École des Mines de Douai, 2008.
17
5 Production scientifique et technique
5.1 Mémoires
M1 - P. Le Grognec, Un modèle d’écrouissage combiné isotrope-cinématique en
plasticité, DEA, École Centrale de Nantes, 1997.
M2 - P. Le Grognec, Contribution à la modélisation du flambement et du post-

flambement des structures élastoplastiques, Thèse de Doctorat, École Centrale de
Nantes, 2001.
5.2 Publications dans des revues à comité de lecture

P1 - A. Le van1 , P. Le Grognec, Modeling and numerical computation of necking in
round bars using a total Lagrangian formulation, Computer Modeling in Engineering
and Sciences, Vol. 2, n°1, pp. 63-72, 2001.
1 : Université de Nantes
P2 - P. Le Grognec, A. Le van1 , Bifurcations primaires et secondaires d’un cylindre

sous compression axiale, Revue Européenne des Éléments Finis, Vol. 12, n°1, pp.
7-41, 2003.
P3 - A. Le van1 , G. De Saxcé2 , P. Le Grognec, General formulation for local integra-

tion in standard elastoplasticity with an arbitrary hardening model, Computers &
Structures, Vol. 81, n°21, pp. 2099-2109, 2003.
2 : Université des Sciences et Technologies de Lille
P4 - P. Le Grognec, A. Le van1 , Modélisation numérique du flambement

élastoplastique d’une poutre sous compression axiale, Revue Européenne des
Éléments Finis, Vol. 12, n°7-8, pp. 887-906, 2003.
P5 - A. El Hakimi1 , S. Hariri1 , A. Laksimi2 , P. Le Grognec, Z. Azari3 , Étude

numérique et semi-analytique de la nocivité de défauts dans les coques sphériques,
Revue Européenne de Mécanique Numérique, Vol. 16, n°1, pp. 79-101, 2007.
1 : École des Mines de Douai
2 : Université de Technologie de Compiègne
3 : Université de Metz
P6 - S. Comas-Cardona1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , Unidirectional

compression of fibre reinforcements. Part 1 : A non-linear elastic-plastic behaviour,
Composites Science and Technology, Vol. 67, n°3-4, pp. 507-514, 2007.
P7 - P. Le Grognec, A. Le van1 , Elastoplastic bifurcation and collapse of axially

loaded cylindrical shells, International Journal of Solids and Structures, Vol. 45,
n°1, pp. 64-86, 2008.
18
P8 - A. El Hakimi1 , P. Le Grognec, S. Hariri1 , Numerical and analytical study of
severity of cracks in cylindrical and spherical shells, Engineering Fracture Mechanics,
Vol. 75, n°5, pp. 1027-1044, 2008.
P9 - P. Le Grognec, A. Le van1 , Some new analytical results for plastic buckling and
initial post-buckling of plates and cylinders under uniform compression, Thin-Walled
Structures, Vol. 47, n°8-9, pp. 879-889, 2009.
P10 - P. Le Grognec, P. Casari1 , D. Choqueuse2 , Influence of residual stresses and

geometric imperfections on the elastoplastic collapse of cylindrical tubes under
external pressure, Marine Structures, Vol. 22, n°4, pp. 836-854, 2009.
1 : Institut Universitaire de Technologie de Saint-Nazaire
2 : Institut Français de Recherche pour l’Exploitation de la Mer (Brest)
P11 - R. Gonçalves1 , P. Le Grognec, D. Camotim2 , GBT-based semi-analytical

solutions for the plastic bifurcation of thin-walled members, International Journal
of Solids and Structures, Vol. 47, n°1, pp. 34-50, 2010.
1 : École Supérieure de Technologie de Barreiro, Portugal
2 : Université Technologique de Lisbonne, Portugal
P12 - F. Zeng1 , P. Le Grognec, M.F. Lacrampe1 , P. Krawczak1 , A constitutive

model for semi-crystalline polymers at high temperature and finite plastic strain :
Application to PA6 and PE biaxial stretching, Mechanics of Materials, Vol. 42, n°7,
pp. 686-697, 2010.
P13 - P. Le Grognec, A. Le van1 , On the plastic bifurcation and post-bifurcation of

axially compressed beams, International Journal of Non-Linear Mechanics, Vol. 46,
n°5, pp. 693-702, 2011.
P14 - M. Deléglise1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , B. Claude2 ,

Modeling of high speed RTM injection with highly reactive resin with on-line
mixing, Composites Part A : Applied Science and Manufacturing, Vol. 42, n°10, pp.
1390-1397, 2011.
2 : Renault
P15 - P. Lapeyronnie1 , P. Le Grognec, C. Binétruy1 , F. Boussu2 , Homogenization

of the elastic behavior of a layer-to-layer angle-interlock composite, Composite
Structures, Vol. 93, n°11, pp. 2795-2807, 2011.
2 : École Nationale Supérieure des Arts et Industries Textiles (Roubaix)
P16 - P. Le Grognec, Q.H. Nguyen1 , M. Hjiaj1 , Exact buckling solution for two-layer
Timoshenko beams with interlayer slip, International Journal of Solids and Struc-
tures, accepté.
1 : Institut National des Sciences Appliquées de Rennes
19
P17 - Q.H. Nguyen1 , M. Hjiaj1 , P. Le Grognec, Exact dynamic analysis of two-layer
Timoshenko beam with interlayer slip, Journal of Sound and Vibration, soumis.
1 : Institut National des Sciences Appliquées de Rennes
5.3 Publications dans des revues sans comité de lecture

P18 - P. Le Grognec, S. Hariri1 , M. Afzali2 , H. Jaffal2 , Simulation numérique de la
propagation de fissures dans les structures sous pression, Soudage et Techniques
Connexes, Vol. 62, n°11-12, pp. 33-36, 2008.
2 : Centre Technique des Industries Mécaniques (Senlis)
P19 - P. Lapeyronnie1 , P. Le Grognec, C. Binétruy1 , F. Boussu2 , Angle-interlock rein-

forcements : Weaving and the mechanical properties of composites, JEC Composites,
Vol. 58, pp. 58-59, 2010.
5.4 Participation à des comités de lecture et conseils d’édition
Évaluation d’articles pour des revues internationales à comité de lecture : International

Journal of Solids and Structures, Mechanics of Materials, Composites Part B : Engineering,
Journal of Zhejiang University - Science A, Journal of Structural Engineering.
5.5 Communications dans des conférences internationales avec comité

de sélection
C1 - A. Le van1 , P. Le Grognec, Modeling and computation of necking instability in
round bars using a total Lagrangian elastoplastic formulation, European Congress
on Computational Methods in Applied Sciences and Engineering, Barcelona, Spain,
September 11-14, 2000.
C2 - P. Le Grognec, A. Le van1 , Buckling and post-buckling of thin elastoplastic

cylindrical shells with finite rotations, ASME International Mechanical Engineering
Congress and Exposition, Orlando, Florida, USA, November 5-10, 2000.
C3 - P. Le Grognec, A. Le van1 , A unified treatment for the elastoplastic bifurcation

of structural elements, 21st International Congress of Theoretical and Applied
Mechanics, Warsaw, Poland, August 15-21, 2004.
C4 - R. Gonçalves1 , P. Le Grognec, D. Camotim2 , Plastic bifurcation FEA of

thin-walled members : Thin shell elements vs. GBT-based beam elements, III
European Conference on Computational Mechanics : Solids, Structures and Coupled
Problems, Mini-Symposium on ”Stability and Non-Linear Behaviour of Thin-Walled
Members and Structures”, Lisbon, Portugal, June 5-8, 2006.
1 : École Supérieure de Technologie de Barreiro, Portugal
2 : Université Technologique de Lisbonne, Portugal
20
C5 - S. Comas-Cardona1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , Modeling of
uniaxial compression of fibre reinforcements using finite strains, 10th ESAFORM
Conference on Material Forming, Zaragoza, Spain, April 18-20, 2007.
C6 - A. El Hakimi1 , S. Hariri1 , P. Le Grognec, Numerical and experimental study

of crack harmfulness in cylindrical and spherical shells under pressure, European
Symposium on Pressure Equipment, Paris, France, October 9-11, 2007.
C7 - P. Le Grognec, S. Hariri1 , M. Afzali2 , H. Jaffal2 , Numerical simulation of crack

propagation in pressure equipments, European Symposium on Pressure Equipment,
Paris, France, October 9-11, 2007.
C8 - A. El Hakimi1 , P. Le Grognec, S. Hariri1 , M. Afzali2 , H. Jaffal2 , Numerical

simulation of crack propagation in pressure equipments, Fatigue Design, Senlis,
France, November 21-22, 2007.
C9 - M. Deléglise1 , P. Le Grognec, B. Claude2 , D. Wilde3 , A. Terenzi4 , J. Kenny4 ,

C. Binétruy1 , P. Krawczak1 , Modeling of the injection of an automative part with a
two component highly reactive resin with on-line mixing in RTM, SAMPE Europe,
Paris, France, March 31-April 2, 2008.
2 : Renault
3 : Alcan Technology & Management, CH
4 : ECNP et Université de Pérouse, Italie
C10 - M. Deléglise1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , B. Claude2 , D.

Wilde3 , A. Terenzi4 , J. Kenny4 , Optimisation par le biais de la simulation numérique
du moulage RTM haute cadence de pièces composites automobiles, Congrès Interna-
tional Carrosserie et Plastiques : ”Nouvelles Exigences de la Carrosserie Automobile
et de ses Équipements”, Douai, France, 11-12 juin 2008.
2 : Renault
4 : ECNP et Université de Pérouse, Italie
C11 - M. Deléglise1 , P. Le Grognec, B. Claude2 , D. Wilde3 , C. Binétruy1 , P.

Krawczak1 , Highly reactive resin viscosity prediction in Liquid Composite Molding
processes, 9th International Conference on Flow Processes in Composite Materials,
Montréal, Québec (Canada), July 8-10, 2008.
2 : Renault
21
C12 - P. Lapeyronnie1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , F. Boussu2 ,
Homogenization of the mechanical behavior of a layer-to-layer angle-interlock
composite, SAMPE Europe, Paris, France, March 23-25, 2009.
C13 - X. Legrand1 , F. Boussu1 , S. Nauman1 , I. Cristian2 , P. Lapeyronnie3 ,

P. Le Grognec, C. Binétruy3 , A forming behaviour of warp interlock composite,
12th ESAFORM Conference on Material Forming, Twente, The Netherlands, April
27-29, 2009.
2 : Université Technique ”Gheorghe Asachi” de Iasi, Roumanie
C14 - P. Le Grognec, P. Casari1 , D. Choqueuse2 , Elastoplastic collapse of cylindrical

tubes under external pressure, 7th EUROMECH Solid Mechanics Conference,
Lisbon, Portugal, September 7-11, 2009.
1 : Institut Universitaire de Technologie de Saint-Nazaire
2 : Institut Français de Recherche pour l’Exploitation de la Mer (Brest)
C15 - F. Zeng1 , P. Le Grognec, M.F. Lacrampe1 , P. Krawczak1 , A constitutive

model for semi-crystalline polymer films in finite elastoplastic strains, Europe/Africa
Regional Meeting of the Polymer Processing Society 2009 (E/A RM PPS 2009),
Larnaca, Cyprus, October 18-21, 2009.
C16 - P. Lapeyronnie1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , F. Boussu2 ,

Experimental study and modeling of the mechanical behaviour of angle-interlock
composites, Congrès International Carrosserie et Plastiques : ”Composites et
Polymères à Haut Module”, Douai, France, 9-10 juin 2010.
C17 - F. Zhang1 , P. Le Grognec, S. Comas-Cardona1 , B. Cosson1 , C. Binétruy1 ,

Numerical simulation of resin flow in fiber reinforcement with stochastic property
field, 10th International Conference on Flow Processes in Composite Materials,
Monte Verità, Ascona, Switzerland, July 11-15, 2010.
C18 - P. Lapeyronnie1 , P. Le Grognec, C. Binétruy1 , F. Boussu2 , Homogenization of

the mechanical behaviour of a layer-to-layer angle-interlock composite, 10th Interna-
tional Conference on Textile Composites, Lille, France, October 26-28, 2010.
22
5.6 Communications dans des conférences nationales avec comité de
sélection
C19 - P. Le Grognec, A. Le van1 , Modes de bifurcation primaires et secondaires d’un
tube cylindrique élastoplastique sous compression axiale, 15ème Congrès Français
de Mécanique, Nancy, 3-7 septembre 2001.
C20 - P. Le Grognec, A. Le van1 , G. De Saxcé2 , Intégration locale en élastoplasticité

avec un écrouissage arbitraire, 16ème Congrès Français de Mécanique, Nice, 1-5
septembre 2003.
2 : Université des Sciences et Technologies de Lille
C21 - A. El Hakimi1 , S. Hariri1 , Z. Azari2 , A. Laksimi3 , P. Le Grognec, Étude

numérique de la nocivité des défauts dans les coques sphériques, 17ème Congrès
Français de Mécanique, Troyes, 29 août-2 septembre 2005.
2 : Université de Metz
3 : Université de Technologie de Compiègne
C22 - S. Comas-Cardona1 , P. Le Grognec, C. Binétruy1 , P. Krawczak1 , Modélisation

de la compression uniaxiale de renforts fibreux en grandes déformations
élastoplastiques, Journée Scientifique et Technique AMAC, ONERA, Châtillon, 13
octobre 2006.
C23 - P. Le Grognec, S. Hariri1 , M. Afzali2 , H. Jaffal2 , Nocivité des défauts et

propagation de fissures dans les équipements sous pression, 18ème Congrès Français
de Mécanique, Grenoble, 27-31 août 2007.
C24 - F. Zeng1 , P. Le Grognec, M.F. Lacrampe1 , P. Krawczak1 , Un modèle de com-

portement de films polymères en grandes déformations élastoplastiques, 2ème Journée
des Jeunes Polyméristes du Nord (JPN2), Villeneuve-d’Ascq, 2 décembre 2009.
5.7 Séminaires
S1 - P. Le Grognec, Instabilités en mécanique non-linéaire des structures, Université
”Dunarea de Jos” de Galati, Roumanie, 18 mars 2008.
S2 - P. Le Grognec, Modélisation du comportement mécanique de films plastiques

multi-couches, Université ”Dunarea de Jos” de Galati, Roumanie, 19 mai 2009.
23
6 Collaborations académiques et industrielles
6.1 Relations nationales et internationales

Université ”Dunarea de Jos” de Galati, Roumanie : Missions d’enseignement
dans le cadre d’un accord Erasmus. Échanges d’étudiants entre le Département
Fabrication, Robotique et Soudage de l’Université de Galati et le Département
Technologie des Polymères et Composites & Ingénierie Mécanique de l’École des
Mines de Douai. Contact : Pr. Catalina Maier.
Université Technologique de Lisbonne, Portugal : Modélisation du flambement

plastique de structures à parois minces. Contact : Pr. Dinar Camotim.
Institut National des Sciences Appliquées, Rennes : Calcul de poutres mixtes

acier-béton avec connexion partielle. Contact : Pr. Mohammed Hjiaj.
École Nationale Supérieure des Arts et Industries Textiles, Roubaix : Étude

du comportement mécanique de composites à renforts interlock 3D. Contact :
François Boussu.
Institut Français de Recherche pour l’Exploitation de la Mer, Brest : Analyse

de l’instabilité d’enceintes sous-marines sous pression hydrostatique externe avec
prise en compte des imperfections géométriques et des contraintes résiduelles.
Contact : Dominique Choqueuse.
6.2 Implication dans des projets de recherche collaboratifs

– Étude et modélisation de propagation automatique de fissures semi-elliptiques
externes dans un cylindre droit, DEA de G. Tetsios à l’Université des Sciences et
Technologies de Lille, Partenaire industriel : CETIM, 2004.
– Modélisation numérique de la propagation de fissures dans les structures sous

pression, Partenaire industriel : CETIM, 2005-2007.
– Dimensionnement fiabiliste d’un réseau de tuyauteries composites utilisées comme

circuit eau de mer équipant les navires de combat, Projet de Recherche Exploratoire
et Innovation, Partenaires industriels : DGA, CETIM, AXS Analyse de Structures,
2008-2011.
– Nouvelles technologies non-tissées versatiles permettant de fabriquer des structures

non-tissées 3D et des sandwiches composites optimisés (NWC-X), Projet FUI, Par-
tenaires industriels : SICOMIN, GLEROY, ASSELIN THIBEAU, ENSAIT, IFTH,
SUBRENAT, DUFLOT, ACHILLE BAYART, TENSYL, NAVI-LINE, STRATI-
FORME, 2009-2012.
6.3 Développement industriel et transfert de technologie
6.3.1 Transfert de technologie

– Vérification du dimensionnement des principaux sous-ensembles d’une cisaille DKP1
pour la sidérurgie, pour le bureau d’études techniques PRODESS (2005).
24
6.3.2 Expertise judiciaire
– Analyse de la rupture d’un caisson hyperbare thérapeutique HYOT 2200/12/2/RD
(de marque HYTECH) pour le cabinet d’expertises TECH PLUS (2005).
25
26
Partie II
Présentation des travaux de
recherche
28
1 Plasticité en grandes transformations
L’essentiel de mes travaux de recherche a un lien plus ou moins direct avec la plasti-
cité. Dans certains cas, la plasticité constitue le fondement même des travaux entrepris,
dans d’autres cas, elle apparaı̂t uniquement comme un outil de modélisation pour une
étude qui porte sur d’autres aspects plus structuraux. Ma toute première activité de re-
cherche, dans le cadre de mon DEA, a justement porté sur la modélisation numérique
de la plasticité. Un premier outil de calcul a vu le jour sous la forme d’un programme
élément fini ”maison”, qui devait faire partie d’un ensemble de programmes (en Fortran),
développés tout au long de ma carrière, que j’ai baptisé NLFEAS (”Non-Linear Finite Ele-
ment Analysis of Structures”). La plasticité est donc une composante essentielle de tous
les outils développés, que ce soit au cours de mon DEA, durant ma thèse (pour l’étude
du flambement plastique) ou encore à diverses autres occasions spécifiques (dans le cadre
de travaux sur le phénomène de rochet ou sur la striction). Elle fait en particulier par-
tie intégrante des nouveaux axes de recherche développés plus récemment concernant les
matériaux polymères et composites. Au vu de tous les contextes dans lequel la plasticité
apparaı̂t, de tous les matériaux concernés (essentiellement métalliques ou polymères), et de
toutes les applications visées, diverses formulations (2D/coque ou 3D) ont été envisagées au
cours du temps (en petites ou grandes transformations), s’appuyant sur différents modèles
(critères et lois d’écrouissage). Les outils de calcul numérique se sont ainsi diversifiés ou
généralisés pour répondre successivement à tous ces besoins. Le principal objectif de mes
travaux en plasticité est de répondre de manière globale et unifiée à la question de la
modélisation de la plasticité : globale, dans le sens où l’on ne se limite pas seulement à
quelques modèles parmi les plus classiques, et unifiée, dans le sens où l’on montre la ca-
pacité de la formulation lagrangienne totale retenue à répondre favorablement à tout un
ensemble de problèmes mettant en jeu de la plasticité dans des conditions parfois sévères,
comme lorsqu’on envisage des phénomènes de ruine par rochet (en plasticité cyclique), par
striction (instabilité matérielle) ou (post-)flambement plastique (instabilité géométrique).
Une fois le socle ”élément fini” du programme établi, la partie primordiale de la mise en
œuvre numérique concerne l’implantation numérique de la loi de comportement, à savoir
l’intégration locale en plasticité. La première partie de cette section résume donc naturelle-
ment les développements théoriques et numériques qui ont permis de modéliser l’ensemble
des problèmes présentés par la suite, concernant spécifiquement la plasticité (qui seront
évoqués dans cette même section) ou alors d’autres phénomènes mettant tout de même en
jeu de la plasticité (qui feront en partie l’objet des prochaines sections). Deux phénomènes
de ruine liés à la nature du comportement plastique sont ensuite abordés pour illustrer la
robustesse et l’efficacité des outils numériques mis en œuvre. Il s’agit en premier lieu du
phénomène de rochet qui apparaı̂t dans des conditions de sollicitation cyclique avec l’uti-
lisation d’un écrouissage cinématique non-linéaire et qui mène à la ruine par déformation
plastique cumulée. En deuxième lieu, on s’intéresse au phénomène de striction qui peut se
définir comme une instabilité purement matérielle et dont la mise en évidence n’est pas
immédiate.
Fort de ces différentes expériences dans le domaine de la plasticité, mon intérêt s’est
récemment porté vers la modélisation du comportement mécanique de films polymères.
La première partie de ce travail a consisté en l’identification et la caractérisation d’un
modèle de comportement élastoplastique pour une classe de polymères sous des conditions
spécifiques de température afin de pouvoir prédire de manière unifiée le comportement de
ces matériaux à la fois en mono- et bi-étirage à chaud. L’objectif à long terme est l’optimi-
sation de la conception de films multi-couches (PE/liant/PA) pour l’emballage alimentaire.
Un modèle en trois phases combinées a été spécialement développé et calibré, en relation
avec la nature micro-structurale des polymères semi-cristallins considérés. L’intégration
29
locale de la plasticité s’appuie sur les méthodes et outils numériques réalisés auparavant
dans le contexte des matériaux métalliques.
1.1 Intégration locale en plasticité et modélisation numérique
1.1.1 Historique
Mes travaux concernant l’intégration locale en plasticité ont évolué progressivement, au

gré des besoins en termes d’outils de calcul. Je vais commencer cette partie en présentant
brièvement les différents contextes successifs dans lesquels je me suis placé pour traiter de
la plasticité.
Dans un premier temps, mon travail de DEA [M1 ] a porté sur la prise en compte
d’un écrouissage mixte (isotrope/cinématique) dans la modélisation numérique de la plas-
ticité 3D en petites transformations quasi-statiques. Dans le cadre des matériaux standard
généralisés, on a retenu le critère de von Mises avec un écrouissage combiné isotrope-
cinématique linéaire. Dans ce contexte particulier, il a été possible de proposer un algo-
rithme d’intégration locale direct (sans calculs itératifs pour la détermination du multipli-
cateur plastique). Le premier programme de calcul par éléments finis développé dans ces
conditions a été validé sur des exemples de difficulté croissante, certains comportant des
solutions analytiques, les autres uniquement des solutions de référence numériques.
À l’occasion de ma thèse [M2 ] portant sur les structures minces élastoplastiques, l’al-
gorithme précédent a été légèrement modifié pour prendre une forme itérative indispen-
sable à la prise en compte de la condition de contraintes planes. Il a alors été possible,
sans modification fondamentale supplémentaire, de généraliser l’écrouissage au cas isotrope
non-linéaire. Parallèlement, on est passé à une description des grandes transformations sui-
vant une formulation lagrangienne totale. Cette description des non-linéarités géométriques
permet, comme on le verra par la suite, de rendre compte de bon nombre de phénomènes
non-linéaires associés à la plasticité et à l’instabilité matérielle ou géométrique, dès lors que
les déformations restent cependant modérées. L’intérêt essentiel d’une telle formulation est
qu’elle garantit d’emblée l’objectivité des grandeurs utilisées pour l’écriture de la loi de
comportement, et que le formalisme employé (en plasticité) reste similaire à celui utilisé
dans le cas des petites déformations.
Dans le cadre 3D, les développements précédents ont été poursuivis en généralisant la
loi de comportement et donc le programme élément fini par la définition d’un écrouissage
cinématique non-linéaire. Une nouvelle formulation plus générale de l’intégration locale
en plasticité pour un écrouissage arbitraire a été mise en œuvre à cette occasion et va-
lidée par des solutions analytiques et numériques de référence (c’est cette formulation la
plus générale, sur laquelle peuvent s’appuyer l’ensemble des applications effectuées, qui
sera décrite dans les sections suivantes). L’utilisation spécifique d’un modèle d’écrouissage
cinématique non-linéaire permet de mieux décrire l’ensemble des phénomènes de plasti-
cité cyclique et de reproduire en particulier par simulation numérique l’effet de rochet
[P3 ,C20 ].
À partir du programme 3D en grandes transformations, une variante a été développée
dans le cas 2D axisymétrique, en vue de s’intéresser à la traction de barreaux cylindriques
jusqu’à la ruine. Le phénomène de striction observé expérimentalement a été ainsi mis en
valeur numériquement par l’intermédiaire d’un écrouissage isotrope non-linéaire approprié
[P1 ,C1 ].
Les développements les plus récents, concernant les films polymères, ont nécessité l’uti-
lisation d’une formulation élément fini différente, du fait des très grandes déformations
30
envisagées (jusqu’à 800%), et ont été implantés par le biais d’une routine UMAT au sein
du logiciel commercial de calcul par éléments finis Abaqus. Malgré tout, la philosophie
de la procédure d’intégration locale s’inspire largement des travaux antérieurs, si ce n’est
que le modèle développé est une combinaison de plusieurs composantes distinctes (pour
chacune des phases) et associe à des modèles élastoplastiques quasi-standard des modèles
de réseau.
1.1.2 Modélisation de la plasticité
On rappelle, dans un premier temps, les équations fondamentales de la plasticité 3D en

petites transformations sur lesquelles va s’appuyer la procédure d’intégration locale décrite
par la suite. On se place dans le cadre de la théorie des matériaux standard généralisés
[HAL, 1975] et on choisit classiquement une décomposition additive des déformations
linéarisées. Les lois d’état permettent d’exprimer les contraintes ainsi que les grandeurs
duales des variables d’écrouissage de la façon suivante :
∂wα ∂wα
σ = D : εe a= a= (1)
∂α ∂α
en supposant le découplage de l’énergie d’écrouissage en deux fonctions indépendantes des
variables d’écrouissage isotrope et cinématique, supposées toutes deux convexes.
De la même façon, le critère de plasticité est supposé prendre la forme générale suivante :
f (σ, a, a) = f1 (σ, a) + f2 (a) = 0 (2)
et permet d’écrire les lois d’évolution d’après le principe de dissipation plastique maximale :
∂f ∂f ∂f
ε̇p = λ̇ − α̇ = λ̇ − α̇ = λ̇ (3)
∂σ ∂a ∂a
On se restreint ainsi au cas de la plasticité associée, pour simplifier la formulation de

l’intégration locale, mais le choix d’un potentiel distinct de la fonction critère ne modifie
pas les grandes lignes de la procédure d’intégration locale présentée. Après différentiation
en temps des lois d’état et élimination des variables d’écrouissage grâce aux lois d’évolution,
on aboutit à un système de 14 équations à 14 inconnues (σ, (a, a) et λ̇) :
∂f
σ̇ = D : ε̇ − λ̇D : ∂σ (σ, a, a)
∂ 2 wα ∂f ∂ 2 wα
ȧ = ∂α2
: α̇ = −λ̇G(a) : ∂a (σ, a, a) avec : G(a) = ∂α2
(α(a))
(4)
∂ 2 wα ∂ 2 wα
ȧ = ∂α2
α̇ = −λ̇G(a) ∂f
∂a (σ, a, a) avec : G(a) = ∂α2
(α(a))
f (σ, a, a) = 0
1.1.3 Intégration locale
La résolution numérique d’un problème de plasticité (par éléments finis) s’effectue

généralement par un calcul incrémental (avec la méthode de Newton-Raphson). La diffi-
culté principale réside en l’intégration locale du modèle de plasticité, à savoir l’évaluation
des variables au nouvel instant en fonction des valeurs précédentes et de l’incrément de
déformation courant, à chaque itération et en tout point de la structure. L’objectif est de
déterminer les nouvelles contraintes et surtout le tenseur tangent consistant qui permet
31
de construire la matrice raideur tangente consistante intervenant dans la procédure de
résolution non-linéaire du problème par éléments finis et assurant une convergence qua-
dratique.
De nombreux algorithmes ont été employés dans divers contextes pour répondre à cette
question d’intégration d’une loi de comportement non-linéaire. Dans certains cas simples,
une intégration locale analytique est possible (en 3D, en plasticité parfaite ou avec un
écrouissage linéaire) [KRI, 1977]. Lorsqu’on impose la condition de contraintes planes et/ou
en présence d’un écrouissage non-linéaire, la détermination du multiplicateur plastique re-
vient au mieux à résoudre une équation algébrique non-linéaire. De manière générale, l’al-
gorithme le plus employé pour la résolution numérique du problème d’intégration locale
s’avère être l’algorithme de retour radial, avec le schéma d’Euler implicite [WIL, 1964].
La résolution des équations non-linéaires impliquées dans l’intégration locale nécessite
alors l’utilisation d’une procédure itérative de type Newton-Raphson. D’autres algorithmes
(qu’on ne citera pas ici par souci de concision) ont été cependant développés parallèlement
de façon à généraliser la procédure à d’autres types de critères ou d’autres lois d’écrouissage,
et à améliorer la convergence et la stabilité des calculs [ORT, 1983] [SIM, 1991]. En pa-
rallèle, Hughes et Pister [HUG, 1978] ont introduit les premiers la notion de linéarisation
consistante et une première expression générale du module tangent consistant a été fournie
par Alfano et Rosati [ALF, 1998] pour un écrouissage mixte.
Notre contribution dans ce domaine est de proposer une méthode générale et
synthétique de détermination des corrections plastiques et du tenseur tangent consistant
valide pour une large classe de modèles non-linéaires de plasticité (avec une fonction seuil
et une énergie d’écrouissage arbitraires). On présente dans ce qui suit, dans le cadre des
petites transformations 3D, les points essentiels de la construction des fonctions/tenseurs
”numériques” (consistants) Gnum , Dnum et Gnum nécessaires à l’intégration locale ainsi que
du tenseur tangent consistant, dont les expressions diffèrent largement de celles de leurs
homologues théoriques. La considération d’une loi d’écrouissage arbitraire fait en particu-
lier apparaı̂tre des termes nouveaux dans les expressions explicites obtenues, vis-à-vis des
travaux antérieurs.
Le point de départ de la procédure d’intégration locale consiste à discrétiser en temps
les relations (4) en utilisant le schéma d’Euler implicite :
∂f
σ − σ n−1 + ∆λD : ∂σ (σ, a, a) − D : ∆ε = 0
∂f
a − an−1 + ∆λG(a) : ∂a (σ, a, a) =0
(5)
a − an−1 + ∆λG(a) ∂f
∂a (σ, a, a) =0
f (σ, a, a) = 0
La résolution itérative du système (5) pour la détermination de l’incrément de multi-

plicateur plastique ∆λ, ainsi que des nouvelles contraintes et variables d’écrouissage (a, a),
nécessite la linéarisation des équations précédentes, qui fait apparaı̂tre les différentes cor-
32
rections plastiques associées (δσ, δa, δa, δλ) à chaque itération :

∂2f ∂2f ∂f
D−1 + ∆λ ∂σ 2 : δσ + ∆λ ∂a∂σ : δa = −Ψ(σ, a, ∆λ) − ∂σ δλ
∂f
avec : Ψ(σ, a, ∆λ) = D−1 : (σ − σ n−1 ) + ∆λ ∂σ (σ, a) − ∆ε
h i
∂2f 2
∂ 3 wα
∆λ ∂σ∂a : δσ + G −1 + ∆λ ∂∂af2 + ∆λG −1 : ∂f
∂a : ∂α3
: G −1 : δa = −Φ(σ, a, ∆λ) − ∂f
∂a δλ
avec : Φ(σ, a, ∆λ) = G −1 (a) : (a − an−1 ) + ∆λ ∂f

∂a (σ, a)
2

∆λ ∂ 3 wα ∂f
1
G(a) + ∆λ ∂∂af2 + G2 (a) ∂α3 ∂a
δa = −Φ(a, ∆λ) − ∂f
∂a δλ
a−an−1
avec : Φ(a, ∆λ) = G(a) + ∆λ ∂f
∂a (a)
∂f ∂f ∂f
∂σ : δσ + ∂a : δa + ∂a δa = −f (σ, a, a)
(6)
La convexité des énergies d’écrouissage permet d’inverser le système linéaire (6) et on
obtient, après calculs, les corrections plastiques suivantes :
∂f
f −( ∂σ )num :Dnum :Ψ−( ∂f
∂a )num num
:G :Φ−Gnum ∂f
∂a
Φ
δλ = ∂f ∂f 2
∂σ
:Dnum : ∂σ (+ ∂f :G
) : ∂f
num ∂a num ∂a num
+Gnum ∂f
∂a ( ) ( )
−1
∂2f −1 ∂2f −1 ∂f ∂ 3 wα −1 ∂f
δσ = Dnum : −Ψ + ∆λ ∂a∂σ : G + ∆λ ∂a2 + ∆λG : ∂a : ∂α3 : G : Φ − ∂σ δλ
num
−1
∂2f ∂2f
δa = Gnum : −Φ + ∆λ ∂σ∂a : D−1 + ∆λ ∂σ 2 : Ψ − ∂f
∂a num δλ

δa = −Gnum Φ + ∂f ∂a δλ
(7)
en définissant les fonctions/tenseurs ”numériques” :
−1 −1
∂2f ∂2f 2
∂ 3 wα ∂2f
Dnum = D−1 + ∆λ ∂σ 2 − (∆λ)2 ∂a∂σ : G −1+ ∆λ ∂∂af2 + ∆λG −1 : ∂f
∂a : ∂α3
: G −1 : ∂σ∂a
−1 −1
2 ∂2f ∂2f ∂2f
∂3wα
Gnum = G −1 + ∆λ ∂∂af2 + ∆λG −1 : ∂f
∂a : ∂α3
: G −1 − (∆λ)2 ∂σ∂a : D−1 + ∆λ ∂σ 2 : ∂a∂σ
G(a)
Gnum = 2f 3 wα ∂f
1+∆λG(a) ∂ ∆λ ∂
+ G(a)
∂a2 ∂α3 ∂a
h i−1
∂f
= ∂f
− ∆λ ∂f : G −1 + ∆λ ∂ 2 f + ∆λG −1 : ∂f : ∂ 3 wα
: G −1 : ∂2f
∂σ ∂σ ∂a ∂a 2 ∂a ∂α3 ∂σ∂a
num −1
∂f ∂f ∂f −1 ∂2f ∂2f
∂a num = ∂a − ∆λ ∂σ : D + ∆λ ∂σ2 : ∂a∂σ
(8)
dont on peut vérifier qu’ils tendent vers leurs homologues théoriques lorsque l’incrément
de multiplicateur plastique tend vers 0.
Cette procédure d’intégration locale affiche une stabilité inconditionnelle avec le schéma
d’Euler implicite, ainsi qu’un taux de convergence quadratique avec la méthode de Newton-
Raphson utilisée pour la résolution itérative des équations discrétisées non-linéaires.
Les énergies d’écrouissage (isotrope et cinématique) sont ici a priori quelconques, et pas
nécessairement des fonctions quadratiques des variables d’écrouissage, comme supposé le
plus souvent. Le tenseur G et la fonction G ne sont donc pas a priori constants. Surtout, on
33
remarque la présence d’un terme supplémentaire vis-à-vis des travaux antérieurs, à savoir
∂ 3 wα
la quantité ∂f
∂a : ∂α3 dans les expressions des tenseurs ”numériques” définis dans (8).
1.1.4 Tenseur tangent consistant
Une fois l’intégration locale accomplie, on s’attache à construire le tenseur tan-

∂σ

gent consistant ∂ε num qui va aussi dépendre des grandeurs introduites précédemment,
définies par les expressions (8). En premier lieu, la différentiation en temps des équations
discrétisées (5) mène au système d’équations suivant :
2
˙ ∂f ∂ f ∂2f
σ̇ = D : ε̇ − ∆λD : ∂σ − ∆λD : ∂σ 2 : σ̇ + ∂a∂σ : ȧ

˙ ∂f ∂2f ∂2f ∂f ∂G
ȧ = −∆λG : ∂a − ∆λ G : ∂σ∂a : σ̇ + G : ∂a2
: ȧ + ∂a : ∂a : ȧ
(9)

˙ ∂f − ∆λ G 2 + ∂2f ∂f ∂G
ȧ = −∆λG ∂a ∂a ∂a ∂a ȧ
∂f ∂f ∂f
f˙ = ∂σ : σ̇ + ∂a : ȧ + ∂a ȧ =0
Ce système (9) est en partie analogue au système d’équations (6) et peut ainsi s’in-
verser de manière similaire. Sans rentrer dans les détails des calculs, l’obtention du taux
d’incrément de multiplicateur plastique ∆λ˙ permet d’aboutir finalement au tenseur tan-
gent consistant :
∂f ∂f
∂σ
Dnum : ∂σ ( )
⊗ ∂σ (
:Dnum )
∂ε num = Dnum − ∂f ∂f
num
∂f
num
∂f ∂f 2 (10)
∂σ
:D ( )
num ∂σ num ∂a num ∂a num +Gnum ∂a
: + :G : ( ) ( )
mettant effectivement en jeu les mêmes fonctions/tenseurs ”numériques” que l’intégration

locale et donc en particulier le même terme supplémentaire vis-à-vis des travaux antérieurs.
La symétrie du tenseur Dnum entraı̂ne visiblement celle du tenseur tangent consistant
(assurée a priori par le choix du schéma d’Euler implicite). Enfin, l’écriture utilisée permet
de retrouver rigoureusement l’expression bien connue du tenseur tangent théorique, dans
laquelle on a remplacé la plupart des tenseurs par leurs homologues ”numériques”.
1.2 Applications
1.2.1 Phénomène de rochet
La procédure d’intégration locale telle qu’elle a été décrite précédemment, dans le

cadre 3D en petites transformations, et sans restriction sur le choix des critères et lois
d’écrouissage, permet d’aborder correctement la plasticité cyclique et de simuler en particu-
lier l’effet de rochet par l’utilisation appropriée d’un écrouissage cinématique non-linéaire.
Le programme élément fini développé à cette occasion a été donc testé sur tout un en-
semble d’exemples, en faisant varier le modèle de plasticité, la géométrie et le chargement.
Son efficacité et sa robustesse ont été démontrées par l’obtention de divers types de rochet
uni- ou multi-axial.
On présente ici quelques résultats significatifs à des fins d’illustration, validés par des
solutions analytiques ou numériques de référence. On suppose un comportement élastique
34
linéaire isotrope, et l’énergie d’écrouissage et le critère de plasticité sont donnés respecti-
vement par [CHA, 1996] [ARM, 1966] :
whard = wqα (α) = 1 2H α : α

2 3
(11)
f (σ, a) = 32 ks − ak + 12 3K
2H a : a − σ0
avec les valeurs numériques listées dans le tableau 1. La fonction critère sert également
de potentiel plastique dont dérivent les lois d’évolution : on se place ainsi dans le cadre
plus naturel de la plasticité associée standard, contrairement à ce qu’on peut trouver
dans la littérature concernant la plasticité cyclique. Par simplicité, on se restreint au seul
écrouissage cinématique non-linéaire, apte à rendre compte du phénomène de rochet.
Module d’Young E Coefficient de Poisson ν Limite élastique σ0

2.1011 P a 0.3 3.108 P a
Module d’écrouissage cinématique H Vitesse de saturation K
3.1010 P a 60
Tableau 1 – Propriétés matérielles pour la modélisation du rochet
Les éléments finis utilisés sont des hexaèdres à 20 nœuds avec 27 points de Gauss par
élément pour l’intégration numérique.
1.2.1.1 Effet de rochet uniaxial Le premier exemple traité concerne le cas d’un
barreau en traction uniaxiale. La contrainte varie cycliquement de σm = −2.108 P a à
σM = 4.5 108 P a. Cette asymétrie entraı̂ne une accumulation progressive de la déformation
plastique qui conduit fatalement à la ruine du barreau. Pour ce problème simple, on dispose
d’une solution analytique du pas du rochet :
H 2
" #
2

1 − am
∆εp = ln K 2 (12)
K H
− a2
K M
avec : q
−1+ 1+2(σm +σ0 ) K
H
am = K
q H (13)
1− 1−2(σM −σ0 ) K
H
aM = K
H
qui coı̈ncide à moins de 1% près avec la solution obtenue numériquement.
1.2.1.2 Effets de rochet multiaxiaux Plusieurs autres effets de rochet ont été re-
produits sur diverses géométries soumises à des états de contraintes plus complexes. Le
cas d’une barre sous traction uniaxiale constante σXX = σ = 2.108 P a et soumise à un
cisaillement alterné symétrique σXY de −2.108 P a à 2.108 P a permet d’illustrer le cou-
plage entre de la traction et de la torsion cyclique. La courbe contrainte-déformation de
la figure 1(a) montre l’accumulation progressive de la déformation plastique longitudinale,
essentiellement due à la présence d’une contrainte non nulle dans cette direction. Le pas
du rochet tend vers une valeur limite, à partir d’un petit nombre de cycles, qui peut être
approchée analytiquement en considérant le cas plus simple de la plasticité non-associée
(en faisant K = 0 uniquement dans la fonction critère) :
4 σ
∆εpXX = √ q ∆εpXY = 0.596∆εpXY (14)
3 H
+ σ0
2
− σ2
K
35
La valeur numérique correspondante 0.612 ne diffère que de 3% de la solution analytique
simplifiée.
Le problème d’un cylindre épais sous pression interne met en jeu un état de contrainte
non homogène, avec en particulier une plastification progressive à partir du rayon interne.
On considère pour ce cas un modèle 2D en déformations planes, ce qui ne change pas
fondamentalement la procédure d’intégration locale. Le cylindre est chargé cycliquement
d’une pression nulle à une valeur maximale donnée. La figure 1(b) montre les variations
de la pression et de la contrainte circonférentielle (prépondérante) avec le déplacement
radial d’un point situé au niveau du rayon interne. La pression maximale étant suffisante
pour replastifier en décharge, le cycle de chargement induit un phénomène de rochet à pas
constant.
(a) Couplage traction/torsion (b) Cylindre sous pression interne
Figure 1 – Courbes de réponse avec rochet multiaxial
Un dernier exemple concerne une poutre sous moment fléchissant (en flexion pure). Là
encore, le rochet semble avoir un pas quasiment constant, et nos résultats concordent avec
ceux de Mahbadi et Eslami [MAH, 2002] obtenus avec un modèle de poutre.
1.2.2 Phénomène de striction
La striction est un phénomène d’instabilité matérielle (de bifurcation) qui a fait l’objet
de nombreuses études, dont les principales concernent le cas fondamental d’un barreau
cylindrique en traction. Les premiers travaux datent des années 1970 [NEE, 1972] et ont
été poursuivis, entre autres, par Simo et Hughes [SIM, 1998] et Brünig [BRü, 1998] avec
différentes formulations élément fini. Parmi les études concernant le phénomène de stric-
tion, on peut mentionner l’analyse de Besdo [BES, 1987] qui fait appel, de manière origi-
nale, à une représentation de la plasticité dans l’espace des déformations. Du point de vue
de la résolution numérique, Okada et al. [OKA, 1990] ont traité le problème de bifurcation
associé à la striction par la méthode des équations intégrales.
L’objectif de notre travail autour de la striction d’un barreau cylindrique en traction
a été de reproduire numériquement ce phénomène à partir du modèle élastoplastique le
plus simple possible, dans le cadre de notre formulation lagrangienne totale. La principale
difficulté rencontrée avec le passage aux grandes transformations réside dans le choix de
la décomposition cinématique la plus appropriée. Parmi les solutions proposées dans la
littérature, et compte tenu des déformations modérées (les déformations envisagées avoi-
sinent au maximum 20 à 30%), on retient une décomposition additive du tenseur de Green-
Lagrange. Le formalisme utilisé pour l’analyse de la plasticité finie est alors en tout point
36
identique à celui présenté précédemment en petites transformations. On reste ici dans le
cadre des matériaux standard généralisés avec le critère de von Mises et un écrouissage
isotrope, comme dans la plupart des travaux antérieurs. On choisit une loi d’écrouissage
non-linéaire à saturation [SIM, 1992], en exprimant l’énergie d’écrouissage à l’aide d’une
exponentielle :

α 1 2 exp (−c3 α)
w (α) = c0 α + c2 α + (c1 − c0 ) α + (15)
2 c3
ce qui contribue à la prédiction du phénomène de striction. Dans les applications suivantes,

les coefficients de la loi d’écrouissage sont calibrés à partir de la courbe de comportement
expérimentale d’un acier standard (XC48C) : c0 = 0 (pas d’écrouissage initial), c1 =
220 M P a, c2 = −560 M P a et c3 = 15.
Les calculs par éléments finis ont été effectués avec un modèle axisymétrique et donc
des éléments quadratiques à 8 nœuds. Concernant les conditions aux limites, deux solutions
sont retenues, toutes deux permettant d’amorcer le phénomène de striction et de décrire
naturellement le comportement post-critique et non la réponse triviale homogène du bar-
reau. La première technique, la plus employée dans la littérature, consiste à laisser libre
tout déplacement transverse des sections aux extrémités du barreau. La solution triviale
(pré-critique) est donc parfaitement homogène et un défaut géométrique sinusoı̈dal axi-
symétrique de très faible amplitude est introduit sur le pourtour du barreau pour favoriser
l’initiation de la striction.
On impose des déplacements longitudinaux aux extrémités du barreau jusqu’à un al-
longement relatif de l’ordre de 23%. Le comportement est instable (avec un point limite
en force, d’où le choix des déplacements imposés). La force associée diminue jusqu’à sen-
siblement la valeur limite d’élasticité (à laquelle s’est initiée la plastification). Le rayon
de la partie centrale du barreau (à l’endroit où la striction se manifeste) diminue aussi
considérablement, de plus de la moitié de sa valeur initiale pour le maximum d’allonge-
ment considéré. Le cas d’un barreau sans imperfection (pour lequel la striction n’est pas
censée apparaı̂tre numériquement) permet de mettre en valeur et de localiser le point de bi-
furcation correspondant au début de striction, et de comparer les deux solutions triviale et
bifurquée (figure 2). La bifurcation intervient à proximité du point limite en force, comme
observé dans [NEE, 1972] et [BRü, 1998] (figure 2(a)). Elle est tout aussi remarquable
sur la figure 2(b) à l’endroit où le rayon passe d’une évolution linéaire à une diminution
brutale.
La figure 3(a) décrit l’évolution de la zone plastique (active) au sein du barreau durant
le phénomène de striction. La plasticité s’initie au centre, du fait du défaut géométrique,
puis s’étend de manière quasi-homogène sur l’ensemble de la pièce. Lors de l’initiation du
phénomène de striction (entre 15 et 16% d’allongement), les extrémités entrent en phase
de décharge élastique et la zone plastique se limite progressivement à la partie centrale la
plus fortement déformée. La déformation plastique équivalente atteint la valeur maximale
de 87% sur la surface latérale de la partie centrale du barreau.
La seconde technique employée pour simuler la striction consiste à fixer transversale-
ment les sections extrêmes du barreau [NEE, 1972] [BRü, 1998]. La faible hétérogénéité
de la réponse pré-critique induite par ces conditions aux limites suffit pour dégénérer le
problème de bifurcation et suivre naturellement la courbe de striction, sans ajouter de
défaut géométrique. Si les déformées successives (et les zones plastiques correspondantes)
diffèrent naturellement entre les deux types de conditions aux limites (figure 3), les courbes
de réponse associées sont presque similaires (figure 2). La différence la plus notable est que
la plastification apparaı̂t tout d’abord au niveau des extrémités (à l’endroit du ”défaut”),
mais son évolution reste ensuite similaire au cas du barreau imparfait avec les bords libres.
37
(a) Courbes force-allongement (b) Courbes rayon-allongement
Figure 2 – Courbes de réponse de la striction d’un barreau cylindrique en traction
(a) Bords libres avec défaut géométrique (b) Bords encastrés sans défaut géométrique
Figure 3 – Évolution des zones plastiques (en noir) durant la striction d’un barreau cy-
lindrique en traction
Les résultats obtenus concordent qualitativement (et même parfois quantitativement,

lorsqu’on s’intéresse à des grandeurs adimensionnées) avec ceux des études antérieures
issues de la littérature, tant au niveau des courbes de réponse que de l’évolution des zones
plastiques, malgré des formulations et des matériaux différents. On montre ainsi la capacité
de notre formulation (lagrangienne totale), avec un modèle de plasticité des plus simples
(indépendamment de la généralité des développements effectués concernant l’intégration
locale), à simuler correctement et efficacement le phénomène de striction dans le cas d’un
barreau cylindrique en traction.
1.3 Modélisation de films polymères en grandes déformations plastiques
1.3.1 Problématique
Les derniers travaux réalisés dans ce domaine concernent la modélisation du com-

portement mécanique de films plastiques. Cette nouvelle activité a débuté dans le cadre
38
d’un Master Recherche [MR2 ] puis d’une thèse [Th2 ], et vient compléter plusieurs
thèses antérieures ayant porté sur la mise en œuvre, l’analyse physico-chimique et la ca-
ractérisation mécanique de ce type de matériaux polymères. Les films considérés sont
multi-couches, formés de polyamide (PA6) et de polyéthylène (PE) dans des proportions
variables (dont l’assemblage assure un bon compromis en termes de tenue mécanique et
d’imperméabilité pour l’application d’emballage alimentaire), ou d’un seul de ces consti-
tuants. Durant ou après le procédé de fabrication (co-extrusion gonflage), les matériaux
mis en jeu peuvent subir des tensions dans les deux directions (longitudinale et trans-
versale) et donc des déformations irréversibles qui viennent renforcer leurs propriétés
mécaniques. La modélisation des mécanismes de déformation des polymères constitutifs
(à haute température) s’avère donc nécessaire pour la prédiction du comportement des
films en mono- et bi-étirage (jusqu’à de très grandes déformations) et l’optimisation de la
conception des assemblages multi-couches. À ces fins, un modèle original à trois phases
a été spécialement développé, calibré à partir d’essais uniaxiaux et validé en bi-étirage
et sur divers autres résultats expérimentaux [P12 ,C15 ,C24 ,S2 ]. La principale difficulté
est de pouvoir rendre compte, de manière unifiée, des deux comportements en mono- et
bi-étirage, ce que les modèles standard ne permettent pas. Ce modèle est censé pouvoir
traduire le comportement en (très) grandes déformations plastiques d’un polymère semi-
cristallin quelconque à une température supérieure à sa transition vitreuse (on se place le
plus souvent à 90◦ C qui correspond à des conditions optimales d’étirage). Les hypothèses
sur lesquelles il s’appuie découlent de l’observation de la morphologie des matériaux et des
phénomènes micro-structuraux mis en jeu (distincts en traction uni- et bi-axiale) et s’ins-
pirent de la base de données des résultats expérimentaux déjà disponibles ou spécialement
effectués dans le cadre de cette étude, dont les principales conclusions sont brièvement
rappelées en introduction à la définition de notre modèle.
Tous les films plastiques considérés dans cette étude (films mono-couches en PA6 ou
PE, films multi-couches) ont été testés expérimentalement en mono- et bi-étirage jusqu’à
la rupture. Les essais uniaxiaux sont réalisés au moyen d’une machine de traction stan-
dard sur des éprouvettes haltères placées dans un four à température constante, avec un
taux de déformation nominale constant. Les déformations sont déduites de la mesure des
déplacements relatifs des mors (valeurs moyennes) mais aussi directement mesurées au
centre des échantillons au moyen d’un système de vidéo-traction (valeurs localisées dans
la partie utile des éprouvettes). Les essais sont effectués la plupart du temps en charge-
ment monotone, ou alors en chargement cyclique pour identifier la partie irréversible des
déformations et discuter d’éventuels effets visqueux (phénomènes d’hystérésis). À ce niveau
de température, l’indépendance des résultats vis-à-vis de la vitesse de chargement (dans
la gamme de vitesses modérées envisagées en pratique) nous conduit à négliger les effets
visqueux. Les cycles successifs de charge/décharge mettent en évidence une évolution du
module élastique sous une contrainte donnée avec le niveau de déformation plastique, soit
un certain couplage inhabituel entre les comportements élastique et plastique. Par ailleurs,
le système de vidéo-traction permet de mesurer à la fois les déformations longitudinale et
transversale dans le plan de l’échantillon. La relation linéaire entre les deux permet d’iden-
tifier un ”coefficient de Poisson” pour les deux matériaux. Sa valeur évolue légèrement selon
la gamme des déformations considérées et tend vers 0.5 pour les déformations les plus im-
portantes. Les matériaux, supposés isotropes, peuvent alors être considérés comme quasi-
incompressibles, ce qui est confirmé par les mesures de variations de volume négligeables
des échantillons et par les résultats de Hong [HON, 2005] concernant le PE.
Les courbes de traction biaxiale utilisées pour la validation proviennent d’une cam-
pagne expérimentale antérieure menée par Sallem-Idrissi [SAL, 2008]. Les essais biaxiaux
ont été accomplis sur un cadre de bi-étirage constitué de quatre pantographes actionnés
par deux vérins hydrauliques à angle droit, au sein d’une enceinte climatique qui maintient
39
la température constante. Les échantillons sont ici de forme carrée et soumis à un étirage
équi-biaxial simultané à faible vitesse. Relativement au cas uniaxial, les courbes de com-
portement biaxial affichent une rupture plus précoce, avant l’éventuel raidissement brutal
qu’on s’attendrait à observer de la même façon qu’en mono-étirage, phénomène auquel la
modélisation est censée apporter un élément de réponse à long terme.
La caractéristique principale commune aux courbes expérimentales de traction uni- et
bi-axiale est la présence de deux contraintes seuils, déjà fréquemment observées dans de
nombreuses études antérieures, qui dépend du degré de cristallinité ou de l’épaisseur des
lamelles cristallines, et de la température ou de la vitesse de déformation [BAL, 1993]
[FEI, 1997]. Ces deux seuils sont globalement dus à la coexistence de deux processus
indépendants à l’échelle moléculaire [SEG, 1990] [GUE, 2008]. Shan et al. [SHA, 2005]
ont remarqué que la présence de deux seuils était plus probable pour les matériaux à taux
de cristallinité faible et lors de sollicitations à faibles vitesses, conformément aux matériaux
et essais considérés dans cette étude.
En résumé, le comportement des deux polymères PA6 et PE considérés (et accessoi-
rement celui des multi-couches), tel qu’il a été quantifié expérimentalement, est de nature
fortement irréversible sans effets visqueux apparents. D’autre part, l’allure générale des
courbes (uniaxiales) avec une extrême déformabilité et un fort durcissement en grandes
déformations, et la quasi-incompressibilité des déformations (élastiques et plastiques), sont
toutes deux caractéristiques d’un comportement élastomérique. L’idée d’un comportement
élastoplastique associé avec une composante hyperélastique s’impose dans la modélisation
recherchée.
1.3.2 Développement d’un modèle élastoplastique à trois phases pour les po-
lymères semi-cristallins
Un modèle original a été spécialement développé pour rendre compte de manière unifiée
des spécificités du comportement élastoplastique uni- et bi-axial des polymères mis en jeu
(PA6 et PE) en très grandes déformations. Il s’inspire des résultats expérimentaux obtenus
et s’appuie sur les observations micro-structurales et les modélisations antérieures relevées
dans la littérature pour ce type de matériaux. Face à la complexité des modèles existants,
l’idée est de proposer un modèle aussi simple que possible, pour faciliter son implantation
dans un programme élément fini, avec le maximum d’hypothèses simplificatrices et un
nombre limité de paramètres matériels, faciles à identifier à partir d’essais mécaniques
simples et dont le sens physique apparaı̂t de manière évidente.
On a montré dans un premier temps l’incapacité d’un modèle élastoplastique standard
à reproduire les comportements à la fois uni- et bi-axial des matériaux considérés. On
utilise une loi élastique (linéaire) de Saint-Venant-Kirchhoff et un critère de plasticité de
type von Mises avec un écrouissage isotrope, incluant l’effet de la pression hydrostatique
(premier invariant du tenseur des contraintes) à la manière de Spitzig et al. [SPI, 1976].
On introduit, à l’instar de Zang et al. [ZAN, 2007], un terme de ”couplage” entre les par-
ties élastique et plastique du comportement afin de reproduire la variation significative
du module élastique en fonction de l’état de déformation plastique relevée durant les es-
sais cycliques de mono-étirage, comme dans [HON, 2005]. La première réserve apparaı̂t
lors de l’identification des paramètres du modèle à partir des essais uniaxiaux qui fait in-
tervenir deux lois polynomiales (l’une pour le module élastique et l’autre pour la courbe
d’écrouissage) avec énormément de paramètres et une signification physique limitée. En
outre, la comparaison des courbes expérimentale et simulée dans le cas biaxial est relative-
ment médiocre, quelle que soit la pondération apportée au terme de pression hydrostatique
dans l’écriture du critère de plasticité, dans la mesure où l’on ne parvient pas à reproduire
40
correctement à la fois les seuils nettement marqués dans les résultats expérimentaux et le
comportement ultérieur (la pente de la fin de la courbe expérimentale).
1.3.2.1 Construction du modèle La modélisation s’appuie sur les hypothèses sui-

vantes : on suppose des matériaux homogènes, isotropes et quasi-incompressibles, des
déformations uniformes (pas de striction) et isothermes à une température au-dessus de
la transition vitreuse, et enfin des effets visqueux négligeables. Dans cette étude, du fait
des très grandes déformations rencontrées (jusqu’à 800% d’allongement en traction uni-
axiale), on décrit la transformation du milieu par les déformations logarithmiques (de Hen-
cky). On suppose ensuite une décomposition additive de ces déformations, conformément
à la décomposition multiplicative de Lee du tenseur gradient de la transformation. D’un
point de vue énergétique, les grandeurs duales des déformations logarithmiques sont les
contraintes de Kirchhoff, qu’on peut identifier aux contraintes de Cauchy du fait de la
quasi-incompressibilité des matériaux, ce qui nous permettra d’implanter directement notre
modèle dans Abaqus (où les grandes déformations sont régies par le couple déformations
logarithmiques/contraintes de Cauchy).
Plusieurs tentatives de modélisation comme celle présentée précédemment avec un
modèle standard ont échoué et motivé, par conséquent, la construction d’un modèle
spécifique. La multiplicité des morphologies et des réponses observées selon les sollicita-
tions appliquées nous invitent à associer plusieurs modèles élémentaires pour caractériser
un seul et même matériau.
À l’instar de quelques travaux récents [DRO, 2008] [HON, 2004] [FU, 1994], on choi-
sit de représenter la morphologie des polymères semi-cristallins en trois phases : la phase
cristalline, la phase amorphe et une troisième phase à l’interface des deux premières aux
caractéristiques intermédiaires, qui peut être vue comme une ”mésophase” [AJJ, 1995]
[AJJ, 1996] qui correspond à une densification de la phase amorphe juste avant la cristal-
lisation. Parmi les descriptions proposées dans la littérature, on retiendra celle du modèle
de cluster. Son principe repose sur le fait que les blocs cristallins sont arrangés en lamelles,
ce qui nous amène à distinguer trois longueurs de chaı̂nes à l’inter-connection des blocs
cristallins : de très courtes chaı̂nes intra-lamellaires, et deux autres cas mettant en jeu des
distances inter-lamellaires soit intermédiaires soit très longues, selon la configuration locale
des lamelles. La considération de ces trois cas de figure revient à définir trois phases, toutes
distribuées uniformément au sein du matériau. Pour s’affranchir de la difficulté d’estimer
expérimentalement les proportions de chacune des phases en présence (dont le taux de
cristallinité), les paramètres matériels macroscopiques seront définis de manière globale
sans corrélation directe avec les aspects micro-structuraux (la dépendance est implicite).
Les trois phases distinctes considérées dans notre modèle sont caractérisées par la distance
moyenne entre les blocs cristallins, c’est-à-dire la longueur moyenne des chaı̂nes amorphes
qui les relient.
1.3.2.1.1 Cas n°1 Dans le cas de très courtes distances, les blocs cristallins contigus
sont fermement raccordés et on peut considérer la phase ainsi formée comme purement
cristalline : un modèle élastoplastique standard est alors retenu, avec le critère de von
Mises et un écrouissage isotrope.
1.3.2.1.2 Cas n°2 Lorsque la distance entre blocs cristallins est intermédiaire, la
plupart des chaı̂nes amorphes ne sont pas directement connectées à l’ensemble des blocs
cristallins environnants et leur comportement mécanique est alors essentiellement gou-
verné par la présence de nombreux enchevêtrements entre paires de chaı̂nes. Sous l’effet du
41
chargement, certains enchevêtrements deviennent rapidement effectifs (les deux chaı̂nes in-
teragissent) et un phénomène de glissement se produit lorsque la contrainte augmente. Ces
enchevêtrements sont supposés fournir tous la même contribution à la résistance globale
du matériau qui peut être caractérisée par une force limite τ correspondant au glissement
naissant entre deux chaı̂nes enchevêtrées. La contrainte macroscopique seuil uniaxiale σ2u
résultant de cette force limite est alors proportionnelle à la densité surfacique nu (uniforme)
du nombre d’enchevêtrements impliqués dans le comportement uniaxial :
σ2u = nu τ (16)
Le comportement uniaxial est donc régi par une loi élastique-plastique parfait avec la
contrainte maximale admissible σ2u définie ci-dessus et un module élastique E2 (englobant
la réponse élastique à la fois des blocs cristallins et de la phase amorphe) dont les effets
dans le modèle complet seront associés à ceux du module élastique du cas n°1.
Pour décrire le comportement multiaxial, on définit une contrainte équivalente de type
von Mises, du fait que le phénomène de glissement à l’origine du critère de plasticité re-
cherché est supposé gouverné par les effets de cisaillement ou de distorsion. D’un autre côté,
la nouvelle contrainte limite doit concorder avec le nombre d’enchevêtrements impliqués
dans le comportement multiaxial, qui dépend clairement du mode de déformation. Seul le
mode de déformation (par exemple, uni- ou bi-axial) et le ”signe” de ces déformations (la
compression n’a aucun effet sur les enchevêtrements) sont censés affecter les déplacements
relatifs entre chaı̂nes et ainsi le nombre d’enchevêtrements mis en jeu. Considérons un état
de déformation multiaxial arbitraire et concentrons-nous sur les déformations principales
strictement positives εI qui sont les seules déformations dans les directions principales
contribuant à l’emboı̂tement des chaı̂nes enchevêtrées. On introduit alors un ”coefficient
multiaxial” K, correspondant à la somme adimensionnalisée de ces déformations :
P
I εI
K= (εI > 0) (17)
maxI εI
La densité nm d’enchevêtrements impliqués dans le comportement multiaxial considéré

est ensuite supposée dépendre linéairement du paramètre K. La contrainte seuil associée
σ2m vérifie les mêmes règles que précédemment, en considérant que toutes les forces limites
τ correspondant à chaque enchevêtrement effectif peuvent être additionnées, sans tenir
compte de leur orientation. Dans le cas de la traction uniaxiale, l’équation (17) entraı̂ne
simplement K = 1, puisque la déformation mesurée dans la direction du chargement est
la seule valeur principale positive. On en déduit alors les relations suivantes :
nm = Knu σ2m = Kσ2u (18)
Un critère de von Mises ”modifié” est spécifiquement défini, basé sur la contrainte
équivalente de von Mises modifiée suivante :
σV M
σV∗ M = (19)
K
et l’unique valeur de contrainte seuil σ2u qu’on notera par la suite σ2 .
Si l’on s’intéresse particulièrement à l’essai de traction équi-biaxiale, le coefficient mul-
tiaxial prend la valeur K = 2, ce qui signifie que le nombre d’enchevêtrements impliqués
dans le phénomène de glissement est double par rapport au cas uniaxial. Tout autre état
de contrainte biaxial, pour lequel les deux déformations utiles sont différentes l’une de
l’autre, correspond à une valeur intermédiaire de K. Le coefficient multiaxial évolue en
fait continûment (et même linéairement) entre les valeurs 1 et 2 correspondant aux cas
limites des états de contraintes uniaxial et équi-biaxial, respectivement.
42
1.3.2.1.3 Cas n°3 Lorsque la région amorphe est particulièrement étendue entre
deux blocs successifs, il n’y a pour ainsi dire aucune liaison entre ces blocs et la phase
concernée est considérée comme purement amorphe : un modèle de réseau (hyperélastique)
de type 8-chaı̂nes est retenu pour ce cas. Les modèles de réseau ont tout d’abord été
développés pour les élastomères [BER, 1998] avant d’être généralisés à d’autres types de
matériaux (voir [MAR, 2002] et [MIE, 2004] pour une revue complète). La théorie des
modèles de réseau s’appuie sur la notion d’élasticité entropique des chaı̂nes moléculaires,
dans le contexte de la mécanique statistique (gaussienne ou non). L’idée générale repose sur
la définition de la densité de probabilité d’une grandeur cinématique telle que l’allongement
des chaı̂nes. L’équation de Boltzmann permet d’en déduire l’entropie puis l’énergie libre
d’une chaı̂ne, dont on peut déduire par dérivation la force duale du paramètre cinématique
retenu. Le principe du minimum de l’énergie potentielle permet finalement de calculer les
contraintes. Les différents modèles issus de cette théorie dépendent donc de l’expression
de l’énergie de déformation qui elle-même dépend de la fonction densité de probabilité.
On retient une approche statistique basée sur la fonction inverse de Langevin [KUH, 1942]
[JAM, 1943], plus apte à décrire l’extensibilité des chaı̂nes en déformations finies qu’une
approche gaussienne. Sans rentrer dans les détails, on peut citer, parmi les modèles qui en
découlent, le modèle de réseau complet (distribution uniforme de chaı̂nes dans une sphère)
difficile à programmer du fait de l’intégration numérique des contraintes, et les modèles p-
chaı̂nes, plus simples, dont le plus employé s’avère être le modèle 8-chaı̂nes [ARR, 1993]. Ce
modèle a la particularité, avec ses 8 chaı̂nes issues des 8 sommets d’un cube inscrit dans
la sphère unité, d’être symétrique vis-à-vis des trois directions principales.
q Ceci permet
X Y Z λ2 +λ2 +λ2
d’affirmer que les 8 chaı̂nes sont étirées avec le même allongement λ = 3 défini
à partir des allongements principaux. Les contraintes principales s’expriment alors de la
manière suivante :
1 √ λ2I −1

∂W λ
σI = λ I = −p + CR N L √ (20)
∂λI 3 λ N
Le calibrage de ce modèle repose sur la détermination des paramètres CR et N , qui

décrivent respectivement la densité des chaı̂nes et leur extensibilité. À partir d’un essai
de compression uniaxiale, Bergström et Boyce [BER, 1998] définissent λlim comme étant
la limite d’allongement à laquelle la contrainte tend vers l’infini (voir figure 4), dont ils
déduisent N par la relation :
1 2 2
N= λlim + (21)
3 λlim
Ensuite, à partir d’un couple de contrainte/allongement équivalents (σeq , λeq ) corres-

pondant à un point quelconque de la courbe d’équilibre en comportement uniaxial, la
valeur de CR peut être déduite :
s  s −1
2 2
σeq λ2eq + λeq λ2eq + λeq
CR = 2 L−1   (22)
λeq − λ1eq 3N 3N
De nombreux modèles élasto-(visco-)plastiques, spécialement mis en œuvre pour des

polymères semi-cristallins, découlent d’un modèle de réseau, auquel ont été associés en
parallèle une ou plusieurs autres composantes élastiques/plastiques ou en lien avec la
dépendance au temps ou à la température [BER, 2002]. Le comportement global est séparé
en deux (ou trois) mécanismes principaux. L’un d’eux représente les déformations pilotées
par les interactions intermoléculaires, qui sont responsables en majorité de la raideur ini-
tiale du matériau et également de la contrainte seuil à laquelle les premières déformations
43
Figure 4 – Définition de la limite d’allongement et d’un point de la courbe d’équilibre
pour la détermination des paramètres du modèle 8-chaı̂nes [BER, 1998]
plastiques apparaissent. L’autre mécanisme s’identifie le plus souvent à un comportement

de réseau avec, entre autres, le phénomène d’orientation des chaı̂nes moléculaires sous al-
longement, qui provoque le durcissement caractéristique de ces matériaux en très grandes
déformations, et éventuellement une prise en compte des effets de relaxation [DUP, 2007].
Dans cette étude, on dispose de trois modèles élémentaires à assembler convenablement
pour construire le modèle complet de notre matériau polymère semi-cristallin. Parmi toutes
les combinaisons possibles (parallèle/série) des trois comportements distincts, on élimine
les moins probables compte tenu des raideurs relatives des cas n°1, n°2 et n°3. En petites
déformations, les modules de raideur des cas n°1 et n°2 sont bien plus élevés que celui
du cas n°3, et l’inverse se produit en grandes déformations. Dans les deux cas, on peut
assimiler la raideur de l’assemblage série des deux raideurs bien distinctes à la raideur la
plus faible. Le modèle retenu finalement suit le schéma ”rhéologique” de la figure 5. Les
modèles des cas n°1 et n°2 sont supposés supporter la même contrainte de telle façon qu’un
arrangement en série est retenu pour l’association de ces deux parties (de par les longueurs
relatives des différents échantillons, il est fort probable qu’on rencontre dans n’importe
quelle direction arbitraire une série alternée de chaı̂nes courtes et moyennes, et non pas
seulement l’une de ces catégories, en parallèle de chaque chaı̂ne longue). La configuration
de la figure 5 est optimale, dans la mesure où elle correspond à l’arrangement le plus simple
capable de prédire l’influence négligeable du cas n°3 et des cas n°1 et n°2 durant les états
de déformations initial et avancé, respectivement. Les 7 paramètres retenus dépendent
implicitement des proportions de chacune des phases en présence et sont plus faciles à
évaluer expérimentalement que les modules intrinsèques de ces phases.
On peut vérifier a posteriori l’hypothèse d’incompressibilité du matériau ainsi modélisé.
Dans le cadre de très petites déformations (élastiques), un coefficient de Poisson ν est
défini comme le rapport entre les déformations transversales et longitudinales de l’essai de
traction uniaxiale et sa valeur avoisine 0.5. Pour les déformations (plastiques) modérées,
les critères de plasticité employés de type von Mises sont garants d’un comportement
44
Figure 5 – Représentation finale du modèle mécanique pour les polymères semi-cristallins
incompressible. Enfin, le modèle 8-chaı̂nes, prépondérant lors de très grandes déformations,

est lui aussi cohérent avec cette hypothèse.
1.3.2.2 Calibration du modèle à partir des essais de traction uniaxiale La

démarche appliquée pour la détermination des 7 paramètres du modèle est décrite au
travers de l’exemple du PA6 dont la courbe contrainte-déformation en traction uniaxiale
est présentée à la figure 6.
Figure 6 – Détermination des paramètres du modèle à partir de la courbe de traction

uniaxiale pour le PA6 à 90◦ C
Comme expliqué précédemment, les cas n°1 et n°2 sont regroupés pour ne former plus
qu’un jeu de paramètres élastoplastiques, à savoir E1−2 , ET 1 , σ1 et σ2 . Sur la figure 6, les
deux premières parties de la courbe (pour 0 6 ε 6 εA et εA 6 ε 6 εB , respectivement)
45
sont supposées linéaires, et leurs pentes respectives E et H sont obtenues par régression
linéaire avec un coefficient de corrélation optimal. Lorsque ε > εB au voisinage de εB (tant
que les déformations restent modérées), la courbe de réponse peut aussi être supposée
linéaire, et une troisième pente E0 peut être définie, également par régression linéaire, qui
correspond au module élastique initial du cas n°3. Une fois les valeurs limites εA , εB , σA
et σB ainsi que les trois pentes précédentes identifiées, les 4 paramètres du cas 1-2 peuvent
être déduits à l’aide des relations suivantes :
E1−2 = E − E0
ET 1 = H − E0
(23)
σ1 = σA − E0 εA
σ2 = σB − E0 εB
Le coefficient de Poisson est obtenu de manière découplée et les deux paramètres CR

et N du modèle 8-chaı̂nes sont évalués conformément à la méthode décrite précédemment,
en déplaçant l’origine des contraintes à σ2 .
1.3.2.3 Implantation du modèle dans un code de calcul par éléments finis Le

modèle 3-phases développé a été implanté dans le code de calcul commercial Abaqus par
l’intermédiaire d’une routine utilisateur UMAT en Fortran. Les films considérés en pratique
sont chargés dans leur plan et leur réponse est uniforme suivant l’épaisseur (au moins dans
le cas des mono-couches). On retient donc, dans un premier temps, une modélisation
purement 2D en respectant la condition de contraintes planes, compte tenu des épaisseurs
très faibles. Par souci de simplicité, cette condition de contraintes planes est imposée aux
deux contraintes en parallèle (associées aux modèles élastoplastique et hyperélastique) et
pas seulement à la contrainte totale, somme des deux précédentes, puisque chacune d’entre
elles joue tour à tour un rôle prépondérant. Du point de vue éléments finis, on préférera sauf
exception des éléments quadratiques avec une intégration numérique complète, c’est-à-dire
3x3 points de Gauss par élément, pour détecter au mieux l’occurence de la plasticité.
L’empilement non symétrique de différentes couches de matériaux induit natu-
rellement un couplage (traction/flexion) entre les différentes couches, qui se mani-
feste (expérimentalement) par des effets de courbure ou d’ondulation 3D. Ce type
de modélisation 2D ne permet alors pas de simuler directement le comportement des
matériaux composites que sont les films multi-couches. Le cas des films multi-couches sera
donc pour l’instant considéré de manière spécifique, et on proposera d’autres modélisations
alternatives pouvant englober le cas des multi-couches comme perspectives de ce travail.
Lors de la mise en œuvre numérique du modèle, les principales difficultés résident
dans l’intégration locale de la plasticité et dans la détermination des tenseurs tangents
élastoplastiques. L’objectif de la routine UMAT à programmer est de calculer, à partir d’un
état de déformation donné, le nouvel état de contrainte associé et le tenseur tangent du
modèle complet. Compte tenu de l’organisation du modèle 3-phases (figure 5), on découple
le problème en deux parties, l’une pour le comportement élastoplastique et l’autre pour
le comportement hyperélastique. Le tenseur des contraintes et le tenseur tangent pour
le modèle global seront tout simplement la somme des tenseurs correspondants des deux
parties en parallèle qui supportent les mêmes déformations.
L’intégration locale de la plasticité s’effectue de manière similaire à ce qui a été présenté
précédemment, en considérant le cas particulier du critère de von Mises (modifié ou non)
et d’un écrouissage isotrope linéaire. Dans ces conditions, le multiplicateur plastique peut
s’identifier à la seule variable d’écrouissage scalaire mise en jeu, qui s’avère être ici la
déformation plastique équivalente. La prise en compte de la condition de contraintes
46
planes dans l’algorithme de retour radial utilisé pour l’intégration locale du comportement
élastoplastique s’effectue classiquement par la projection itérative de la solution sur l’es-
pace réduit des contraintes planes. Le modèle 3-phases comporte deux spécificités quant
à la définition de la partie élastoplastique vis-à-vis d’un modèle standard classique, qui
nécessitent quelques aménagements concernant la mise en œuvre numérique. La présence
de deux sous-modèles (et donc de deux seuils) nous oblige à distinguer non pas deux
(élastique ou plastique) mais trois cas de figure quant à la position de la contrainte cou-
rante par rapport aux contraintes limites. La présence du coefficient multiaxial modifie
le critère de von Mises (pour le cas n°2) mais son évaluation à partir des déformations
connues à l’incrément courant ne pose aucune difficulté particulière. De manière générale,
ces deux points ne modifient pas fondamentalement la méthodologie de résolution mais
nous obligent cependant à réécrire notre propre programme d’intégration locale.
La détermination des contraintes et du tenseur tangent pour la partie hyperélastique
du modèle découle de l’écriture intrinsèque de la loi de comportement à partir de l’énergie
de déformation [CHA, 2003]. Pour faciliter l’écriture de ces différentes grandeurs, on intro-
duit le tenseur des dilatations de Cauchy-Green gauche B = F.FT . De manière générale,
l’énergie de déformation est supposée dépendre des trois invariants classiques du tenseur
des déformations/dilatations ou de son déviateur. Dans le cas du modèle 8-chaı̂nes (in-
compressible), elle ne dépend que du premier invariant I1 = trace(B). En utilisant ces
notations et compte tenu de ces propriétés, le tenseur des contraintes et le tenseur tangent
déviatoriques 3D peuvent s’exprimer de la manière suivante :

S = 2 I − 13 I ⊗ I : ∂W ∂I1 B
(24)
1
∂2W

C dev ⊗ I : ∂W : I − 31 I ⊗ I

=2 I− 3I ∂I1 H + 2 ∂I 2 B ⊗ B
1
1
avec Hijkl = 2 (δik Bjl + Bik δjl + δil Bjk + Bil δjk ).
Sous la condition de contraintes planes, les expressions se simplifient légèrement. Cette
condition lève l’indétermination sur la pression hydrostatique p due à l’incompressibilité du
comportement. On a alors simplement p = SZZ et les contraintes dans le plan s’écrivent :
∂W ∂W ∂W
σXX = 2 (BXX − BZZ ) σY Y = 2 (BY Y − BZZ ) σXY = 2 BXY (25)
∂I1 ∂I1 ∂I1
En découplant les parties planes (avec l’exposant .cp ) et les composantes hors plan (dans
la direction Z) des tenseurs (ou matrices) utilisés, on peut transformer de manière similaire
l’expression du tenseur tangent 3D (24) pour obtenir le tenseur tangent en contraintes
planes :
∂2W

cp ∂W cp ∂W cp cp cp cp cp cp
C =2 H +2 BZZ I ⊗ I + 2 2 (B − BZZ I ) ⊗ (B − BZZ I ) (26)
∂I1 ∂I1 ∂I1

cp cp cp cp
avec Hijkl = 12 δik Bjl + Bik δjl + δil Bjk + Bilcp δjk .
Les expressions (24) et (26) font apparaı̂tre des termes de convection qui proviennent
naturellement de la dérivation de quantités eulériennes. Ces termes supplémentaires obte-
nus dans les calculs de variations constituent l’expression de la dérivée de Jaumann, qui
s’avère être le choix d’Abaqus en termes de dérivée objective relativement à la formula-
tion lagrangienne réactualisée utilisée pour la résolution incrémentale en grandes trans-
formations. En effet, on peut montrer que c’est précisément la dérivée de Jaumann de la
contrainte de Kirchhoff qui se trouve être conjuguée d’un point de vue énergétique au taux
de déformation de Hencky.
47
1.3.3 Validation du modèle et applications
1.3.3.1 Traction uni- et bi-axiale des films mono-couches Les résultats de si-
mulation du mono- et bi-étirage du PA6 et du PE sont comparés aux courbes contrainte-
déformation expérimentales correspondantes dans les figures 7(a) et 7(b), respectivement.
Les courbes uniaxiales sont naturellement en très bon accord, du fait de leur utilisation
pour la calibration. Mais le résultat le plus satisfaisant est la très bonne concordance
obtenue pour les courbes biaxiales.
(a) PA6 (b) PE
Figure 7 – Comparaison des résultats de simulation et expérimentaux à 90◦ C dans les

deux cas de chargement
Afin de quantifier la variabilité de la réponse du matériau vis-à-vis de celle de ses

paramètres, on effectue une courte analyse paramétrique relative aux constantes du modèle
8-chaı̂nes, dont l’identification est certainement la plus hasardeuse parmi l’ensemble des
variables du modèle. La limite d’extensibilité des chaı̂nes macro-moléculaires N a une faible
influence sur les résultats de simulation, en particulier pour les fortes valeurs (déjà pour
l’ordre de grandeur des valeurs calibrées pour nos deux matériaux). On montre d’ailleurs,
en se limitant au comportement uniaxial, que lorsque la valeur de N devient suffisamment
importante, le modèle 8-chaı̂nes tend vers le modèle néo-hookéen (basé sur la théorie
statistique de Gauss) dont la loi de comportement ne dépend plus que de CR :

2 1
σX = CR λX − (27)
λX
En revanche, la valeur de CR a une influence significative sur les résultats obtenus,

comme le montre la figure 8 dans le cas du PE, qui nécessite une vigilance particulière
dans le choix du point arbitraire retenu pour la calibration du modèle.
1.3.3.2 Applications aux films multi-couches À partir des modèles respectifs des
deux matériaux PA6 et PE, on peut simuler le comportement mécanique de films multi-
couches. Si l’on s’affranchit des effets hors plan, ce qui revient à considérer une absence de
couplage entre les différentes couches successives, la simulation du comportement d’un film
multi-couches peut alors s’effectuer par l’application de la loi des mélanges sur les résultats
de simulation du même problème pour les deux matériaux pris séparément. L’hypothèse
fondamentale consiste à supposer que les déformations sont uniformes sur toute l’épaisseur
du film. La contrainte moyenne dans le film multi-couches (nominale ou vraie) peut alors
s’écrire comme la moyenne des contraintes équivalentes dans chacune des couches pondérées
48
Figure 8 – Courbes simulées de traction uni- et bi-axiale du PE avec différentes valeurs
de CR
par leurs fractions linéiques :
σ = χP E σP E + χP A6 σP A6 (28)
Les courbes de comportement uni- et bi-axiales ainsi simulées pour l’ensemble des multi-
couches considérés se répartissent entre les deux courbes limites correspondant aux mono-
couches, dans l’ordre des fractions respectives de chacun des constituants, et coı̈ncident as-
sez bien avec les résultats expérimentaux (figure 9). En traction uniaxiale, les résultats sont
très satisfaisants, ce qui signifie que le comportement d’un film multi-couches correspond
bien à la juxtaposition des comportements de chacune des couches indépendamment les
unes des autres. Dans le cas de la traction biaxiale, les légères différences observées peuvent
s’expliquer par le cumul de plusieurs types d’imprécisions et l’absence de prise en compte de
certains effets 3D, comme l’apparition de plis localisés, observés expérimentalement, lors de
déformations modérées. Ces plis sont dus aux décalages entre les contraintes seuils des deux
matériaux, aux effets transverses (coefficients de Poisson) distincts et éventuellement à la
légère anisotropie des deux matériaux. Ils jouent un rôle significatif en traction biaxiale en
diminuant la raideur globale lorsqu’ils tendent à disparaı̂tre, alors qu’ils sont majoritaire-
ment longitudinaux en traction uniaxiale (dans le sens du chargement) et subsistent jusqu’à
la rupture finale. L’analyse précise de ce type d’effets hors plan nécessite une modélisation
”tridimensionnelle” plus générale du problème qui sera discutée dans les perspectives de
ce travail.
1.3.3.3 Film troué en traction uniaxiale Les applications suivantes mettent en jeu
des aspects structuraux (avec des états de contraintes non uniformes) de façon à valider
notre modèle non plus seulement localement mais dans son environnement élément fini.
Elles concernent exclusivement le matériau PE, à température ambiante (23◦ C), pour
permettre la comparaison avec des essais effectués préalablement. À cette température,
le PE est toujours au-dessus de sa transition vitreuse et on peut donc utiliser le modèle
développé, à condition de le calibrer sur la nouvelle courbe uniaxiale à cette température.
49
(a) Traction uniaxiale (b) Traction biaxiale
Figure 9 – Comparaison des résultats expérimentaux et de l’application de la loi des

mélanges sur différents assemblages multi-couches à 90◦ C
Selon le point choisi arbitrairement pour la calibration du modèle 8-chaı̂nes, la valeur de CR

varie assez fortement. Pour mieux représenter la courbe uniaxiale du côté des déformations
modérées (inférieures à 0.5), on choisit pour CR la valeur extrême obtenue par extrapolation
linéaire à un niveau de déformation nul, soit CR = 2.88 M P a. On utilisera également,
pour comparaison, la valeur minimale mesurée en moyenne dans l’intervalle des grandes
déformations, à savoir CR = 2.12 M P a.
La première application consiste en la traction uniaxiale d’une éprouvette
précédemment trouée en son centre. Ce problème est par excellence simple à modéliser et à
reproduire expérimentalement, et à la fois exigeant de par les forts gradients de contraintes
observés au voisinage du trou. Il représente une étape intermédiaire, avant de s’intéresser à
la fissuration des films. L’éprouvette utilisée est ici rectangulaire, de plus grande taille que
lors des essais uniaxiaux standard, de façon à pouvoir garder un rapport d’échelle suffisant
entre les dimensions de l’échantillon et le rayon du trou circulaire.
La simulation a été effectuée jusqu’à un déplacement maximal correspondant à la valeur
atteinte expérimentalement avant rupture. La figure 10 présente la déformée (à l’échelle
1) à ce niveau de chargement avec les isovaleurs des contraintes de von Mises. Au voisi-
nage du trou, la valeur maximale de contrainte de von Mises, à l’extrémité inférieure de
l’arc de cercle, correspond à un état de contrainte purement uniaxial dans la direction du
chargement, avec une amplitude trois fois supérieure à la valeur nominale loin du trou. Ce
résultat est remarquable, car similaire à ce qu’on obtient analytiquement dans le cas de
l’élasticité linéaire (en petites déformations), pour un trou infiniment petit.
Les résultats de simulation sont validés au travers de la comparaison de la courbe
force-déplacement expérimentale aux courbes équivalentes obtenues numériquement pour
les deux valeurs extrêmes du paramètre CR retenues précédemment (figure 11). Les deux
courbes numériques coı̈ncident parfaitement avec la courbe expérimentale durant la phase
élastique et le début de plasticité. Le résultat le plus significatif est de retrouver une allure
de courbe simulée semblable à la solution expérimentale, avec en particulier un pic de
force suivi d’un minimum (ou deux points d’inflexion successifs) qui pourrait s’identifier à
une sorte de striction dont l’initiation serait facilitée par la présence du trou. La fin de la
courbe est également bien reproduite avec une très bonne estimation de la force maximale
admissible par la structure. Le choix de ces deux valeurs de paramètre permet en tout
cas d’encadrer la courbe expérimentale par deux courbes simulées assez proches l’une de
l’autre.
À titre de comparaison, la figure 11 présente également les résultats de simulation
obtenus pour un modèle élastoplastique standard (calibré avec la même courbe de trac-
50
(a) Modèle global (b) Zoom autour du trou
Figure 10 – Distribution des contraintes de von Mises pour le déplacement imposé maxi-
mal
tion uniaxiale que le modèle 3-phases). La courbe associée au modèle standard affiche
également un pic relativement aigu mais décroı̂t ensuite brutalement jusqu’à des valeurs
de contraintes proches de zéro. Les écarts considérables entre les courbes issues des deux
modèles sont dus à la présence ou non du modèle 8-chaı̂nes qui décrit à lui seul le raidis-
sement du matériau en grandes déformations, et à celle du coefficient multiaxial qui joue
un rôle non négligeable dans la zone intermédiaire entre les petites déformations élastiques
et les grandes déformations plastiques. La description correcte, dans notre approche, du
comportement biaxial du matériau (qui se manifeste au voisinage du trou) est le garant
de la qualité du modèle défini et justifie toute l’attention qu’on a portée à cet égard.
1.3.3.4 Fissuration d’un film (essai DENT) On s’intéresse dans un deuxième

temps à la fissuration des films, en s’appuyant sur des essais réalisés au cours d’une étude
préalable [POI, 2005] sur des éprouvettes DENT (Double-Edge Notched Tension) avec
différentes tailles de fissure initiale. En présence d’une fissure, les contraintes se concentrent
naturellement au voisinage du front, davantage que dans le cas précédent du trou circu-
laire, et les déformations (plastiques) s’intensifient alors autour de la fissure qui se propage
pour des niveaux de chargement nominal bien inférieurs à ceux atteints lors de la rupture
de films initialement vierges de défauts. Parmi les différentes techniques disponibles pour
modéliser ce phénomène de fissuration, on choisit la méthode des éléments de cohésion,
déjà utilisée dans le contexte des polymères par Gao et Klein [GAO, 1998] et Nguyen et al.
[NGU, 2004], entre autres. Les méthodes basées sur la mécanique de la rupture et le calcul
des facteurs d’intensité de contrainte ou de l’intégrale J nécessitent un effort particulier de
maillage et de remaillage automatique tout au long de l’évolution de la fissure. L’utilisa-
tion d’un modèle d’endommagement, forcément couplé à la modélisation de la plasticité,
nécessiterait un nouvel effort de construction de la loi de comportement. La trajectoire
de la fissure étant connue à l’avance (rectiligne), on choisit une solution intermédiaire qui
51
Figure 11 – Comparaison des courbes force-déplacement expérimentale et numériques
(pour le modèle 3-phases avec plusieurs valeurs de CR et pour un modèle
élastoplastique standard) du problème de traction uniaxiale d’un film troué
permet de décrire l’endommagement du film jusqu’à la rupture tout en se focalisant sur

la zone susceptible de se dégrader, sans nouvelle difficulté ni de maillage ni de program-
mation. On caractérise la zone où se propage la fissure (une mince couche d’éléments) par
un comportement cohésif (élastique endommageable) en utilisant les éléments de cohésion
disponibles dans Abaqus et, de manière découplée, le reste de la structure par notre modèle
3-phases. Le nouveau comportement fait apparaı̂tre de nouveaux paramètres qui devront
être calibrés à partir d’essais de référence.
Le comportement des éléments de cohésion est défini (dans chaque direction) par une
courbe de type traction-séparation, avec un comportement élastique (linéaire) jusqu’à une
contrainte maximale admissible, suivi d’un comportement linéaire ou non avec endom-
magement progressif jusqu’à la rupture (lorsque la contrainte s’annule). Plusieurs critères
d’initiation et lois d’évolution de l’endommagement peuvent être utilisés dans le cadre de
sollicitations multiaxiales. Dans notre cas, étant donné que seul le mode d’ouverture entre
en jeu dans la décohésion du matériau, le choix du critère d’initiation n’est pas déterminant
et on retient arbitrairement le critère de la contrainte maximale. Quant à l’évolution de
l’endommagement, elle est décrite par l’intermédiaire de l’énergie, avec un adoucissement
linéaire et un choix de critère à nouveau arbitraire. La rupture est alors logiquement définie
pour une énergie dissipée maximale.
Seules les grandeurs matérielles relatives à la direction normale à la surface de cohésion
sont identifiées. On affecte la valeur du module élastique initial du modèle 3-phases à la
raideur élastique des éléments de cohésion dans la direction normale. La détermination
des valeurs de l’énergie de rupture et surtout de la contrainte maximale admissible (plus
influente) fait l’objet d’une optimisation par rapport aux résultats expérimentaux de l’essai
correspondant à la plus petite taille de ligament restant. Dans les deux directions tangentes,
on choisit de prendre les mêmes valeurs que celles calibrées dans la direction normale, en
supposant le comportement de la zone de cohésion isotrope.
52
L’évolution de la déformée et la répartition des contraintes de von Mises sont
représentées à la figure 12 dans le cas du ligament de plus petite taille.
Figure 12 – Simulation de la propagation de fissure : déformées et contraintes de von

Mises
À des fins de validation, on trace enfin les paires de courbes force-déplacement pour les
cinq tailles initiales de ligament considérées (figure 13). Si le début de courbe élastique
paraı̂t bien concorder entre les deux approches (numérique et expérimentale), le seuil
de plasticité semble être atteint plus tard avec la simulation. Les forces limites sont
généralement semblables mais atteintes pour des déplacements inférieurs. Toutes ces ob-
servations concernent identiquement l’ensemble des cinq cas de figure, ce qui tend à prou-
ver que les réponses sont ”proportionnelles” entre elles, aussi bien d’un point de vue
expérimental que numérique.
53
Figure 13 – Comparaison des résultats de simulation et d’essai de fissuration avec
différentes longueurs initiales de ligament
1.4 Bilan et perspectives
Mes travaux de recherche sur la plasticité reposent sur une formulation générale de
l’intégration locale établie dans le cadre des matériaux standard généralisés pour un modèle
de plasticité (critère et écrouissage) arbitraire. L’intérêt de cette formulation est d’avoir
obtenu des expressions exactes pour la correction itérative de l’incrément de multiplicateur
plastique (et les corrections associées des contraintes et variables d’écrouissage), valables
pour une large classe de modèles, sans hypothèses restrictives. Parallèlement, on fournit
une expression générale du tenseur tangent consistant (symétrique) faisant intervenir un
terme original (dérivée d’ordre 3 de l’énergie d’écrouissage) vis-à-vis des travaux antérieurs
supposant a priori une forme particulière de la loi d’écrouissage. L’utilisation du schéma
d’Euler implicite et du tenseur tangent consistant assurent la stabilité et une convergence
quadratique pour la résolution incrémentale du problème de plasticité.
Les résultats présentés ne concernent que la plasticité associée mais peuvent être fa-
cilement étendus au cas de la plasticité non-associée, sans modification majeure de la
procédure. L’utilisation d’un potentiel F différent de la fonction critère f dans l’écriture
des lois d’évolution conduit à un tenseur tangent d’expression similaire mais cette fois non
symétrique. La généralisation à des modèles de plasticité plus élaborés (avec davantage de
variables internes que les deux variables d’écrouissage mises en jeu dans le cas présent),
voire à des modèles d’endommagement, est possible mais plus délicate.
La procédure d’intégration locale a été présentée dans le cadre des petites transforma-
tions. Mais l’utilisation d’une description lagrangienne totale des grandes transformations
permet de conserver le formalisme précédent en plasticité finie. Le choix d’un écrouissage
isotrope non-linéaire spécifique a permis de traiter correctement le problème d’un barreau
cylindrique en traction, de mettre en valeur le phénomène de bifurcation qu’est la striction
et de décrire efficacement la solution post-critique jusqu’à un stade avancé, grâce à l’in-
troduction de différentes types d’imperfections. Depuis, des méthodes numériques vouées
à la résolution de problèmes de bifurcation et développées dans le cadre de l’analyse du
54
flambement ont été introduites dans le programme de plasticité finie 3D. Une perspective
immédiate serait d’appliquer ces méthodes de branchement au calcul de striction d’un bar-
reau sans défaut d’aucune sorte (tel que la solution triviale soit homogène), afin d’affiner la
solution et de considérer le problème comme un réel problème de bifurcation. Par ailleurs, le
choix d’un écrouissage cinématique non-linéaire particulier a permis de mettre en évidence
plusieurs phénomènes de rochet uni- et multi-axiaux, avec des résultats comparables aux
solutions analytiques et numériques de référence.
Les travaux les plus récents dans ce domaine concernent la modélisation du compor-
tement de films polymères en grandes déformations élastoplastiques. Cette étude couvre
l’ensemble de la démarche allant de l’identification d’un modèle approprié à la valida-
tion expérimentale en passant par la calibration des paramètres et la mise en œuvre
numérique. L’originalité de ce travail vient de la construction d’un nouveau modèle à
trois phases susceptible de prédire de manière unifiée le comportement quasi-statique
uni- et bi-axial (et multiaxial quelconque par extension) de polymères semi-cristallins
à une température au-dessus de la transition vitreuse. Sous des hypothèses peu restric-
tives et à partir de considérations morphologiques et expérimentales, le modèle s’orga-
nise en trois parties représentant chacune une phase particulière du matériau (cristalline,
amorphe ou intermédiaire). Chaque phase est caractérisée par un modèle élastoplastique
ou hyperélastique simple mais c’est la mise en commun de ces différents sous-modèles qui
donne toute son originalité au modèle complet. Alors que de nombreux modèles dans la
littérature ne considèrent que les deux phases cristalline et amorphe, la définition d’une
troisième phase intermédiaire, caractérisée par un critère de plasticité faisant intervenir la
multiaxialité des déformations, constitue l’un des apports principaux de la modélisation.
La calibration du modèle à partir des essais de traction uniaxiale est particulièrement
aisée du fait du faible nombre de paramètres mis en jeu (seulement 7 au total) et de
la signification physique de ces paramètres, qui ne dépendent qu’implicitement des pro-
portions de chacune des phases. Dans le prolongement de ce travail, on pourrait s’atta-
cher à corréler les paramètres du modèle avec certaines caractéristiques micro-structurales.
L’intérêt d’une telle analyse serait de pouvoir calibrer une fois pour toutes le comporte-
ment d’un type de matériau donné quelles que soient les proportions des différentes phases
sans procéder à autant d’essais que de taux de cristallinité, par exemple. Dans un même
ordre d’idées, il est envisageable d’étudier l’influence de la température sur les paramètres
mécaniques. Toutes ces considérations ne remettent pas en cause la définition générale
du modèle élastoplastique développé mais compliquent nettement l’identification des pa-
ramètres matériels. À l’inverse, la prise en compte d’effets visqueux éventuels pour pouvoir
traduire des taux de déformation plus élevés nécessite de modifier fondamentalement la
nature du modèle.
La mise en œuvre numérique du modèle n’est pas originale en tant que telle mais
nécessite un effort de programmation important du fait de l’originalité du modèle. Si
l’intégration locale de la plasticité s’inspire largement des précédents développements, de
nouvelles difficultés apparaissent avec le modèle 8-chaı̂nes quant à l’écriture du tenseur
tangent, et pour l’ensemble du modèle du fait des nouveaux tenseurs eulériens et de la
configuration lagrangienne réactualisée utilisés dans Abaqus. Les calculs par éléments finis
effectués avec ce programme décrivent correctement le comportement des matériaux PA6
et PE pour différentes géométries sous diverses sollicitations jusqu’à la rupture (survenant
pour de très grandes déformations). Les premiers résultats sur la fissuration des films sont
prometteurs et motivent une réflexion plus approfondie sur la modélisation de l’endomma-
gement (en incluant un critère d’initiation), pour pouvoir décrire le phénomène de rupture
dans son ensemble.
L’intégration de la loi de comportement dans Abaqus se limite actuellement au cas 2D
en contraintes planes. La simulation du comportement des films multi-couches n’est pas
55
directement envisageable avec cette modélisation et a été pour l’instant uniquement traitée
de manière détournée. Une perspective évidente sur le plan numérique est de proposer une
nouvelle routine utilisateur pour le cas 3D (voire pour une modélisation de type coque)
afin de pouvoir traiter numériquement des problèmes de multi-couches sans hypothèses
simplificatrices dans la direction de l’épaisseur et mettre en valeur les effets 3D occasionnés
par le gradient de propriétés dans l’épaisseur. L’objectif à long terme est de pouvoir prédire
par le calcul la rupture des films multi-couches et optimiser ainsi les assemblages pour une
meilleure résistance mécanique.
Après les matériaux métalliques et les polymères, citons pour terminer la contribu-
tion à une étude sur la modélisation de renforts fibreux (secs) en compression uniaxiale,
mettant en jeu de grandes transformations irréversibles. Les résultats expérimentaux en
chargement cyclique exprimés en termes de contraintes et déformations moyennes lagran-
giennes ont permis d’identifier un comportement simple, assimilable à de l’élastoplasticité
[P6 ,C5 ,C22 ]. La poursuite de ce travail consisterait à corréler le modèle de plasti-
cité ”macroscopique” aux propriétés géométriques et matérielles des renforts à l’échelle
mésoscopique par une démarche d’homogénéisation.
56
2 Homogénéisation périodique de matériaux composites
La deuxième thématique abordée dans ce mémoire concerne l’homogénéisation

périodique. Les matériaux composites occupent une place de plus en plus importante dans
de nombreux domaines de l’industrie, avec en particulier l’avènement de nouveaux compo-
sites plus fonctionnels et plus fiables, et de moins en moins coûteux. Jusqu’à maintenant,
les structures composites les plus utilisées en pratique sont les stratifiés uni- ou multi-
directionnels, reconnus pour leurs excellentes propriétés mécaniques. Ils se comportent de
façon remarquable face à des chargements dans le plan des renforts suivant la direction
des fibres. Cependant, en présence de sollicitations hors plan (compression, cisaillement,
impact), les stratifiés ne sont plus aussi performants et on tend à les remplacer par des
composites 3D (renforcés dans la direction de l’épaisseur) qui font naturellement preuve
d’une meilleure tolérance à l’endommagement et retardent entre autres la rupture par
délaminage. Grâce à tous leurs avantages, ces matériaux à renforts 3D sont souvent syno-
nymes d’innovation et stimulent la recherche et le développement de nouveaux produits
dans de nombreux domaines d’application tels que l’aérospatiale, l’aéronautique, l’auto-
mobile, le génie civil et les sports et loisirs. D’un point de vue scientifique, l’étude de tels
matériaux composites ouvre de nouvelles perspectives du fait de l’originalité de la géométrie
des renforts et/ou des matériaux constituants. Dans ce contexte, et dans le cadre de divers
projets et collaborations, des travaux de recherche ont donc été engagés dans le domaine
des composites à renforts 3D.
D’un point de vue modélisation mécanique, tous les matériaux composites en général, et
particulièrement les composites 3D du fait de leur architecture souvent complexe, affichent
des coûts rédhibitoires quand il s’agit de représenter la géométrie d’une pièce de grande
taille en incluant les hétérogénéités à une échelle bien moindre. La réponse à ce problème
s’avère être l’homogénéisation. Le principe d’une méthode d’homogénéisation consiste à
identifier le comportement macroscopique effectif d’un matériau hétérogène, par des cal-
culs préliminaires sur une cellule unité ou un Volume Élémentaire Représentatif, de façon
à pouvoir modéliser la structure considérée à l’aide d’un matériau homogène équivalent,
et effectuer ainsi un dimensionnement plus efficace et beaucoup moins coûteux par l’utili-
sation d’un modèle de taille plus raisonnable. Compte tenu des matériaux considérés par
la suite, parmi les méthodes existantes, on fait le choix de l’homogénéisation périodique
et on vise à remplacer une cellule 3D par un comportement équivalent de type plaque (la
périodicité n’apparaı̂t que suivant les deux directions du plan de la structure composite).
L’homogénéisation asymptotique d’une plaque périodique a déjà fait l’objet de nombreuses
études et se trouve juste employée ici dans un contexte original. Pour couvrir le maximum
d’applications, on choisit une modélisation numérique basée sur la méthode des éléments
finis. La démarche de calcul mise en œuvre ne nécessite pas de développements particu-
liers concernant les éléments finis, si bien que le programme réalisé (en langage Python)
pour la modélisation géométrique, la résolution et le post-traitement des résultats s’ap-
puie entièrement sur les possibilités standard du code commercial Abaqus. En pratique,
on s’intéresse principalement au comportement homogénéisé élastique linéaire (à savoir les
modules ou raideurs intégrées, dont le nombre dépend du degré d’anisotropie du matériau
équivalent), mais aussi, dans une moindre mesure, à l’endommagement local (en particulier
la décohésion fibres/matrice).
Deux matériaux particuliers à renforts 3D (tous deux novateurs en comparaison des
composites stratifiés classiques) sont considérés, chacun présentant des avantages ou
intérêts spécifiques. Le premier matériau auquel on s’intéresse est un composite renforcé
d’une structure tissée interlock 3D. La présence de fils liants dans l’épaisseur d’un tel
matériau contribue fortement à l’amélioration des propriétés mécaniques dans cette direc-
tion, vis-à-vis des textiles 2D traditionnels. Les travaux concernant les composites interlock
57
ont fait l’objet d’une thèse [Th3 ] en collaboration avec l’École Nationale Supérieure des
Arts et Industries Textiles de Roubaix, et s’appuient sur les compétences conjointes des
deux établissements dans la conception et la mise en forme de ce type de matériaux.
Une étude préliminaire a pu montrer l’inadéquation des méthodes de calcul analytiques
basées sur des schémas micro-mécaniques (comme la loi des mélanges ou le schéma de
Mori-Tanaka) du fait de la complexité de la géométrie du renfort, et en particulier de la
présence de fils dans toutes les directions. La méthode d’homogénéisation numérique uti-
lisée s’avère être suffisamment générale, par rapport aux modélisations semi-analytiques
ou numériques rencontrées dans la littérature, pour intégrer la réalité du matériau (dont
on maı̂trise le procédé de fabrication, du tissage à l’infusion), à savoir prendre en compte
les éventuels défauts et aléas de la géométrie (parmi lesquels l’ondulation des fils) et mesu-
rer ainsi la sensibilité des propriétés mécaniques macroscopiques vis-à-vis des paramètres
géométriques de la micro-structure.
Le second matériau étudié consiste en un sandwich 3D Napcor , dont le procédé de fa-
brication et le comportement mécanique sont conjointement étudiés dans le cadre du pro-
jet NWC-X. Il s’agit d’une structure sandwich dont l’âme en mousse est renforcée par des
fibres transversales (orthogonales aux peaux ou éventuellement obliques) provenant direc-
tement des peaux par aiguilletage transverse. Cette technologie récente permet de renforcer
la structure vis-à-vis des sollicitations transverses. Contrairement au matériau précédent,
la simplicité de la géométrie permet ici l’obtention analytique de certains résultats d’ho-
mogénéisation et donc une validation plus complète des outils numériques développés. Pour
ce matériau, non seulement les modules (ou raideurs) en traction et flexion dans le plan
(caractérisant le modèle de plaque de Kirchhoff-Love) doivent être évalués, mais aussi le(s)
terme(s) de cisaillement transverse (apparaissant dans le modèle de Mindlin-Reissner),
du fait de l’importance considérable du glissement transverse dans un sandwich en flexion
simple. Une première méthodologie pour l’analyse de l’endommagement a été mise au point
sur ce matériau simple au cours d’un Master Recherche [MR3 ]. Des éléments de cohésion
ont été définis avec une loi de comportement adéquate pour rendre compte, en traction
dans un premier temps, du décollement fibres/matrice avec pour objectif de définir, à partir
de quelques calculs en chargement simple (tractions uniaxiale, équi-biaxiale et cisaillement
simple), un critère d’initiation de l’endommagement et les différents états stabilisés pos-
sibles. Une thèse [Th4 ] sur l’homogénéisation du sandwich Napcor est actuellement en
cours. Elle s’appuie, comme pour le matériau précédent, sur une caractérisation précise de
la géométrie et des matériaux constituants.
L’ensemble des travaux réalisés autour de l’homogénéisation, les méthodes analytiques
exploitées ainsi que les outils de calcul développés, constituent de manière générale une
introduction à l’analyse du flambement des matériaux composites, auquel je m’intéresse
désormais tout particulièrement. La première application concerne le sandwich Napcor qui
pose naturellement des problèmes d’instabilité lorsqu’il est comprimé suivant l’épaisseur.
Cette partie de la thèse sur le flambement local des renforts sera présentée dans la dernière
section du mémoire.
Dans ce qui suit, on décrit rapidement les grandes lignes de la méthode d’ho-
mogénéisation numérique mise en œuvre, ainsi que quelques notions relatives à la
modélisation de l’endommagement. Puis on présente les résultats les plus significatifs, va-
lidés expérimentalement ou numériquement, concernant successivement les deux matériaux
composites 3D envisagés.
58
2.1 Méthodes d’homogénéisation numérique
2.1.1 Homogénéisation asymptotique d’une structure périodique
Le principe de l’homogénéisation d’un matériau hétérogène consiste à calculer ses

constantes effectives de façon à pouvoir le remplacer par un matériau homogène équivalent
dans les applications de calcul de structures. Cela permet d’éviter des temps de
modélisation et de calcul en 3D d’autant plus importants que les hétérogénéités sont
fines vis-à-vis des dimensions de la structure considérée, sans parler des problèmes de
place mémoire. Les composites considérés sont exclusivement des structures minces et
périodiques suivant les deux directions du plan. Par conséquent, la méthode mise en
œuvre ici va consister en une homogénéisation non pas 3D mais 2D, qui remplace le so-
lide hétérogène par un modèle de plaque équivalent, en s’appuyant sur une cellule unité
dont l’épaisseur est celle du matériau et qui représente le plus petit élément permettant de
reproduire le matériau complet par périodicité. L’approche utilisée s’inscrit dans le cadre
de l’homogénéisation asymptotique, dont les principaux précurseurs sont Bensoussan et al.
[BEN, 1978], Bakhvalov et Panasenko [BAK, 1989] et Sanchez-Hubert et Sanchez-Palencia
[SAN, 1992], entre autres. Elle s’inspire particulièrement de la démarche employée par
Buannic et al. [BUA, 2003] pour des structures sandwich et permet d’identifier un modèle
équivalent de plaque de Kirchhoff-Love à partir du calcul des raideurs effectives en traction
et en flexion.
La formulation de l’homogénéisation asymptotique d’une structure mince périodique
dans le plan fait intervenir deux rapports d’échelle qui jouent le rôle de pa-
ramètres infinitésimaux dans les développements asymptotiques. Il s’agit du rapport
épaisseur/longueur caractéristique de la plaque (minceur) et du rapport longueur de la
cellule élémentaire/longueur caractéristique de la plaque (périodicité), qui seront supposés
ici confondus car du même ordre de grandeur. La méthode d’homogénéisation asympto-
tique consiste à développer les champs de déplacement, déformation et contrainte en séries
entières de ce paramètre e qu’on fait tendre ensuite vers 0 [CAI, 1984]. La solution en
déplacements du problème d’élasticité 3D est en particulier recherchée sous la forme :
u = u(0) (x1 , x2 )+eu(1) (x1 , x2 , y1 , y2 , y3 )+e2 u(2) (x1 , x2 , y1 , y2 , y3 )+... avec yi = xi /e (29)
en introduisant les deux descriptions macroscopique ”x” à l’échelle de la structure et mi-

croscopique ”y” à l’échelle des hétérogénéités. La résolution complète du problème 3D peut
donc se décomposer en une séquence de problèmes microscopiques 3D sur une cellule unité
et une séquence de problèmes macroscopiques 2D sur la structure entière. On s’intéresse
ici uniquement à la solution du problème microscopique au premier ordre, qui permet
de déterminer les raideurs effectives de la plaque homogène équivalente dans le cadre de
l’hypothèse de Kirchhoff-Love.
La démarche employée revient à supposer un état de déformation moyen dans la cellule
unité, représenté par les tenseurs des macrodéformations de membrane E et de courbure K,
et à en déduire les contraintes intégrées (et par conséquent les raideurs effectives) obtenues
avec des conditions aux limites périodiques. Le champ de déformation local à l’équilibre
se divise en deux parties : la déformation macroscopique E + y3 K et une déformation
corrective (de moyenne volumique nulle) qui apparaı̂t du fait des hétérogénéités et qui
dérive d’un champ de déplacement périodique uper [MIC, 1999]. Le problème revient alors
à déterminer les champs de déplacement uper , de déformation ε et de contrainte σ locales
59
tels que :
divσ = 0
σ=D:ε
εij = Eij + y3 Kij + ε∗ij (uper ) avec i, j = 1, 2 (30)
σ.n = 0 sur ∂Y±
uper y1 − et y2 − périodique et σ.n anti-périodique
Une fois ce problème résolu, on définit les grandeurs duales des macrodéformations, à
savoir les résultantes N et les moments M des contraintes moyennes, pour pouvoir écrire
la loi de comportement homogénéisée sous la forme matricielle suivante :
( ) " #( )
{N } [A] [B] {E}
= (31)
{M } [B]T [C] {K}
dans laquelle les matrices [A], [C] et [B] forment la matrice globale des raideurs effec-
tives, comprenant respectivement les constantes relatives au comportement en traction, au
comportement en flexion et au couplage entre les deux.
Le fait d’imposer comme données d’entrée les macrodéformations permet de résoudre
le problème avec des éléments finis standard. On utilise alors le code commercial Abaqus
pour la résolution numérique par éléments finis. Dans les faits, on applique dans un calcul
préliminaire les efforts volumiques et surfaciques appropriés sur une cellule homogène pour
générer successivement chaque macrodéformation souhaitée. Puis on applique cette ma-
crodéformation comme condition initiale à la cellule hétérogène, qui se déforme sous des
conditions aux limites périodiques jusqu’à trouver son état d’équilibre. Le post-traitement
des contraintes pour calculer les valeurs moyennes intégrées n’étant pas immédiat, on se
rabat sur l’utilisation de l’énergie de déformation qui est directement disponible comme
grandeur de sortie sur Abaqus. L’expression de cette énergie de déformation en fonction
des grandeurs moyennes/homogénéisées fournit une seule équation :
Z
1 S
W = σ : εdV = {E}T [A] {E} + {K}T [C] {K} + 2 {E}T [B] {K} (32)
2 V 2
L’obtention des 21 termes (par symétrie) de la matrice de rigidité globale nécessite

alors la réalisation de 21 calculs avec 21 macrodéformations indépendantes (les 6 ma-
crodéformations élémentaires et des combinaisons de celles-ci). À partir de la matrice
raideur globale, et dans le cas fréquent d’un découplage entre traction et flexion (lorsque
la matrice [B] est nulle), on peut déduire des expressions de modules d’Young, coefficients
de Poisson et modules de cisaillement effectifs pour les comportements de traction et de
flexion, indépendamment l’un de l’autre, en écrivant les deux matrices de souplesse :
   
f f f
1/E1m m /E m
−ν21 2 0 1/E 1 −ν 21 /E 2 0
1 12 
[A]−1 =  −ν12 m /E m 1/E m 0 [C] −1
= −ν f
/E f
1/E f
0
 
1 2 3 12 1 2
T T
  
f
0 0 1/Gm12 0 0 1/G12
(33)
Au lieu d’utiliser l’interface graphique de pré-traitement Abaqus/CAE, on crée un
programme script en langage Python afin d’automatiser toutes les étapes de travail, à
savoir la rentrée des données, la modélisation géométrique, le maillage, la définition des
propriétés matérielles, la génération des chargements et conditions aux limites, le calcul
élément fini à proprement parler et le post-traitement des résultats jusqu’à l’obtention des
raideurs équivalentes.
60
2.1.2 Modélisation de l’endommagement
En marge de l’étude du comportement élastique, on s’est intéressé à la modélisation de

l’endommagement des composites à renforts fibreux considérés. On cherche à déterminer
l’évolution des constantes élastiques effectives de ces matériaux au cours de leur endom-
magement. L’objectif à long terme est de pouvoir rendre compte de manière macrosco-
pique ”homogénéisée” de la détérioration du comportement due localement à la décohésion
fibres/matrice, qui s’avère être le mode de rupture le plus probable pour de tels matériaux.
L’aboutissement d’une telle analyse est censé mener à l’identification d’un critère de rup-
ture et/ou d’une loi d’endommagement pour le matériau homogène équivalent. Les pre-
miers travaux réalisés dans ce contexte consistent à traduire correctement le phénomène
d’initiation et de propagation de fissure à l’interface fibres/matrice des matériaux étudiés.
De nombreuses méthodes sont disponibles dans la littérature pour modéliser
numériquement la fissuration en général. On distingue les approches basées sur la
mécanique de la rupture élastique (facteurs d’intensité de contraintes) et élastoplastique
(intégrale J), la méthode de fermeture virtuelle de fissure (VCCT), qui s’appuient
généralement sur la méthode des éléments finis classique. La méthode des éléments fi-
nis étendus (XFEM) permet également de décrire la propagation d’une fissure dans une
direction arbitraire en s’affranchissant des difficultés liées au remaillage au voisinage du
front de fissure durant son évolution. De manière différente, la méthode des éléments de
cohésion permet de modéliser l’endommagement et la fissuration qui s’ensuit suivant un
chemin prédéfini comme par exemple l’interface entre différents matériaux. On retient cette
solution la plus adaptée à la description de la décohésion entre les fibres et la matrice d’un
matériau composite et par conséquent la plus facile à mettre en œuvre.
Le principe de la méthode revient à mailler spécifiquement l’interface, d’épaisseur très
faible, par des éléments finis de cohésion au comportement élastique endommageable, sui-
vant une loi de type traction-séparation dans les trois directions de l’espace (définissant
ainsi trois modes de décohésion). Une loi de type traction-séparation suppose un comporte-
ment initial élastique linéaire suivi d’un endommagement qui évolue jusqu’à la décohésion
totale. Dans un premier temps, les trois contraintes (ou efforts) normale tn , et de cisaille-
ment ts et tt à l’interface sont reliées aux déformations (ou séparations) conjuguées δn , δs et
δt par une matrice de rigidité élastique. L’initiation de l’endommagement fait référence au
début de la dégradation de la réponse des éléments de cohésion et repose sur la vérification
d’un critère en contraintes ou en déformations, comparant la composante maximale ou une
fonction (quadratique) des différentes composantes à des valeurs seuils. La loi d’évolution
de l’endommagement décrit le comportement des éléments de cohésion une fois le critère
d’initiation satisfait, c’est-à-dire la vitesse à laquelle la rigidité de l’interface diminue. Cette
évolution peut être pilotée en déplacement avec en particulier une séparation maximale
δim ou en énergie avec une valeur limite Gm i pour chaque composante i = n, s, t. Cette
énergie maximale correspond à l’aire sous la courbe de traction-séparation représentée à la
figure 14. Tout comme pour l’initiation, la description de l’évolution de l’endommagement
nécessite la définition d’un critère pour associer les différentes composantes dans le cas
d’un comportement avec plusieurs modes couplés. Outre l’utilisation fréquente d’une loi
puissance, on peut citer le critère de Benzeggagh-Kenane [BEN, 1996], approprié lorsque
les énergies critiques dans les deux directions de cisaillement sont égales.
Le logiciel Abaqus utilisé pour les calculs numériques dispose d’éléments de cohésion et
d’une bibliothèque de lois et critères associés. Il laisse le choix entre deux techniques pour la
modélisation géométrique de la zone cohésive, à savoir l’utilisation d’éléments de cohésion
ou de surfaces de cohésion. La seconde technique est plus facile à mettre en place car elle ne
nécessite pas l’ajout d’éléments spécifiques pour la description du comportement cohésif :
elle repose sur la définition d’interactions entre deux domaines contigus au niveau de la
61
Figure 14 – Représentation schématique d’une loi de comportement de type traction-
séparation
surface de raccord. Ce type de modélisation a déjà été utilisé avec succès pour l’analyse
de la rupture à l’interface fibres/matrice d’un composite carbone/carbone [RAO, 2008] et
d’un composite carbone/époxy [RAG, 2005].
Pour les deux applications de l’interlock et du sandwich Napcor , on adopte parmi les
choix possibles un critère d’initiation de l’endommagement en contrainte quadratique :
2 2 2
htn i ts tt
+ + =1 (34)
t0n t0s t0t
Les contraintes seuils sont plus faciles à identifier que les déformations et le caractère
quadratique permet de prendre en compte l’ensemble des modes de décohésion de manière
équilibrée. On choisit ensuite une loi puissance en termes d’énergie pour décrire l’évolution
de l’endommagement avec une courbe contrainte-déformation linéaire (comme à la figure
14) :
Gn α Gs α Gt α

+ + =1 (35)
Gmn Gm
s Gmt
2.2 Application au composite interlock
Un renfort 3D est constitué de fils continus dans les trois directions, avec en particulier
des fils orientés suivant l’épaisseur qui renforcent les propriétés mécaniques dans cette di-
rection. On ne s’intéresse ici qu’aux interlocks qui sont des renforts tissés, et en particulier
aux interlocks d’angle couche-à-couche qui offrent une variété de configurations possibles
et sont encore assez peu connus relativement aux interlocks orthogonaux. La réticence
des industriels à utiliser des renforts 3D provient justement de cette méconnaissance de
leur comportement mécanique. L’analyse d’un tel comportement nécessite des outils plus
généraux et performants que les méthodes usuelles utilisées avec les stratifiés, par exemple,
du fait de la complexité de leur architecture. Cependant, elle s’avère indispensable pour
pouvoir quantifier l’amélioration des propriétés dans l’épaisseur vis-à-vis d’un renfort 2D,
et évaluer l’éventuelle réduction associée des propriétés planes. Nos travaux effectués dans
ce domaine répondent à ce besoin en proposant une modélisation mécanique de ce type
de matériaux en corrélation avec une analyse expérimentale, incluant une étude des effets
inhérents au processus complet de fabrication du composite (tissage et procédé d’infu-
sion) sur le comportement et des essais de caractérisation mécanique. La définition du
62
modèle repose en particulier sur une représentation fidèle du matériau basée sur des me-
sures précises, que ce soit du point de vue géométrique ou des propriétés des constituants
[P15 ,P19 ,C12 ,C16 ,C18 ].
2.2.1 Modélisation géométrique
De façon à pouvoir traiter un maximum de configurations, le programme Python

développé pour cette application propose une architecture paramétrée. On définit une
cellule unité de taille minimale qui contient seulement deux fils dans les directions aussi
bien de trame que de chaı̂ne, et un nombre de fils quelconque selon l’épaisseur. Les résultats
présentés par la suite concernent un matériau 8-couches avec une géométrie idéalisée dans
laquelle les fils de chaı̂ne sont droits et les fils de trame liants inclinés suivant une fonction
spline cubique, assurant ainsi la continuité géométrique de leur pente et courbure sur toute
la cellule. Les paramètres géométriques de cette cellule sont essentiellement les dimensions
des sections (lenticulaires) des deux types de fils, les espacements horizontaux et verticaux
entre fils ainsi que les angles d’inclinaison. La construction du modèle géométrique est
fastidieuse (d’où l’automatisation) du fait du grand nombre de fils qui nécessite de nom-
breuses partitions de la cellule parallélépipédique. On obtient finalement la cellule unité
de la figure 15 où seuls les fils apparaissent pour une meilleure visualisation.
Figure 15 – Fils de trame et de chaı̂ne formant la cellule unité du composite interlock 3D
2.2.2 Propriétés matérielles
La matrice (en résine époxy) est isotrope alors que les fils sont isotropes transverses
suivant leur axe propre (tout comme les fibres de carbone les constituant). Les propriétés
des fils imprégnés (modules longitudinal et transverse, modules de cisaillement et coeffi-
cients de Poisson) sont déterminées par des lois des mélanges [BER, 1996], en différenciant
les fils de chaı̂ne intacts et les fils de trame usés par le tissage. Le module longitudinal des
fibres est évalué expérimentalement par des essais de traction sur fils imprégnés, alors que
les autres propriétés sont issues des travaux de Cox et Dadkhah [COX, 1995].
63
L’attribution des propriétés aux fils liants est problématique du fait de leur orientation
variable. La solution retenue est de définir un repère local évoluant le long des fils discrétisés
en petits segments (voir figure 15).
2.2.3 Maillage et calcul
Le type d’élément fini choisi pour la discrétisation géométrique de la cellule entière

(que ce soient les fils ou la matrice) est un élément tétraédrique à 10 nœuds (avec une
interpolation quadratique). Des partitions supplémentaires de la géométrie sont nécessaires
de façon à pouvoir affiner le maillage localement (au croisement des deux types de fils) et
faire converger les résultats pour un modèle de taille raisonnable. La figure 16 représente
le maillage finalement retenu pour le matériau considéré, constitué de 380000 éléments soit
1574000 degrés de liberté environ.
Figure 16 – Maillage de la cellule unité du composite interlock 3D
Malgré l’optimisation du maillage, deux types de problèmes apparaissent au cours du

calcul : un temps de calcul rédhibitoire aux alentours de 40 heures pour l’ensemble des
calculs préliminaires (génération des champs de déformation) et une mémoire virtuelle
nécessaire trop importante (30 Go) pour les calculs périodiques. L’utilisation d’un cluster
s’avère donc nécessaire pour augmenter la mémoire virtuelle disponible et, par la même
occasion, réduire les temps de calcul.
2.2.4 Résultats numériques
Le calcul numérique mène à la matrice de rigidité suivante :

 
51386 3113 3.88 −11700 −3362 3.4

 151925 56 −8638 −34406 −13.9 

6492 −3.67 −12.1 −1118 
 
[D] =  (36)

−18.4 

 35751 4858
 
 87346 −27.6 
sym 4333
64
Les termes de la matrice [B] ainsi que les termes A13 , A23 , C13 et C23 sont négligeables
par rapport aux autres, à l’exception de B11 et B22 . Ce faible couplage est majoritairement
dû au déséquilibre de l’architecture en termes de volume de fils entre les deux directions (il
disparaı̂t dans le cas de configurations où les fractions volumiques de fils sont équivalentes
dans les deux directions). On évalue tout de même les modules effectifs en traction et
flexion, sur lesquels les termes de couplage n’ont aucune influence :
E1m = 17303 M P a E1f = 16318 M P a

E2m = 51116 M P a E2f = 39867 M P a
Gm12 = 2189 M P a Gf12 = 1993 M P a (37)
m f
ν12 = 0.0205 ν12 = 0.0556
m = 0.0606 f
ν21 ν21 = 0.135
où 1 représente la direction chaı̂ne et 2 la direction trame.

Ces modules obtenus numériquement peuvent être, pour certains, estimés analytique-
ment moyennant quelques approximations. La présence de fils dans les deux directions du
plan ne permet pas d’appliquer en toute rigueur la loi des mélanges classique. On évalue
tout de même les modules en traction dans chacune des deux directions par cette formule,
en négligeant l’effet des fils transverses. Dans la direction de trame, la forte ondulation
des fils liants longitudinaux nous invite à corriger la formule précédente pour tenir compte
de son effet sur le module homogénéisé. La formule de Krenchel [KRE, 1964] permet de
prendre en compte cette ondulation par l’introduction d’un coefficient correctif dans la loi
des mélanges qui traduit la projection des propriétés dans le plan du composite et leur
intégration le long des fils. Les résultats obtenus avec ces deux formules pour les directions
chaı̂ne et trame, respectivement, sont acceptables dans le cas de l’interlock étudié (l’erreur
maximale vis-à-vis des valeurs numériques est de l’ordre de 10%). Cela provient de la faible
ondulation des fils de trame due à l’éloignement des fils de chaı̂ne entre eux. Lorsque les
fils de chaı̂ne sont plus nombreux et rapprochés (avec des fractions volumiques de fils de
chaı̂ne et de trame similaires), l’ondulation des fils liants est plus prononcée, ce qui met
en défaut la formule de Krenchel. De plus, l’entrelacement des fils, plus marqué, renforce
l’influence des fils transverses sur le module dans la direction longitudinale. De manière
générale, ni la loi des mélanges ni même la formule de Krenchel ne sont appropriées pour
les composites à renforts interlock, car elles ne rendent pas compte de la structure monoli-
thique, générée par l’entrelacement des fils de chaı̂ne et de trame, qui renforce le matériau
composite. Le modèle de Mori-Tanaka est censé pouvoir fournir de meilleures estimations
de l’ensemble des modules de traction pour un tel composite à renforts fibreux. La compa-
raison avec les résultats numériques est cependant bien médiocre, ce qui peut être dû à la
fois aux fractions volumiques trop importantes, à l’ondulation des fils liants ou à la forme
très aplatie des sections lenticulaires. Dans tous les cas, aucune méthode analytique simple
ne permet de calculer les modules relatifs à la flexion, qui diffèrent énormément de ceux
en traction. Ces investigations analytiques menées en parallèle des calculs numériques ne
permettent donc pas véritablement de valider la méthode d’homogénéisation numérique,
mais justifient plutôt le recours à des approches numériques (par éléments finis) à défaut
de solution analytique satisfaisante.
2.2.5 Comparaison 3D/2D
Pour évaluer la qualité des modules effectifs et vérifier ainsi le bien-fondé de la méthode
d’homogénéisation dans le cas d’une structure interlock, on procède à la comparaison de
résultats de calcul provenant, d’une part, d’un modèle 3D hétérogène et, d’autre part,
d’un modèle 2D de plaque homogène avec les modules équivalents de l’équation (37). Les
65
résultats considérés pour la confrontation des deux modèles sont les déplacements obtenus
pour différents cas de chargement (traction, cisaillement ou flexion) sur une plaque formée
d’un ensemble de quatre cellules dans les directions de trame ou de chaı̂ne. Cette taille de
structure est le meilleur compromis qui permette d’ignorer les effets de bord (qui dépendent
de la manière dont le chargement a été appliqué ainsi que des conditions aux limites) sans
toutefois exiger un temps de calcul ou une capacité mémoire rédhibitoires. Des calculs sur
deux cellules voire une seule cellule sont réalisés à des fins de comparaison.
Parmi l’ensemble des résultats obtenus, on ne présente ici que les cas de la traction et
de la flexion simple dans le sens trame des fils liants, à titre d’illustration. En traction, les
déplacements dans la direction du chargement sont tracés le long d’une génératrice pour
une, deux et quatre cellules, et comparés à la solution analytique linéaire de la plaque
homogénéisée caractérisée par le module longitudinal effectif en traction dans le sens trame
(figure 17). On note que les courbes numériques oscillent régulièrement, ce qui est dû à
l’alternance périodique de points à proximité des fils de chaı̂nes et éloignés de ces mêmes
fils lorsqu’on décrit la génératrice choisie pour le tracé des déplacements. Quand on passe
d’une cellule à quatre cellules, l’erreur sur le déplacement en bout de plaque est diminuée
de moitié (elle est de l’ordre de 20% pour quatre cellules). Cela montre l’importance de
considérer des structures de grandes dimensions, afin de garantir la séparation des échelles
microscopique et macroscopique.
Figure 17 – Comparaison numérique 3D/analytique 2D des déplacements en traction

d’une, deux et quatre cellules dans la direction trame d’un composite in-
terlock
En flexion simple, avec deux cellules, la différence entre les résultats numériques 3D et
analytiques 2D n’excède pas 20% (figure 18), contre environ 10% en flexion pure. L’aug-
mentation systématique de l’erreur en flexion simple vis-à-vis de la flexion pure s’explique
par les effets du cisaillement transverse qui ne sont pas représentés dans la solution analy-
tique de plaque de Kirchhoff-Love. Ceux-ci sont tout de même relativement faibles et ne
justifient pas l’utilisation d’un modèle plus complet.
De manière générale, l’ensemble des résultats montrent une bonne cohérence des pro-
priétés élastiques effectives avec les simulations 3D, ou tout au moins une tendance à la
convergence des résultats numériques vers les solutions analytiques lorsque la taille de la
structure augmente.
66
Figure 18 – Comparaison numérique 3D/analytique 2D des déplacements en flexion
simple de deux cellules dans la direction trame d’un composite interlock
2.2.6 Comparaison numérique/expérimentale
Des essais de traction et de flexion 3 points dans les deux directions sur des échantillons
de composite interlock permettent d’une part de déduire des modules ”équivalents”
expérimentaux et de les comparer aux modules effectifs numériques (tableau 2), et d’autre
part de tracer les courbes force-déplacement ou contrainte-déformation mesurées pendant
les essais et de les comparer aux courbes obtenues numériquement (figure 19).
Sens Traction (exp) Traction (num) Flexion (exp) Flexion (num)

Chaı̂ne 12315 M P a 17303 M P a (+30%) 17302 M P a 16318 M P a (−8%)
Trame 64474 M P a 51116 M P a (−20%) 58105 M P a 39867 M P a (−32%)
Tableau 2 – Modules numériques et expérimentaux équivalents d’un composite interlock

en traction et en flexion
Dans la direction des fils droits de chaı̂ne, seuls les modules de flexion sont très proches.
Dans la direction des fils liants de trame, quel que soit le cas de chargement, les modules
numériques sous-estiment assez fortement les valeurs expérimentales. L’origine des écarts
entre résultats numériques et expérimentaux est à rechercher au niveau des hypothèses
effectuées dans la définition du modèle géométrique idéalisé. L’erreur la plus grossière se
traduit par une incohérence entre les fractions volumiques de fibres dans le composite,
mesurées par calcination, les fractions volumiques de fibres dans les fils et les fractions vo-
lumiques de fils dans la cellule unité définies par le modèle géométrique. Cette incohérence
provient de la négligence de certains aspects, comme l’imbrication des fils, due au tis-
sage et accentuée par le procédé d’infusion, qui se manifeste par un certain désalignement
vertical des fils de chaı̂ne d’une même colonne. La prise en compte de cette imbrication
dans la modélisation géométrique permettrait de diminuer le volume global de la cellule
en agençant les fils de manière plus compacte et donc d’augmenter légèrement les modules
numériques. La variation des sections de fil au niveau des entrelacements, qui n’a pas été
envisagée dans notre modèle, peut aussi modifier considérablement la réponse du matériau
[SHE, 2007]. En revanche, la prise en compte de l’endommagement des fils de trame ne joue
67
(a) Traction
(b) Flexion simple
Figure 19 – Comparaison expérimentale/numérique du comportement d’un composite in-

terlock dans les directions trame et chaı̂ne
manifestement pas un rôle prépondérant sur la précision des résultats numériques par rap-
port aux valeurs expérimentales, comme le montrent les courbes obtenues numériquement
avec des modules calculés pour des fils de trame intacts.
2.2.7 Endommagement d’un composite interlock
Les composites à structure 3D, et en particulier les interlocks, possèdent une tenue
améliorée au délaminage, aux chocs et aux impacts. Leur réponse à ces sollicitations est
toutefois complexe du fait de l’entrelacement des fils et des différents modes d’endommage-
ment mis en jeu. Les principaux phénomènes d’endommagement d’un composite interlock
68
d’angle couche-à-couche en carbone/époxy sont la fissuration de la matrice, la rupture
des fils et leur arrachement [COX, 1994]. La complexité des mécanismes de rupture réside
dans leur chronologie et leurs interactions mutuelles. Au vu des observations répertoriées
dans les travaux antérieurs de la littérature, il apparaı̂t que la décohésion fils/matrice dans
un composite interlock est l’un des premiers modes d’endommagement rencontré. Ce sont
principalement les fils liants qui, sous un chargement (de traction ou de compression), se
déforment et engendrent un phénomène de décohésion à l’interface avec la résine, menant
éventuellement à la fissuration de la matrice. Cette décohésion affecte également les fils
droits transverses, à cause de leur entrelacement avec les fils liants.
Après l’étude du comportement élastique, il convient donc de s’intéresser à l’endomma-
gement des composites interlock et en particulier à la décohésion fils/matrice qui participe
grandement à la ruine du matériau. Cette étude s’appuie sur la même cellule unité que
précédemment avec le modèle de décohésion présenté auparavant. Elle requiert cependant
une modélisation géométrique spécifique pour pouvoir disposer des fils (de trame et de
chaı̂ne) et de la matrice séparément et introduire des surfaces cohésives à l’interface entre
les deux. La seconde difficulté réside dans le choix des paramètres de cohésion. Ne disposant
pas de données expérimentales propres aux matériaux considérés, on s’inspire de valeurs
issues de la littérature employées pour des matériaux similaires [RAG, 2005]. La valeur de
l’exposant α dans la loi puissance a été fixée à 2 conformément à ce qui est fréquemment
observé dans la littérature. On suppose le découplage entre les composantes normale et
tangentielles dans le comportement élastique et la raideur K est supposée identique pour
les trois modes. De la même façon, les contraintes et énergies maximales admissibles σmax
et G se voient affecter les mêmes valeurs quel que soit le mode considéré.
L’application de la méthode d’homogénéisation périodique à un instant arbitraire au
cours de l’endommagement du matériau pose des difficultés d’ordre pratique. Cela nécessite
la définition à chaque nouvel instant de nouvelles conditions aux limites s’appuyant sur
la géométrie courante dépendant de l’état d’avancement du processus de décohésion. Afin
d’effectuer en une seule étape le calcul d’endommagement incrémental non-linéaire et donc
de faciliter son déroulement, on fait le choix de déplacements imposés, avec lesquels les
résultats s’apparentent à ce qu’on obtiendrait pour des conditions aux limites périodiques
(tout au moins dans le cas de nos deux matériaux). Le déplacement est imposé suivant
la direction trame et on cherche à tracer la courbe contrainte-déformation macroscopique
du matériau en traction uniaxiale (figure 20). Les trois paramètres de cohésion se voient
attribuer différents jeux de valeurs pour montrer leur influence sur le comportement ho-
mogénéisé.
Dans tous les cas, on observe de manière assez originale un pic de contrainte, cor-
respondant aux décohésions simultanées entre la plupart des fils de trame et la matrice.
On remarque sur la figure 21 les concentrations de contraintes et la décohésion au niveau
des croisements entre fils de trame et fils de chaı̂ne et à l’extrémité des fils de trame. La
croissance régulière des courbes qui s’ensuit indique une certaine stabilisation de l’état
de décohésion. L’entrelacement des fils de trame avec les fils de chaı̂ne semble bloquer le
phénomène de décohésion en empêchant la propagation des fissures initiées dans les zones
décrites précédemment. L’avantage des interlocks vis-à-vis de la tolérance au dommage,
grâce à l’entrelacement des fils, est ainsi confirmé. On note enfin, de façon remarquable,
que la variation de l’énergie de rupture permet de retarder l’ouverture des interfaces sans
modifier la répartition des zones désolidarisées dans l’état stabilisé et donc la fin de la
courbe contrainte-déformation.
69
Figure 20 – Courbes d’endommagement d’un composite interlock en traction uniaxiale
dans la direction trame
(a) Contraintes longitudinales dans les fils liants (b) Déplacements relatifs (décohésion) à l’interface
au pic de contrainte macroscopique entre les fils liants et la matrice
Figure 21 – Endommagement d’un composite interlock en traction uniaxiale dans la di-

rection trame
2.3 Application au sandwich Napcor
Les composites de type sandwich sont largement utilisés dans l’industrie (dans le do-
maine du transport et en particulier dans l’aéronautique), associant légèreté et bonnes
performances mécaniques. Les sandwichs traditionnels ”2D” sont composés d’une âme
(généralement en mousse de polyuréthane, polychlorure de vinyle, balsa ou polystyrène)
qui sépare deux peaux constituées de tissus ou de mats de fibres de verre ou de carbone. La
raideur importante des peaux situées à une certaine distance du plan moyen de la struc-
ture confère aux sandwichs classiques de bonnes propriétés mécaniques en flexion. Mais
leur résistance mécanique est bien moindre en cisaillement transverse, pour lequel c’est
l’âme qui joue un rôle prépondérant, ainsi que l’adhérence âme/peaux [LAS, 2006]. Pour
améliorer la raideur en cisaillement et retarder le délaminage (décohésion âme/peaux),
70
différentes solutions technologiques peuvent être envisagées. On s’est intéressé à la solu-
tion consistant à implanter des renforts dans la direction de l’épaisseur du sandwich, ce
qui n’ajoute pas une masse considérable au produit et augmente fortement les propriétés
dans cette direction (en compression et cisaillement).
Dans ce contexte de sandwich 3D à renforts dans l’épaisseur, la technologie Napcor
est un procédé de fabrication (continu et donc facilement industrialisable) basé sur l’ai-
guilletage transverse pour la réalisation des renforts qui s’avèrent donc solidaires des peaux
du sandwich. Cette technologie se distingue des autres (Z-pinning, tufting, stitching) par
les fortes liaisons mécaniques qu’elle induit entre les renforts et les peaux mais aussi par
la faible augmentation de la masse totale du produit (puisque les fibres des renforts pro-
viennent des peaux) uniquement due au remplissage des piquages par de la résine. On
distingue le cas où les renforts sont implantés de manière orthogonale (comme à la figure
22) et le cas plus général du Napcor X où l’implantation est orientée d’un angle différent
de 90◦ .
(a) Représentation schématique (b) Photographie
Figure 22 – Sandwich Napcor
La modélisation du comportement d’une structure sandwich Napcor en 3D pose des

difficultés pratiques (temps de modélisation et de calcul, capacité en mémoire virtuelle et
stockage) au même titre que pour l’interlock, même si l’architecture du composite est ici
particulièrement simple. On cherche donc à homogénéiser le matériau pour faciliter ensuite
les calculs de structure. Les résultats de l’homogénéisation (à savoir les modules effectifs
pour les différents types de comportement, dans les différentes directions) permettent d’op-
timiser la conception d’un tel sandwich vis-à-vis de certains paramètres géométriques. Dans
le cas du Napcor droit (à renforts orthogonaux), le diamètre des renforts et leur densité
surfacique sont limités par la technologie, mais on peut tout de même s’interroger sur
l’influence relative de ces deux grandeurs sur les propriétés mécaniques. Dans le cas du
Napcor X, l’agencement des renforts inclinés donne lieu à une multitude de configurations
possibles et l’angle d’inclinaison est un paramètre dimensionnant essentiel.
2.3.1 Modélisation géométrique
Les résultats numériques présentés par la suite concerneront pour l’instant la mousse
renforcée et non le sandwich complet avec les peaux. Les renforts sont supposés cylin-
driques à section circulaire. Dans le cas du Napcor droit, on choisit une cellule unité pa-
rallélépipédique avec un arrangement hexagonal (en quinconce) des renforts pour obtenir
un matériau équivalent isotrope dans le plan. Trois densités de renforts sont considérées :
1.2%, 4.3% et 9.4%. La définition d’une cellule unité pour le Napcor X repose sur plu-
sieurs hypothèses, en se limitant ainsi à un type de configuration parmi d’autres : on ne
représente qu’un seul piquage en X par cellule et les renforts de ce piquage sont supposés
ne pas s’interpénétrer. On considère trois angles d’inclinaison différents (45◦ , 60◦ et 75◦
71
par rapport aux peaux) pour les applications numériques avec une fraction volumique de
renforts constante (4.3%). Les deux types de cellule unité utilisés pour modéliser la mousse
renforcée du Napcor droit et du Napcor X sont représentés à la figure 23.
(a) Napcor droit (b) Napcor X
Figure 23 – Cellules unité de la mousse renforcée
2.3.2 Propriétés matérielles
En pratique, on choisit une mousse polyuréthane et des peaux constituées de mats de

fibres de verre. Les renforts sont des composites unidirectionnels constitués de fibres de
verre (issues des peaux) et de résine polyester. Leur comportement homogénéisé (issu d’un
premier calcul d’homogénéisation analytique avec le modèle micro-mécanique de Mori-
Tanaka) peut être considéré comme isotrope transverse. Les propriétés matérielles des
différents constituants du Napcor droit sont résumées dans le tableau 3.
Composant Propriétés
Mousse (PU) E = 12 M P a
ν = 0.1
Renforts (UD) E11 = E22 = 4548 M P a
E33 = 6834 M P a
ν12 = 0.492
ν13 = ν23 = 0.39
G13 = G23 = 1538 M P a
Mat (fibre de verre) E = 9000 M P a
ν = 0.3
Tableau 3 – Propriétés matérielles des constituants du sandwich Napcor droit
Selon l’angle de pénétration, les aiguilles entraı̂nent avec elles plus ou moins de fibres
dans le piquage, si bien que les renforts inclinés n’ont pas les mêmes propriétés mécaniques
que les renforts droits (elles dépendent de l’angle via la fraction volumique de fibres).
72
2.3.3 Résultats numériques
Les résultats présentés ici concernent les cellules unité de mousse renforcée en l’absence
de peaux. À titre d’illustration, la matrice de rigidité effective obtenue numériquement pour
le Napcor droit à 9.4% de renforts est la suivante :
 
204.06 25.79 0.038 0.0683 0.0224 0.0034

 204.35 0.485 0.0224 0.0022 0.53  
89.58 0.0034 0.53 2.678 
 
[D] =  (38)

3463.95 424.87 −4.627 


 
 3463.9 5.184 
sym 1540.88
Dans ce cas, tous les termes de la matrice [B] sont bien négligeables par rapport aux
autres termes de la matrice [D], ce qui prouve bien le découplage entre les comportements
en traction et en flexion de la mousse renforcée, assuré par la symétrie de la structure
suivant l’épaisseur. On en déduit les modules effectifs du tableau 4, respectivement en
traction et en flexion, pour les différentes fractions volumiques considérées.
Densité 1.2% 4.3% 9.4%

E1m = E2m 12.24 M P a 12.89 M P a 14.04 M P a
m = νm
ν12 0.1039 0.1132 0.1265
21
Gm12 5.552 M P a 5.823 M P a 6.288 M P a
E1f = E2f 12.165 M P a 12.805 M P a 13.96 M P a
f f
ν12 = ν21 0.1027 0.1095 0.1266
Gf12 5.533 M P a 5.862 M P a 6.37 M P a
Tableau 4 – Modules effectifs obtenus par homogénéisation périodique de la mousse ren-

forcée du Napcor droit
On vérifie bien que le matériau équivalent est isotrope dans le plan (aussi bien en
traction qu’en flexion). Les modules associés aux deux comportements sont d’ailleurs très
similaires, ce qui provient certainement de la géométrie particulière de la mousse renforcée,
invariante suivant l’épaisseur. Globalement, les propriétés évoluent de façon linéaire en
fonction de la fraction volumique de renforts. Le module d’Young dans le plan de la mousse
augmente, en présence de renforts, de 2%, 7.4% et 17%, respectivement, pour une densité de
renforts de 1.2%, 4.3% et 9.4%. Cette évolution est significative, mais pas déterminante.
L’intérêt des renforts est d’améliorer les performances suivant l’épaisseur et non dans
le plan, d’autant plus que ce sont les peaux qui jouent un rôle prépondérant dans le
comportement en traction et en flexion du sandwich complet. L’un des avantages des
renforts inclinés, vis-à-vis des renforts orthogonaux, est tout de même de cumuler une
amélioration des propriétés planes et hors plan.
La même démarche a été menée sur des cellules unité avec renforts inclinés. Les résultats
obtenus en termes de modules sont présentés dans le tableau 5. La direction 1 se trouve
dans le plan d’inclinaison des renforts.
De manière générale, on note que les modules d’Young et de cisaillement de la mousse
renforcée en traction et en flexion diminuent progressivement au fur et à mesure que l’angle
d’inclinaison des renforts s’accroı̂t. Les coefficients de Poisson, à l’inverse, augmentent
avec l’angle d’inclinaison des renforts. Les autres propriétés sont moins affectées par un
changement d’inclinaison des renforts. On remarque, par ailleurs, que pour des angles
73
θ (◦ ) 45 60 75
E1m 16.08 M P a 13.33 M P a 12.87 M P a
E2m 13.05 M P a 13.03 M P a 13 M P a
m
ν12 0.0742 0.0937 0.1076
m
ν21 0.0603 0.092 0.1087
Gm12 5.89 M P a 5.85 M P a 5.81 M P a
E1f 15.8 M P a 14.01 M P a 13.07 M P a
E2f 12.85 M P a 12.83 M P a 12.81 M P a
f
ν12 0.1 0.105 0.1092
f
ν21 0.081 0.0967 0.107
Gf12 6.18 M P a 6.01 M P a 5.84 M P a
Tableau 5 – Modules effectifs obtenus par homogénéisation périodique de la mousse ren-

forcée du Napcor X pour différentes inclinaisons des renforts
relativement faibles (45◦ et 60◦ ), la mousse renforcée n’est plus isotrope dans le plan. Cela
s’explique naturellement par le fait que tous les renforts sont inclinés invariablement dans
une direction spécifique du plan (la direction 1) et non dans l’autre (la direction 2).
L’apport des renforts inclinés est difficile à quantifier du fait du caractère anisotrope
de la mousse renforcée et de la faible fraction volumique de renforts considérée (4.3%).
Toutefois, dans la direction d’inclinaison des renforts, on peut noter une augmentation
maximale du module d’Young de 34%, par rapport à celui de la mousse seule, pour l’angle
minimal de 45◦ .
Contrairement au cas de l’interlock, la simplicité de la géométrie du sandwich Napcor
rend possible l’évaluation analytique des différents modules effectifs. Dans le cas du Napcor
droit, des calculs ont été effectués sur la mousse renforcée (basés sur la loi des mélanges
dans le sens de l’épaisseur et sur le modèle de Mori-Tanaka dans le plan) puis sur le
sandwich (en utilisant la théorie des stratifiés et les modules équivalents de la mousse
renforcée). Les valeurs obtenues concordent parfaitement avec les résultats numériques, ce
qui valide la méthode d’homogénéisation numérique. Des calculs sont en cours concernant
le Napcor X, s’inspirant entre autres de la démarche de Lascoup et al. [LAS, 2010], basée
sur l’utilisation de moyennes volumiques.
2.3.4 Comparaison 3D/2D
Des calculs par éléments finis d’une plaque hétérogène en 3D ont été réalisés (pour
la plus forte fraction volumique de renforts droits), et les résultats ont été comparés à
des fins de validation aux solutions analytiques de plaque de Kirchhoff-Love impliquant les
modules effectifs obtenus par homogénéisation périodique. Les plaques considérées ont pour
dimensions 10 × 10 cellules unité (figure 24(a)) et les comportements testés sont la traction
pure, la flexion pure et la flexion simple. Les courbes de la figure 24(b) témoignent de la
très bonne concordance des résultats en flexion simple. Les effets du cisaillement transverse
ne se font donc pas du tout sentir sur le comportement de la mousse renforcée en flexion
simple. C’est la présence des peaux dans les sandwichs qui accentue considérablement ces
effets. Leur prise en compte dans la modélisation nécessite un effort d’homogénéisation
supplémentaire, et une discussion sur les différentes méthodes envisagées pour l’obtention
des raideurs effectives en cisaillement transverse est proposée dans les perspectives.
74
(a) Maillage et déformée (b) Comparaison des flèches analytiques et
numériques
Figure 24 – Validation numérique en flexion simple de l’homogénéisation périodique de

la mousse renforcée du sandwich Napcor
2.3.5 Endommagement du sandwich Napcor
La plus grande faiblesse des structures sandwich classiques réside dans l’interface entre
les peaux et l’âme, quelle que soit la sollicitation mécanique appliquée [LAS, 2006]. Les
performances de tels sandwichs sont donc limitées par le phénomène de décohésion (ou
délaminage) inévitable, qui s’initie dans les zones sèches dépourvues de résine. L’avantage
des sandwichs 3D est de contribuer à la cohésion âme/peaux par la présence de renforts.
C’est particulièrement le cas avec le sandwich Napcor puisque les renforts sont solide-
ment enchevêtrés dans les peaux. En revanche, les interfaces âme/renforts constituent de
nouveaux lieux probables de décohésion (figure 25).
Figure 25 – Décohésion le long d’un renfort du sandwich Napcor en flexion 3 points
Peu de travaux dans la littérature concernent l’étude de l’endommagement d’un sand-

wich à renforts dans l’épaisseur. Dans le prolongement de l’homogénéisation du compor-
tement élastique, on s’intéresse à la décohésion entre les renforts et l’âme du sandwich
Napcor droit. Les premiers résultats ont été obtenus en 2D (en contraintes planes) en ne
considérant encore une fois que la mousse renforcée. L’utilisation d’un modèle 2D permet
75
de mailler très finement le voisinage des interfaces, qui joue un rôle prépondérant dans la
précision des résultats, sans parvenir pour autant à des temps de calcul rédhibitoires.
Comme pour l’interlock, on impose des déplacements sur les différentes faces suivant
une ou plusieurs directions, afin de définir des états de chargement de type traction uni-
axiale, traction biaxiale ou cisaillement. On charge la structure de manière incrémentale
jusqu’au maximum de décohésion (état stabilisé) et on trace les courbes contrainte-
déformation appropriées.
On retient les mêmes hypothèses concernant le modèle de décohésion que pour le cas de
l’interlock. Peu d’informations sont fournies dans la littérature concernant les propriétés
de cohésion d’une interface entre ce type de renfort et ce type de mousse. La détermination
expérimentale des trois paramètres matériels du modèle est difficile, d’autant plus que le
phénomène de décohésion consiste en fait en une fissuration de la mousse à proximité des
renforts. Le tableau 6 liste les paramètres finalement utilisés dont la valeur, ou tout au
moins l’ordre de grandeur, a été définie en recoupant des données de la littérature concer-
nant des matériaux similaires. La valeur de l’énergie de rupture correspond en particulier
à celle relevée pour la fissuration d’une mousse PU de la même densité [MCI, 1979].
Paramètres Valeurs
t0n = t0s = t0t 1 MPa
Gn = Gm
m
s = Gt
m 88.9 J.m−2
Knn = Kss = Ktt 109 N.mm−3
Tableau 6 – Paramètres de décohésion pour l’interface âme/renforts du sandwich Napcor
À titre d’illustration, on présente ici les résultats obtenus en traction uniaxiale pour
la fraction volumique maximale (9.4%). La figure 26 représente les déplacements et les
contraintes suivant la direction du chargement lorsque le matériau a atteint son état sta-
bilisé (certaines zones de contact subsistent entre les renforts et l’âme).
(a) Déplacements dans la direction de chargement (b) Contraintes dans la direction de chargement
Figure 26 – Décohésion de la mousse renforcée du sandwich Napcor en traction uniaxiale

(état stabilisé)
76
La courbe contrainte-déformation macroscopique (figure 27) prend une allure similaire
à celle obtenue par Raghavan et Ghosh [RAG, 2005] et ne fait pas apparaı̂tre de pic de
contrainte, sans doute du fait d’une décohésion plus progressive. À titre de comparai-
son, on trace sur le même graphique les courbes de comportement de la même cellule
(chargée) sans tenir compte de la décohésion et de la cellule vide (sans renforts) soumises
aux mêmes sollicitations. On distingue trois étapes dans le comportement du matériau
homogène équivalent :
• une partie linéaire qui correspond au comportement élastique de la cellule unité
intègre, avant l’initiation de l’endommagement : cette portion de la courbe coı̈ncide
parfaitement avec la courbe de référence de la cellule chargée ;
• une partie non-linéaire qui correspond à l’adoucissement du comportement de la zone
cohésive et à la progression de la fissuration entre l’âme et les renforts ;
• une dernière partie dans laquelle le comportement de la cellule unité redevient linéaire
(la phase d’endommagement est terminée), mais avec un module élastique équivalent
(pente de la courbe) inférieur au module initial, du fait de la décohésion partielle (la
décohésion est d’ailleurs presque totale au point que la portion finale de la courbe
est très proche de la courbe de référence de la cellule vide (sans renforts)).
Figure 27 – Courbe de décohésion du sandwich Napcor en traction uniaxiale
Le même type de comportement macroscopique est obtenu pour une sollicitation en

traction équi-biaxiale. Toutefois, par symétrie de révolution, la décohésion survient simul-
tanément sur l’ensemble de l’interface âme/renforts. Cela se traduit par un palier dans la
courbe contrainte-déformation entre les deux portions linéaires.
Jusqu’à récemment, l’ensemble de mes activités de recherche concernaient principale-

ment des matériaux homogènes (isotropes), que ce soit dans les applications de striction,
de rochet, de flambement élastique ou plastique. Ces mêmes activités s’orientent actuel-
lement vers les matériaux composites, qui nécessitent l’utilisation conjointe de méthodes
d’homogénéisation. Avant d’entreprendre l’identification de comportements non-linéaires
par l’homogénéisation de matériaux élastoplastiques et/ou en grandes transformations, les
premiers travaux effectués dans ce cadre ont permis de se familiariser avec quelques ap-
77
proches analytiques et surtout les méthodes numériques associées à l’homogénéisation des
matériaux composites en élasticité linéaire. Deux types de matériaux périodiques à ren-
forts 3D ont été considérés : les composites à structure interlock et les sandwichs renforcés
dans l’épaisseur. Les études réalisées ont porté sur l’influence du procédé de fabrication et
sur la prise en compte précise des caractéristiques géométriques et matérielles du compo-
site dans l’identification du comportement macroscopique équivalent. D’un point de vue
numérique, elles ont été l’occasion de développer des outils de calcul d’homogénéisation
par éléments finis pour des structures périodiques, suffisamment généraux pour s’appli-
quer à divers matériaux et représenter précisément la géométrie des renforts. Un effort a
été particulièrement apporté sur l’automatisation de la modélisation géométrique et du
calcul. Concernant les interlocks, les résultats sont encourageants. On relève encore cer-
tains écarts significatifs entre les modules effectifs homogénéisés et les résultats numériques
3D ou expérimentaux, mais qui s’expliquent soit par la taille limitée des modèles utilisés
pour la comparaison, ou par certaines hypothèses trop restrictives (comme l’absence de
prise en compte de l’imbrication des fils). Le modèle numérique est suffisamment souple
pour prendre en compte dans une étude future de nouveaux paramètres de modélisation
et améliorer la concordance des résultats. Concernant le sandwich Napcor , les premiers
résultats numériques sont bien meilleurs (en accord avec les solutions analytiques) bien
que la phase de validation expérimentale n’ait pas encore été finalisée. La simplicité de la
géométrie dans ce cas facilite les développements et ouvre la voie à de nombreuses pers-
pectives. D’un point de vue analytique, l’évaluation des modules effectifs de la mousse
renforcée du Napcor X est en cours. À partir de là, l’ensemble des modules de traction
et de flexion des deux types de sandwich, à renforts droits ou inclinés, peuvent être es-
timés analytiquement en combinant les schémas micro-mécaniques utilisés à la théorie des
stratifiés pour prendre en compte l’effet des peaux. D’un point de vue numérique, tous
les calculs pour l’instant effectués sur la mousse renforcée, peuvent être reproduits sur le
sandwich complet, en incluant les peaux dans le modèle.
Une attention toute particulière doit être portée sur la prise en compte des effets
du cisaillement transverse. Contrairement au cas de l’interlock et de la mousse renforcée
seule, le sandwich met en évidence des écarts considérables entre les réponses numérique
(avec un modèle 3D hétérogène) et analytique (basée sur la théorie de Kirchhoff-Love avec
les modules effectifs calculés par homogénéisation périodique) d’un problème de flexion
simple. Plusieurs solutions sont envisageables pour la détermination des raideurs en ci-
saillement transverse. La première idée serait de poursuivre la résolution du problème
d’homogénéisation asymptotique en cherchant la solution du problème microscopique au
deuxième ordre. Cette technique présente l’inconvénient de demander énormément de
développements numériques supplémentaires et nécessite l’utilisation d’un programme
spécifique. En outre, si la solution au second ordre constitue une meilleure approxima-
tion de la solution du problème 3D hétérogène, elle ne s’identifie pas à un modèle de
plaque de Mindlin-Reissner, qui fait apparaı̂tre naturellement les raideurs en cisaillement
transverse. En vue de déterminer ces raideurs, Cecchi et Sab [CEC, 2007] ont proposé une
méthode d’homogénéisation conduisant à l’obtention d’une plaque homogène équivalente
de Mindlin-Reissner, qui constitue une extension aux plaques périodiques orthotropes de la
méthode de Reissner pour les plaques homogènes isotropes. L’évaluation des constantes de
cisaillement transverse requiert la résolution d’un nouveau problème 3D sur la cellule unité
pour lequel le chargement volumique imposé dépend des contraintes issues du problème
3D résolu précédemment pour l’obtention du modèle de plaque de Kirchhoff-Love. Cette
méthode pose malheureusement quelques difficultés informatiques et remet en cause l’uti-
lisation d’un code commercial comme Abaqus. D’autres méthodes ont été développées,
parmi lesquelles on peut citer les travaux de Lebée [LEB, 2010], qui s’appuient sur la
définition d’une théorie de plaque enrichie (basée sur le modèle de gradient du moment).
78
Dans le cas de structures 3D exigeantes (comme des sandwichs à âme pliable en chevrons),
pour lesquelles le modèle de Mindlin-Reissner ne donne pas satisfaction, la nouvelle théorie
permet de construire une plaque homogène équivalente en meilleur accord avec le compor-
tement de la structure hétérogène. Dans notre cas, on recherche plutôt une solution simple,
s’appliquant éventuellement aux seules structures sandwich, permettant d’obtenir les rai-
deurs en cisaillement transverse par un calcul élément fini standard sur une cellule unité, et
ainsi de construire le modèle de plaque de Mindlin-Reissner équivalent. La méthode la plus
naturelle pour l’obtention de ces raideurs, et qui servira de référence, consiste à effectuer
un calcul de flexion simple sur une plaque complète formée d’un ensemble suffisamment
important de cellules unité. À la flèche mesurée, on peut déduire la flèche obtenue ana-
lytiquement par le modèle de Kirchhoff-Love. La différence de ces deux flèches représente
la flèche due uniquement à l’effort tranchant. Celle-ci dépend linéairement de la variable
d’espace dans la direction de chargement, si bien qu’on peut déduire la raideur dans cette
direction par simple régression linéaire.
Parmi les travaux déjà réalisés, on s’est intéressé à l’endommagement des matériaux
considérés et particulièrement à la décohésion entre renforts et matrice. Ce phénomène
fait partie intégrante du processus global qui engendre la rupture de ces matériaux. La
principale difficulté rencontrée lors de la modélisation de la fissuration aux interfaces entre
renforts et matrice est l’identification des propriétés mécaniques de ces interfaces. Peu
de travaux dans la littérature portent sur cette question. L’obtention de ces valeurs re-
quiert la mise en place d’essais de caractérisation (pull-out ou push-out) qui s’avèrent
délicats à mettre en œuvre et qui nécessitent la réalisation d’un matériau modèle. Les pre-
miers calculs effectués sont uniquement comparés de manière qualitative à des résultats de
référence de la littérature. Une validation expérimentale s’avère nécessaire pour valider la
modélisation retenue. Concernant l’interlock, les porosités aux croisements des fils doivent
nécessairement accentuer le phénomène de décohésion dans ces zones déjà fortement sol-
licitées et leur prise en compte dans la modélisation semble inévitable pour une bonne
corrélation numérique/expérimentale. Concernant le Napcor , le passage à un modèle 3D
est obligatoire pour les renforts inclinés et/ou pour des sollicitations de flexion. L’objectif à
long terme est de caractériser un critère d’endommagement macroscopique, puis de décrire
finalement l’évolution de cet endommagement pour des trajets de chargement quelconques
jusqu’à la ruine.
Dans le seul cas du sandwich Napcor , on s’est également intéressé au flambement
en compression dans l’épaisseur. Les premiers résultats analytiques constituent l’une des
applications des méthodes de bifurcation présentées dans la partie suivante.
79
3 Instabilités géométriques : flambement et post-
flambement élastique/plastique des structures minces
ou élancées
Le phénomène d’instabilité géométrique, élastique ou plastique, est l’une des princi-

pales causes de ruine pour une pièce soumise à des sollicitations mécaniques. Le terme
générique de flambement est couramment utilisé pour désigner l’occurence d’un point sin-
gulier dans l’évolution de l’équilibre quasi-statique d’une structure par perte de stabilité
géométrique : il concerne spécifiquement les structures minces ou élancées soumises de
façon prédominante à des efforts de compression. On peut ensuite distinguer deux types
de singularités : les points de bifurcation (ou multifurcation) caractérisés par la perte d’uni-
cité de la solution (on dit alors qu’il y a flambement par bifurcation) et les points limites
pour lesquels l’instabilité s’exprime par la perte de monotonie de la courbe d’équilibre
force-déplacement (on parle alors de claquage).
Mes travaux dans ce domaine ont débuté lors de ma thèse [M2 ]. L’objectif était de
mettre en œuvre des outils de modélisation permettant de rendre compte de manière
efficace et précise du flambement (et éventuellement du post-flambement) de structures
diverses. En effet, si de nombreux codes de calcul commerciaux proposent déjà des outils
numériques pour l’analyse du flambement, il s’agit essentiellement d’approches linéarisées
dont la validité n’est pas toujours vérifiée (surtout en plasticité). Concernant le post-
flambement, la plupart des logiciels de calcul des structures par éléments finis prétendent
s’y intéresser alors qu’en pratique, un recours aux imperfections (souvent hasardeux) s’im-
pose pour pouvoir dégénérer la solution et ainsi déterminer la réponse post-critique (dans le
cas d’un problème de bifurcation). Dans tous les cas, lorsque les points de bifurcation ou les
points limites sont fortement singuliers (par exemple lors d’une bifurcation sub-critique ou
d’un claquage abrupt), un certain nombre de programmes montrent encore actuellement
des difficultés de convergence et ne fournissent pas de solution complète pour certains
problèmes réputés comme étant critiques. L’objectif du travail initié durant ma thèse est
donc de fournir une solution numérique répondant favorablement à tous les problèmes
soulignés ci-dessus. La réalisation complète d’un code de calcul par éléments finis a permis
de maı̂triser les aspects à la fois mécaniques et purement numériques, à savoir :
• la prise en compte des non-linéarités matérielles et géométriques (plasticité en
grandes transformations avec une formulation lagrangienne totale) ;
• l’évaluation des points critiques sans linéarisation préalable et l’obtention des courbes
d’équilibre post-critiques sans utiliser de défauts d’aucune sorte ;
• la construction d’un élément fini robuste auquel s’ajoutent des méthodes numériques
performantes pour faire face aux très fortes non-linéarités rencontrées.
En parallèle de ces développements numériques, une méthode de résolution analytique
d’un problème de bifurcation (plastique) a été mise au point. Le travail a consisté en
l’exploitation des résultats théoriques fondamentaux existants concernant la bifurcation
plastique et la formalisation d’une méthodologie de résolution générale, pour des applica-
tions variées. L’intérêt de cette méthode, face à l’approche numérique précédente, est de
fournir une solution rapide et explicite, quoiqu’incomplète (seuls le flambement et le post-
flambement immédiat sont obtenus) et limitée à des géométries et chargements simples.
La poursuite de ces travaux depuis ma thèse jusqu’à ce jour a consisté à améliorer et
compléter les outils initialement développés afin d’enrichir les premiers résultats obtenus
et d’étendre les champs d’application à d’autres contextes (autres géométries et charge-
ments, prise en compte des contraintes résiduelles, ...) et d’autres matériaux (composites,
en particulier).
80
Dans cette section, on présente logiquement les principaux développements théoriques
et numériques effectués durant ma thèse puis au cours des années qui ont suivi. Un certain
nombre d’applications permettent d’illustrer chacune des deux approches sur un large
éventail de géométries, chargements et matériaux, et de montrer ainsi la généralité et la
robustesse des méthodes mises en œuvre.
3.1 Méthode générale de résolution analytique d’un problème de flam-

bement élastique/plastique
La motivation première de cette partie de mes travaux concerne le flambement plas-

tique, pour lequel on dispose encore de peu de solutions analytiques, contrairement au cas
du flambement élastique. D’un point de vue théorique, le flambement plastique (au sens
où l’instabilité survient après le début de plastification) présente des difficultés spécifiques
de par l’existence de continua de points de bifurcation [CIM, 1987] et l’indépendance
des notions de stabilité et de bifurcation (démontrée avec l’utilisation du concept de
”solide élastique de comparaison” [HIL, 1958]). Le premier résultat significatif est dû à
Shanley [SHA, 1947] et porte sur une colonne rigide à deux degrés de liberté (ressorts
élastoplastiques). Pour ce système discret comme pour un système continu, il montre que
la première bifurcation plastique survient à la valeur critique du module tangent et que
la post-bifurcation (à charge croissante) s’effectue avec décharge naissante. Pour l’ana-
lyse du post-flambement, Hutchinson [HUT, 1974] étend ensuite les résultats de Koiter en
élasticité en introduisant des développements en séries fractionnaires. Plus récemment, El
Koulani et Léger [KOU, 1997] résolvent le problème de bifurcation plastique de manière
plus globale et montrent ainsi l’existence de la branche post-critique, avec la validité et la
convergence des développements précédemment introduits.
La plupart des résultats de bifurcation plastique dans la littérature concernent la charge
critique du module tangent et le développement asymptotique de la charge λ en fonction
du paramètre de perturbation ξ s’écrit alors de manière générale :
λ = λT + λ1 ξ + λ2 ξ 1+β + ... (39)
avec λ1 positif (bifurcation à charge croissante), λ2 négatif (présupposant un point limite

sur la branche bifurquée, à savoir une charge maximale admissible), et l’exposant β compris
entre 0 et 1.
Notre travail a tout d’abord consisté à appliquer de manière spécifique la théorie de la
bifurcation plastique (définie initialement dans un cadre 3D) et les méthodes associées de
développement asymptotique aux cas de structures 1D et 2D [C3 ]. L’hypothèse d’états de
contraintes particuliers permet d’expliciter les modules des lois de comportement apparais-
sant entre autres dans l’équation de bifurcation. De cette façon, on dispose d’une démarche
simple, efficace, et applicable à un grand nombre de problèmes, qui permet de retrouver
certains résultats analytiques (charge critique λT et mode de bifurcation) de manière ori-
ginale et systématique, et de les enrichir par de nouveaux termes dans les développements
asymptotiques, comme la pente de la courbe d’équilibre post-critique au point de bifurca-
tion λ1 , représentative de la résistance mécanique de la structure dans son comportement
en post-flambement.
3.1.1 Théorie de la bifurcation plastique
L’ensemble des développements analytiques ont été effectués dans le cadre d’une for-
mulation lagrangienne totale, propice à l’écriture de l’équation de bifurcation et à la mise
en valeur des termes non-linéaires responsables du phénomène de flambement. Compte
81
tenu des faibles déformations pré-critiques, le choix de la décomposition cinématique en
plasticité n’est pas déterminant et on retient une décomposition additive, similaire au
cas des petites déformations. Le matériau est isotrope et on suppose un comportement
élastique linéaire régi par la loi de Saint-Venant-Kirchhoff. On se place dans le contexte
des matériaux standard généralisés [HAL, 1975] avec le choix du critère de von Mises et un
écrouissage isotrope linéaire. Le choix de ces différentes hypothèses nous permet d’abou-
tir de manière générale à des expressions simplifiées servant à la résolution du problème
de flambement, mais la même démarche peut être envisagée dans d’autres conditions,
sous réserve de la difficulté des développements mathématiques. En particulier, le choix
en plasticité de la théorie incrémentale s’impose, dans la mesure où la décharge élastique
joue un rôle prépondérant dans les problèmes de bifurcation plastique, malgré la meilleure
adéquation de la théorie de la déformation avec les résultats expérimentaux, observée para-
doxalement dans de nombreuses applications. Cependant, les résultats ci-dessous peuvent
tous être obtenus de manière similaire dans le cas de la théorie de la déformation, en
remplaçant globalement les modules tangents par les modules sécants équivalents.
Dans l’hypothèse d’une bifurcation à la valeur critique du module tangent (qui constitue
le minimum de chaque continuum et donc la véritable charge de flambement en pratique),
on peut développer le chargement et le déplacement à l’ordre 1 en fonction du paramètre
de perturbation, de la manière suivante :
(
λ = λT + λ1 ξ + o(ξ)
(40)
U = Uf (λ) + ξX + o(ξ)
La charge critique λT et le mode de flambement X s’obtiennent par la résolution de

l’équation de bifurcation [NGU, 2000] :
Z
∀ δU, ∇T δU : K (Uf (λT )) : ∇XdΩo = 0 (41)
Ωo
qui fait intervenir le tenseur tangent nominal K élastique ou plastique selon le comporte-
ment considéré.
Sous les hypothèses supplémentaires que λT soit valeur propre simple de (41), d’un
coefficient de transversalité positif strictement et d’une charge plastique active au point
critique, la bifurcation est dite angulaire et la pente de la branche bifurquée au point
critique est donnée par :

M (X, λT ) : ∇X
λ1 = max − (42)
X∈Ωpf (tc ) M (X, λT ) : ∇Uf ,λ (X, λT )
qui fait intervenir le tenseur M défini dans le paragraphe suivant.
3.1.2 Hypothèses sur les états de contraintes
L’écriture explicite de l’équation de bifurcation nécessite la connaissance des modules

nominaux. À partir des lois d’état et d’évolution, on aboutit à l’expression du tenseur
tangent élastoplastique matériel :
∂Σ
Dp = =D−N⊗N (43)
∂E
avec :
∂f
D : ∂Σ
N= q (44)
∂f ∂f
H + ∂Σ :D: ∂Σ
82
puis à celle du tenseur tangent élastoplastique nominal :
∂Π ∂Σ T
Kp = = F. .F + (I.Σ)T = Ke − MT ⊗ M (45)
∂F ∂E
avec :
Ke = F.D.FT + (I.Σ)T M = N.FT (46)
En supposant un comportement pré-critique avec un état de contrainte biaxial :

 
−λ 0 0
Π = −λeX ⊗ eX − ηλeY ⊗ eY =  0 −ηλ 0  (λ > 0, η ∈ R) (47)
 
0 0 0
et des déformations pré-critiques faibles (ce qui est fréquemment vérifié en pratique), les
tenseurs matériels et nominaux s’identifient en partie et on a :
∂Σ
Kp ≈ ∂E + (I.Σ)T = Dp − λei ⊗ eX ⊗ eX ⊗ ei − ηλei ⊗ eY ⊗ eY ⊗ ei
µ
M≈N = √ [I − 3eX ⊗ eX + η (I − 3eY ⊗ eY )]
(H+3µ)(1−η+η 2 )
  (48)
−2 + η 0 0
µ
=√ 0 1 − 2η 0 
 
(H+3µ)(1−η+η 2 )

0 0 1+η
Les structures considérées par la suite sont soit 1D soit 2D. Dans les deux cas, des
hypothèses complémentaires sur la nature des contraintes permettent de remplacer le ten-
seur 3D Dp par un tenseur réduit C p . Dans le cas 1D, les contraintes normales dans les
directions transversales sont supposées nulles, ce qui donne :
p p p p p p p p p p
p p DijY Y (DY Y ZZ DZZkl −DZZZZ DY Y kl )+DijZZ (DZZY Y DY Y kl −DY Y Y Y DZZkl )
Cijkl = Dijkl + p p p p
DY Y Y Y DZZZZ −DY Y ZZ DZZY Y
(49)
(i, j) 6= (Y, Y ), (Z, Z) (k, l) 6= (Y, Y ), (Z, Z)
dont on retient les seules valeurs utiles :

p p p p
CXXXX = ET CXY XY = CXZXZ = CY ZY Z = µ (50)
Dans le cas 2D, l’état de contrainte plan induit une autre expression du tenseur réduit :
p p
p p DijZZ DZZkl
Cijkl = Dijkl − p (i, j) 6= (Z, Z) (k, l) 6= (Z, Z) (51)
DZZZZ
dont on retient les trois modules réduits suivants :
ET
p 1+4η(η−1)+3
α = CXXXX =E E
ET
(5−4ν)(1+η 2 )−2(4−5ν)η−(1−2ν)[(1−2ν)(1+η 2 )−2(2−ν)η] E
ET
p 2−5η+2η 2 −[2(1−2ν)(1+η 2 )−(5−4ν)η ] E
β = CXXY Y =E ET (52)
(5−4ν)(1+η 2 )−2(4−5ν)η−(1−2ν)[(1−2ν)(1+η 2 )−2(2−ν)η] E
ET
(2−η)2 +3η 2
γ = CYp Y Y Y = E E
ET
(5−4ν)(1+η 2 )−2(4−5ν)η−(1−2ν)[(1−2ν)(1+η 2 )−2(2−ν)η] E
Les résultats précédents sont suffisamment généraux pour s’appliquer à différentes

configurations de géométrie et de chargement. La structure étudiée peut être élancée (1D)
83
ou mince (2D), soumise à un chargement de compression prédominante uniaxiale (avec
η = 0) ou biaxiale. Dans les applications considérées, la structure est totalement plas-
tifiée de manière homogène dans son comportement pré-critique au point de bifurcation,
ou alors totalement élastique (les expressions précédentes sont toujours valables en rem-
plaçant simplement les modules élastoplastiques par leurs équivalents élastiques). Dans le
cas élastoplastique, les deux relations essentielles sont les suivantes :
Z
∀ δU, ∇T δU : (C p − λei ⊗ eX ⊗ eX ⊗ ei − ηλei ⊗ eY ⊗ eY ⊗ ei ) : ∇XdΩo = 0 (53)
Ωo
!
3(∇X)XX − divX + η (3(∇X)Y Y − divX)
λ1 = max − (54)
3 ∇Uf ,λ XX − div Uf ,Tλ + η 3 ∇Uf ,Tλ Y Y − div Uf ,Tλ
T

X∈Ωo
Les principales applications effectuées à ce jour de la méthodologie présentée ci-dessus

sont rapportées dans les paragraphes suivants, en mettant l’accent à chaque fois sur
quelques résultats originaux, sans rentrer dans le détail des développements permettant de
les obtenir, et sur la qualité de ces résultats en les comparant à des solutions numériques
issues entre autres du programme élément fini développé en parallèle, qui fera l’objet de
la seconde partie de ces travaux sur le flambement.
3.1.3 Poutre de Timoshenko
La poutre droite comprimée constitue un problème de référence dans le domaine du

flambement, autant en élasticité qu’en plasticité. Si la première valeur de charge critique en
élasticité a été obtenue par Euler au début du 18ème siècle, les premiers travaux sur le flam-
bement plastique d’une poutre datent de la fin du 19ème siècle avec Engesser, Considère,
et plus tard von Karman. La charge critique d’une poutre de Bernoulli est désormais bien
connue, même en plasticité. La pente de la branche bifurquée est donnée, entre autres, par
Nguyen [NGU, 2000]. Le cas d’une poutre de Timoshenko, incluant les effets du cisaille-
ment transverse, a déjà été exploité en élasticité, mais très peu encore en plasticité.
Cette première application naturelle de notre démarche a pour objectif de retrouver de
manière systématique les résultats établis pour les poutres de Bernoulli, mais surtout de
fournir des solutions originales dans le cas d’une poutre de Timoshenko [P13 ]. On considère
donc une poutre de Timoshenko encastrée en X = 0 et comprimée par une charge λS en
X = L (les résultats sont bien entendu généralisables à d’autres conditions aux limites).
La cinématique associée, retenue aussi bien pour le mode de flambement X que pour le
déplacement virtuel δU, permet d’expliciter l’équation de bifurcation et d’aboutir, après
intégration dans la section et intégration par parties, au système d’équations différentielles
suivant :
ET SU,XX = 0
kµS (Θ,X −V,XX ) + λc SV,XX = 0 (55)
ET IΘ,XX +kµS (V,X −Θ) = 0
avec les conditions aux limites sthéniques en X = L : U,X (L) = 0, kµS (Θ(L) − V,X (L))+
λc SV,X (L) = 0 et Θ,X (L) = 0. En y ajoutant les conditions aux limites cinématiques
U(0) = V(0) = Θ(0) = 0, on obtient la valeur critique :
π 2 ET I
4L2
λT S = π 2 ET I
(56)
1 + 4kµSL 2
84
et le mode de bifurcation associé :

 U=0

π 2 ET I
V = 2e 1 + 4kµSL 1 − cos πX

2 2L (57)

πe
sin πX

Θ = 4L 2L
Compte tenu de la solution fondamentale du problème de la poutre comprimée

entièrement plastifiée, la pente de la branche bifurquée vaut :
3π 2 ET I π 2 ET I

λ1 S = = 3 1 + λT S (58)
4L2 4kµSL2
En faisant tendre k vers l’infini, les trois résultats précédents se simplifient pour don-
ner les quantités bien connues dans le cas d’une poutre de Bernoulli. Les pentes obtenues
pour les deux théories sont littéralement identiques, alors que les valeurs critiques diffèrent
légèrement. À des fins de validation, une géométrie de poutre à section rectangulaire creuse
a été définie spécialement pour mettre en valeur numériquement la diminution de la valeur
critique de flambement plastique due aux effets du cisaillement transverse. Les valeurs ana-
lytiques obtenues par les deux théories sont comparées aux solutions numériques obtenues
par éléments finis dans le tableau 7.
Épaisseur λT S (N ) λT S (N ) λT S (N ) Erreur (%)

(mm) (Bernoulli) (Timoshenko) (EF) (Timoshenko vs. EF)
2 3.823 106 3.596 106 3.596 106 0.0005
3 5.735 106 5.395 106 5.444 106 0.9
5 9.558 106 8.991 106 9.11 106 1.3
Tableau 7 – Charges critiques de flambement plastique d’une poutre à section rectangu-

laire creuse comprimée
3.1.4 Plaque sous compression/traction biaxiale
Le cas d’une plaque sous chargement uni- ou bi-axial constitue également une appli-
cation courante dans le domaine du flambement, de par la simplicité de la géométrie.
Durban et Zuckerman [DUR, 1999], entre autres, ont établi un certain nombre de so-
lutions semi-analytiques pour les charges critiques de flambement plastique de plaques
rectangulaires sous diverses conditions aux limites. Notre démarche permet de retrouver
systématiquement l’ensemble de ces résultats ainsi que les pentes initiales des branches
bifurquées, en respectant la cinématique des plaques de Kirchhoff-Love [P9 ].
Sans rentrer dans les détails des calculs, l’équation de bifurcation peut se ramener à
une seule équation aux dérivées partielles mettant en jeu la seule composante hors plan
du mode de bifurcation :
e2 [αW,XXXX +γW,Y Y Y Y +(2β + 4µ)W,XXY Y ] + 12λc (W,XX +ηW,Y Y ) = 0 (59)
Dans le cas de bords simplement appuyés, l’expression analytique de la contrainte

critique est la suivante :
p4 q4 p q 2 2
π 2 e2 α a4 + γ b4 + (2β + 4µ) a2 b2
λT = p2 q2
(60)
12 2 + η 2
a b
85
où les nombres de demi-ondes p et q dans les deux directions du mode de flambement sont
choisis de façon à minimiser cette expression et vérifient par conséquent :
4 2 2 4
α ap4 + 2αη ap2 qb2 + ((2β + 4µ)η − γ) qb4 = 0
4 2 2 4 (61)
(2β + 4µ − αη) ap4 + 2γ ap2 qb2 + γη qb4 = 0
Dans le cas particulier d’une plaque carrée (a = b), avec η ∈ [−0.1, 1], le minimum est
obtenu pour p = q = 1. On montre alors aisément que l’ajout d’une faible contrainte de
traction dans la direction perpendiculaire a un effet stabilisateur vis-à-vis du problème de
flambement sous compression uniaxiale. À l’inverse, une contrainte de compression diminue
la valeur critique jusqu’à la valeur limite du cas équi-biaxial.
Dans le cas uniaxial, pour une plaque rectangulaire quelconque, les nombres p et q
correspondant à la valeur critique minimale sont :
!
q r
γ
p = Int 4 α ab = Int 4 4
E
a
1+3 ET b (62)
q=1
et la contrainte critique devient :
 h i 2 2 
ET ET pb a
π 2 e2  2 4 1 − (1 − 2ν) E + 1 + 3 E a + 4 pb
λT = E + (63)

12b2 1 + ν
 2 ET 
5 − 4ν − (1 − 2ν) E
ce qui permet de retrouver, dans le cas particulier d’une plaque carrée, un résultat similaire
à celui obtenu par Cimetière [CIM, 1987] de façon différente.
Lorsqu’un des bords (en X = 0) est encastré, les autres restant simplement appuyés, on
écrit le mode de manière plus générale, de façon à vérifier automatiquement les conditions
aux limites sur les bords Y = 0 et Y = b :
qπY
W = F(X) sin (q ∈ N∗ ) (64)
b
ce qui mène à l’équation différentielle en F :
qπ 2 qπ 4
2
e αF,XXXX −(2β + 4µ) F,XX +γ F + 12λc F,XX = 0 (65)
b b
De par l’ordre de grandeur des paramètres géométriques et matériels, la solution

générale peut s’écrire :
F(X) = k1 cos(rX) + k2 sin(rX) + k3 cos(sX) + k4 sin(sX) (66)
avec : s rh
2 2
2
i 4
12λ−(2β+4µ)( eqπ
b )
− 12λ−(2β+4µ)( eqπ
b )
−4αγ ( eqπ
b )
r= 2e2 α
s rh i2
(67)
eqπ 2 eqπ 2 4
12λ−(2β+4µ)( b ) + 12λ−(2β+4µ)( b ) −4αγ eqπ
b ( )
s= 2e2 α
Les conditions aux limites sur le bord encastré X = 0 (W = 0 et W,X = 0) et sur

le bord opposé X = a (W = 0 et αW,XX +βW,Y Y = 0) forment un système d’équations
linéaires, dont il faut annuler le déterminant pour s’assurer de l’existence d’un mode non
nul :
r s2 − r2 cos(ra) sin(sa) + s r2 − s2 cos(sa) sin(ra) = 0

(68)
86
La résolution de cette équation mène à une valeur critique minimale dont le mode as-
socié présente une unique demi-onde (q = 1) selon l’axe eY . La comparaison des résultats
avec le cas des bords simplement appuyés (figure 28) affiche des solutions identiques entre
les deux types de conditions aux limites pour certaines valeurs du rapport a/b et globale-
ment lorsque ce rapport tend vers l’infini.
Figure 28 – Contraintes critiques de flambement plastique d’une plaque rectangulaire

sous compression uniaxiale
La pente initiale λ1 de la branche bifurquée s’obtient toujours à partir de la solution

triviale (plane) du problème. Dans le cas de bords simplement appuyés, on trouve :
p 2 2
π 2 e2 a2
(2 − η) + qb2 (2η − 1)
λ1 = ET (69)
2 5 − 8η + 5η 2 − (1 − 2ν) (1 − 4η + η 2 ) EET
avec les mêmes valeurs de p et q vérifiant les relations (61). Dans le cas uniaxial, l’expression
précédente se simplifie : 2
π 2 e2 ap2 − 2b12
λ1 = ET (70)
5 − (1 − 2ν) EET
où p est donné par l’équation (62).
3.1.5 Cylindre sous compression axiale
La troisième géométrie envisagée est le cylindre circulaire (creux) qui, de même que les
poutres ou les plaques, pose naturellement la question du flambement dans de nombreuses
applications industrielles. Parmi les chargements induisant des instabilités géométriques,
on ne considère ici que la compression axiale plus propice à fournir des solutions analy-
tiques simples. Deux types de déformées peuvent survenir lors du flambement d’un cy-
lindre sous compression axiale : dans le cas de cylindres minces, le flambement (élastique)
mène fréquemment à des modes non-axisymétriques, alors que pour des épaisseurs plus
importantes, le flambement devient plastique et donne lieu le plus souvent à des modes
axisymétriques.
87
Parmi les travaux analytiques de référence, on peut citer Batterman qui obtint la
contrainte critique d’un cylindre pour un mode de flambement axisymétrique avec des
bords simplement appuyés [BAT, 1965] (le cas des bords encastrés donne des résultats
similaires) ou avec des bords libres en considérant un cylindre semi-infini [BAT, 1968].
Plus récemment, Mao et Lu [MAO, 2001] ont établi des solutions semi-analytiques dans
le cas plus général de contraintes de compression combinées, axiales et circonférentielles,
en incluant les effets de cisaillement transverse. Les résultats qui vont suivre sont basés
plus simplement sur le modèle de Koiter, équivalent à la cinématique de Kirchhoff-Love,
et ont pour objectif de retrouver les résultats axisymétriques, purement analytiques, de
Batterman pour diverses conditions aux limites, complétés par la donnée de la pente des
branches bifurquées [P2 ,P9 ].
Les trois équations aux dérivées partielles obtenues par intégration de l’équation de
bifurcation sont initialement complexes, mais se simplifient largement dans l’hypothèse
d’une réponse axisymétrique pour donner principalement une équation différentielle en la
composante radiale du mode de flambement, formellement similaire à l’équation de Donnell
[DON, 1976] en élasticité :
e2 β2

1
α W,XXXX +λc W,XX + 2 γ − W=0 (71)
12 R α
L’obtention d’une solution analytique pour des bords libres nécessite de considérer
un cylindre infini. Par symétrie, on représente uniquement un cylindre semi-infini avec le
bord libre en X = 0 et les conditions de symétrie reportées à l’infini. La résolution des
équations locales avec les conditions aux limites à l’infini et en X = 0 (U = 0, W,XX = 0
e2
et α 12 W,XXX +λc W,X = 0) fournit la solution suivante :
 q
3e q pX pX


 U = −e R p sin √
Re
exp − √
3Re
V=0h (72)
 W = e √3 cos √pX − sin √pX
 i
pX
exp − √

Re Re 3Re
avec : s h i
E E
27 5−4ν−(1−2ν)2 ET ET
p= 4
E
2
1+3 ET (73)
ET
1−(1−2ν)
q=2 ET
E
1+3 E
et en particulier la contrainte critique :

v
ET
Ee
u
λLT = E
u
t h i (74)
R 3 5 − 4ν − (1 − 2ν)2 ET E
Dans le cas des bords simplement appuyés, le mode est tout simplement sinusoı̈dal
suivant la longueur finie du cylindre :

qle nπX

 U = nπR cos l − 1

V=0 (75)
 W = e sin nπX

l
et la contrainte critique vaut :

v
2
2
2 ET
e β 1 l 2Ee
u
λA 2 2
− E i = 2λLT
u
T = αn π + γ = t (76)
12l2 α n2 π 2 R 2
h
R 3 5 − 4ν − (1 − 2ν) 2 ET
E
88
en choisissant la valeur réelle n = π√2lp
3Re
comme valeur approchée de l’entier minimisant
la contrainte critique, ce qui est valide pour des cylindres suffisamment longs.
Dans ce dernier cas, la pente initiale λ1 est évaluée à partir de la solution fondamentale,
plus simple car homogène, d’un cylindre comprimé avec bords libres, ce qui constitue une
approximation raisonnable lorsque les déformations pré-critiques restent faibles :
r h i
5 + (8ν − 1) E + 4 3 5 − 4ν − (1 − 2ν)2 EET EET
ET
2ET e
λA
1 = h i (77)
R 1 + 3 EET 5 − (1 − 2ν) EET
La solution est plus complexe dans le cas d’un cylindre infini avec bords libres :
2ET e 5 + (8ν − 1) EET

λL1 = h i max [−f (τ, ρ)] (78)
R 1 + 3 ET 5 − (1 − 2ν) ET τ >0
E E
ρ ∈ [−1/2, 1/2]
avec :
√
f (τ, ρ) = exp − √τ3

3 cos τ − (1 − Cρ) sin τ
r h i
E E
16 3 5−4ν−(1−2ν)2 ET ET
C= >0 (79)
ET
5+(8ν−1) E h √ i
1
3 (16 + 4C + C 2 ) exp − √13 tan−1 − 3 4+C
p
max(−f ) = − min(f ) = 4 4−C
sous l’hypothèse d’un cylindre suffisamment mince et d’un rapport ET /E suffisamment

petit.
Mais l’originalité de ce cas de figure s’avère être la différence de pente relevée lorsque le
mode s’exprime avec l’un ou l’autre des signes possibles. La formule précédente (équations
78 et 79) est valide lorsque le bord libre se déplace vers l’extérieur. À l’inverse, lorsque le
bord libre rentre vers l’intérieur du cylindre, les calculs mènent à une autre expression de
la pente initiale pour le même point critique :
2ET e 5 + (8ν − 1) EET

λL-int
1 = h i max f (τ, ρ) (80)
R 1 + 3 ET 5 − (1 − 2ν) ET τ >0
E E
ρ ∈ [−1/2, 1/2]
avec : √
max(f ) = 3 6= max(−f ) (81)
La dissymétrie de la solution, due à la nature même du mode de flambement sinusoı̈dal

amorti, fournit au point critique deux branches bifurquées de pentes distinctes. Le compor-
tement en post-flambement ”vers l’intérieur” est toujours plus raide, ce qui peut s’expli-
quer par la présence de contraintes circonférentielles compressives, induisant une décharge
élastique au niveau du bord libre. Ces deux pentes, ainsi que celle du cas des bords appuyés,
sont comparées dans la figure 29 en faisant varier le rapport ET /E.
3.1.6 Poutre mixte acier/béton avec connexion partielle
Les deux dernières applications de la méthode générale de résolution de l’équation

de bifurcation concernent des structures composites en élasticité linéaire. La formulation
89
Figure 29 – Pentes initiales des branches bifurquées d’un cylindre élastoplastique sous
compression axiale
variationnelle utilisée permet sans mal de considérer un système formé de plusieurs entités,
impliquant le comportement de plusieurs matériaux.
La première de ces applications concerne les poutres composites à interaction partielle
[P16 ], fréquemment rencontrées dans le domaine du génie civil, formées de deux poutres
contiguës de matériaux distincts (le plus souvent acier et béton) raccordées par une dis-
tribution de ressorts longitudinaux pour rendre compte des phénomènes de glissement à
l’interface (figure 30). Depuis les travaux précurseurs de Newmark et al. [NEW, 1951],
bon nombre d’auteurs ont étudié ce type de structure, d’un point de vue analytique ou
numérique, en considérant différentes théories de poutres et en incluant éventuellement
des non-linéarités matérielles ou géométriques. Les résultats les plus récents portent sur
l’introduction des effets de cisaillement transverse, à savoir l’utilisation du modèle de Ti-
moshenko. Concernant le flambement sous compression axiale, les rares solutions ana-
lytiques disponibles dans la littérature reposent sur la théorie de Bernoulli [KRY, 2009]
[SCH, 2010]. Dans le cas de poutres de Timoshenko, Schnabl et Planinc [SCH, 2011] ont
établi les équations générales du problème et fourni une stratégie de résolution, sans ex-
pliciter de solutions analytiques exploitables. Enfin, Xu et Wu [XU, 2007] ont quant à eux
abouti à des formules analytiques explicites dans le cas de bords simplement appuyés, mais
en se limitant au cas où la rotation de la section, autrement dit le cisaillement transverse,
est identique dans les deux poutres.
Dans notre cas, on considère un comportement des poutres élastique linéaire ainsi
qu’une loi linéaire entre le cisaillement interfacial et le glissement correspondant. En re-
vanche, le modèle de Timoshenko est retenu avec un cisaillement propre pour chacune des
deux poutres, ce qui est davantage conforme à la réalité de deux poutres de géométries et
matériaux différents. Dans ces conditions, l’analyse de bifurcation 3D permet d’aboutir au
système d’équations différentielles suivant pour les cinq champs modaux inconnus (deux
déplacements longitudinaux Ua et Ub , deux rotations Θa et Θb , et un seul déplacement
90
r
ey (Eb , Gb , kb , Sb , I b )
(ksc ) L
2hb (b)
λ
2ha (a)
r r
ez ex
(Ea , Ga , ka , Sa , I a )
Figure 30 – Poutre bi-couche comprimée avec interaction partielle
transverse V commun aux deux couches) :
Ea Sa Ua ,XX −ksc G = 0
Eb Sb Ub ,XX +ksc G = 0
ka Ga Sa (Θa ,X −V,XX ) + kb Gb Sb (Θb ,X −V,XX ) + (Ea Sa + Eb Sb ) λLc V,XX = 0 (82)
Ea Ia Θa ,XX +ka Ga Sa (V,X −Θa ) + ksc ha G = 0
Eb Ib Θb ,XX +kb Gb Sb (V,X −Θb ) + ksc hb G = 0
où G = Ua −ha Θa −Ub −hb Θb représente la composante modale du glissement à l’interface.

La résolution de ces équations mène à tout un ensemble de solutions explicites pour diverses
conditions aux limites. Par exemple, le cas des bords simplement appuyés donne lieu au
déplacement critique suivant :

π 2 EI ∞ f 2 GS
π 2 EI EI f 2 EI
ksc EI c ∞ GS ksc L2
L2 f2
+ EI + d c f2
+ 2d
L L2 EI ∞ GS ∞ ES EIEI ∞ GS π ES
λc = 2 (83)
ES π4 EIf
1 + ksc L2 EI
c∞
+ 1 + π 2 EI

1 + ksc L2
+ ksc h̄2
2 2d c L2 GS 2d
4f
L GS π ES EI π ES GS
c
avec :
h̄ = ha + hb ES = Ea Sa + Eb Sb d = Ea Sa Eb Sb
ES Ea Sa +Eb Sb
√
EI = Ea Ia + Eb Ib EI ∞ = EI + h̄2 ESd EI
f = Ea Ia Eb Ib
√
GS = ka Ga Sa + kb Gb Sb GS
g = ka Ga Sa kb Gb Sb
2
−1 −1
h2b
2
c = h̄2 h2a + hb
EI d = h̄2
GS ha
+
Ea Ia Eb Ib ka Ga Sa kb Gb Sb
c + h̄2 ES EI Ea Ia Eb Ib

EI
c ∞ = EI d
GS = ka Ga Sa + kb Gb Sb
(84)
91
associé à un mode de la forme :
Ua = EhaaSβaa +hπb2βEba Sa sin πX


 1+ E S + L
ksc L2

 b b




Ub = − hEa βSa +hπb2βEb S sin πX



 L
1+ E b Sb + b b



 a a ksc L2

(85)
V = α 1 − cos πX



 L




Θa = βa sin πX




 L




Θb = βb sin πX

L
Cette formule générale a été exploitée dans divers cas particuliers (dont les deux cas
extrêmes de deux poutres solidaires d’une part, déconnectées dans la direction longitudi-
nale d’autre part) et donne les résultats attendus. Elle permet également de retrouver la
solution de Bernoulli, en faisant tendre les raideurs de cisaillement transverse vers l’infini.
De façon à mieux évaluer et valider l’effet du cisaillement transverse, la valeur critique d’une
telle poutre composite a été estimée numériquement par un calcul de flambement linéarisé
par éléments finis en contraintes planes (les sections de poutres sont supposées rectangu-
laires). Les deux poutres sont supposées particulièrement épaisses et très différentes l’une
de l’autre de façon à s’écarter du cas particulier symétrique pour lequel les deux poutres
semblables affichent le même cisaillement transverse (les caractéristiques géométriques et
matérielles sont résumées dans le tableau 8). La comparaison des résultats numériques avec
les solutions analytiques obtenues pour les différentes hypothèses cinématiques (figure 31)
montre bien la nécessité de prendre en compte le cisaillement transverse indépendamment
dans chacune des couches et la qualité de la solution originale obtenue par notre méthode.
Poutre Longueur Largeur Hauteur Module d’Young Coefficient de Poisson

(a) 1m 100 mm 20 mm 200000 M P a 0.3
(b) 1m 100 mm 200 mm 5000 M P a 0.3
Tableau 8 – Paramètres géométriques et matériels pour la validation numérique du flam-

bement d’une poutre bi-couche avec interaction partielle
3.1.7 Flambement du sandwich Napcor en compression suivant l’épaisseur
La dernière application traite de la compression du sandwich Napcor droit dans la di-

rection de l’épaisseur. La présence de renforts transverses permet d’augmenter de manière
significative la résistance mécanique d’un tel composite dans cette direction, à savoir en
cisaillement transverse et en compression. Cette dernière sollicitation est pourtant suscep-
tible de poser des problèmes d’instabilité compte tenu de la géométrie des renforts. Des
essais de compression ont d’ailleurs montré d’importants écarts entre les pentes des courbes
contrainte-déformation expérimentales et les modules homogénéisés par une simple loi des
mélanges. Ces différences ne sont pas forcément dues à l’inadéquation de la loi des mélanges,
qui s’applique parfaitement dans une telle configuration géométrique, mais plutôt à la
sensibilité du comportement aux (nombreuses) imperfections associée au phénomène de
flambement sous-jacent lorsqu’on comprime le matériau sandwich. L’objectif de cette par-
tie est donc de déterminer une solution analytique de la valeur critique de flambement
du sandwich Napcor comprimé suivant l’épaisseur afin de la confronter aux résultats
92
1,40E+07
1,20E+07
1,00E+07
Buckling load (N)
8,00E+06
Timoshenko
Xu and Wu (2007)
6,00E+06
Bernoulli
FEM
4,00E+06
2,00E+06
0,00E+00
0,00E+00 2,00E+03 4,00E+03 6,00E+03 8,00E+03 1,00E+04 1,20E+04
Interface shear bond stiffness (MPa)
Figure 31 – Comparaison des charges critiques numériques et analytiques (sous

différentes hypothèses cinématiques) d’une poutre bi-couche avec interaction
partielle pour différentes raideurs à l’interface
expérimentaux et de fournir ainsi une valeur explicite du chargement limite dans cette
direction.
Le flambement du matériau composite peut être vu comme le flambement des ren-
forts (assimilables à des poutres élancées) au sein du milieu homogène et isotrope qu’est
l’âme en mousse du sandwich. La présence de la mousse, malgré son très faible module,
influence considérablement la valeur critique (elle joue le rôle de stabilisateur), ce qui
fait des valeurs critiques d’Euler classiques pour les poutres (en l’absence de mousse) de
très mauvaises estimations de la solution. De nombreux travaux ont déjà porté sur la re-
cherche d’une meilleure solution (analytique) pour un tel problème de micro-flambement,
considéré comme l’un des principaux modes de ruine des composites renforcés de fibres.
Les travaux de Rosen sont précurseurs et fournissent une solution simple, mais avec l’hy-
pothèse très forte d’un cisaillement uniforme dans la matrice, ce qui limite son domaine
de validité. À partir de là, des solutions intermédiaires ont vu le jour, parmi lesquelles on
peut citer celle de Drapier et al. [DRA, 1996], qui part d’hypothèses simplifiées sur l’allure
des déformations dans le renfort et la matrice pour déterminer par homogénéisation la
charge critique de flambement d’un pli unidirectionnel d’un composite stratifié. Liu et al.
[LIU, 2008] sont parmi les seuls à s’être intéressé au flambement d’un matériau de type
sandwich renforcé dans l’épaisseur, similaire au Napcor . Ils supposent dans leur modèle
analytique que les renforts peuvent être assimilés à des poutres simplement appuyées et que
l’âme peut être représentée par la superposition de deux distributions de ressorts élastiques,
l’une horizontale et l’autre verticale. Leur modèle est ainsi la généralisation de l’approche
93
utilisée initialement par Timoshenko dans le cas du flambement d’une poutre sur une fon-
dation élastique. L’ensemble des modèles simplifiés établis (dont les deux précédents) ne
donnent malheureusement de bons résultats que dans des cas spécifiques (par exemple pour
de très faibles ou très fortes fractions volumiques de renforts) et ne constituent en aucun
cas une réponse satisfaisante dans le cas général. Certains auteurs se sont alors attaché à
la recherche de solutions ”exactes”, considérant le milieu environnant comme continu et
sans apporter de simplification quant au champ des déformations. Les principaux résultats
concernent les composites à fibres longues et sont donnés dans un cadre 2D. Parnes et Chis-
kis [PAR, 2002] comparent leurs propres résultats aux solutions de Rosen dans le contexte
des composites chargés de nano-fibres. Plus tôt, Zhang et Latour [ZHA, 1994] ont montré
que pour des fractions volumiques suffisamment importantes (lorsqu’il y a interaction entre
les renforts), le mode prédominant est le mode dit de cisaillement. Après avoir obtenu une
solution 2D explicite, ils ont étudié le cas 3D de manière semi-analytique [ZHA, 1997],
afin d’identifier d’éventuelles règles de correspondance entre les géométries 2D et 3D (en
termes de dimensions des renforts et de fractions volumiques). Dans le but d’obtenir une
solution purement analytique, on se limite à une modélisation 2D, où l’équivalence entre
le matériau réel 3D et le modèle 2D est définie par les deux règles suivantes : (i) la largeur
des renforts dans le modèle 2D est telle qu’en l’absence de mousse, on retrouve le même
déplacement critique que celui obtenu pour les renforts cylindriques circulaires seuls en
3D ; (ii) la largeur de la mousse est calculée de façon à conserver, dans le cas 2D, la même
fraction volumique de renforts qu’en 3D.
3.1.7.1 Solution analytique pour la valeur critique La méthode générale de

résolution de l’équation de bifurcation s’applique ici sur la structure composite formée
d’une cellule unité, qui permet de reproduire par périodicité le matériau 2D équivalent
au sandwich Napcor . De façon plus précise, on ne modélise effectivement que la mousse
renforcée, sans les peaux dont l’influence sera éventuellement prise en compte au travers
des conditions aux limites (le modèle géométrique est représenté à la figure 32).
Figure 32 – Modèle géométrique 2D d’une cellule unité de mousse renforcée Napcor
À l’instar des auteurs précédents, les renforts sont modélisés par des poutres (de Ber-
noulli), alors que la mousse est représentée par un milieu 2D (en déformations planes).
L’application de notre méthode est donc ici assez originale dans la mesure où au raccorde-
ment des deux matériaux seront associés une condition aux limites (pour la mousse) avec
une équation de champ (pour la poutre). La cinématique de la cellule unité considérée est
caractérisée par deux champs de déplacement Um (X, Y ) et Vm (X, Y ) en tout point de la
mousse et deux champs de déplacement Ur (X) et Vr (X) en tout point de la ligne moyenne
des renforts. Pour simplifier les calculs, on fait l’hypothèse que les pré-contraintes dans la
94
mousse (dues à la compression de l’ensemble) n’influencent pas le flambement des renforts,
si bien qu’on remplace les modules nominaux par les modules matériels dans la partie de
l’équation de bifurcation relative à la mousse. Après quelques développements, intégrations
par parties et simplifications (dont certains s’inspirent des travaux précédents concernant
les poutres), on obtient d’une part un déplacement modal Ur identiquement nul (comme
dans le cas d’une poutre seule) et d’autre part les trois équations aux dérivées partielles
suivantes :
(λm + 2µm ) Um ,XX +µm Um ,Y Y + (λm + µm ) Vm ,XY = 0

(λm + 2µm ) Vm ,Y Y +µm Vm ,XX + (λm + µm ) Um ,XY = 0
2Er e3 Er eλc (86)
3 Vr ,XXXX + L Vr ,XX −2eµm (Um ,XY +Vm ,XX )|Y =H
+2λm Um ,X |Y =H + 2 (λm + 2µm ) Vm ,Y |Y =H = 0
où λc représente le déplacement critique (dans le sens de la compression). Les deux

premières équations s’identifient aux équations locales d’équilibre 2D de la mousse. La
troisième correspond à l’équation classique du flambement d’une poutre, modifiée par
des termes caractérisant l’effet des contraintes à l’interface mousse/renfort sur le flambe-
ment du renfort. Son expression n’est pas nouvelle mais elle découle ici naturellement de
l’équation de bifurcation appliquée à l’ensemble du modèle. La solution recherchée est sup-
posée périodique et symétrique en Y , ce qui évite d’écrire une nouvelle équation similaire
à l’interface Y = −H.
En pratique, les peaux du sandwich sont comprimées par des plaques rigides, et trans-
mettent à la mousse renforcée des déplacements imposés. Les deux extrémités de chaque
renfort peuvent être donc considérées comme encastrées et en encastrement glissant, res-
pectivement. La condition aux limites de bord libre pour la mousse (en X = ±L) ne
représente a priori pas l’entraı̂nement de la mousse sur les bords par la translation d’en-
semble des peaux, entraı̂nées elles-mêmes par la flexion des renforts. Elle donne cependant
de très bons résultats lorsque les renforts sont suffisamment denses et permet d’aboutir à
une solution analytique explicite par séparation des variables. Cette solution correspond
au mode de cisaillement précédemment obtenu par Parnes et Chiskis [PAR, 2002] et Zhang
et Latour [ZHA, 1994], entre autres, à la différence près que ces auteurs s’intéressent à des
composites à fibres longues. Dans leur cas, le mode de flambement sinusoı̈dal des renforts a
une longueur d’onde a priori indéterminée, dont la valeur est finalement obtenue par mini-
misation du chargement critique associé, alors que cette longueur d’onde pour les renforts
du sandwich est directement liée à l’épaisseur du matériau.
La résolution des équations (86) avec ces différentes conditions aux limites mène à la
solution suivante pour le mode de bifurcation :

πH πH πX

 Vr = K1 cosh 2L + K2 H sinh 2L sin 2L

Um = K3 sinh πY πY
2L + K2 Y cosh 2L cos 2L
πX (87)
 V = K cosh πY + K Y sinh πY sin πX

m 1 2L 2 2L 2L
avec :
K1 = 2LπH cosh πH 2 2 πH

2L + 4L (3 − 4νm ) − eHπ sinh 2L
K2 = eπ 2 cosh πH πH
2L − 2Lπ sinh 2L
(88)
K3 = 2Lπ (H + 3e − 4eνm ) cosh πH 2 πH
2L − eHπ sinh 2L
et le déplacement critique :
λc = 24L2 Em 4L2 + e2 π 2 (νm − 1) cosh2 πH 3 3 2

2L + 2Er e π L 4νm + νm − 3
+48L3 Em eπ (2νm − 1) cosh πH πH 4 3 4

2L sinh 2L + 96L Em (1 − νm ) − Er e π H (1 +ν )
m
2 2 πH πH

−24L Em eHπ / 6Er eLπ (1 + νm ) 2L (4νm − 3) cosh 2L sinh 2L − πH
(89)
95
3.1.7.2 Validations numérique et expérimentale Ce résultat a été validé par com-
paraison à des solutions numériques obtenues par éléments finis. De façon à vérifier l’en-
semble des hypothèses retenues pour le modèle simplifié 2D, les calculs numériques sont
effectués en 3D sur l’ensemble du sandwich (mousse renforcée + peaux). Pour limiter
le temps de calcul, on ne considère à nouveau qu’une cellule unité et on applique des
conditions périodiques sur les faces latérales pour respecter la périodicité du mode de
flambement. La figure 33 représente le premier mode de flambement obtenu par un cal-
cul linéarisé. Pour la forte valeur de densité de renforts considérée (9.4%), le déplacement
critique vaut 0.205 mm. La solution analytique correspondante est de 0.174 mm et se
trouve donc être environ 15% inférieure à la valeur numérique. Cette différence s’explique
entre autres par la présence des peaux dans le modèle numérique 3D mais surtout par
l’approximation liée aux règles d’équivalence entre les modèles 2D et 3D.
Figure 33 – Premier mode de flambement du sandwich Napcor en compression dans

l’épaisseur
Lorsqu’on trace les courbes de réponse expérimentales en compression pour plusieurs

échantillons (figure 34), on remarque bien la dispersion des résultats en termes de module et
de contrainte maximale, due à la présence d’imperfections géométriques et au phénomène
d’instabilité sous-jacent. Le point de bifurcation sur chacune des courbes se situe aux
alentours du changement de pente (lors de la perte de linéarité) et précède nettement le
point limite dont la présence est certainement due à des effets secondaires tels que la rupture
successive des renforts. Tous ces points de bifurcation ont la particularité d’avoir le même
déplacement critique (on peut montrer numériquement que des défauts sous la forme de
forces pertubatrices ne modifient pas non plus la valeur du déplacement critique obtenue),
qui coı̈ncide parfaitement avec la valeur numérique. La figure 34 présente également la
courbe de réponse numérique obtenue de manière incrémentale, à des fins de comparaison.
Les résultats sont moins satisfaisants lorsque la densité des renforts dans le matériau
diminue. Avec des renforts moins nombreux, l’influence des peaux se fait davantage sentir
dans la déformée modale et donc dans la valeur de déplacement critique. Mais les imperfec-
tions liées entre autres au procédé de fabrication Napcor sont manifestement la principale
cause des écarts relevés. La construction d’un matériau modèle, avec un perçage manuel
et précis des renforts, est prévue, de façon à limiter la présence de défauts aléatoires aux-
quels le flambement est très sensible. L’aspect le plus important semble être la maı̂trise de
96
Figure 34 – Courbes pression-déplacement expérimentales et numérique de la compres-
sion dans l’épaisseur du sandwich Napcor
la fraction volumique de fibres au sein des renforts, dont le module d’Young longitudinal
dépend fortement.
3.2 Développements supplémentaires pour une poutre de Bernoulli
La démarche décrite et illustrée précédemment permet d’obtenir la valeur critique

et le premier terme du développement asymptotique de la solution bifurquée. Dans ce
paragraphe, on présente quelques développements qui ont permis d’identifier des termes
supplémentaires dans ce développement en séries fractionnaires [P13 ]. Pour des raisons
de simplicité, on s’intéresse uniquement à l’exemple de la poutre de Bernoulli, pour lequel
on réécrit les équations de manière spécifique, en suivant le formalisme utilisé par Nguyen
[NGU, 2000]. La méthode employée permet également de déterminer de manière asymp-
totique la forme et la taille de la zone de décharge mise en jeu durant le comportement
post-critique.
On s’intéresse donc à la bifurcation à la valeur critique du module tangent d’une poutre
encastrée-libre à section rectangulaire pleine, supposée fléchir vers les Y négatifs. La zone
de décharge élastique est séparée de la zone plastique à un instant t par la frontière Y =
d(X, t), X ∈ [0, a], comme représenté sur la figure 35.
Dans un premier temps, les quatre opérations suivantes sont effectuées :
(a) La définition d’une flèche relative v(X, t) = V (X, t) − V (L, t).
(b) L’introduction de la fonction r(X, t) définie par :
v̇(X, t) = λ̇(t)r(X, t) (90)
sous réserve de la condition sur la charge critique λ̇ > 0.
(c) Le changement de variables suivant, de façon à écrire les équations sur des
intervalles fixes, indépendants de la zone de décharge :
X
• pour X ∈]0, a[ : X = as ↔ s= a avec s ∈]0, 1[
97
Figure 35 – Zone de décharge élastique durant le flambement plastique d’une poutre can-
tilever
X−a
• pour X ∈]a, L[ : X = a + (L − a)(s − 1) ↔ s= L−a + 1 avec s ∈]1, 2[
(d) Le choix de la grandeur adimensionnelle t = a/L comme paramètre infinitésimal

pour les développements asymptotiques.
Ceci étant posé, on peut montrer que les fonctions λ(t), v(s, t), r(s, t) et d(s, t) sont
régies par le système d’équations suivant :
• pour s ∈]0, 1[ :
P (d)r” + t2 L2 = 0 (a)
Q(d)r” + t2 L2 (λr + v) = 0 (b)
λ̇r = v̇ − v 0 st (c)
où les fonctions P (d) et Q(d) sont des polynômes donnés.
• pour s ∈]1, 2[ :
ET Ir” + (1 − t)2 L2 (λr + v) = 0 (d)
λ̇r = v̇ − v 0 2−s
1−t (e)
• avec les conditions aux limites et à l’interface :
d(1, t) = 2e (f ) r0 (0, t) = 0 (g) r(2, t) = 0 (h)

v 0 (0, t) = 0 (i) v(2, t) = 0 (j)
r (1− , t) = r (1+ , t) (k) − +
v (1 , t) = v (1 , t) (l)
(1 − t)r0 (1− , t) = tr0 (1+ , t) (m)
• et les conditions initiales :

e
λ(0) = λc (n) v(s, 0) = 0 (o) d(s, 0) = (p)
2
98
Le système non-linéaire (a) − (p) a été résolu par une méthode asymptotique, en
développant en séries entières l’ensemble des quantités inconnues. Les calculs ont été menés
jusqu’à l’ordre 10. Compte tenu de la nullité de certaines valeurs, et en posant ξ = 2v(0) e le
rapport de la flèche maximale sur la demi-épaisseur, on aboutit aux deux développements
suivants : 5 7 9
λ = λ0 + λ5 t5! + λ7 t7! + λ9 t9!
5 7 9 (91)
ξ = ξ5 t5! + ξ7 t7! + ξ9 t9!
Enfin, en éliminant de proche en proche le paramètre ”temps” t, il est possible d’expri-

mer l’amplitude du chargement comme un développement en série à exposants fraction-
naires du paramètre de perturbation choisi :
7 9
9
λ = λT 1 + c1 ξ + c2 ξ 5 + c3 ξ 5 + o ξ 5 (92)
avec :
 2  4
5 5
15  15π EET 65 EET − 48 15π EET
c1 = 3 c2 = − √  c3 =  √  (93)
7 4 2 1 − ET 42 EET 4 2 1 − EET
E
Ces résultats concordent bien entendu avec ceux obtenus par la méthode générale
précédente, pour ce qui est de la valeur critique du module tangent et du terme linéaire
(pente initiale de la branche bifurquée). On retrouve également la valeur du premier terme
non-linéaire (c2 ) déjà obtenue par ailleurs de manière différente. En revanche, le terme non-
linéaire suivant dans le développement asymptotique n’a pour ainsi dire jamais été formulé
correctement. Seuls Cheng et al. [CHE, 1995] ont proposé une valeur pour ce quatrième
terme du développement en utilisant une approche spécifique permettant de décliner di-
rectement le développement final à puissances fractionnaires. Leur solution littérale diffère
cependant légèrement de la nôtre, ce qui induit des écarts importants lors des applications
numériques.
De façon à valider ce nouveau résultat, on a comparé les branches bifurquées définies
analytiquement par les développements asymptotiques aux différents ordres successifs à la
solution incrémentale obtenue par éléments finis (figure 36). La comparaison a été effectuée
dans le domaine de validité des développements (91) et (92), là où les deux développements
donnent en particulier des solutions similaires (pour des valeurs du rapport de la flèche sur
l’épaisseur ξ et du temps cinématique t infiniment petites, en post-flambement immédiat).
Les résultats numériques sont en parfait accord avec la solution analytique complète. On
montre ainsi l’importance du dernier terme à l’ordre 9/5 dans la description de la branche
bifurquée, qui permet de se rapprocher nettement de la courbe numérique par rapport à
la solution à l’ordre inférieur, contrairement au terme de Cheng, Lu et Fang. La méthode
employée ici fournit donc un résultat original et définitif, dans la mesure où l’on peut
ajouter qu’en pratique, il n’est pas nécessaire de pousser le développement plus loin.
3.3 Modélisation numérique du flambement et du post-flambement

élastoplastiques
Les résultats précédents permettent l’analyse du flambement et du post-flambement

immédiat, mais ne rendent pas compte du comportement post-critique lointain. Le recours
à la simulation numérique est l’unique alternative pour décrire de manière plus globale la
réponse d’une structure donnée au phénomène de flambement jusqu’à la ruine éventuelle,
qui plus est, pour des géométries et chargements quelconques. Un programme de calcul par
éléments finis consacré à la résolution des problèmes de flambement a donc été mis en œuvre
99
Figure 36 – Courbes d’équilibre post-critiques d’une poutre comprimée issues de la va-
leur critique du module tangent : comparaison des différents développements
fractionnaires avec les résultats numériques
dans la continuité des précédents programmes développés dans le cadre de la plasticité,

comme nouveau module de NLFEAS. Les méthodes numériques employées spécifiquement
pour le traitement des instabilités ont été associées au module 3D en grandes transfor-
mations plastiques développé initialement pour des applications de type striction, entre
autres. Cependant, les structures envisagées pour l’analyse des instabilités géométriques
étant la plupart du temps minces ou élancées, la solution optimale permettant de modéliser
efficacement ces deux types de géométries s’avère être une modélisation de type coque. Un
modèle élément fini coque a donc été spécialement conçu, en s’appuyant toujours sur une
formulation lagrangienne totale, apte à rendre compte des non-linéarités géométriques et
matérielles rencontrées lors du post-flambement voire dès le comportement pré-critique.
Dans un problème de flambement type (qui met en jeu majoritairement de la flexion), les
déformations (plastiques) restent relativement modérées alors que les rotations peuvent
être réellement importantes. Le choix d’une formulation lagrangienne totale est donc idéal
dans ce cadre d’étude, quoique malgré tout peu employé dans les codes commerciaux, du
fait d’un spectre d’utilisation souvent plus large.
Si le programme envisagé s’intéresse particulièrement au flambement et post-
flambement, il doit permettre d’en fournir les caractéristiques principales, autant en
élasticité qu’en plasticité, à savoir la nature des points critiques rencontrés (point limite
ou point de bifurcation), les grandeurs associées (charge, contrainte, déformation et/ou
déplacement critique), ainsi que le comportement post-critique. L’évaluation des points
critiques peut s’effectuer dans l’absolu par des méthodes directes, mais celles-ci sont soit
linéarisées, ce qui limite le champ d’application du programme à des problèmes au com-
portement pré-critique ”linéaire”, au même titre que la démarche analytique présentée
précédemment (contrairement aux problèmes classiques de bifurcation, bon nombre de
problèmes de claquage mettent en jeu des déformations pré-critiques importantes), soit
100
infiniment coûteuses en temps de calcul (sans linéarisation préalable, la détermination di-
recte de points singuliers nécessite l’utilisation de systèmes de taille double). On choisit
alors une méthode indirecte, qui doit fournir de manière incrémentale l’ensemble de la
courbe d’équilibre du problème posé, incluant les éventuels points limites et/ou points de
bifurcation. Avec une telle approche numérique, on s’affranchit des hypothèses restrictives
précédentes (de la nécessité des déformations pré-critiques faibles, entre autres), et on
décrit en une seule opération l’ensemble du comportement de la structure, de la réponse
pré-critique initiale jusqu’à la solution post-critique lointaine.
Les fortes non-linéarités géométriques et matérielles cumulées, ajoutées au caractère
spécialement critique d’un problème de flambement, nécessitent l’utilisation de méthodes
numériques performantes. Une méthode de longueur d’arc mixte (en forces ou déplacements
imposés) a été spécialement mise en œuvre pour faire face aux comportements non-
monotones les plus critiques. Mais la particularité du programme développé tient sur-
tout dans l’utilisation de méthodes de branchement pour identifier les points critiques, et
décrire à partir des points de bifurcation les différentes courbes souhaitées. Ainsi, à la fois
la détection du flambement mais aussi la description du post-flambement s’effectue sans
l’apport d’aucun défaut. Si toutes sortes d’imperfections (géométriques, en particulier)
sont inévitables en pratique, il est souvent difficile de les quantifier et donc de mesurer
correctement leur incidence sur le comportement en flambement et post-flambement des
structures. Par conséquent, excepté dans un cas particulier, où différents types de défauts
ont été évalués et considérés dans la modélisation pour mieux estimer les chargements li-
mites admissibles, l’ensemble des applications effectuées avec le programme élément fini ont
consisté en la détermination des courbes d’équilibre de structures ”parfaites”. Ces courbes
idéalisées, riches en singularités qui sont la source des principales difficultés numériques,
forment le squelette sur lequel s’appuient les courbes réelles, dégénérées par la présence de
défauts et dont l’obtention s’en trouve facilitée. Le travail réalisé est donc une première
étape essentielle dans l’analyse du comportement non-linéaire d’une structure.
Le programme mis en œuvre a été tout d’abord validé sur un large éventail de
problèmes plus ou moins critiques dont on connaı̂t des solutions de référence analytiques ou
numériques disponibles dans la littérature. Par ailleurs, il a été exploité pour la validation
de certains résultats originaux obtenus avec la méthode analytique exposée précédemment.
Enfin, et surtout, il a permis d’aller plus loin dans la description du comportement post-
critique en faisant apparaı̂tre des comportements spécifiques (modes secondaires) et en
donnant une réponse plus complète à la question de l’instabilité post-critique et de la
ruine éventuelle qui en découle.
3.3.1 Élément fini coque élastoplastique en rotations finies
Sans rentrer dans les détails, voici quelques aspects essentiels de la modélisation élément
fini, qui s’appuie, entre autres, sur les travaux de Brank et al. [BRA, 1997]. On représente
une structure de type coque par sa surface moyenne et un champ de vecteurs directeurs
inextensibles dans la direction de l’épaisseur. On retient une formulation de type Mindlin-
Reissner qui tient compte des effets de cisaillement transverse, compte tenu des épaisseurs
assez importantes parfois envisagées. Le champ de déplacement en tout point de la coque
fait intervenir en particulier la rotation du vecteur directeur ~t = R.T~ où le tenseur rotation
R est associé au tenseur antisymétrique Θ et à son dual le vecteur rotation θ~ d’amplitude
θ par la formule exponentielle de Rodrigues :
sin θ 1 − cos θ ~ ~
R = cos θI + Θ+ θ⊗θ (94)
θ θ2
101
La cinématique de la coque en rotations finies est donc particulièrement pilotée par la
relation suivante :
~t = R.T~ = cos θT~ + sin θ θ~ ∧ T~ (95)
θ
Les déformations de Green-Lagrange et les contraintes de Piola-Kirchhoff sont écrites
dans des bases duales contravariante et covariante, respectivement, définies sur la surface
moyenne en négligeant les variations de métrique dans l’épaisseur. Les déformations sont
supposées linéaires en fonction de la variable de position dans l’épaisseur, en négligeant le
terme d’ordre 2.
Dans le cadre des matériaux standard généralisés (en plasticité associée), les
déformations élastiques et les contraintes sont reliées par la loi linéaire élastique de Saint-
Venant-Kirchhoff, qui fait intervenir un tenseur réduit du fait de la condition de contraintes
planes et éventuellement un état de contraintes résiduelles à la configuration de référence.
On considère une décomposition additive des déformations et, parmi les critères employés
dans les précédents modules numériques, on ne retient que le critère de von Mises, en se
limitant à de l’écrouissage isotrope linéaire dans les applications qui vont suivre (ces choix
sont similaires aux hypothèses effectuées dans les travaux analytiques et facilitent ainsi
la comparaison). La théorie incrémentale est encore préférée, même dans ce contexte de
flambement plastique où les solutions de la théorie de la déformation s’accordent mieux
avec les résultats expérimentaux.
L’utilisation du principe des travaux virtuels permet la discrétisation des équations
non-linéaires du problème, par le biais d’éléments finis coque isoparamétriques à 8 nœuds
et 5 degrés de liberté par nœud. L’interpolation quadratique associée concerne à la fois la
position de référence, le déplacement sur la surface moyenne et les vecteurs directeurs T~
et ~t. Ce choix, qui simplifie la mise en œuvre numérique, mène à plusieurs types d’erreur
de discrétisation géométrique (entre autres choses, les normales à la surface discrétisée en
chaque nœud dépendent de l’élément considéré et diffèrent en général du vecteur directeur)
dont l’influence sur les résultats numériques est négligeable. L’intégration numérique sur
la surface moyenne et dans l’épaisseur de la coque s’effectue par points de Gauss.
Le système matriciel d’équations non-linéaires est résolu par une procédure itérative
de Newton-Raphson, qui requiert le calcul de la matrice raideur tangente consistante.
Cette matrice est obtenue en parallèle des contraintes et déformations plastiques à chaque
itération par intégration locale. Le schéma d’Euler implicite et un algorithme de retour
radial sont employés pour intégrer les lois d’état et d’évolution, et ainsi évaluer les gran-
deurs inconnues à l’incrément courant en fonction des mêmes grandeurs à l’incrément
précédent et des déformations courantes. La principale difficulté consiste ici en la condi-
tion de contraintes planes qui complique l’intégration locale et nécessite l’utilisation d’une
démarche itérative pour l’estimation du multiplicateur plastique, même dans le cas d’un
écrouissage linéaire (voir [SIM, 1986] pour plus de détails).
3.3.2 Méthodes numériques pour le traitement des fortes non-linéarités
3.3.2.1 Pilotage par longueur d’arc Une paramétrisation par longueur d’arc a été
spécialement envisagée au sein du programme élément fini afin de pouvoir décrire une
courbe force-déplacement quelconque (non-monotone) comportant des points limites de
type ”snap-through” ou ”snap-back”, ce que ne permettent pas les paramétrisations par
forces ou déplacements imposés. D’un point de vue mécanique, on considère non seulement
le cas où les forces appliquées sont connues mais aussi le cas de déplacements donnés,
beaucoup moins traité dans la littérature, de façon à pouvoir résoudre les deux types de
problèmes distincts.
102
Le vecteur des déplacements
n o nodaux {U} peut être préalablement séparé en deux

parties : la première U contient les degrés de liberté inconnus et la seconde U les
e
degrés de liberté imposés. De la même façon, les forces internes {Ψ} et externes {Φ} (ainsi
que leur différence qui constitue le vecteur résidu {R} à annuler) sont partagées entre les
forces données (associées aux déplacements inconnus) et les forces inconnues (réactions) là
où les déplacements sont imposés. La matrice raideur tangente [KT ] est alors partitionnée
en quatre sous-matrices :
( ) ( ) " #
U Φ eT K
K
e e
T
e e
{U} = {Φ} = [KT ] = (96)
U Φ K
e T KT
Selon la nature du problème mécanique posé, on suppose que les déplacements imposés
ou les forces extérieures appliquées sont proportionnelles :
ref
U = λU
(97)
Φ e ref
e = λΦ
ref e ref les quantités imposées de référence.

où λ constitue le paramètre de contrôle, et U et Φ
Parmi les méthodes couramment employées, on retient la méthode de longueur d’arc qua-
dratique qui consiste, à chaque incrément, étant donné une longueur d’arc, à ajouter aux
équations d’équilibre l’une des équations de contrainte suivantes [RIK, 1979] [CRI, 1991] :
e 2 ref 2

2
∆U + ∆λ U = ∆l2
(98)
e 2

∆U + ∆λ2 Cref
2 = ∆l2
en introduisant le facteur d’échelle Cref et en notant ∆U

e et ∆λ les différences entre
l’incrément courant et l’incrément précédent.
Alors que les méthodes linéaires imposent le sens de progression sur la courbe
d’équilibre, et parfois échouent au sens où elles forcent à rebrousser chemin aux points
limites les plus aigus, la méthode quadratique utilisée fournit deux solutions possibles et
laisse le choix d’un critère adéquat pour suivre l’évolution de la courbe d’équilibre dans
la bonne direction. Cependant, un problème survient lorsque l’équation quadratique à
résoudre fournit des racines complexes. Une modification du schéma de résolution standard
a été apportée à l’instar de Lam et Morley [LAM, 1992] pour faire face efficacement à
ces problèmes numériques. L’idée principale consiste à projeter le résidu à annuler sur
le vecteur force extérieur et à éliminer au mieux la partie complémentaire orthogonale
responsable des racines complexes. Dans le cas de forces imposées, la procédure itérative
à adopter est la suivante :
Boucle itérative
[· · · ]
Résolution de l’équation quadratique issue de la méthode standard.
En cas de racines complexes :
eTΦ
(i) Calcul de g = R e ref /Φ
e ref T Φ
e ref et H = R e ref .
e − gΦ
(ii) Calcul des termes suivants :

−1 −1
e H = −K
δU e H et
e
U
e Φref = K
e Φ
e e ref (99)
T T
103
(iii) Résolution d’une nouvelle équation quadratique : αη 2 + 2βη + γ = 0, dont les
2 , U
coefficients α, β et γ dépendent de g, Cref e Φref et δ U
e H et qui donne invariablement des
racines réelles η1 et η2 . Choix d’une valeur adéquate de η distante de 0.05 |η2 − η1 | de η1
ou η2 rendant le polynôme ci-dessus faiblement positif.
(iv) Résolution d’une dernière équation quadratique : Aξ 2 + 2B(η)ξ + C(η) = 0,
dont le coefficient A dépend de Cref 2 et Ue Φref , et B(η) et C(η) sont des fonctions de η, g,
2 , U
Cref e Φref et δ Ue H , et qui donne des racines réelles compte tenu du choix de η. Choix de
la racine (réelle) ξ qui permet de progresser dans la bonne direction.
(v) Corrections à l’itération courante : δλ = g + ξ et δ U e = ηδ U e H + ξUe Φref .
Le paramètre de pondération η permet d’éliminer au maximum la contribution du terme
orthogonal néfaste au bon déroulement du calcul et évite l’apparition répétée de racines
complexes dans les itérations ultérieures (contrairement à d’autres méthodes plus brutales).
Le paramètre ξ joue, quant à lui, le rôle du nouveau paramètre de contrôle et permet
d’avancer de la longueur d’arc voulue.
[· · · ]
Cette méthode de longueur d’arc modifiée s’applique également

au cas des déplacements
ref 2

ref 2
imposés en remplaçant Φref par KT U , C
e e par U et U
ref
e Φref par U e U ref =
−1
ref
−K
e K e U .
e
T T
La méthode de pilotage par longueur d’arc implantée dans le programme élément fini
répond de manière unifiée aux conditions de forces et déplacements imposés. Accompagnée
d’un contrôle automatique du pas de temps, elle donne de très bons résultats de convergence
et par conséquent des temps de calcul raisonnables, même pour les problèmes réputés les
plus critiques.
3.3.2.2 Méthodes de branchement La gestion des points de bifurcation rencontrés

au cours du calcul incrémental s’effectue au moyen de méthodes de branchement, qui per-
mettent de détecter les points singuliers, de les identifier en tant que points de bifurcation
éventuels, et de bifurquer sur la branche souhaitée [RIK, 1991] [SEY, 1994]. La méthode
implantée dans le programme élément fini suit les quatre étapes suivantes :
1. À la fin de chaque incrément, la matrice raideur tangente courante est factorisée (par
la méthode de Crout) de façon à évaluer son degré de singularité et détecter le passage
d’un ou plusieurs points critiques. Le nombre de points critiques traversés sur la branche
courante est directement lié à la variation du nombre de valeurs propres négatives de KT ,
c’est-à-dire du nombre de pivots négatifs de la matrice factorisée.
2. Chaque point critique détecté doit être isolé afin d’identifier sa nature, à savoir s’il
s’agit d’un point limite ou d’un point de bifurcation. Dans le cas d’un vecteur force de
référence, un moyen simple de distinguer un point limite d’un point de bifurcation repose
sur l’estimation du paramètre de raideur introduit par Bergan et al. [BER, 1978] :
n oT n o
Φe ref Ue Φref
k=n oT n o (100)
Ue Φref Ue Φref
Le signe de k est supposé changer au passage d’un point limite, et non au passage d’un
point de bifurcation. Le cas du contrôle en déplacements se traite de manière similaire
104
avec :
e Uref T U
n o n o
K T
e U ref
k=− n oT n o (101)
U
e U ref Ue U ref
3. S’il s’agit d’un point de bifurcation, on redéfinit l’incrément courant en visant la

proximité de ce point pour pouvoir ultérieurement bifurquer sans mal sur l’une ou l’autre
des branches qui en sont issues. L’estimation du point critique (pour lequel la valeur propre
considérée s’annule) s’effectue par interpolation de la longueur d’arc entre les incréments
précédent et courant ou, plus précisément, par une technique de dichotomie.
4. Enfin, l’utilisation d’une prédiction modale permet l’aiguillage vers la branche bi-
furquée [SEY, 1994] : le mode propre Z e associé à la valeur propre nulle, solution de
KeTZe = 0, est calculé numériquement et la première itération du premier incrément post-
e
critique est basée sur les prédictions suivantes :
e = ± Z
e
δλ = 0 δU (102)
Z
e
Les méthodes de calcul de solutions post-flambées basées sur l’introduction de défauts

dans la modélisation numérique permettent uniquement de s’intéresser aux modes pri-
maires, qui plus est au premier mode, et souffrent intrinsèquement d’un manque de rigueur
du fait de la méconnaissance des défauts adéquats et de leur influence sur le comporte-
ment global de la structure. Les méthodes de branchement, plus systématiques, autorisent
ici la recherche complète de toutes les solutions du problème (stables ou instables) et en
particulier des modes secondaires, effectivement rencontrés en pratique dans diverses ap-
plications courantes. De plus, comme évoqué précédemment, elles fournissent une solution
de référence non dégénérée du problème de bifurcation idéalisé.
3.3.3 Validation du programme élément fini
Les travaux dans ce domaine ont donné lieu à un certain nombre d’applications [S1 ]
dont les plus remarquables vont être discutées dans la suite. Avant de présenter les résultats
obtenus en termes de bifurcation et post-bifurcation pour les deux géométries fondamen-
tales que sont la poutre et le cylindre, voici deux solutions relatives au phénomène de
claquage, qui vont permettre de valider dans un premier temps la robustesse de notre
formulation élément fini et la méthode de longueur d’arc retenue.
On considère premièrement un cylindre élastoplastique pincé par deux forces ponc-
tuelles en son milieu et bloqué radialement au niveau des deux extrémités. Ce problème
est reconnu comme particulièrement critique s’il est mené jusqu’au contact des deux points
sollicités (fermeture du cylindre), dans la mesure où il met en jeu de grandes rotations et
surtout un claquage abrupt qui coı̈ncide avec la formation d’un pli longitudinal.
Les déformées successives obtenues (figure 37) concordent parfaitement avec les
résultats antérieurs de la littérature, ce qui constitue une bonne validation de notre
élément fini dans son comportement en grandes rotations, le tout en élastoplasticité, avec
la détection et le franchissement d’un point limite relativement brutal.
La validation de notre méthode de longueur d’arc a été plus spécialement effectuée
sur le problème d’un panneau cylindrique sous force ponctuelle (au centre), avec les bords
rectilignes simplement appuyés tandis que les arêtes curvilignes sont libres, en élasticité.
Sous certaines conditions sur les paramètres géométriques (une épaisseur suffisamment
105
Figure 37 – Déformées successives d’un cylindre élastoplastique pincé
faible), ce problème fait apparaı̂tre des maxima de charge et de déplacement successifs

(autrement dit, un ”snap-back”) qui empêche l’utilisation d’un pilotage aussi bien en forces
imposées qu’en déplacements imposés.
On effectue le calcul avec deux pilotages par longueur d’arc, l’un contrôlé en force,
l’autre en déplacement. Les deux méthodes permettent de décrire le phénomène complet
de claquage avec très peu d’incréments et conduisent aux mêmes résultats (figure 38).
3.3.4 Poutres élastoplastiques
Les premiers résultats numériques de bifurcation concernent le problème de la poutre

comprimée [P4 ]. De par la simplicité relative de la géométrie et du problème dans son
ensemble, de nombreuses solutions analytiques sont disponibles dans la littérature, et la
démarche présentée précédemment a permis de les retrouver et de les compléter à divers
titres. Comme évoqué à cette occasion, les solutions obtenues analytiquement sont ce-
pendant restreintes à la valeur critique, au mode de bifurcation et au post-flambement
immédiat. On a en effet montré que la validité du développement asymptotique de la
branche bifurquée était limitée à de très faibles déflexions, en particulier à cause de
l’évolution rapide de la forme de la zone de décharge. La modélisation par éléments fi-
nis semble alors la seule alternative pour traduire le comportement en post-flambement
jusqu’à un stade post-critique plus avancé. Les résultats présentés ci-dessous ont donc
comme principal intérêt de décrire la totalité des courbes de réponse, sans hypothèses res-
trictives, et de mettre en valeur l’effet des non-linéarités géométriques sur le comportement
de la poutre. Dans le même temps, les méthodes de bifurcation permettent de démontrer
numériquement certaines propriétés intrinsèques d’un problème de flambement plastique,
comme l’existence de continua de points de bifurcation.
On choisit de s’intéresser à une poutre de section rectangulaire particulièrement mince,
de façon à pouvoir utiliser notre élément fini coque. Un seul élément fini suffit dans le
106
Figure 38 – Panneau cylindrique sous force ponctuelle : relation force(kN)-
déplacement(mm) au centre
sens de la largeur, alors que 50 éléments sont retenus suivant la longueur pour représenter
correctement les rotations finies durant le post-flambement. On utilise, sauf exception, 4
points de Gauss par élément (2 dans chaque direction du plan de l’élément), et 4 points de
Gauss selon l’épaisseur. La section de la poutre étant petite, on affecte à la limite élastique
du matériau une valeur particulièrement faible pour permettre de plastifier la poutre avant
d’avoir atteint la charge critique du module tangent. Enfin, les modules élastique et tangent
sont choisis suffisamment proches l’un de l’autre de façon à séparer les premiers continua
de points de bifurcation, ce qui facilite l’analyse du problème.
En plasticité, à chaque mode de bifurcation correspond un continuum qui s’étend de
la valeur critique du module tangent à la valeur critique élastique en passant par la va-
leur critique du module réduit pour laquelle le flambement s’effectue à charge constante.
Pour une poutre simplement appuyée, ces trois valeurs critiques (pour le premier mode de
flambement) s’écrivent dans l’ordre croissant :
π 2 ET I π 2 ER I π 2 EI
λT = λR = λE = (103)
L2 L2 L2
avec le module réduit qui vaut, dans le cas d’une section rectangulaire :
4EET
ER = √ √ 2 (104)
E + ET
Chaque point du continuum est potentiellement un point de bifurcation dont la solution

post-bifurquée en vitesses initiales met en jeu une zone de décharge dont la forme et la
taille dépendent de la charge critique. À la valeur critique du module tangent, le concept
de ”solide élastique de comparaison” permet d’utiliser la même approche qu’en élasticité :
la charge critique λT correspond bien à une singularité (valeur propre nulle) de la matrice
107
raideur tangente (on observe la même absence de décharge dans le comportement pré-
critique et post-critique immédiat) et la branche bifurquée est précisément prédite par le
mode propre associé. En revanche, si l’on souhaite bifurquer à partir d’un autre point du
continuum, la singularité est bien présente, mais pour la matrice raideur tangente intégrant
la zone de décharge à la bifurcation (qu’on ne connaı̂t pas a priori), différente de la matrice
raideur tangente calculée sur la branche fondamentale d’équilibre. En pratique, on ne peut
donc détecter que la valeur critique du module tangent. Par contre, il est possible de
s’arrêter sur un point quelconque du continuum et de bifurquer à partir de ce point avec
une prédiction approximative basée sur le mode propre correspondant à la plus petite
valeur propre de la matrice raideur tangente en valeur absolue.
Cette technique a permis de tracer le réseau complet des branches bifurquées à partir de
l’ensemble des points du continuum pour le premier mode de flambement (figure 39). Les
résultats numériques obtenus sont conformes à la théorie, dans le sens où la bifurcation
s’effectue à charge croissante pour λc ∈ [λT , λR [ et à charge décroissante pour λc ∈
] λR , λE ].
Dans la zone de post-flambement initial (petites déformations), toutes les branches
bifurquées convergent vers la même asymptote λ = λR , comme le prédit la solution
analytique globale d’El Koulani [KOU, 1996] (figure 39(a)). Un résultat intéressant ap-
paraı̂t dans la zone post-critique avancée (figure 39(b)), à savoir que les branches bi-
furquées tendent toutes vers une nouvelle asymptote. Ce comportement est certainement
dû aux non-linéarités géométriques qui prévalent sur les non-linéarités matérielles lorsque
les déformations deviennent plus importantes.
(a) Post-flambement initial (b) Post-flambement lointain
Figure 39 – Diagramme de bifurcation élastoplastique d’une poutre comprimée (premier

mode)
En complément des résultats précédents, intéressons-nous désormais aux zones de

décharge dans la poutre. On considère toujours le cas d’une poutre simplement appuyée
à section rectangulaire, mais en inversant le rôle de la largeur et de l’épaisseur. On force
la coque à fléchir dans son plan en bloquant tous les déplacements hors plan, de façon
à visualiser plus facilement l’allure des zones de décharge durant le flambement. Le nou-
veau maillage contient seulement 30 éléments suivant la longueur et 4 éléments suivant la
hauteur, avec 7 points de Gauss par direction dans le plan (toujours pour une meilleure vi-
sualisation) et seulement 2 points de Gauss pour l’intégration numérique dans l’épaisseur
puisque le comportement est supposé uniforme suivant cette direction. En théorie, les
points en charge plastique à un instant donné sont ceux qui affichent des vitesses de
déformation plastique instantanées strictement positives. De par l’approche incrémentale
108
du calcul, les zones de charge plastique et de décharge élastique sont identifiées de manière
discrète. Puisque la zone de décharge ne fait que croı̂tre (au moins au voisinage du point de
bifurcation), les points en charge plastique à l’incrément n sont détectés numériquement
par comparaison des déformations plastiques équivalentes entre les incréments n et n + 1.
À la charge critique du module tangent, la zone de décharge s’étend à partir d’un point
au fur et à mesure qu’on avance sur la branche bifurquée. En appliquant la procédure
ci-dessus, les zones de décharge obtenues par éléments finis sont en très bonne adéquation
avec les résultats théoriques pour un même niveau de chargement (figure 40).
Figure 40 – Zones de charge plastique/décharge élastique dans une poutre comprimée sur
la branche bifurquée issue de la valeur critique du module tangent (premier
mode)
À la charge critique du module réduit, la frontière entre les zones plastiques et élastiques
est théoriquement horizontale et n’évolue pas sur la branche bifurquée, tout au moins dans
un premier temps. La figure 41 compare les zones de décharge obtenues analytiquement et
numériquement aux points de bifurcation aux charges critiques du module réduit des deux
premiers modes. La frontière est bien horizontale (par morceaux, dans le cas du second
mode) et les valeurs théoriques de l’ordonnée de cette frontière coı̈ncident avec les résultats
numériques avec une erreur relative de moins de 1% et 5% respectivement pour les modes
1 et 2.
3.3.5 Poutres à parois minces
Le programme élément fini a contribué à l’analyse du flambement plastique de poutres

à sections minces, réalisée dans le cadre d’une collaboration avec l’Université Technologique
de Lisbonne. Les solutions numériques obtenues ont servi de validation aux résultats d’un
modèle basé sur une théorie de poutre généralisée [P11 ,C4 ]. Contrairement aux poutres
de section pleine qui mettent en jeu uniquement un comportement global de type poutre
à l’échelle de la structure, les poutres creuses (à parois minces) subissent des déformations
localisées, à l’échelle de la section, qui s’ajoutent aux déformations ”macroscopiques” glo-
bales de la poutre. Un dimensionnement correct de ce type de poutre nécessite la prise en
compte de ces effets et ne peut s’effectuer à l’aide d’un modèle de poutre classique. Une
alternative à l’utilisation d’un modèle de dimension plus élevée réside en la théorie des
109
Figure 41 – Zones de charge plastique/décharge élastique dans une poutre comprimée à
la bifurcation au module réduit (modes 1 et 2)
poutres généralisées dont le principe consiste à décrire à la fois le comportement global

et local par un modèle de poutre enrichi. Aux champs macroscopiques (déplacements et
rotations) s’ajoutent des degrés de liberté relatifs aux différents modes de déformation
de la section identifiés préalablement : des modes naturels (extension, flexion, torsion et
distorsion) et des modes de déformation locale des parois.
Cette approche permet, par le biais d’éléments finis poutres spécifiques, ou le plus
souvent par des solutions analytiques ou semi-analytiques, de traduire correctement la
réponse de ces poutres à diverses sollicitations. L’utilisation de notre élément fini coque a
permis de valider la méthode dans le cadre du flambement plastique, sur deux exemples,
en décrivant explicitement la géométrie des sections.
La première application est une poutre simplement appuyée avec une section en C.
Le matériau (alliage d’aluminium) est caractérisé par une loi de Ramberg-Osgood. La
figure 42 montre une bonne adéquation des contraintes critiques issues du présent modèle
avec celles obtenues par la théorie des poutres généralisées, autant en élasticité qu’en
plasticité (pour cet exemple, la théorie incrémentale et la théorie de la déformation, en
plasticité, fournissent des résultats similaires), avec une erreur maximale de l’ordre de 7%.
La figure 43, quant à elle, décrit les déformées modales obtenues avec des éléments finis
coque, d’une part, et reconstruites à partir des solutions de type poutre, d’autre part, pour
différentes longueurs. Selon la longueur, la poutre est susceptible de flamber suivant un
mode combinant des contributions locale et distortionnelle symétrique (L = 15.8 cm), un
mode purement distortionnel symétrique comprenant une, deux ou trois demi-ondes (pour
110
L = 33.6 cm, L = 55 cm et L = 92 cm, respectivement), ou un mode couplant de la
distorsion, de la flexion et de la torsion (L = 104.3 cm).
Figure 42 – Contraintes critiques de bifurcation plastique d’une poutre à section en C
La seconde application concerne une poutre à section rectangulaire creuse en acier

inoxydable. Les figures 44 et 45 fournissent également une comparaison des contraintes
critiques et des modes de flambement plastique, respectivement. Avec cet exemple, les va-
leurs obtenues par l’approche de type coque sous-estiment plus largement (avec une erreur
de l’ordre de 19% au maximum) les résultats de poutre avec la théorie incrémentale, mais se
positionnent malgré tout au-dessus des résultats relatifs à la théorie de la déformation. Ces
écarts s’expliquent par des différences évidentes de modélisation. Alors que la théorie des
poutres permet de définir rigoureusement l’état de contrainte uniaxial ainsi que les condi-
tions aux limites souhaitées, la modélisation de type coque fait naturellement apparaı̂tre
des concentrations de contraintes qui agissent comme autant d’imperfections susceptibles
de diminuer fortement les valeurs critiques de flambement plastique. De plus, l’approche
incrémentale utilisée rend compte de l’effet des déformations pré-critiques (absentes de
l’approche linéarisée retenue dans le cadre de la théorie des poutres généralisées).
La figure 46 illustre la sensibilité aux imperfections d’un tel problème (pour L = 15 cm)
en comparant la valeur critique de poutre (GBT) avec le tracé de la courbe d’équilibre ob-
tenue avec le modèle de coque. Un point de bifurcation est obtenu numériquement, malgré
la présence de défauts qui se manifestent par une déviation de la courbe fondamentale
d’équilibre par rapport à la verticale. L’influence des défauts est d’autant plus importante
que le flambement est instable (le point de bifurcation est immédiatement suivi d’un point
limite).
3.3.6 Cylindres sous compression axiale
Le cylindre sous compression axiale fait partie des problèmes de flambement les plus
critiques, dans la mesure où il est fortement sensible aux défauts de toutes sortes. Du fait de
l’intérêt qu’il suscite en pratique, on trouve un large panel de résultats expérimentaux dans
la littérature, pour différentes géométries, différents matériaux et différentes conditions aux
limites. En parallèle, de nombreuses solutions numériques ont été proposées, pour pallier à
l’insuffisance des résultats analytiques disponibles. En effet, les rares solutions analytiques
111
Figure 43 – Comparaison des déformées modales d’une poutre à section en C pour
différentes longueurs
concernent majoritairement les modes axisymétriques, alors qu’en pratique d’autres types
de modes sont observés, même en plasticité, dépendant fortement des conditions aux li-
mites et du mode d’application du chargement. Indépendamment du mode, la solution
post-flambée d’un cylindre sous compression axiale est la plupart du temps très instable,
ce qui explique sa grande sensibilité vis-à-vis des imperfections. La charge critique de bi-
furcation ne donne en général qu’une estimation grossière de la charge limite applicable
à la structure, à laquelle s’initie réellement la perte de stabilité et donc la ruine de la
structure. Pour ce problème, même en l’absence de défauts, la solution post-critique ap-
porte infiniment plus d’informations que la simple connaissance du mode et de la valeur
critique de flambement. L’évolution des déformées post-critiques montre par exemple sou-
vent des phénomènes de localisation, correspondant parfois à des modes secondaires. Pour
toutes ces raisons, la nécessité d’un outil de calcul numérique est indéniable si l’on veut
résoudre et analyser correctement ce type de problème. L’objectif des travaux numériques
réalisés autour du flambement plastique de cylindres sous compression axiale est de re-
trouver, avec le programme développé, l’ensemble des résultats remarquables obtenus par
nos prédécesseurs. Une étude paramétrique a été menée en faisant varier plusieurs ca-
ractéristiques du problème (géométriques, matérielles, ou ayant trait aux conditions aux
limites au sens large), avec la volonté de classifier les solutions obtenues.
112
Figure 44 – Contraintes critiques de bifurcation plastique d’une poutre à section rectan-
gulaire creuse
Figure 45 – Comparaison des déformées modales d’une poutre à section rectangulaire

creuse pour différentes longueurs
Du fait de la forte sensibilité aux imperfections reconnue pour ce problème, on relève

d’importants écarts entre les résultats numériques et expérimentaux (la charge critique
expérimentale est bien souvent diminuée de moitié par rapport à la valeur théorique).
De nombreux auteurs ont tenté d’expliquer ces différences. D’un point de vue matériel,
les résultats de la théorie de la déformation s’écartent nettement de ceux de la théorie
incrémentale, et semblent mieux s’accorder avec les résultats expérimentaux. L’utilisa-
tion d’une surface seuil avec coins [CHR, 1979] réalise un compromis entre les deux
méthodes précédentes, en se rapprochant des valeurs expérimentales sans trahir la nature
intrinsèque de la plasticité. Avec un tel modèle phénoménologique, Tvergaard [TVE, 1983]
a analysé l’effet des paramètres géométriques et matériels sur le flambement d’un cy-
lindre sous compression axiale. Il considère une imperfection initiale sous la forme d’un
113
Figure 46 – Courbe d’équilibre d’une poutre élastoplastique à section rectangulaire creuse
comprimée : sensibilité aux imperfections
déplacement sinusoı̈dal axisymétrique et cherche une bifurcation possible vers un mode

non-axisymétrique. Lorsque le cylindre est relativement mince, il trouve effectivement
un point de bifurcation suivi de près par un point limite. Pour des cylindres plus épais,
la charge limite est atteinte et s’accompagne d’une localisation du mode axisymétrique.
La bifurcation est retardée ou disparaı̂t complètement (la ruine devient axisymétrique à
partir d’une certaine épaisseur). Plus tard, une analyse de cette transition entre modes
axisymétrique et non-axisymétrique a été effectuée par Goto et Zhang [GOT, 1999] sur
un cylindre parfait avec la théorie incrémentale. Ils ont montré que l’épaisseur du cy-
lindre et les conditions aux limites sont déterminantes quant à l’obtention de certains
modes. Avec des bords encastrés et un cylindre modérément épais, ils obtiennent deux
points de bifurcation successifs, le premier menant à un mode axisymétrique et le second
à un mode non-axisymétrique localisé. Lorsque la coque est suffisamment mince, les deux
points de bifurcation coalescent. Parmi les travaux récents dans ce domaine, on peut citer
Bardi et al. [KYR, 2005] [BAR, 2006a] [BAR, 2006b] qui ont réalisé à la fois des me-
sures expérimentales et des calculs numériques, en comparant les résultats fournis par les
deux théories de la plasticité. La meilleure corrélation a été obtenue pour la théorie de
la déformation avec un défaut axisymétrique d’amplitude faible devant l’épaisseur. Mais
aucun résultat de bifurcation n’est analysé et certains choix de modélisation, relevant du
cas par cas, prêtent encore à discussion.
Dans ce travail, malgré l’influence significative des défauts sur le comportement d’un
cylindre sous compression axiale, on traite le problème, davantage singulier, d’un cylindre
parfait, afin d’établir des résultats de référence (courbes squelettes) à partir desquelles on
peut ensuite quantifier l’effet des défauts éventuels. Des tests préliminaires de convergence
ont abouti à un maillage optimal de seulement 200 éléments pour la portion de cylindre
modélisée. Avec ce maillage, le programme numérique a d’abord été validé par comparaison
aux solutions analytiques présentées précédemment. Une grande variété de modes a été
ensuite obtenue en faisant varier l’ensemble des paramètres du modèle. Les résultats les
plus originaux concernent l’obtention de modes secondaires dans des conditions spécifiques
[P2 ,P7 ,C2 ,C19 ].
114
Une étude préliminaire a montré que les grandeurs déterminantes pour l’obtention d’un
type de mode donné sont le rapport rayon sur épaisseur, le rapport longueur sur rayon
influençant plus faiblement la nature du mode, et le module d’écrouissage (davantage que le
module élastique). L’analyse paramétrique a donc porté sur l’épaisseur (à rayon constant),
le module d’écrouissage isotrope, ainsi que les conditions aux limites au sens large (bords
libres, simplement appuyés ou encastrés, déplacements ou forces imposées) afin d’exhiber
la plus grande variété de modes possible. Les résultats numériques sont obtenus à partir
des propriétés géométriques et matérielles des tableaux 9 et 10.
Longueur l Rayon R Épaisseur t

5m 1.5 m 7 cm, 5 cm, 3 cm, 2 cm, 12 mm, 6 mm
selon le cas considéré
Tableau 9 – Paramètres géométriques considérés pour le flambement plastique d’un cy-

lindre sous compression axiale
Module Coefficient Limite Module d’écrouissage

d’Young E de Poisson ν élastique σ0 isotrope h
2.1 1011 P a 0.3 4.108 P a 2.10 P a, 5.1010 P a, 5.108 P a
10
selon le cas considéré
Tableau 10 – Propriétés matérielles considérées pour le flambement plastique d’un cy-

lindre sous compression axiale
Dans le cas des bords libres, en forces imposées, on obtient un mode axisymétrique, ou
alors non-axisymétrique localisé aux extrémités (sous la forme de fossettes vers l’intérieur
du cylindre), dont le nombre d’ondes circonférentielles augmente lorsque l’épaisseur di-
minue (comme le montre la figure 47). Le mode d’application de la charge influence
considérablement la réponse lorsque les bords sont libres. Les modes non-axisymétriques
précédents, lorsque les extrémités sont libres de tout mouvement, sont généralement rem-
placés par des modes axisymétriques lorsqu’on impose des déplacements longitudinaux sur
l’ensemble de la circonférence aux extrémités, ce qui empêche les ondes circonférentielles
localisées de se développer. Les courbes force-déplacement montrent que, pour les valeurs
choisies, la plastification (complète) du cylindre précède tout juste la bifurcation, elle-même
suivie de près par un point limite synonyme de ruine pour la structure.
Les conditions de bords encastrés sont plus proches des conditions expérimentales
classiques. Le module d’écrouissage joue ici un rôle primordial : on choisit tout
d’abord la valeur la plus importante de façon à privilégier l’obtention de modes non-
axisymétriques. Ces modes sont sinusoı̈daux selon les directions à la fois longitudinales
et circonférentielles (modes ”diamant”), mais les fossettes se manifestent souvent davan-
tage près des extrémités, sans qu’on puisse cependant parler de localisation. Les nombres
d’ondes axiales et circonférentielles augmentent tous deux lorsque l’épaisseur diminue (voir
la figure 48). Les courbes d’équilibre ont la même allure que dans le cas des bords libres
(figure 49).
Lorsque le module d’écrouissage est plus faible (de l’ordre de grandeur des valeurs ren-
contrées en pratique), le mode de flambement est axisymétrique (c’est la solution la plus
courante en plasticité) et immédiatement s’ensuit une localisation de la déformée (seuls
s’affirment les deux ”bourrelets” aux extrémités alors que les autres s’estompent, comme
en élasticité) qui coı̈ncide avec le passage de la charge limite. Si l’épaisseur du cylindre est
115
(a) t = 7 cm (b) t = 5 cm
Figure 47 – Modes de bifurcation non-axisymétriques d’un cylindre élastoplastique sous

compression axiale (bords libres, forces imposées)
Figure 48 – Modes primaires (”diamant”) d’un cylindre élastoplastique sous compression

axiale (bords encastrés, forces imposées)
suffisamment importante, la ruine est alors axisymétrique. Pour des épaisseurs modérées
(lorsque l’épaisseur est trop faible, le mode est non-axisymétrique, mais surtout élastique),
la bifurcation vers un mode non-axisymétrique survient de manière retardée, après locali-
sation axisymétrique. On a alors affaire à un mode secondaire (bifurcation sur une branche
post-bifurquée) appelé ”patte d’éléphant” du fait de sa forme spécifique. Il est obtenu ma-
joritairement dans le cas de déplacements imposés qui facilitent la localisation du mode
axisymétrique. Son nombre d’ondes circonférentielles dépend toujours essentiellement et
de la même façon du rapport rayon sur épaisseur (voir la figure 50). Pour la coque la plus
fine considérée (donnant lieu à un mode secondaire), on relève un comportement de type
snap-back, comme le montre la figure 51.
Le cas des bords simplement appuyés donne des résultats similaires aux bords en-
castrés, sinon qu’il ne fait pas l’objet de modes secondaires. Avec un module d’écrouissage
suffisamment fort, on obtient également des modes non-axisymétriques, mais beaucoup
116
Figure 49 – Cylindre élastoplastique sous compression axiale (bords encastrés, forces im-
posées) : courbes d’équilibre (modes primaires)
Figure 50 – Modes secondaires (”patte d’éléphant”) d’un cylindre élastoplastique sous

compression axiale (bords encastrés, déplacements imposés)
plus localisés aux extrémités (surtout en forces imposées). Ces résultats traduisent le rap-
port intuitif entre le raidissement de la structure (par blocage des rotations ou contrôle
des déplacements longitudinaux) et le nombre de fossettes dans la déformée modale. Dans
le cas des déplacements imposés, certaines fossettes apparaissent également dans le com-
portement post-critique, simultanément avec le passage de points limites dans la courbe
force-déplacement.
117
Figure 51 – Cylindre élastoplastique sous compression axiale (bords encastrés,
déplacements imposés) : courbes d’équilibre (modes secondaires)
Le tableau 11 résume les différents types de modes et de comportements post-critiques

observés selon l’épaisseur du cylindre et les conditions aux limites.
3.3.7 Cylindres sous pression externe
Après le cylindre sous compression axiale, on considère le cas du cylindre sous pres-
sion externe, qui induit également des contraintes de compression (non pas axiales mais
circonférentielles) et pose donc des problèmes d’instabilité. Contrairement aux applica-
tions précédentes, très académiques, on s’intéresse ici particulièrement au comportement
d’enceintes sous-marines soumises à la pression hydrostatique, c’est-à-dire à des tubes cy-
lindriques sollicités à la fois par une pression externe et une compression axiale (due à l’effet
de fond, mais qui sera négligée en pratique), dans le cadre d’une collaboration avec l’IFRE-
MER [P10 ,C14 ]. Devant les écarts relevés entre les pressions limites expérimentales et
les valeurs critiques calculées par éléments finis de manière linéarisée en élasticité, l’objec-
tif de ce travail est d’améliorer les prédictions effectuées en tenant compte à la fois de la
plasticité, des imperfections géométriques et des contraintes résiduelles. Ces trois aspects,
individuellement les uns des autres, ont déjà fait l’objet de bon nombre d’études relatives au
flambement. L’effet des contraintes résiduelles sur la charge critique d’un cylindre sous com-
pression axiale a été par exemple mis en évidence par des essais de flambement sous poids
propre (ce qui permet de libérer les déplacements longitudinaux à l’extrémité supérieure du
cylindre et donc les contraintes axiales internes) [MAN, 2000] [LAN, 2000]. Concernant les
cylindres sous pression externe, l’étude de différentes imperfections géométriques a montré
que l’eccentricité ou les variations d’épaisseur ont une influence secondaire par rapport à
l’ovalité [YEH, 1986] [YEH, 1988]. Peu de travaux à ce jour considèrent l’influence de ces
différents aspects combinés dans une seule et même modélisation. On peut citer l’ouvrage
de Kyriakides et Corona [KYR, 2007] sur le flambement de pipelines offshore sous divers
chargements, en présence d’imperfections incluant les contraintes résiduelles.
118
Mode non-axisymétrique
Mode axisymétrique
sinusoı̈dal amorti
sinusoı̈dal amorti
intermédiaires)
(Résultats
Bords
libres
Mode axisymétrique (+ localisation)

Mode secondaire ”patte d’éléphant”
Post-bifurcation : légère localisation
Mode axisymétrique sinusoı̈dal
encastrés
”diamant”
Bords
Mode
Mode axisymétrique sinusoı̈dal
Post-bifurcation : localisation
Bords simplement
intermédiaires)
”diamant”
(Résultats
appuyés
Mode
Modérément
Cylindres
Mince
Épais
épais
Tableau 11 – Modes de bifurcation et comportements post-critiques d’un cylindre sous

compression axiale en fonction de l’épaisseur et des conditions aux limites
Compte tenu des non-linéarités matérielles et géométriques interagissant dans ce

problème, une analyse par éléments finis s’avère indispensable pour bien cerner le com-
portement complet de telles structures, en plasticité, avec tous les défauts envisagés, et
prédire les points de bifurcation éventuels (en l’absence de défauts) et surtout les pressions
limites menant à la ruine. Dans ce cas de figure (pour un cylindre sous pression externe),
119
le choix de la théorie de plasticité n’influence pas les résultats : la théorie incrémentale
retenue est donc censée donner des résultats en accord avec les valeurs expérimentales, dès
lors que les conditions expérimentales sont bien représentées.
On considère le cas particulier d’un tube cylindrique dont les dimensions et les pro-
priétés matérielles (aluminium) ont été mesurées et sont listées dans les tableaux 12 et
13, encastré aux extrémités (pour tenir compte de la présence des fonds) et soumis à une
pression externe uniforme.
Longueur l Rayon R Épaisseur h

970 mm 53.375 mm 6.75 mm
Tableau 12 – Paramètres géométriques du tube cylindrique sous pression externe
Module Coefficient Limite Module d’écrouissage

d’Young E de Poisson ν élastique σ0 isotrope H
71000 M P a 0.32 365 M P a 625 M P a
Tableau 13 – Propriétés matérielles du tube cylindrique sous pression externe
Un tel tube a été testé expérimentalement jusqu’à la ruine par implosion, observée
à une pression de 366 bars. Les déformations locales ont été mesurées tout au long du
chargement au moyen de jauges extensométriques axiales et circonférentielles placées dans
la partie centrale du tube, susceptible de se déplacer davantage. Les réponses des jauges
sont toutes identiques et linéaires jusqu’à environ 250 − 300 bars, puis elles divergent les
unes des autres, caractérisant ainsi l’ovalisation progressive de la section centrale, bien
avant le flambement à proprement parler. Cette ovalisation précoce est le signe d’une
ovalité initiale de la section idéalement circulaire, qui devra être prise en compte dans la
modélisation numérique.
Les contraintes résiduelles (axiales et circonférentielles) issues du procédé de fabrication
ont été mesurées par une démarche expérimentale spécifique. Celles-ci sont supposées ne
dépendre que de la position dans l’épaisseur. Les mesures sont donc effectuées sur un
échantillon (anneau circulaire) par enlèvement de matière successif. L’enlèvement d’une
couche de matière à l’extérieur de l’anneau relâche les contraintes résiduelles et conduit à
des déformations qui sont mesurées à l’intérieur de l’anneau au moyen de jauges axiales et
circonférentielles. Dans l’hypothèse de contraintes planes et d’un cylindre qui reste toujours
circulaire (sans ovalisation ni flexion), les contraintes dans le tube sont simplement déduites
des déformations mesurées par la loi linéaire élastique de Hooke. Les contraintes initiales
sont ainsi calculées de manière séquentielle et lissées pour une implantation numérique
plus aisée (figure 52).
Les résultats sont conformes à ceux de la littérature obtenus dans des conditions simi-
laires. Les contraintes résiduelles dans les deux directions sont négatives (en compression)
sur la partie interne et positives (en traction) sur la partie externe du cylindre. L’évolution
n’est pas linéaire dans l’épaisseur, du fait en particulier d’une légère diminution au niveau
des rayons interne et externe, due au procédé d’extrusion des tubes. L’amplitude maxi-
male des contraintes résiduelles est suffisamment large pour modifier significativement la
charge limite admissible du cylindre. Les contraintes les plus nocives sont bien entendu les
contraintes circonférentielles négatives, dans la moitié intérieure du cylindre, qui s’ajoute-
ront aux contraintes de compression dues au chargement extérieur.
120
(a) Contraintes résiduelles longitudinales (b) Contraintes résiduelles circonférentielles
Figure 52 – Variations des contraintes résiduelles dans l’épaisseur du cylindre sous pres-
sion externe
Plusieurs calculs sont effectués à l’aide de notre programme, en élasticité et en plasticité,

tout d’abord pour un cylindre parfait, puis en incluant les différents défauts, à savoir les
imperfections géométriques (en faisant varier le degré d’ovalisation initial qui n’a pu être
mesuré avec précision) et les contraintes résiduelles. En élasticité, sans aucun défaut, un
point de bifurcation super-critique est obtenu à la pression critique Pc = 39.55 M P a. Le
comportement post-critique stable, à charge croissante, corrobore la faible sensibilité aux
imperfections de ce problème en élasticité. Si la plasticité est prise en compte, elle ne se
manifeste, pour un cylindre parfait, que lors du post-flambement, dans la partie centrale,
là où les contraintes sont majorées du fait des encastrements aux extrémités qui jouent le
rôle de défauts intrinsèques (les contraintes ne sont pas parfaitement homogènes durant
le comportement pré-critique). La bifurcation est donc toujours élastique mais la réponse
post-bifurquée fait apparaı̂tre un point limite à la pression Pmax = 39.9 M P a qui constitue
le chargement limite admissible et mène à la ruine plastique du cylindre.
On considère ensuite un cylindre de section elliptique dont l’ovalité est caractérisée par
la différence de rayon des deux axes principaux de l’ellipse, qui varie arbitrairement de
∆R = 0 à 0.4 mm, en respectant les tolérances dimensionnelles des tubes extrudés de la
sorte. Lorsque le défaut considéré est suffisamment important, le problème est dégénéré
et ne fait plus apparaı̂tre la première singularité, à savoir le point de bifurcation. Les
résultats suivants concernent donc uniquement la plasticité, qui seule fait apparaı̂tre un
comportement post-critique instable (au-delà de la pression limite). La figure 53 montre
l’évolution de cette charge limite en fonction de la taille du défaut considéré (la pression
critique élastique de référence et la valeur expérimentale ont été ajoutées, à des fins de
comparaison). L’effet des contraintes résiduelles est enfin pris en compte. Dans le cas
présent, ces contraintes initiales agissent uniquement par l’augmentation de la taille de la
zone plastique qu’elles induisent (elles n’ont en particulier aucune influence en élasticité),
ce qui a pour effet de diminuer la pression limite admissible, d’autant plus que les défauts
géométriques sont importants. Les résultats associés sont également tracés dans la figure
53.
À titre de comparaison, la figure 54 confronte la déformée post-critique avancée du cy-
lindre sous pression externe calculée en plasticité à l’allure du tube testé expérimentalement
en fin d’essai. Si le mode de flambement en plasticité est classiquement le même qu’en
élasticité, les déformations post-critiques sont plus localisées en plasticité : elles s’accen-
tuent spécialement dans la partie centrale du tube, en charge plastique, ce qui concorde
avec les résultats expérimentaux.
121
Figure 53 – Pression externe limite d’un cylindre élastoplastique en fonction de l’ampli-
tude des imperfections géométriques, avec ou sans contraintes résiduelles
(a) Solution numérique en plasticité (b) Allure expérimentale
Figure 54 – Déformées post-critiques du tube cylindrique sous pression externe
Par l’ensemble de ces calculs, on a montré que chacun de ces aspects (plasticité, défauts
géométriques, contraintes résiduelles) pris séparément ne porte pas tellement préjudice à
la structure dans le cas présent. Compte tenu des paramètres géométriques et matériels,
122
en l’absence de défauts, le comportement est élastique (et donc peu sensible aux imper-
fections). Les contraintes résiduelles n’affectent pas non plus le comportement du cylindre
en élasticité. L’étude effectuée a permis inversement de montrer le rôle combiné significatif
de la plasticité, des défauts de forme et des contraintes initiales dans la réponse du cy-
lindre sous pression externe. La présence de défauts (géométriques) et/ou de contraintes
résiduelles donne lieu à des concentrations de contraintes qui entraı̂nent un comportement
plastique tout au moins local (compte tenu des grandeurs matérielles et géométriques
considérées) avec du coup une plus grande sensibilité aux imperfections et un effet avéré
des contraintes résiduelles, plus important en présence de défauts géométriques. Aucun
de ces aspects ne peut donc être négligé dans la modélisation puisqu’avec les trois cu-
mulés, la pression limite n’excède pas 75% de la charge critique élastique si on choisit
la plus forte amplitude de défaut considérée dans cette étude, qui représente seulement
0.75% du rayon moyen. Inversement, avec la plus faible amplitude de défaut considérée
(∆R = 0.05 mm), inévitable en pratique, la pression limite calculée coı̈ncide parfaitement
avec la valeur expérimentale relevée.
Cette étude est sur le point d’être poursuivie dans le cadre de tubes composites.
L’ensemble de mes travaux de recherche sur le flambement s’appuient sur le

développement de deux outils bien distincts : une méthode de résolution analytique et un
programme élément fini. La démarche employée pour résoudre analytiquement un problème
de flambement s’appuie sur la théorie 3D de la bifurcation plastique et exploite l’équation
de bifurcation écrite sous forme variationnelle. Dans l’hypothèse de faibles déformations
pré-critiques (pour pouvoir linéariser certains termes), on aboutit à des expressions simples
de la charge critique et du mode de flambement pour bon nombre d’applications qui
reposent sur l’utilisation éventuelle de modules réduits adaptés à la formulation choisie
(poutre, coque) et sur la définition d’états de pré-contraintes issus du chargement ap-
pliqué. L’obtention de formules explicites dépend de l’expression de la loi de comporte-
ment. Quelques hypothèses et développements supplémentaires permettent d’identifier la
pente de la branche bifurquée et d’évaluer ainsi la stabilité immédiate de la bifurcation. La
généralité de la méthode et la souplesse de l’approche variationnelle utilisée ont permis de
traiter avec succès des problèmes de poutres, de plaques et de cylindres, avec différentes
cinématiques et sous différentes sollicitations et conditions aux limites, mais aussi des cas de
matériaux composites. Si l’utilisation de la méthode peut s’étendre encore à de nouvelles
géométries, les premières perspectives concernent les deux applications composites. Le
problème de poutre mixte avec connexion partielle a été pour l’instant traité en élasticité.
La généralisation en plasticité (que ce soit avec les théories de Bernoulli ou de Timoshenko)
ne devrait pas poser de difficultés supplémentaires. Le problème du sandwich Napcor a été
pour l’instant traité avec la cinématique de Bernoulli pour les renforts. Il serait intéressant
de quantifier à quel point la cinématique de Timoshenko permet de se rapprocher de la
solution numérique (et donc expérimentale) où les renforts sont représentés en 3D.
Dans le cas le plus simple d’une poutre élastoplastique de Bernoulli comprimée, un
traitement particulier a permis de poursuivre le développement asymptotique de la solution
bifurquée. Les calculs effectués ont mis en évidence un nouveau terme (à l’ordre 9/5) vis-
à-vis des travaux antérieurs, mais ont montré que la validité d’un tel développement se
limitait aux très faibles déflexions, avant que la zone de décharge n’ait atteint les extrémités
de la poutre. Une description plus complète du post-flambement (pas seulement immédiat)
nécessite un outil de calcul numérique qui puisse intégrer dans sa formulation l’influence
des non-linéarités géométriques sur la réponse post-critique lointaine.
123
Un programme de calcul incrémental en grandes transformations élastoplastiques a été
réalisé dans l’optique de s’intéresser à des géométries ou chargements plus complexes et
d’analyser le problème d’instabilité jusqu’à un stade avancé de déformation. Il est basé
sur une formulation élément fini coque, qui s’avère un bon compromis pour représenter la
plupart des structures (minces ou élancées) sujettes au flambement. Le développement d’un
programme ”maison” a été motivé par la nécessité de disposer de méthodes numériques
performantes pour faire face aux singularités les plus sévères. Une méthode de longueur
d’arc mixte avec contrôle en force ou déplacement a été implantée dans le programme,
ainsi qu’une méthode de branchement pour la détermination systématique de l’ensemble
du squelette des courbes d’équilibre sans utiliser de défauts arbitraires. La formulation
numérique proposée, basée sur la théorie incrémentale de la plasticité, permet de décrire
de nombreux résultats de référence, mettant en jeu des modes de bifurcation primaires mais
aussi secondaires (à partir de points critiques sur les branches post-bifurquées primaires)
ainsi que des points limites et le phénomène de localisation éventuelle des déformées post-
critiques.
Les méthodes numériques développées pour ce programme ont été parallèlement
intégrées aux autres modules de NLFEAS pour permettre de nouvelles applications. En
particulier, un modèle élément fini membrane a été déduit de l’élément coque, et de nou-
velles lois de comportement hyperélastiques ont été introduites, de façon à élargir notre
champ d’application aux structures gonflables. Dans ce contexte, on a obtenu quelques
résultats où l’instabilité géométrique se traduit par la formation de plis en présence d’ef-
fets de compression localisée.
Dans le prolongement de tous ces travaux, le démarrage d’une thèse est prévu sur
l’étude analytique, numérique et expérimentale du flambement global et local de structures
sandwich simples (le cas du sandwich Napcor pourra être considéré ultérieurement) sous
différents types de chargement comme la compression plane ou la flexion. Outre la recherche
de solutions analytiques et la détermination de résultats numériques 2D ou 3D avec les
outils précédemment développés, l’idée serait de mettre en œuvre un modèle numérique
de coque enrichi, intégrant dans l’épaisseur les modes de déformation dus aux instabilités
locales.
124
Conclusions et perspectives
Les travaux présentés dans ce mémoire sont voués majoritairement à l’étude de l’in-
stabilité des structures mécaniques qui constitue la dernière et principale partie de ce
rapport. Les matériaux concernés sont successivement les matériaux métalliques et les
matériaux composites, ce qui m’a amené à m’intéresser respectivement à la plasticité et
à l’homogénéisation. Les deux premières parties de ce rapport traitent donc des travaux
menés en plasticité et autour de l’homogénéisation, qui ont donné lieu à des résultats origi-
naux dans différents cadres d’application (sur différents matériaux spécifiques) et soulèvent
de nombreuses questions qui représentent autant de perspectives. Ils permettent aussi et
surtout, en tant qu’outils de modélisation, de s’intéresser au flambement plastique (des
structures métalliques) et au flambement de structures hétérogènes.
Ces trois notions sont donc intimement liées, comme le prouvent les nombreux pro-
grammes informatiques développés, combinant ces composantes les unes avec les autres.
Parmi les programmes mis en œuvre au cours de ces années, on a déjà mis l’accent sur
les solveurs (programmes de calcul par éléments finis ou routines d’intégration) développés
à des fins scientifiques. D’autres outils informatiques ont été réalisés pour faciliter les
pré- et post-traitements dans différents types d’analyse. Il s’agit de programmes script de
modélisation géométrique et de maillage censés remplacer l’utilisation fastidieuse d’une
interface graphique lorsque la géométrie est trop complexe (et éventuellement évolutive au
cours du calcul) ou lorsque plusieurs calculs s’enchaı̂nent et doivent être post-traités de
manière groupée. Ce type d’outil répond à la demande des industriels en leur proposant
une modélisation automatique, simple et efficace, qui demande uniquement des paramètres
d’entrée et fournit des résultats directement exploitables.
Rappelons, en conclusion, les réalisations et résultats principaux correspondant à cha-
cune des trois composantes de mon activité de recherche. En plasticité, l’apport fonda-
mental de mon travail consiste en la formulation générale de l’intégration locale pour un
modèle de plasticité standard avec un critère de plasticité et un modèle d’écrouissage ar-
bitraires. De cette formulation ont découlé plusieurs programmes de calcul par éléments
finis qui se distinguent par le choix de la ”dimension” (3D, 2D) et donc du type d’élément
fini implanté et par la prise en compte ou non des grandes transformations. Outre le
flambement, ces programmes ont permis de modéliser de nombreux comportements et de
mettre en évidence certains phénomènes comme la striction ou le phénomène de rochet
sur des exemples académiques. Le travail le plus complet et le plus récent dans ce domaine
porte sur la modélisation des films polymères. L’apport principal dans ce contexte est
la construction d’un modèle original qui fait intervenir plusieurs phases avec de la plas-
ticité mais aussi de l’élasticité non-linéaire. Si la difficulté principale consiste encore en
l’intégration locale des équations de comportement du modèle, cette étude s’inscrit dans
une dimension plus globale de la modélisation du comportement de matériaux, par l’ana-
lyse micro-structurale des matériaux effectuée au préalable et dont s’inspire la définition
du modèle, et par la réalisation d’essais expérimentaux pour la calibration des paramètres
du modèle et la validation des résultats numériques.
Concernant l’homogénéisation, l’apport essentiel est la mise en œuvre d’outils de
modélisation et de calcul de modules effectifs pour des structures minces périodiques. Les
différentes approches retenues permettent toujours de faire appel à un code de calcul par
éléments finis standard sans développements d’éléments spécifiques. Tous les programmes
sont ainsi développés au sein d’Abaqus et se distinguent les uns des autres à deux niveaux :
• le pré-traitement : il consiste en majeure partie en la modélisation de la géométrie
et le maillage, qui nécessitent des développements spécifiques pour chaque nouveau
matériau (et donc chaque nouvelle cellule unité) considéré ;
125
• le calcul et le post-traitement : le choix des calculs par éléments finis à effectuer
(en particulier des conditions aux limites et des chargements) et la manière de post-
traiter les résultats dépend des modules effectifs recherchés.
Les applications traitées jusqu’alors concernent deux composites à renforts 3D : un
composite renforcé d’une structure interlock d’angle couche-à-couche et un sandwich à ren-
forts orthogonaux ou inclinés dans l’épaisseur. L’essentiel des calculs réalisés concerne, pour
l’instant, le comportement élastique linéaire et tend à identifier les modules effectifs d’un
modèle de plaque équivalente de Kirchhoff-Love (dans le contexte de l’homogénéisation
asymptotique) pour les structures minces hétérogènes que sont nos deux matériaux. De
premiers résultats ont été obtenus en parallèle sur l’évolution de ces modules en présence
de décohésion fibres/matrice.
Mes résultats les plus nombreux et les plus avancés concernent le flambement et le
post-flambement. Dans ce cadre, deux types d’outil ont été spécialement mis au point pour
contribuer à la résolution de problèmes mettant en jeu des instabilités géométriques. Une
démarche de calcul analytique basée sur la résolution de l’équation de bifurcation et repo-
sant sur diverses hypothèses simplificatrices (petites déformations pré-critiques, écrouissage
linéaire en plasticité) a permis d’aboutir à des expressions explicites de la charge critique,
du mode de flambement mais aussi de la pente initiale de la branche bifurquée, pour bon
nombre de problèmes de flambement plastique de poutres (pour différentes cinématiques),
de plaques, de cylindres, et même de structures composites (poutres mixtes avec connexion
partielle, mousse renforcée du sandwich Napcor ). En parallèle, un programme élément
fini coque a été mis en œuvre pour aller plus loin dans la description du comportement en
post-flambement de structures éventuellement plus complexes. La valeur ajoutée princi-
pale d’un tel programme vis-à-vis des codes disponibles dans le commerce est la procédure
de résolution, avec une méthode de longueur d’arc robuste et mixte (avec contrôle en
forces ou déplacements) et une méthode de branchement automatique qui permettent de
traiter de nombreux problèmes de flambement par claquage ou bifurcation dont certains
problèmes de référence fortement non-linéaires réputés comme particulièrement critiques
(snap-back, bifurcations secondaires). Presque tous les calculs numériques sont effectués
sur des structures parfaites, sans défauts d’aucune sorte. De tels calculs idéalisés sont in-
dispensables avant d’entamer une analyse de sensibilité aux imperfections, en fournissant
le squelette des courbes d’équilibre sur lequel les solutions en présence de défauts s’ap-
puieront. L’introduction future de défauts dans les modèles (comme cela a déjà été fait
dans le cas du cylindre sous pression externe avec des imperfections géométriques et des
contraintes résiduelles) n’est censée poser aucune difficulté supplémentaire, dans la mesure
où elle transforme un problème de bifurcation mathématiquement singulier en un problème
dégénéré a priori beaucoup moins critique. Le développement de ces deux approches ana-
lytique et numérique présente l’intérêt de pouvoir valider mutuellement les résultats de
chacune d’elles.
De nombreuses perspectives découlent des différents axes de recherche présentés. À
la fin de chaque partie du mémoire, j’ai déjà évoqué les perspectives immédiates propres
à chacune des thématiques, que je ne rappellerai pas ici. Je vais plutôt insister sur les
opportunités en termes de nouveaux sujets de recherche qui émergent des précédents tra-
vaux et de la combinaison de plusieurs des trois axes présentés que sont la plasticité,
l’homogénéisation et le flambement. En parallèle des travaux sur les composites à renforts
fibreux, des études sont menées sur des renforts secs. Du point de vue du comportement
mécanique, les premiers résultats ont mené à l’identification pour un renfort fibreux d’un
modèle macroscopique ”élastoplastique” à partir de courbes expérimentales de compression
uniaxiale. L’idée serait désormais de s’intéresser au renfort à l’échelle de la fibre pour tenter
de reconstruire le modèle macroscopique à partir d’une modélisation du renfort hétérogène.
Si l’homogénéisation périodique (3D) est aisée lorsque le comportement à l’échelle micro-
126
scopique (et donc macroscopique) est linéaire, l’introduction de non-linéarités pose un
certain nombre de difficultés, comme il a pu être constaté avec les premières tentatives de
modélisation de l’endommagement. Dans le cas d’un renfort sec, de nouvelles difficultés
proviennent essentiellement du fait que le milieu est non cohésif. Il existe alors diverses
façons de caractériser le comportement équivalent macroscopique du milieu fibreux qui
peut être représenté par un milieu de Cosserat périodique. Si dans certains cas, le milieu
homogène équivalent peut être considéré lui-même comme étant un milieu de Cosserat,
les résultats expérimentaux nous invitent à décrire le comportement effectif comme celui
d’un milieu continu de Cauchy (classique) avec une loi de comportement élastoplastique.
Une telle homogénéisation doit nous permettre de retrouver le comportement identifié
expérimentalement. Ce type d’investigation nécessiterait le développement de nouveaux
outils de modélisation sur la base de ceux déjà développés au sein d’Abaqus pour les
composites.
L’étude réalisée avec l’IFREMER sur le flambement de cylindres sous pression externe
est sur le point de se poursuivre. Des matériaux métalliques, on passe à des matériaux
composites. Le calcul du flambement de tels tubes composites nécessite au préalable
l’homogénéisation du comportement mécanique, mais les deux aspects restent a priori
découplés. Lorsqu’au flambement global s’ajoute la possibilité d’un flambement local
(à l’échelle de l’hétérogénéité périodique), les deux notions (homogénéisation et flam-
bement) interagissent. À l’heure actuelle, nos premiers résultats sur le flambement de
matériaux périodiques concernent le sandwich Napcor . Il s’agit de solutions analytiques
(ou numériques) pour des conditions de chargement et des conditions aux limites parti-
culières (compression dans l’épaisseur). Du point de vue analytique, des améliorations
sont possibles en s’affranchissant de certaines hypothèses simplificatrices, c’est-à-dire
en considérant plus précisément l’influence des peaux, ou en modifiant par exemple la
cinématique des poutres renforts. Mais le vrai challenge est numérique. La géométrie des
renforts et des peaux entraı̂ne fréquemment l’apparition de flambement local quel que soit
le chargement appliqué. L’intérêt d’une homogénéisation du comportement élastique est
perdu si l’on doit reconsidérer la structure complète hétérogène pour détecter ce flam-
bement local. Une thèse va donc être bientôt entreprise sur l’intégration des modes de
flambement locaux dans le processus d’homogénéisation, en considérant des structures
sandwich traditionnelles dans un premier temps. L’idée générale est de vouloir construire
un modèle de plaque enrichi, pour lequel aux déplacements macroscopiques s’ajoutent de
nouveaux degrés de liberté faisant référence aux modes de déformation locaux, dont on
connaı̂t l’expression analytique.
Dans le même ordre d’idées, même s’il ne s’agit pas ici d’homogénéisation, la collabora-
tion avec l’INSA de Rennes sur le flambement des poutres mixtes avec connexion partielle
est riche en perspectives. Les calculs ont pour l’instant été réalisés en élasticité, l’originalité
provenant de la théorie cinématique choisie (Timoshenko) pour chacune des deux poutres.
La dimension ”plasticité” peut être facilement ajoutée. En effet, si les expressions se com-
pliquent en passant de l’élasticité à l’élastoplasticité, la démarche de calcul analytique reste
inchangée. Outre cette perspective immédiate, la température jouant un rôle primordial
dans les applications visées en génie civil, une première réflexion a été lancée sur l’analyse
du comportement thermo-mécanique de ce type de poutre composite et particulièrement
sur le flambement thermo-élastique/plastique qui en résulte.
Pour terminer, j’ai mentionné dans ce mémoire, à différentes occasions, quelques
développements numériques concernant l’endommagement ou la fissuration. Il s’agit de la
fissuration par fatigue sous chargement cyclique d’équipements sous pression métalliques
pour laquelle un programme de modélisation basé sur le concept de bloc-fissure a été
réalisé. Dans le cadre des films polymères, l’utilisation d’éléments de cohésion a per-
mis de modéliser le déchirement d’un film pré-fissuré à titre d’illustration. Enfin, avec
127
les matériaux composites, les objectifs sont plus ambitieux puisque l’idée de départ est
d’”homogénéiser” la décohésion fibres/matrice à l’échelle microscopique en un modèle d’en-
dommagement équivalent. Les premiers résultats, quoiqu’encourageants, sont encore loin
de pouvoir répondre à cette question. La poursuite de cette réflexion n’est pas prévue dans
l’immédiat mais fera peut-être partie de travaux ultérieurs qui permettront alors, dans le
prolongement des travaux sur le flambement, de maı̂triser globalement le comportement
de matériaux tels que les sandwichs.
128
Références
[AJJ, 1995] A. Ajji, K.C. Cole, M.M. Dumoulin et J. Brisson, Amorphous orientation
of poly(ethylene terephthalate) by X-ray diffraction in combination with
Fourier transform infra-red spectroscopy, Polymer , vol. 36, n◦ 21, pp. 4023–
4030 (1995)
[AJJ, 1996] A. Ajji, J. Guèvremont, K.C. Cole et M.M. Dumoulin, Orientation and
structure of drawn poly(ethylene terephthalate), Polymer , vol. 37, n◦ 16,
pp. 3707–3714 (1996)
[ALF, 1998] G. Alfano et L. Rosati, A general approach to the evaluation of consistent
tangent operators for rate-independent elastoplasticity, Computer Methods
in Applied Mechanics and Engineering, vol. 167, n◦ 1-2, pp. 75–89 (1998)
[ARM, 1966] P.J. Armstrong et C.O. Frederick, A mathematical representation of the
multiaxial Bauschinger effect, Technical Report RD/B/N731, General Elec-
tricity Generating Board, Berkeley Nuclear Laboratories (1966)
[ARR, 1993] E.M. Arruda et M.C. Boyce, A three-dimensional constitutive model for the
large stretch behavior of rubber elastic materials, Journal of the Mechanics
and Physics of Solids, vol. 41, n◦ 2, pp. 389–412 (1993)
[BAK, 1989] N.S. Bakhvalov et G.P. Panasenko, Homogenization : Averaging Processes
in Periodic Media (Kluwer Academic Publishers, 1989)
[BAL, 1993] V. Balsamo et A.J. Müller, The phenomenon of double yielding under ten-
sion in low-density polyethylene, linear low-density polyethylene and their
blends, Journal of Materials Science Letters, vol. 12, n◦ 18, pp. 1457–1459
(1993)
[BAR, 2006a] F.C. Bardi et S. Kyriakides, Plastic buckling of circular tubes under axial
compression – Part I : Experiments, International Journal of Mechanical
Sciences, vol. 48, n◦ 8, pp. 830–841 (2006)
[BAR, 2006b] F.C. Bardi, S. Kyriakides et H.D. Yun, Plastic buckling of circular tubes
under axial compression – Part II : Analysis, International Journal of Me-
chanical Sciences, vol. 48, n◦ 8, pp. 842–854 (2006)
[BAT, 1965] S.C. Batterman, Plastic buckling of axially compressed cylindrical shells,
AIAA Journal , vol. 3, n◦ 2, pp. 316–325 (1965)
[BAT, 1968] S.C. Batterman, Free-edge plastic buckling of axially compressed cylindrical
shells, Journal of Applied Mechanics, vol. 35, pp. 73–79 (1968)
[BEN, 1978] A. Bensoussan, J.L. Lions et G. Papanicolaou, Asymptotic Analysis for
Periodic Structures (North-Holland, 1978)
[BEN, 1996] M.L. Benzeggagh et M. Kenane, Measurement of mixed-mode delamination
fracture toughness of unidirectional glass/epoxy composites with mixed-
mode bending apparatus, Composites Science and Technology, vol. 56, n◦ 4,
pp. 439–449 (1996)
[BER, 1978] P.G. Bergan, G. Horrigmoe, B. Krakeland et B. Soreide, Solution techniques
for non-linear finite element problem, International Journal for Numerical
Methods in Engineering, vol. 12, n◦ 11, pp. 1677–1696 (1978)
[BER, 1996] J.M. Berthelot, Matériaux Composites : Comportement Mécanique et Ana-
lyse des Structures (Masson, 1996)
[BER, 1998] J.S. Bergström et M.C. Boyce, Constitutive modeling of the large strain
time-dependent behavior of elastomers, Journal of the Mechanics and Phy-
sics of Solids, vol. 46, n◦ 5, pp. 931–954 (1998)
129
[BER, 2002] J.S. Bergström, S.M. Kurtz, C.M. Rimnac et A.A. Edidin, Constitutive mo-
deling of ultra-high molecular weight polyethylene under large-deformation
and cyclic loading conditions, Biomaterials, vol. 23, n◦ 11, pp. 2329–2343
(2002)
[BES, 1987] D. Besdo, Total Lagrangian strain-space-representation of the elasto-
plasticity of metals, C.S. Desai et al. (editor), 2nd International Conference
on Constitutive Laws for Engineering Materials : Theory and Applications,
Tucson, Arizona, USA, 1357–1364 (January 5-8, 1987)
[BRü, 1998] M. Brünig, Numerical analysis and modeling of large deformation and ne-
cking behavior of tensile specimens, Finite Elements in Analysis and Design,
vol. 28, n◦ 4, pp. 303–319 (1998)
[BRA, 1997] B. Brank, D. Peric et F.B. Damjanic, On large deformations of thin elasto-
plastic shells : Implementation of a finite rotation model for quadrilateral
shell element, International Journal for Numerical Methods in Engineering,
vol. 40, n◦ 4, pp. 689–726 (1997)
[BUA, 2003] N. Buannic, P. Cartraud et T. Quesnel, Homogenization of corrugated core
sandwich panels, Composite Structures, vol. 59, n◦ 3, pp. 299–312 (2003)
[CAI, 1984] D. Caillerie, Thin elastic and periodic plates, Mathematical Methods in the
Applied Sciences, vol. 6, n◦ 1, pp. 159–191 (1984)
[CEC, 2007] A. Cecchi et K. Sab, A homogenized Reissner-Mindlin model for orthotropic
periodic plates : Application to brickwork panels, International Journal of
Solids and Structures, vol. 44, n◦ 18-19, pp. 6055–6079 (2007)
[CHA, 1996] J.L. Chaboche, Unified cyclic viscoplastic constitutive equations : De-
velopment, capabilities, and thermodynamic framework, A.S. Krausz et
K. Krausz (editors), Unified Constitutive Laws of Plastic Deformation, 1–68
(Academic Press, 1996)
[CHA, 2003] G. Chagnon, Modélisation de l’effet Mullins dans les élastomères, Thèse de
doctorat, École Centrale de Nantes (2003)
[CHE, 1995] Y.S. Cheng, W.D. Lu et H. Fang, Plastic post-buckling of a simply suppor-
ted column with a solid rectangular cross-section, Applied Mathematics and
Mechanics, vol. 16, n◦ 8, pp. 713–722 (1995)
[CHR, 1979] J. Christoffersen et J.W. Hutchinson, A class of phenomenological corner
theories of plasticity, Journal of the Mechanics and Physics of Solids, vol. 27,
n◦ 5-6, pp. 465–487 (1979)
[CIM, 1987] A. Cimetière, Sur la modélisation et le flambage des plaques
élastoplastiques, Thèse de doctorat, Université de Poitiers (1987)
[COX, 1994] B.N. Cox, M.S. Dadkhah, W.L. Morris et J.G. Flintoff, Failure mechanisms
of 3D woven composites in tension, compression, and bending, Acta Metal-
lurgica et Materialia, vol. 42, n◦ 12, pp. 3967–3984 (1994)
[COX, 1995] B.N. Cox et M.S. Dadkhah, The macroscopic elasticity of 3D woven com-
posites, Journal of Composite Materials, vol. 29, n◦ 6, pp. 785–819 (1995)
[CRI, 1991] M.A. Crisfield, Non-Linear Finite Element Analysis of Solids and Structures
– Volume 1 : Essentials (John Wiley & Sons, 1991)
[DON, 1976] L.H. Donnell, Beams, Plates and Shells (Mc Graw-Hill, 1976)
[DRA, 1996] S. Drapier, C. Gardin, J.C. Grandidier et M. Potier-Ferry, Structure ef-
fect and microbuckling, Composites Science and Technology, vol. 56, n◦ 7,
pp. 861–867 (1996)
130
[DRO, 2008] A.D. Drozdov et J.deC. Christiansen, Thermo-viscoelastic and viscoplastic
behavior of high-density polyethylene, International Journal of Solids and
Structures, vol. 45, n◦ 14-15, pp. 4274–4288 (2008)
[DUP, 2007] R.B. Dupaix et M.C. Boyce, Constitutive modeling of the finite strain be-
havior of amorphous polymers in and above the glass transition, Mechanics
of Materials, vol. 39, n◦ 1, pp. 39–52 (2007)
[DUR, 1999] D. Durban et Z. Zuckerman, Elastoplastic buckling of rectangular plates in
biaxial compression/tension, International Journal of Mechanical Sciences,
vol. 41, n◦ 7, pp. 751–765 (1999)
[FEI, 1997] J.L. Feijoo, J.J. Sanchez et A.J. Müller, The phenomenon of double yielding
in oriented high density polyethylene films, Journal of Materials Science
Letters, vol. 16, n◦ 21, pp. 1721–1724 (1997)
[FU, 1994] Y. Fu, W.R. Busing, Y. Jin, K.A. Affholter et B. Wunderlich, Structure
analysis of the noncrystalline material in poly(ethylene terephthalate) fibers,
Macromolecular Chemistry and Physics, vol. 195, n◦ 2, pp. 803–822 (1994)
[GAO, 1998] H. Gao et P. Klein, Numerical simulation of crack growth in an isotropic
solid with randomized internal cohesive bonds, Journal of the Mechanics
[GOT, 1999] Y. Goto et C. Zhang, Plastic buckling transition modes in moderately thick
cylindrical shells, Journal of Engineering Mechanics, vol. 125, n◦ 4, pp. 426–
434 (1999)
[GUE, 2008] O. Gueguen, J. Richeton, S. Ahzi et A. Makradi, Micromechanically ba-
sed formulation of the cooperative model for the yield behavior of semi-
crystalline polymers, Acta Materialia, vol. 56, n◦ 7, pp. 1650–1655 (2008)
[HAL, 1975] B. Halphen et Q.S. Nguyen, Sur les matériaux standard généralisés, Journal
de Mécanique, vol. 14, n◦ 1, pp. 39–63 (1975)
[HIL, 1958] R. Hill, A general theory of uniqueness and stability in elastic-plastic solids,
Journal of the Mechanics and Physics of Solids, vol. 6, n◦ 3, pp. 236–249
(1958)
[HON, 2004] K. Hong, A. Rastogi et G. Strobl, A model treating tensile deformation
of semi-crystalline polymers : Quasic-static stress-strain relationship and
viscous stress determined for a sample of polyethylene, Macromolecules,
vol. 37, n◦ 26, pp. 10165–10173 (2004)
[HON, 2005] K. Hong, A model treating tensile deformation of semi-crystalline polymers,
Thèse de doctorat, Albert-Ludwigs-Universität Freiburg im Breisgau (2005)
[HUG, 1978] T.J.R. Hughes et K.S. Pister, Consistent linearization in mechanics of solids
and structures, Computers & Structures, vol. 8, n◦ 3-4, pp. 391–397 (1978)
[HUT, 1974] J.W. Hutchinson, Plastic buckling, Advances in Applied Mechanics, vol. 14,
pp. 67–144 (1974)
[JAM, 1943] H.M. James et E. Guth, Theory of the elastic properties of rubber, Journal
of Chemical Physics, vol. 11, n◦ 10, pp. 455–481 (1943)
[KOU, 1996] A. El Koulani, Continuation dans les problèmes à frontières libres de type
bifurcations plastiques, Thèse de doctorat, Université de Metz (1996)
[KOU, 1997] A. El Koulani et A. Léger, Comportement post-critique des poutres
élastoplastiques : Existence et régularité des branches bifurquées (Comptes-
Rendus de l’Académie des Sciences, Paris, 1997), t. 324
[KRE, 1964] H. Krenchel, Fibre Reinforcement : Theoretical and Practical Investigations
of the Elasticity and Strength of Fibre-Reinforced Materials (Akademisk
Forlag, 1964)
131
[KRI, 1977] R.D. Krieg et D.B. Krieg, Accuracies of numerical solution methods for
the elastic-perfectly plastic model, Journal of Pressure Vessel Technology,
vol. 99, n◦ 4, pp. 510–515 (1977)
[KRY, 2009] A. Kryzanowski, S. Schnabl, G. Turk et I. Planinc, Exact slip-buckling
analysis of two-layer composite columns, International Journal of Solids
and Structures, vol. 46, n◦ 14-15, pp. 2929–2938 (2009)
[KUH, 1942] W. Kuhn et F. Grün, Beziehungen zwischen elastischen Konstanten
und Dehnungsdoppelbrechung hochelastischer Stoffe, Kolloid Zeitschrift,
vol. 101, n◦ 3, pp. 248–271 (1942)
[KYR, 2005] S. Kyriakides, F.C. Bardi et J.A. Paquette, Wrinkling of circular tubes un-
der axial compression : Effect of anisotropy, Journal of Applied Mechanics,
vol. 72, pp. 301–305 (2005)
[KYR, 2007] S. Kyriakides et E. Corona, Mechanics of Offshore Pipelines – Volume 1 :
Buckling and Collapse (Elsevier, 2007)
[LAM, 1992] W.F. Lam et C.T. Morley, Arc-length method for passing limit points in
structural calculation, Journal of Structural Engineering, vol. 118, n◦ 1,
pp. 169–185 (1992)
[LAN, 2000] E.R. Lancaster, C.R. Calladine et S.C. Palmer, Paradoxical buckling beha-
vior of a thin cylindrical shell under axial compression, International Jour-
nal of Mechanical Sciences, vol. 42, n◦ 5, pp. 843–865 (2000)
[LAS, 2006] B. Lascoup, Z. Aboura, K. Khellil et M.L. Benzeggagh, On the mechanical
effect of stitch addition in sandwich panel, Composites Science and Tech-
nology, vol. 66, n◦ 10, pp. 1385–1398 (2006)
[LAS, 2010] B. Lascoup, Z. Aboura, K. Khellil et M.L. Benzeggagh, Homogenization
of the core layer in stitched sandwich structures, Composites Science and
Technology, vol. 70, n◦ 2, pp. 350–355 (2010)
[LEB, 2010] A. Lebée, Homogénéisation de plaques périodiques épaisses, application aux
panneaux sandwichs à âme pliable en chevrons, Thèse de doctorat, Univer-
sité Paris-Est (2010)
[LIU, 2008] T. Liu, Z.C. Deng et T.J. Lu, Analytical modeling and finite element simu-
lation of the plastic collapse of sandwich beams with pin-reinforced foam
cores, International Journal of Solids and Structures, vol. 45, n◦ 18-19,
pp. 5127–5151 (2008)
[MAH, 2002] H. Mahbadi et M.R. Eslami, Cyclic loading of beams based on the Prager
and Frederick-Armstrong kinematic hardening models, International Jour-
nal of Mechanical Sciences, vol. 44, n◦ 5, pp. 859–879 (2002)
[MAN, 2000] P. Mandal et C.R. Calladine, Buckling of thin cylindrical shells under axial
compression, International Journal of Solids and Structures, vol. 37, n◦ 33,
pp. 4509–4525 (2000)
[MAO, 2001] R. Mao et G. Lu, Plastic buckling of circular cylindrical shells under combi-
ned in-plane loads, International Journal of Solids and Structures, vol. 38,
n◦ 5, pp. 741–757 (2001)
[MAR, 2002] G. Marckmann, E. Verron, L. Gornet, G. Chagnon, P. Charrier et P. Fort, A
theory of network alteration for the Mullins effect, Journal of the Mechanics
[MCI, 1979] A. McIntyre et G.E. Anderson, Fracture properties of a rigid polyurethane
foam over a range of densities, Polymer , vol. 20, pp. 247–253 (1979)
132
[MIC, 1999] J.C. Michel, H. Moulinec et P. Suquet, Effective properties of composite ma-
terials with periodic microstructure : A computational approach, Computer
Methods in Applied Mechanics and Engineering, vol. 172, n◦ 1-4, pp. 109–
143 (1999)
[MIE, 2004] C. Miehe, S. Göktepe et F. Lulei, A micro-macro approach to rubber-like
materials – Part I : The non-affine micro-sphere model of rubber elasticity,
Journal of the Mechanics and Physics of Solids, vol. 52, n◦ 11, pp. 2617–
2660 (2004)
[NEE, 1972] A. Needleman, A numerical study of necking in circular cylindrical bar,
Journal of the Mechanics and Physics of Solids, vol. 20, n◦ 2, pp. 111–127
(1972)
[NEW, 1951] N.M. Newmark, C.P. Siess et I.M. Viest, Tests and analysis of composite
beams with incomplete interaction, Proceedings of the Society of Experi-
mental Stress Analysis, vol. 9, n◦ 1, pp. 75–92 (1951)
[NGU, 2000] Q.S. Nguyen, Stabilité et Mécanique Non-Linéaire (Hermès, 2000)
[NGU, 2004] T.D. Nguyen, S. Govindjee, P. Klein et H. Gao, A rate-dependent cohe-
sive continuum model for the study of crack dynamics, Computer Methods
in Applied Mechanics and Engineering, vol. 193, n◦ 30-32, pp. 3239–3265
(2004)
[OKA, 1990] H. Okada, H. Rajiyah et S.N. Atluri, A full tangent stiffness field-boundary-
element formulation for geometric and material non-linear problems of solid
mechanics, International Journal for Numerical Methods in Engineering,
vol. 29, n◦ 1, pp. 15–35 (1990)
[ORT, 1983] M. Ortiz, P.M. Pinsky et R.L. Taylor, Operator split methods for the nu-
merical solution of the elastoplastic dynamic problem, Computer Methods
in Applied Mechanics and Engineering, vol. 39, n◦ 2, pp. 137–157 (1983)
[PAR, 2002] R. Parnes et A. Chiskis, Buckling of nano-fibre reinforced composites : A
re-examination of elastic buckling, Journal of the Mechanics and Physics of
Solids, vol. 50, n◦ 4, pp. 855–879 (2002)
[POI, 2005] C. Poisson, Optimisation de films complexes PE/liant/PA à propriétés
d’usage maı̂trisées, Thèse de doctorat, Université des Sciences et Technolo-
gies de Lille/École des Mines de Douai (2005)
[RAG, 2005] P. Raghavan et S. Ghosh, A continuum damage mechanics model for uni-
directional composites undergoing interfacial debonding, Mechanics of Ma-
terials, vol. 37, n◦ 9, pp. 955–979 (2005)
[RAO, 2008] M.V. Rao, P. Mahajan et R.K. Mittal, Effect of architecture on mechanical
properties of carbon/carbon composites, Composite Structures, vol. 83, n◦ 2,
pp. 131–142 (2008)
[RIK, 1979] E. Riks, An incremental approach to the solution of snapping and buck-
ling problems, International Journal of Solids and Structures, vol. 15, n◦ 7,
pp. 529–551 (1979)
[RIK, 1991] E. Riks, On formulations on path-following techniques for structural stabi-
lity analysis, New Advances in Computational Structural Mechanics, 65–80
(Elsevier, 1991)
[SAL, 2008] N. Sallem-Idrissi, Comportement mécanique et évolution structurale induite
dans les films multi-couches à base de polyamide 6 et de polyéthylène, Thèse
de doctorat, Université des Sciences et Technologies de Lille/École des Mines
de Douai (2008)
133
[SAN, 1992] J. Sanchez-Hubert et E. Sanchez-Palencia, Introduction aux Méthodes
Asymptotiques et à l’Homogénéisation (Masson, 1992)
[SCH, 2010] S. Schnabl et I. Planinc, The influence of boundary conditions and axial
deformability on buckling behavior of two-layer composite columns with
interlayer slip, Engineering Structures, vol. 32, n◦ 10, pp. 3103–3111 (2010)
[SCH, 2011] S. Schnabl et I. Planinc, The effect of transverse shear deformation on the
buckling of two-layer composite columns with interlayer slip, International
Journal of Non-Linear Mechanics, vol. 46, n◦ 3, pp. 543–553 (2011)
[SEG, 1990] R. Seguela et F. Rietsch, Double yield point in polyethylene under tensile
loading, Journal of Materials Science Letters, vol. 9, n◦ 1, pp. 46–47 (1990)
[SEY, 1994] R. Seydel, Practical Bifurcation and Stability Analysis. From Equilibrium
to Chaos (Springer-Verlag, 1994)
[SHA, 1947] F.R. Shanley, Inelastic column theory, International Journal of Aeronautical
Sciences, vol. 14, n◦ 5, pp. 261–267 (1947)
[SHA, 2005] G.F. Shan, W. Yang, B.H. Xie, Z.M. Li, J. Chen et M.B. Yang, Double yiel-
ding behaviors of polyamide 6 and glass bead filled polyamide 6 composites,
Polymer Testing, vol. 24, n◦ 6, pp. 704–711 (2005)
[SHE, 2007] M. Sherburn, Geometric and mechanical modelling of textiles, Thèse de
doctorat, University of Nottingham (2007)
[SIM, 1986] J.C. Simo et R.L. Taylor, A return mapping algorithm for plane stress elas-
toplasticity, International Journal for Numerical Methods in Engineering,
vol. 22, n◦ 3, pp. 649–670 (1986)
[SIM, 1991] J.C. Simo et S. Govindjee, Non-linear B-stability and symmetry preserving
return mapping algorithms for plasticity and viscoplasticity, International
Journal for Numerical Methods in Engineering, vol. 31, n◦ 1, pp. 151–176
(1991)
[SIM, 1992] J.C. Simo, Algorithms for static and dynamic multiplicative plasticity that
preserve the classical return mapping schemes of the infinitesimal theory,
Computer Methods in Applied Mechanics and Engineering, vol. 99, n◦ 1,
pp. 61–112 (1992)
[SIM, 1998] J.C. Simo et T.J.R. Hughes, Computational Inelasticity (Springer, 1998)
[SPI, 1976] W.A. Spitzig, R.J. Sober et O. Richmond, The effect of hydrostatic pressure
on the deformation behavior of maraging and HY-80 steels and its impli-
cations for plasticity theory, Metallurgical Transactions A, vol. 7, n◦ 11,
pp. 1703–1710 (1976)
[TVE, 1983] V. Tvergaard, On the transition from a diamond mode to an axisymmetric
mode of collapse in cylindrical shells, International Journal of Solids and
Structures, vol. 19, n◦ 10, pp. 845–856 (1983)
[WIL, 1964] M.L. Wilkins, Calculation of elastic-plastic flow, B. Adler et al. (editor),
Methods of Computational Physics, volume 3, 211–263 (Academic Press,
1964)
[XU, 2007] R. Xu et Y. Wu, Static, dynamic, and buckling analysis of partial interaction
composite members using Timoshenko’s beam theory, International Journal
of Mechanical Sciences, vol. 49, n◦ 10, pp. 1139–1155 (2007)
[YEH, 1986] M.K. Yeh et S. Kyriakides, On the collapse of inelastic thick-walled tubes
under external pressure, Journal of Energy Resources Technology, vol. 108,
pp. 35–47 (1986)
134
[YEH, 1988] M.K. Yeh et S. Kyriakides, Collapse of deepwater pipelines, Journal of
Energy Resources Technology, vol. 110, pp. 1–11 (1988)
[ZAN, 2007] S.L. Zang, J. Liang et C. Guo, A constitutive model for spring-back pre-
diction in which the change of Young’s modulus with plastic deformation
is considered, International Journal of Machine Tools and Manufacture,
vol. 47, n◦ 11, pp. 1791–1797 (2007)
[ZHA, 1994] G. Zhang et R.A. Latour, An analytical and numerical study of fiber mi-
crobuckling, Composites Science and Technology, vol. 51, n◦ 1, pp. 95–109
(1994)
[ZHA, 1997] G. Zhang et R.A. Latour, A three-dimensional micromechanical model of the
compressive behavior of unidirectional FRP composites, Journal of Ther-
moplastic Composite Materials, vol. 10, n◦ 2, pp. 173–184 (1997)
135
136
Partie III
Recueil de publications
138
Liste de publications
i - A. Le van, P. Le Grognec, Modeling and numerical computation of necking in
round bars using a total Lagrangian formulation, Computer Modeling in Engineering
and Sciences, Vol. 2, n°1, pp. 63-72, 2001 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 140
ii - A. Le van, G. De Saxcé, P. Le Grognec, General formulation for local integration

in standard elastoplasticity with an arbitrary hardening model, Computers &
Structures, Vol. 81, n°21, pp. 2099-2109, 2003 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 150
iii - S. Comas-Cardona, P. Le Grognec, C. Binétruy, P. Krawczak, Unidirectional

compression of fibre reinforcements. Part 1 : A non-linear elastic-plastic behaviour,
Composites Science and Technology, Vol. 67, n°3-4, pp. 507-514, 2007 . . . . . . . p. 161
iv - P. Le Grognec, A. Le van, Elastoplastic bifurcation and collapse of axially loaded

cylindrical shells, International Journal of Solids and Structures, Vol. 45, n°1, pp.
64-86, 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 169
v - A. El Hakimi, P. Le Grognec, S. Hariri, Numerical and analytical study of severity

of cracks in cylindrical and spherical shells, Engineering Fracture Mechanics, Vol.
75, n°5, pp. 1027-1044, 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 192
vi - P. Le Grognec, A. Le van, Some new analytical results for plastic buckling and
initial post-buckling of plates and cylinders under uniform compression, Thin-Walled
Structures, Vol. 47, n°8-9, pp. 879-889, 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 210
vii - P. Le Grognec, P. Casari, D. Choqueuse, Influence of residual stresses and

geometric imperfections on the elastoplastic collapse of cylindrical tubes under
external pressure, Marine Structures, Vol. 22, n°4, pp. 836-854, 2009 . . . . . . . . p. 221
viii - R. Gonçalves, P. Le Grognec, D. Camotim, GBT-based semi-analytical solutions

for the plastic bifurcation of thin-walled members, International Journal of Solids
and Structures, Vol. 47, n°1, pp. 34-50, 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 240
ix - F. Zeng, P. Le Grognec, M.F. Lacrampe, P. Krawczak, A constitutive model for

semi-crystalline polymers at high temperature and finite plastic strain : Application
to PA6 and PE biaxial stretching, Mechanics of Materials, Vol. 42, n°7, pp. 686-697,
2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 257
x - P. Le Grognec, A. Le van, On the plastic bifurcation and post-bifurcation of

axially compressed beams, International Journal of Non-Linear Mechanics, Vol. 46,
n°5, pp. 693-702, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 269
xi - M. Deléglise, P. Le Grognec, C. Binétruy, P. Krawczak, B. Claude, Modeling

of high speed RTM injection with highly reactive resin with on-line mixing, Com-
posites Part A : Applied Science and Manufacturing, Vol. 42, n°10, pp. 1390-1397,
2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 279
xii - P. Lapeyronnie, P. Le Grognec, C. Binétruy, F. Boussu, Homogenization of the

elastic behavior of a layer-to-layer angle-interlock composite, Composite Structures,
Vol. 93, n°11, pp. 2795-2807, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 287
139
140
141
142
143
144
145
146
147
148
149
Computers and Structures 81 (2003) 2099–2109
www.elsevier.com/locate/compstruc
General formulation for local integration in standard

elastoplasticity with an arbitrary hardening model
a,*
Anh Le van , Gery de Saxce b, Philippe Le Grognec a
a
Laboratoire de G
enie Civil, Facult
e des Sciences et des Techniques, 2 rue de la Houssini
ere, 44322 Nantes Cedex 3, France
b
Laboratoire de Mecanique de Lille, URA CNRS 1441, Universit e de Lille I, 59655 Villeneuve d’Ascq Cedex, France
Received 31 December 2002; accepted 1 May 2003
Abstract
This paper describes a general method for deriving the plastic corrections and the consistent tangent modulus for a
wide range of arbitrary non-linear hardening models within the framework of standard small strains elastoplasticity.
The features of the proposed formulation are: (i) the local solution is obtained through an iterative procedure. The
plastic corrections are given in closed forms exhibiting one scalar function denoted by Galg and three fourth-order
tensors Dalg , Galg , Lalg , which are shown to be the algorithmic discrete counterparts of usual theoretical continuum
quantities, (ii) the consistent tangent modulus has a symmetrical expression involving the same quantities. Finite ele-
ment computations are performed using a particular non-linear kinematic hardening model and allow to exhibit the
ratcheting phenomenon usually observed on mechanical components subjected to cyclic loadings.
2003 Elsevier Ltd. All rights reserved.
Keywords: Elastoplasticity; Local integration; Plastic correction; Consistent tangent modulus; Ratcheting effect
1. Introduction In some cases an analytical local integration is pos-

sible. Thus, Krieg and Krieg [16] gave the exact local
The numerical response of an elastoplastic structure solution in perfect plasticity with von Mises yield func-
is usually determined by an incremental method where tion, by assuming a given constant and deviatoric in-
the solution is computed at every time step by using the crement of strain rate. Later, dealing with the plane
finite element method combined with a Newton-type stress situation, Simo and Govindjee [29] showed that
iterative scheme. Within each iteration it is necessary to the linear kinematic hardening leads to an algebraic
perform the two following important tasks characteristic equation of degree four which can be solved analytically.
of any elastoplastic algorithm: first, the local integration Recently, Alfano et al. [1] successfully generalized the
of the non-linear constitutive equations in order to treatment in [29] to include the significant case of iso-
compute the stress for a given strain increment and to tropic hardening as well as elastoviscoplasticity of the
build up the internal force vector, second, the compu- Perzyna type, and also reduced the solution procedure
tation of the consistent tangent modulus in order to to that of a quartic equation in the plastic multiplier.
form the structural tangent matrix. Since these tasks In view of numerical purposes when analytical solu-
represent the main part of the elastoplastic calculation, tions are not available, the first local integration algo-
a lot of work were devoted to their analysis. rithm using Euler implicit scheme––referred to as radial
return algorithm––was proposed by Wilkins [38] for
perfect plasticity with von Mises yield criterion. Later, it
was extended to linear isotropic and kinematic harden-
ings by Krieg and Key [15]. The further case of non-
*
Corresponding author. linear isotropic hardening combined with a special form
0045-7949/03/$ - see front matter 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0045-7949(03)00256-6
150
2100 A. Le van et al. / Computers and Structures 81 (2003) 2099–2109
of non-linear kinematic hardening was investigated by The above review makes no claim to completeness, the
Simo and Taylor [34]. In the slightly different frame of reader can find more details on the historical back-
non-smooth mechanics, the catching up algorithm was ground to the local integration subject as well as more
proposed by Moreau [19,20] and used later on by references in the quoted papers.
Nguyen [22]. In this paper, we attempt to go further by proposing
In order to handle other criteria than von Mises one, a more general, synthetic method to derive the plastic
Ortiz et al. [24] proposed the closest point algorithm as a corrections and the consistent tangent modulus for a
generalization of the radial return algorithm to arbitrary wide class of non-linear hardening models. The for-
convex yield functions. The difficult case of plane stress mulation is built for small strain elastoplasticity within
was solved by Simo and Taylor [35]. Besides, other the framework of the generalized standard materials
formulations using either the generalized trapezoidal or theory described in [10,28]. The assumptions made are
the generalized midpoint rule were presented by Ortiz general enough for the proposed formulation to be
and Popov [25], Ortiz and Martin [23], and Simo and valid with somewhat arbitrary yield functions and
Govindjee [30], with a view to encompassing the previ- hardening energies. The local solution of the elasto-
ous algorithms and combining numerical stability and plastic equations involves the one-step fully implicit
accuracy. The cutting-plane algorithm successively pro- integration scheme combined with a local Newton it-
posed by Simo and Ortiz [33] in large deformations and erative procedure, as usually done in the literature.
Ortiz and Simo [26] in small strains is another stress However, compared with Ref. [31] for instance, here the
update procedure lying on a different conception and expressions for the plastic corrections and the consis-
may entail less computational efforts. After Hughes and tent tangent modulus are obtained in closed (tensorial
Pister [12] introduced the notion of consistent lineari- or matrical) forms ready for computer use, rather than
zation, the role of consistent tangent modulus was em- in operator forms. The presented formulation high-
phasized by Hughes and Taylor [13]. Later, a general lights the role of one scalar function and three fourth-
local integration algorithm and the corresponding con- order tensors, which are shown to be the algorithmic
sistent tangent modulus for rate-independent plasticity discrete counterparts of usual theoretical continuum
were presented by Simo and Hughes [31,32], Ortiz and quantities. Also, consideration of general expressions
Martin [23], and Simo and Govindjee [30]. More re- for the hardening energies gives rise to some additional
cently, Hartmann and Haupt [11] considered a non-lin- terms in the plastic corrections, which are not reported
ear kinematic hardening model due to Armstrong and in the previous works.
Frederick [3] and used the implicit Euler time integration The finite element computations carried out on the
procedure to solve the non-linear equation system for example of a particular non-linear kinematic hardening
the stress computation. By the same backward scheme, model allow to assess the validity of the proposed al-
Matzenmiller and Taylor [18] gave the stress update for gorithm and to exhibit different types of ratcheting
general isotropic yield criteria containing all the invari- usually observed on mechanical components subjected
ants of the stress tensor. Completely explicit expressions to cyclic loadings.
for the plastic corrections and the consistent tangent
modulus were given by Doghri [6–8] for J2 -plasticity
possibly combined with damage. A first general ap- 2. Local integration
proach to the derivation of the explicit expression for the
consistent tangent modulus was provided by Alfano Let us consider a three-dimensional continuum body
and Rosati [2] in rate-independent elastoplasticity with undergoing a quasi-static infinitesimal elastic–plastic
mixed hardening. More recently, Palazzo et al. [27] suc- transformation. The linearized strain tensor is decom-
ceeded in obtaining explicit representations for the stress posed into an elastic and plastic part, denoted by ee and
update and the consistent tangent operator for general ep respectively, according to the relationship
isotropic elastoplastic models, by means of a novel linear
combination of dyadic and square tensor products. e ¼ ee þ ep ð1Þ
Finally, it should be mentioned that one of the first
solution for general inelastic analysis is the effective- The state and evolution equations for the elastoplastic
stress-function algorithm proposed by Bathe and co- problem are obtained according to the generalized
workers [4,14]. The complete stress state is obtained by standard materials theory which is based on thermody-
solving a governing scalar equation called the effective namical foundations [10,28]. The following assump-
stress function equation and the consistent tangent tions––not so restrictive––are made on the internal state
matrix is derived thereof. This method is actually ap- of the body.
plicable for more complex inelastic situations, like
Drucker–Prager soil model, thermoelastoplasticity and Assumption 1. The internal state at any point in the
creep solutions, as well as large strain elastoplasticity [9]. body and at any time is characterized by the plastic
151
A. Le van et al. / Computers and Structures 81 (2003) 2099–2109 2101
strain ep and two other internal variables to account for of

r_ ¼ D : e_ k_D : ðr; a; aÞ
the deformation history, one second-order tensor hard- or
ening variable denoted by a and one scalar hardening o2 wa _ of
a_ ¼ : a ¼ k_GðaÞ : ðr; a; aÞ
variable denoted by a. oa 2
oa ð6Þ
o2 wa of
a_ ¼ a_ ¼ k_GðaÞ ðr; a; aÞ
Assumption 2. The hardening energy whard splits up into oa2 oa
two independent functions of the hardening variables f ðr; a; aÞ ¼ 0

whard ða; aÞ ¼ wa ðaÞ þ wa ðaÞ ð2Þ where GðaÞ is a symmetric tensor of order four and GðaÞ
a scalar function defined by
where wa and wa are strictly convex functions. Definition
The state laws give us the stress r and the thermo- o2 wa o2 w a
GðaÞ ðaðaÞÞ; GðaÞ ðaðaÞÞ ð7Þ
dynamic conjugates a and a, duals of the hardening oa2 oa2
variables a and a respectively
In the sequel, G is referred to as the plastic (hard-
r ¼ D : ee ¼ D : ðe ep Þ ening) tensor––as opposed to the elastic tensor D––and
G as the plastic (hardening) scalar function. In writing
owa (7), use has been made of the strict convexity of the
a ¼ ðaÞ ð3Þ
o a hardening energies wa and wa according to which the
owa state laws (3b) and (3c) can be inverted giving ða; aÞ in
a¼ ðaÞ terms of ða; aÞ.
oa
Relations (6) are now discretized by means of the
implicit Euler scheme, giving rise to 14 scalar equations
where D denotes the elasticity fourth-order tensor de-
with 14 unknowns ðr; a; a; DkÞ
fined as the second derivative of the elastic energy
1 e
: D : e . The yield criterion expressed in terms of the of
2
r rn1 þ DkD : ðr; a; aÞ D : De ¼ 0
stress r and the hardening variables ð a; aÞ is satisfied at or
any point in plastic loading of
a an1 þ DkGðaÞ : ðr; a; aÞ ¼ 0
oa ð8Þ
f ðr; a; aÞ ¼ 0 ð4Þ of
a an1 þ DkGðaÞ ðr; a; aÞ ¼ 0
oa
The evolution laws result from generalization of the f ðr; a; aÞ ¼ 0
maximum plastic dissipation rule and provide the rates
of plastic strain and hardening variables by differenti- In the above equations, the subscripts for quantities at
ating the yield function f present time tn have been omitted for brevity, thus r for
instance designates rn . The quantity Dk can be consid-
of ered as an incremental plastic multiplier. The backward
e_ p ¼ k_ implicit Euler scheme ensures the numerical stability and
or
the symmetry of the consistent tangent modulus, as has
of
a_ ¼ k_ ð5Þ been shown in number of works. To solve the non-linear
oa system (8) by means of Newton iterative method, we
of make the following hypothesis which is satisfied in
a_ ¼ k_
oa practice.
where k_ is the plastic multiplier. Assumption 3. The yield function f ðr; a; aÞ is decom-
Given a strain increment De ¼ e en1 at a point in posed as the sum of one term depending on ðr; aÞ and
the body and at present time tn , let us assume that there another one depending on the scalar variable a
is plastic loading and determine the elastoplastic solu-
tion in terms of De and the other known quantities at f ðr; a; aÞ ¼ f1 ðr; aÞ þ f2 ðaÞ ð9Þ
previous time tn1 .
By differentiating the state laws (3) with respect to The form (9) is not so restrictive in practice and is
time and eliminating the conjugate hardening variables general enough to enable one to deal with new plasticity
by means of the evolution laws (5), we get 14 equations models. Note that variables ðr; aÞ do not necessarily
with 14 unknowns, viz. the stress r, the hardening appear in the form of their difference r a, as we will see
variables ða; aÞ and the plastic multiplier k_ below in the case of non-linear kinematic hardening.
152
2

Since the elastic and hardening energies are strictly e G1 þ Dk o f þ DkG1 : of o3 wa
C : : G1
convex, D and Gð aÞ are invertible. Taking into account oa2 oa oa3
(9) at each local iteration for solving (8), the plastic
corrections ðdr; d a; da; dkÞ for ðr; a; a; DkÞ are given by and
2
2 2
of of of
D1 þ Dk 2 : dr þ Dk : d
a f
A D1 þ Dk 2
or oaor or
of
¼ Wðr; a; DkÞ dk are invertible like G1 and D1 . In such conditions, the
or system (13) is solved as follows
o2 f o2 f
Dk : dr þ G1 þ Dk 2 e : p þ f
dr ¼ D L : q
oroa oa
ð14Þ
1 of o w 3 a
1
f T : p þ G
da ¼ L e : q
þ DkG : : :G : da
oa oa3 ð10Þ
with De ¼ ðf
AB e :C e T Þ1 , f
e 1 : B L ¼ f e :G
A 1 : B e
of
¼ Uðr; a; DkÞ dk e ¼ ðC
and G eB e :f
T 1
A :B eÞ .
1
oa
1 o2 f Dk o3 wa of
þ Dk 2 þ 2 da From (14), the incremental plastic multiplier Dk, the
GðaÞ oa G ðaÞ oa3 oa
stress r and the hardening variables ða; aÞ are updated at
of
¼ Uða; DkÞ dk each local iteration using the following plastic correc-
oa tions [36,37]
of of of
: dr þ : da þ da ¼ f ðr; a; aÞ
or oa oa dr ¼ Dalg : W Lalg : U

where W and U are second-order tensors and U a scalar of of
Dalg : þ Lalg : dk
function defined by or oa
da ¼ LTalg : W Galg : U ð15Þ
Definition
of of of
Wðr; a; DkÞ ¼ D1 : ðr rn1 Þ þ Dk ðr; aÞ De LTalg : þ Galg : dk
or or oa

of of
Uðr; a; DkÞ ¼ G1 ð
aÞ : ð
a an1 Þ þ Dk ðr; aÞ ð11Þ da ¼ Galg U þ dk
o
a oa
a an1 of
Uða; DkÞ ¼ þ Dk ðaÞ where the scalar function Galg and the fourth-order
GðaÞ oa
tensors Dalg , Galg and Lalg are defined by
Note that the inner parentheses in the left-hand side of
Eq. (10b) are essential since the colon products therein are Definition
not associative. Moreover, since the components of the "
3 a
fourth-order tensor ofoa
: ooaw3 in a fixed orthonormal basis o2 f 2
2 o f o2 f
read (with summation implied over repeated subscripts) Dalg ¼ D1 þ Dk ðDkÞ : G1 þ Dk 2
or 2
oaor o
a
3 a
of o3 wa of ow # 1
: ¼ of o3 wa
1
o2 f
o a oa3 ijkl o
a mn oa3 nmijkl þ DkG :1
: :G 1
:
oa oa3 o
or a
of o3 wa "
¼
a mn oamn oaji oalk
o o2 f of o3 wa
Galg ¼ G1 þ Dk 2 þ DkG1 : : : G1
of o3 wa oa oa oa3
¼ : ð12Þ 1 #1
o
a oa3 klij 2
2 o f 1 o2 f o2 f
3 a ðDkÞ : D þ Dk 2 :
the fourth-order tensor of
oa
: ooaw3 is symmetric. Therefore, oroa or oaor
Eq. (10a) and (10b) form a symmetric system with un- GðaÞ
knowns dr and d a. These equations, uncoupled from Galg ¼ 2
1 þ DkGðaÞ ooaf2 þ GðaÞ
Dk o w of
3 a
(10c) with unknown da, take the following form oa3 oa
1
f e : da ¼
A : dr þ B p o2 f o2 f
ð13Þ Lalg ¼ D1 þ Dk 2 : Dk : Galg
e T e
or oaor
B : dr þ C : da ¼
q
ð16Þ
In practice, the incremental plastic multiplier Dk is small
enough like the strain increment. Thus, the fourth-order One can check that the transpose of Lalg can be put in
tensors the form
153
LTalg ¼ Ce 1 : BeT : De plastic scalar function G, see definition (7), respectively. As

1 shown by definition (16), when the incremental plastic
o2 f of o3 wa
¼ G1 þ Dk 2 þ DkG1 : : 3 : G1 multiplier Dk tends to zero, all the algorithmic quantities
oa o
a oa tend to their theoretical counterparts.
2
of In this study, we assume that the hardening energies
: Dk : Dalg
oroa waðaÞ and wa ðaÞ in Eq. (2) can be given general expres-
sions. Thus, the plastic (hardening) tensor G and the
Eventually, inserting relations (15) into (10d) provides
plastic (hardening) scalar function G, defined by relation
the correction for the plastic multiplier
(7), are not constant a priori. In particular, this leads to
3 a
of of T the term ofoa
: ooaw3 in Eq. (16a) and (16b), which is not
dk ¼ f : Dalg þ : Lalg : W
or o
a reported in the literature, see e.g. Box 3.5 in [31].
Although to date usual models involve quadratic ex-
of of of
: Lalg þ : Galg : U Galg U pressions for wa so that of
3 a
: ooaw3 is zero, theoretical con-
or o
a oa oa
," siderations give rise to the latter term which must be
of of of of taken into account in general.
: Dalg : þ2 : Lalg :
or or or o
a In the case of arbitrary yield functions and hardening
2 # energies, fourth-order tensors Dalg , Galg and Lalg must
of of of
þ : Galg : þ Galg ð17Þ be computed numerically from definition (16), and this
oa o
a oa requires inversions of 6 · 6 symmetric matrices.
Recall that the right-hand sides of (15)–(17) are com- Remark. In some cases, these numerical computations
puted with values obtained at the previous local itera- can be avoided as recently shown by Palazzo et al. [27].
3 a
tion. Since the fourth-order tensor of
oa
: ooaw3 is symmetric, These authors, by using an original linear combination
Dalg and Galg are symmetric too. To sum up, the local of dyadic and square tensor products, have successfully
solution of the elastoplastic problem is described by the derived explicit representations for the stress update and
following algorithm. the consistent tangent operator, for general isotropic
yield functions depending upon all the three invariants
Local integration loop of the relative stress r a.
(a) Compute
1 Furthermore, using the symmetry of Dalg and Galg ,
o2 f of o3 wa relations (15a), (15b) and (17) can be recast as follows
G1 þ Dk 2 þ DkG1 : : : G 1
oa a oa3
o "
of of
and derive Dalg from (16a). dk ¼ f : Dalg : W : Galg : U
or alg oa alg
(b) Compute #,"
1 of of of
o2 f Galg U : Dalg :
D1 þ Dk 2 oa or or alg
or 2 #
of of of
þ : Galg : þ Galg ð18Þ
and derive Galg from (16b) and Lalg from (16d). oa oa alg oa
(c) Compute Galg from (16c).
(d) Compute the plastic corrections dk by (17), and
ðdr; da; daÞ by (15), and update the solution "
ðDk; r; a; aÞ.
o2 f o2 f
(e) Check the convergence, based on dk for instance, d r ¼ Dalg : W þ Dk : G1 þ Dk 2

o aor oa
to exit the loop. 1 #
End of the loop 1 of o3 wa 1 of
þ DkG : : :G :U dk
oa o a3 o r alg
The proposed local integration procedure combines the " 1
quadratic convergence rate of Newton scheme with the o2 f o2 f
da ¼ Galg : U þ Dk : D1 þ Dk 2
unconditional stability of implicit Euler scheme. Expres- oroa or
sions (15)–(17) for the plastic corrections show the major #
of
role of tensors Dalg , Galg , Lalg and scalar function Galg in :W dk
the inversion of the local equation set (8). The quantities oa alg
Dalg , Galg and Galg represent the algorithmic or discrete ð19Þ
finite-step counterparts of the usual theoretical continuum of of
quantities: the elastic tensor D, the plastic tensor G and the where or alg
and o
a alg
are second-order tensors defined by
154
Definition By comparison with the theoretical tangent tensor oore

derived from system (6) for the continuous
of of of o2 f problem,
the
¼ Dk : G1 þ Dk 2 algorithmic or consistent tangent tensor oore alg must be

or alg or
oa o
a
different because of additional terms in (22).
1
of o3 wa o2 f By noting that system (22a) and (22b) is analogous to
þ DkG1 : : : G 1
:

oa oa 3
oroa system (10a) and (10b) (replace tensor W in (10a) by e_
1 and make U ¼ 0 in (10b)), it can be inverted using (14)
of of of o2 f o2 f
¼ Dk : D1 þ Dk 2 : as follows
oa alg oa or or o
aor
ð20Þ _r ¼ Dalg : e_ Dalg : of þ Lalg : of D_ k
or oa
Expression (18) for dk, strictly equivalent to (17), is quite ð23Þ
of of _
analogous to the theoretical expression for k_ derived a_ ¼ LTalg : e_ LTalg : þ Galg : Dk
or oa
from the continuous problem:
where tensors Dalg , Galg and Lalg are defined in (16).
of
: D : e_ Inserting relations (22c) and (23) into (22d) leads to
k_ ¼ or
of 2 ð21Þ
of
or
: D : of
or
þ of
oa
aÞ : of
: Gð oa
þ GðaÞ oa
"
_Dk ¼ of of T of of
where the sum appearing in the denominator : Dalg þ oa : Lalg or
: Dalg :
or
or
2
h of : GðaÞ : of þ GðaÞ of þ2
of
: Lalg :
of of
þ : Galg :
of
oa oa oa or oa oa oa
2 !#
of
is usually referred to as the plastic modulus. When the þ Galg : e_ ð24Þ
incremental oa
of plastic
of
multiplier Dk tends to zero, tensors
or alg
and oa alg
tend to their theoretical continuum
counterparts of
and of
, respectively. Eventually, by putting (24) in (23a) we obtain the con-
or oa
sistent tangent modulus

3. Consistent tangent modulus or of of
¼ Dalg Dalg : þ Lalg :
oe alg or oa

The local integration is now assumed to be achieved of of
: Dalg þ : LTalg
so that the local solution ðr; a; a; DkÞ at present time tn of or oa
the system (3)–(5) is available and a formal implicit re- ,"
of of of of
lation has been established between stress and strain, : Dalg : þ2 : Lalg :
r ¼ r^ðe; state at tn1 Þ, where the expression ‘‘state at tn1 ’’ or or or oa
designates the known values of all the variables at pre- 2 #
of of of
vious time tn1 . The concepts developed in the previous þ : Galg : þ Galg ð25Þ
oa oa oa
section will allow us to readily derive the consistent
^
tangent modulus oore alg ¼ oore as the algorithmic coun-
where symbol denotes dyadic tensor product defined
terpart of the theoretical tangent modulus [13,21,34].
by ðc dÞijkl ¼ cij dkl , 8 second-order tensors c and d.
Differentiating relations (8) with respect to time yields
2 Relation (25) can be recast into a simpler
expression
2
_r ¼ D : e_ D_ kD : of DkD : o f : r_ þ o f : a_
using algorithmic tensors oofr alg and of oa alg
introduced
or or2 o
aor in (20)
2
a_ ¼ D_ kG :
of of _ " #
Dk G : : r or of of
oa oro
a ¼ Dalg Dalg : : Dalg
oe alg or alg or alg
o2 f _ of oG _
þG: 2 :aþ : :a ,"
oa o
a o a of of of of
2 : Dalg : þ : Galg :
_ of o f of oG or or alg oa oa alg
a_ ¼ DkG Dk G 2 þ a_
oa oa oa oa 2 #
of
of _ of _ of þ Galg ð26Þ
f_ ¼ : r þ :
a þ a_ ¼ 0 oa
or oa oa
ð22Þ
The consistent tangent modulus is symmetric since Dalg
In the above relations, computations are made with the is so. Expression (26) shows strong analogy with the
values ðDk; r; a; aÞ deduced from the local integration. continuum theoretical tangent tensor
155
or D : oofr oofr : D (5) are derived from potential f given by (28), whereas
¼D 2 ð27Þ the yield function is given by (28) with K ¼ 0. Here, the
oe of
: D : oofr þ of : Gð aÞ : of þ GðaÞ of
o r oa oa oa
model (28) is used in the frame of standard associative

Also, the consistency of tensor oore alg is directly con- plasticity which means that the potential coincide with
firmed according to the limit properties discussed in the the yield function given by (28). Although any combi-
previous section when Dk tends to zero. nation of non-linear isotropic and non-linear kinematic
hardenings is possible, the numerical results presented
in this paper only involve a pure kinematic non-linear
Remarks
hardening, as it is enough to describe the ratcheting
(i) The proposed formulation can be extended to non-
phenomenon.
associated plasticity by using a potential F different from
The material properties have the same values for all
the yield function f in the evolution laws (5). Then one
the numerical examples, see Table 1.
observes that the new tensors Dalg and Galg ––equivalent
Throughout, the finite element computations are
to (16a) and (16b)––remain symmetric but the new
carried out using three-dimensional 20-node hexahedral
tangent modulus––equivalent to (25) or (26)––does not,
elements with 3 · 3 · 3 ¼ 27 Gauss integration points.
as expected.
(ii) Only two hardening variables have been consid-
ered as usually done in plasticity, one second-order 4.1. Uniaxial ratcheting effect
tensor and one scalar function. In more elaborate plas-
ticity models or in damage mechanics where a larger The first numerical computation is performed on a
number of internal variables is required, it is more dif- bar of constant section under uniaxial loading. The
ficult to derive closed-form expressions like (18), (19) uniaxial stress is applied cyclically from rm ¼ 2 108
and (26) for the local iteration corrections and the Pa to rM ¼ 4:5 108 Pa corresponding to an asym-
consistent tangent modulus. metric loading process. The stress–strain curve resulting
from the tension-compression cyclic loading is shown in
Fig. 1.
4. Numerical results The non-zero mean load has led to accumulation of
the plastic strain which is known as the uniaxial rat-
The above local integration procedure is now incor- cheting. The ratcheting step, corresponding to the ac-
porated in a three-dimensional finite element program. cumulation of plastic strain during one cycle, only
The description of a given hardening model must include depends on the material properties and loading process,
the first and second derivatives of the yield function f , and is given by
Eq. (9), as well as the derivatives of the hardening energy " 2 #
1 H
a2m
whard , Eq. (2), in order to define the plastic quantities (7) p
De ¼ ln 2 K
ð29Þ
and the algorithmic tensors (16) and (20). K H
a2
K M
Elasticity is assumed to be isotropic, characterized by
where
the Young modulus E and the Poisson ratio m, and the qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hardening energy and the yield function are given by 1 þ 1 þ 2ðrm þ r0 Þ HK
[3,5] am ¼ K
H
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð30Þ
1 2H 1 1 2ðrM r0 Þ HK
whard ¼ waðaÞ ¼ a : a
aM ¼
rffiffiffi 2 3 ð28Þ K
H
3 1 3K
f ðr; aÞ ¼ ks
ak þ a:
a r0
2 2 2H The computed value Dep ¼ 1:88 103 is identical with
the analytical solution within 1%.
where s denotes the deviator of stress r, r0 the initial
yield stress, H the kinematic hardening modulus and K 4.2. Multiaxial ratcheting effect
the saturation velocity. To investigate cyclically loaded
structures, the model (28) is commonly used in the frame Now consider a bar of rectangular section subjected to
of non-associative plasticity, that is, the evolution laws a constant tension r11 ¼ r and symmetrically alternated
Table 1
Material properties
Young modulus E Poisson ratio m Initial yield stress r0 Kinematic hardening Saturation velocity K
modulus H
2 · 1011 Pa 0.3 3 · 108 Pa 3 · 1010 Pa 60
156
1.75 by calculating the evolution laws (5) with potential f

1.5
σ σ given by (28), whereas the yield function is given by (28)
with K ¼ 0 (this is what is done in non-associated
1.25
plasticity). One obtains the following relation between
1 the variations of the two plastic strains ep11 and ep12
0.75 4 r
Dep11 ¼ pffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi Dep12 ¼ 0:596Dep12 ð31Þ
3
σ /σ0
0.5 H
þr r
K 0
2
0.25
0
The numerical factor 0.612 obtained is 3% higher than
the value given by (31).
-0.25
-0.5
4.3. Cylinder under internal pressure
-0.75
0 0 .005 0.01 0.015
ε
Let us consider a thick-walled cylinder of internal
Fig. 1. Uniaxial ratcheting, axial stress r versus axial strain e. and external radii Ri ¼ R ¼ 1 m and Re ¼ 2R ¼ 2 m.
Boundary conditions are applied so as to realize the
plane strain conditions. The mesh of one quarter of the
shear stress r12 , going from s to s, which represents the cylinder contains 503 nodes and 60 elements (10 ele-
coupling between traction and cyclic twisting. Here the ments along the radial direction by 6 elements along the
stresses r and s take the same value 2 · 108 Pa. circumferential direction). The cylinder is subjected to a
The stress–strain curve, Fig. 2, resulting from the cyclic internal pressure, going from p ¼ 0 to a given
cyclic loading shows that combining a constant tensile maximum pressure p ¼ pmax ¼ 430 MPa. Fig. 3 depicts
stress with a symmetric alternated shear yields accu- the variation of pressure p and loop stress rhh versus the
mulation of the axial plastic strain. The multiaxial rat- evolution of inner radial displacement.
cheting is essentially due to the non-zero value of the Since pmax is sufficiently high to induce plasticity
axial stress. during the unloading, a ratcheting phenomenon occurs
In this case, a constant ratcheting step is reached with a nearly constant ratcheting step, similar to Fig. 1.
after a few cycles corresponding to a transitional stage, However, the situation here is different from the previ-
subsequent to the progressive accommodation encoun- ous ones, as the stress state is not homogeneous in the
tered in the case of linear kinematic hardening. cylinder and the plastic zone spreads out progressively in
Since it is difficult to obtain an analytical relation like the radial direction. The peak in rhh in Fig. 3 is not
Eq. (29), we shall confine ourselves to an approximate proper to non-linear kinematic hardening, it also occurs
expression for the limit ratcheting step. This can be done for the same cylinder in perfect plasticity.
1.5
p/σ0
1 1.25
0.75 1 σθθ /σ0
0.5 0.75
0.25
0.5
σ12 /σ0
0.25
0
0 p
-0.25
-0.5
τ -0.25
τ
-0.5
-0.75 τ
σ σ -0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12
-1 u(R)/R
00 .005 0.01 0.015 0.02
ε11
Fig. 3. Thick-walled cylinder, internal pressure p and loop
Fig. 2. Multiaxial ratcheting, shear r12 versus axial strain e11 . stress rhh versus inner radial displacement uðRÞ.
157
4.4. Beam under bending moment these are available. It is also necessary to experimentally
identify the mechanical properties of the theoretical
The last numerical example is related to a beam of model in order to assess its validity. This important
square section (the width and the height are equal, work is not considered in this paper.
b ¼ h ¼ 2 cm, and the length is l ¼ 10 cm), subjected to
a bending moment M applied at each end so that the
beam undergoes a pure bending. The beam is discretized
as a three-dimensional body with 20-node hexahedral 5. Conclusions
elements. The mesh is made of eight elements in length
and six elements in height to take account of the non- In this paper, a general local integration procedure in
homogeneous stress state in the section. small strain elastoplasticity has been analyzed together
Loading and unloading cycles are computed between with the corresponding consistent tangent modulus. The
Mm ¼ 600 N m and MM ¼ 800 N m. The load–dis- proposed formulation––established within the frame-
placement curve, Fig. 4, where M0 is the moment cor- work of the generalized standard materials theory––
responding to the onset of plastification, clearly shows utilizes a minimal number of assumptions, and is general
that ratcheting occurs with a nearly constant ratcheting enough to be applied to arbitrary hardening models. The
step. The result is similar to that obtained by Mahbadi salient features of the present work are summarized
and Eslami [17] using a one-dimensional beam model. below.
The maximum deflection at the center of the beam in- The yield function has been decomposed as the sum
creases regularly and always takes positive values (ex- of one term depending on the stress and the back-stress
cept for the first cycle), see Table 2. Thus, the beam tensor, and another one depending on a scalar. Here the
remains in the upper half-plane during the bending stress and the back stress do not necessarily appear in
process. the form of their difference, as seen in the case of non-
All the numerical results obtained above have been linear kinematic hardening.
compared to analytical or numerical models only, when When the hardening energy waðaÞ in relation (2) take
general expressions other than a quadratic one––simi-
larly to wa ðaÞ in non-linear isotropic hardening––it has
been shown that there appears the additional term
2.25 of o3 wa
oa
: oa3 in Eqs. (16a) and (16b), which does not exist in
2
1.75
the literature. This term is a symmetric fourth-order
1.5 tensor which should be taken into account in general
1.25 theoretical considerations.
1 The plastic corrections have been given in closed
0.75
forms, Eqs. (15)–(19). The consistent tangent modulus
0.5
M/M 0
has a symmetrical expression, Eqs. (25) or (26). All these

0.25
0
expressions have been given in tensorial forms which
-0.25 are ready to be converted to matrical expressions for
-0.5 computational use.
-0.75
v The expressions for the plastic corrections and the
-1
M M consistent tangent modulus involve the scalar function
-1.25
Galg and the fourth-order tensors Dalg , Galg and Lalg , the
-1.5
0 0.005 0.01 0.015 0.02 first two being symmetric. These algorithmic quantities
v/l represent the discrete finite-step counterparts of the
usual theoretical continuum quantities: the plastic scalar
Fig. 4. Beam under pure bending, load–displacement curve:
function G, the elastic tensor D and the plastic tensor G,
moment M versus maximum deflection v.
respectively. When the time step tends to zero, all the
algorithmic quantities tend to their theoretical counter-
parts.
Table 2 The numerical computations using a particular non-
Beam under pure bending, extremum deflections at extremum linear kinematic hardening model have provided results
moments which are in good agreement with analytical or other
M=M0 2 )1.5 2 numerical solutions available in the literature. They also
v=‘ 1.24 · 102 )1.53 · 103 1.43 · 102 show that non-linear kinematic hardening models are
effective in reproducing the important ratcheting phe-
M=M0 )1.5 2 )1.5 nomenon, responsible for one of the structural failure
v=‘ 3.65 · 104 1.61 · 102 2.16 · 103
cases.
158
References [18] Matzenmiller A, Taylor RL. A return mapping algorithm

for isotropic elastoplasticity. Int J Numer Meth Eng 1994;
[1] Alfano G, De Angelis F, Rosati L. Closed-form solution of 37:813–26.
the finite-step constitutive problem in plane stress j2 - [19] Moreau JJ. Proximite et dualite dans un espace hilbertien.
elastoviscoplasticity. ECCMÕ99, European Conference Bull Soc Math France 1965;93:273–99.
on Computational Mechanics, M€ unchen, Germany, 31 [20] Moreau JJ. On unilateral constraints, friction and plastic-
August–3 September, 1999. ity. Lecture notes, CIME, Bressanone, Cremonese, Roma,
[2] Alfano G, Rosati L. A general approach to the evaluation 1973.
of consistent tangent operators for rate-independent elas- [21] Nagtegaal JC. On the implementation of inelastic constit-
toplasticity. Comput Meth Appl Mech Eng 1998;167:75–89. utive equations with special reference to large deformation
[3] Armstrong PJ, Frederick CO. A mathematical representa- problems. Comput Meth Appl Mech Eng 1982;33:469–84.
tion of the multiaxial Bauschinger effect. Technical Report [22] Nguyen QS. On the elastic–plastic initial-boundary value
RD/B/N731, General Electricity Generating Board, Berke- problem and its numerical integration. Int J Numer Meth
ley Nuclear Laboratories, 1966. Eng 1977;11:817–32.
[4] Bathe KJ. Finite element procedures. Englewood Cliffs, [23] Ortiz M, Martin JB. Symmetry-preserving return mapping
NJ: Prentice Hall; 1996. algorithms and incrementally extremal paths: a unification
[5] Chaboche JL. Unified cyclic viscoplastic constitutive of concepts. Int J Numer Meth Eng 1989;28:1839–53.
equations: development, capabilities, and thermodynamic [24] Ortiz M, Pinski PM, Taylor RL. Operator split methods
framework. In: Krausz AS, Krausz K, editors. Unified for the numerical solution of the elastoplastic dynamic
constitutive laws of plastic deformation. Academic Press; problem. Comput Meth Appl Mech Eng 1983;39:137–57.
1996. p. 1–68. [25] Ortiz M, Popov EP. Accuracy and stability of integration
[6] Doghri I. Fully implicit integration and consistent tangent algorithms for elastoplastic constitutive relations. Int J
modulus in elastoplasticity. Int J Numer Meth Eng Numer Meth Eng 1985;21:1561–76.
1993;36:3915–32. [26] Ortiz M, Simo JC. An analysis of a new class of integration
[7] Doghri I. Numerical implementation and analysis of a algorithms for elastoplastic constitutive equations. Int J
class of metal plasticity models coupled with ductile Numer Meth Eng 1986;23:353–66.
damage. Int J Numer Meth Eng 1995;38:3403–31. [27] Palazzo V, Rosati L, Valoroso N. Computational issues of
[8] Doghri I. Mechanics of deformable solids: linear and general isotropic elastoplastic models. In: VI International
nonlinear, analytical and computational aspects. Springer; Conference on Computational Plasticity (COMPLAS VI),
2000. Barcelona, Spain, 11–14 September, 2000.
[9] Eterovic AL, Bathe KJ. A hyperelastic-based large strain [28] Reddy BD, Martin JB. Internal variable formulations of
elasto-plastic constitutive formulation with combined iso- problems in elastoplasticity: constitutive and algorithmic
tropic–kinematic hardening using the logarithmic stress aspect. Appl Mech Rev 1994;47(9):429–56.
and strain measures. Int J Numer Meth Eng 1990;30:1099– [29] Simo JC, Govindjee S. Exact closed-formed solution of the
114. return mapping algorithm for plane stress elasto-visco-
[10] Halphen B, Nguyen QS. Sur les materiaux standard plasticity. Eng Comput 1988;3:254–8.
generalises. J Mecanique 1975;14(1):39–63. [30] Simo JC, Govindjee S. Non-linear b-stability and symme-
[11] Hartmann S, Haupt P. Stress computation and consistent try preserving return mapping algorithms for plasticity and
tangent operator using non-linear kinematic hardening viscoplasticity. Int J Numer Meth Eng 1991;31:151–76.
models. Int J Numer Meth Eng 1993;36:3801–14. [31] Simo JC, Hughes TJR. Computational inelasticity. Springer;
[12] Hughes TJR, Pister KS. Consistent linearization in me- 1998.
chanics of solids and structures. Comput Struct 1978; [32] Simo JC, Hughes TJR. General return mapping algorithms
9:391–7. for rate-independent plasticity. In: Desai CS et al., editors.
[13] Hughes TJR, Taylor RL. Unconditionally stable algo- 2nd Int Conf on Constitutive Laws for Eng Materials:
rithms for quasi-static elasto-viscoplastic finite element Theory and Applications, vol. I. Tucson, Arizona, USA,
analysis. Comput Struct 1978;8:169–73. 5–8 January, 1987. p. 221–31.
[14] Kojic M, Bathe KJ. The ‘‘effective-stress-function’’ algo- [33] Simo JC, Ortiz M. A unified approach to finite deformation
rithm for thermo-elasto-plastic and creep. Int J Numer elastoplasticity based on the use of hyperelastic constitutive
Meth Eng 1987;24:1509–32. equations. Comput Meth Appl Mech Eng 1985;49:221–45.
[15] Krieg RD, Key SW. Implementation of a time dependent [34] Simo JC, Taylor RL. Consistent tangent operators for rate-
plasticity theory into structural computer programs. In: independent elastoplasticity. Comput Meth Appl Mech
Stricklin JA, Saczalski KJ, editors. Constitutive equations Eng 1985;48:101–18.
in viscoplasticity: computational and engineering aspects, [35] Simo JC, Taylor RL. A return mapping algorithm for
AMD-20. New York: ASME; 1976. plane stress elastoplasticity. Int J Numer Meth Eng 1986;
[16] Krieg RD, Krieg DB. Accuracies of numerical solution 22:649–70.
methods for the elastic–perfectly plastic model. J Press [36] Le van A, de Saxce G. Exact expressions for the plastic
Vess Technol, ASME 1977;99:510–5. corrections and the consistent tangent modulus in rate-
[17] Mahbadi H, Eslami MR. Cyclic loading of beams based on independent elastoplasticity. In: VI International Conference
the Prager and Frederick–Armstrong kinematic hardening on Computational Plasticity (COMPLAS VI), Barcelona,
models. Int J Mech Sci 2002;44:859–79. Spain, 11–14 September, 2000.
159
[37] Le van A. Une formulation lagrangienne totale en [38] Wilkins ML. Calculation of elastic–plastic flow. In: Adler
elastoplasticite finie. Habilitation Thesis. Ecole Centrale B et al., editors. Methods of computational physics 3. New
de Nantes, Universite de Nantes, France, 1999. York: Academic Press; 1964.
160
COMPOSITES
SCIENCE AND
TECHNOLOGY
Composites Science and Technology 67 (2007) 507–514
www.elsevier.com/locate/compscitech
Unidirectional compression of fibre reinforcements. Part 1:

A non-linear elastic-plastic behaviour
S. Comas-Cardona *, P. Le Grognec, C. Binetruy, P. Krawczak
Polymers and Composites Technology and Mechanical Engineering Department, Ecole des Mines de Douai,
941 rue Charles Bourseul, BP 10838, 59508 Douai Cedex, France
Received 4 January 2006; received in revised form 19 July 2006; accepted 20 August 2006
Available online 9 October 2006
Abstract
Liquid composite moulding processes are increasingly used to manufacture composite structures. Such processes combine compres-
sion of the fibre reinforcements (in dry and/or lubricated states) and resin flow. A better modelling of the compression of fibre reinforce-
ments would improve the accuracy of hydro-mechanical coupling modelling involved in these processes. Several models are available in
the literature, but none of them consider permanent deformations (such as plasticity). This article presents a methodology to measure
plasticity of fibre reinforcements in dry and impregnated states. Also a non-linear elastic-plastic model, including large deformations,
is proposed for unidirectional compression. Results on glass fibre reinforcements are presented and discussed.
Keywords: Unidirectional compression; Finite strains; A. fabrics/textiles; Elastoplasticity
1. Introduction qualified of ‘‘controlled’’ during RTM, for example, when

the top half of the mould comes into close contact with the
Structural composite materials are increasingly produced reinforcement and compresses it until reaching the desired
using liquid composite moulding (LCM) manufacturing thickness.
techniques. During such processes, dry fibre reinforcements Induced deformations happen for most of LCM pro-
are placed into a rigid half mould, which is closed either with cesses. During monolithic composite RTM manufacturing,
another half mould for resin transfer moulding (RTM) or the use of pressure-driven flows on dry fibre reinforcements
with a flexible bag tooling for vacuum assisted resin transfer can generate a force on the fabric surface that may deform
moulding (VARTM) for instance. The common feature [1]. In the case of sandwich structures manufacturing using
between such processes is that the fibre reinforcement will RTM, besides the initial compression during mould closing
be compacted. Then a flow is created to force the liquid resin of both core and skin reinforcements, the core can also be
through the fibre reinforcement. Positive pressure is used for further compressed (e.g., deformable cores such as poly-
RTM whereas vacuum is used during VARTM. Once the meric foams) [2,3] and even shifted away during the filling
resin has impregnated the fibre reinforcements and has stage [3,4]. During those processes, involving hydro-
cured, the part can be demolded. mechanical coupling, once the reinforcement is impreg-
Two types of fibre reinforcement compressive deforma- nated, it will continue to deform and eventually, when
tions can occur and be observed during processing: con- the injection is completed, it may not fully recover [3,4].
trolled and induced deformations. The deformation is Induced deformations can also be present during processes
such as VARTM. Recent studies have shown that the part
*
Corresponding author. Tel.: +33 3 27 71 21 87; fax: +33 3 27 71 29 81. thickness evolves during the resin filling. Moreover the final
E-mail address: comas-cardona@ensm-douai.fr (S. Comas-Cardona). part thickness is not necessarily uniform and is always
0266-3538/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2006.08.017
161
508 S. Comas-Cardona et al. / Composites Science and Technology 67 (2007) 507–514
Nomenclature
A sample planar area in the deformed configura- h sample height in the current configuration
tion h0 initial sample height
A0 sample planar area in the reference configura- H hardening compression modulus of the fibre
tion reinforcement in the z-direction
C elastoplastic compression tangent modulus of I identity tensor
the fibre reinforcement in the z-direction M material point
Ce elastic compression tangent modulus of the fibre n normal vector on the deformed configuration
reinforcement in the z-direction n0 normal vector on the reference configuration
E Green strain tensor S surface in the deformed configuration
Ee elastic strain tensor S0 surface in the reference configuration
Ep plastic strain tensor u displacement
df force acting on a deformed area element x, y, z cartesian coordinates
df0 transformed force acting on the corresponding a relative shortening in the z-direction
reference area element k nominal stress applied to the fibre reinforcement
F resultant force applied to the sample in the in the z-direction
z-direction P first Piola–Kirchhoff stress tensor
F deformation gradient tensor R second Piola–Kirchhoff stress tensor
larger at the gate than at the vent because of the mechan- During VARTM, when the vacuum is initially applied
ical equilibrium between the reinforcement stress, liquid to the fabrics, the latter are submitted to compression.
pressure, vacuum and atmospheric pressure [5,6]. Then, when the resin infuses the part, the fibre reinforce-
A proper modelling of the processes previously cited, ment is gradually unloaded. Some recent studies focused
especially estimating the final thicknesses of the manufac- on those loading and unloading behaviours. Rudd et al.
tured parts, for instance, requires constitutive laws for applied another compression to a sample that had already
compression of fibre reinforcements in dry and lubricated been compacted. They noted that the reinforcement did not
states. Such models should also take into account the dif- return to its initial thickness before recompaction. They
ferent loading and unloading behaviours of the reinforce- concluded that the reinforcements displayed a pseudoplas-
ment. Numerous experimental studies have been tic behaviour [9]. Another study showed that elastic and
performed for transverse compression of fibre reinforce- permanent deformations are very significant for continu-
ments [7–12]. It is well known that the compression curves ous filament random mat and plain weave fabric compres-
(pressure vs. fibre volume fraction or thickness) display a sions, while viscoelasticity remains less important [17]. Also
non-linear behaviour. Also, depending on the compression because of the limited compression speeds, involved in
speed, the compression curves can be shifted revealing vis- VARTM for instance, viscoelastic effects can be considered
coelastic behaviour. However, Saunders et al. mentioned limited. Even though numerous previous studies focused
that, for a dry plain-woven fabric, the compression exhibits on fibre reinforcement compression modelling, none of
no changes with compression speeds ranging from 0.05 to them seemed to include permanent deformations.
1 mm/min [11]. Also Rudd et al., working on dry glass fibre Finally, during composites manufacturing, in order to
reinforcements (continuous filament random mat and obtain the desired fibre volume fraction, the fibre reinforce-
quasi-unidirectional), observed that there was no signifi- ments are highly compressed. A 50% reduction of the fab-
cant change of compaction behaviour for compression ric thickness (between the natural thickness and the final
speeds between 2 and 10 mm/min even with the presence thickness) is quite usual. Therefore, the modelling of fibre
of binder [9]. reinforcements behaviour in compression does not comply
Researchers attempted to model fibre reinforcement with the small strains assumption and requires the use of a
submitted to transverse compression using power laws large deformation formulation.
[8]. Other studies proposed non-linear elastic models usu- Usually during composites manufacturing, the fibre
ally based on micromechanics, where fibre bundles or fibres reinforcements can be compressed and sheared in the three
are assumed to behave such as elastic beams [12–14]. Those directions of space. However, in this study we focus on the
models lead to a relationship between applied pressure and unidirectional compression behaviour in order to discard
fibre volume fraction and have several empirical parame- shearing effects. Therefore, a non-linear elastic-plastic con-
ters to fit to the experimental data. Other studies focused stitutive law for fibre reinforcement in compressive loading
on the viscoelastic and relaxation behaviours displayed and unloading including finite strains is proposed. An
by the fibre reinforcement using a Maxwell–Wiechert visco- experimental methodology is given in order to accurately
elastic model [15] or a non-linear viscoelastic model [16]. measure the model’s parameters. Results of the modelling
162
S. Comas-Cardona et al. / Composites Science and Technology 67 (2007) 507–514 509
will be given for loading and unloading of a glass fibre the particular sample geometry and deformation involved
woven fabric in dry state to check the validity of the con- in the considered test.
stitutive law. For the compression of few plies of fibre reinforcements
between parallel compression platens (Fig. 2), the following
2. Large deformation formulation assumptions can be made:
2.1. General framework • the in-plane (x, y) deformations are neglected with
respect to the through-thickness (z) deformation,
When dealing with finite deformations, i.e. strains because the thickness of the sample is much smaller than
greater than 10%, the reference and deformed states can the diameter of the platens. Therefore, the compression
no longer be superposed. Strains and stresses can be can be considered unidirectional
defined in both configurations, leading to Eulerian or • the compression is homogeneous within the thickness
Lagrangian approaches. The readers may refer to Simo direction (z), thus the displacement field u can be
and Hughes [18], for instance, for more details and funda- expressed as:
mentals on large deformations.
Using a total Lagrangian formulation, the nominal uðMÞ ¼ uz z ¼ ðaz þ bÞz ð6Þ
stress P (or first Piola–Kirchhoff stress) and the material Choosing arbitrarily uz(0) = 0 along with uz(h0) = Dh
stress R (or second Piola–Kirchhoff stress), which, respec- leads to a = Dh/h0 and b = 0.
tively, relate forces in the current and reference configura-
tions to areas in the reference configuration (Fig. 1), are The deformation gradient tensor F can be expressed
defined as: using the displacement field:
df ¼ P n0 dS 0 ð1Þ 2 3
1 0 0
df 0 ¼ R n0 dS 0 ð2Þ 6 7
F ¼ I þ ru ¼ 4 0 1 0 5 ð7Þ
The force in the reference configuration is deduced from 0 0 1þa
the one in the current configuration using the suitable map-
Using Eq. (5), the Green strain tensor becomes:
ping, involving the deformation gradient, as follows: 2 3
0 0 0
df ¼ F df 0 ð3Þ 1 6 0 7
E ¼ ðFT F IÞ ¼ 4 0 0 5 ð8Þ
which yields: 2 2
0 0 a þ a2
P¼FR ð4Þ
Let us assume a homogeneous nominal stress state within
The deformation is measured using the Green strain tensor the sample:
E (energy conjugate to the second Piola–Kirchhoff stress), 2 3
which can be defined in terms of the displacement u or 0 0 0
6 7
the deformation gradient tensor F: P ¼ 40 0 0 5 ð9Þ
1 1 0 0 k
E ¼ ðru þ rT u þ rT u ruÞ ¼ ðFT F IÞ ð5Þ
2 2 where k is the applied pressure. The stress tensor P verifies
both equilibrium and boundary conditions.
2.2. Compression between parallel platens
The constitutive law must relate the material stress R to

the Green strain E. Those tensors have to be expressed for
Fig. 1. Reference and deformed configurations for large deformation

formulation. Fig. 2. Unidirectional compression geometry and displacement field.
163
The constitutive law involves the material stress R that is 3.2. Unloading compression
written:
2 3 During unloading, the behaviour is assumed to be non-
0 0 0
6 linear elastic. Therefore, stress and strain are related
R ¼ F1 P ¼ 4 0 0 0 75 ð10Þ through the following relationship:
0 0 k
1þa
k a2
d ¼ C e ðEezz Þd a þ ð16Þ
1þa 2
3. Non-linear elastic-plastic model
In order to find the unknown functions C e ðEezz Þ, H(Ezz) and
Epzz ðEzz Þ, whether for loading or unloading steps, compres-
3.1. Loading compression
sion tests will be performed following procedures detailed
below. Those coefficients will be measured for fibre rein-
Considering the unidirectional compression (loading
forcements in dry and impregnated states.
step) of fibre reinforcements, in order to be consistent with
the non-linear behaviour of the fibre reinforcements
4. Experimental setup and data analysis
observed experimentally [1,8–10,15], a constitutive law that
includes non-linearities must be used:
Compression tests on fibre reinforcement samples are
dRzz ¼ CðEzz ÞdEzz ð11Þ carried out on a material testing machine (Zwick). A glass
Substituting Eqs. (8) and (10) into Eq. (11) leads to: twill-weave fibre reinforcement is tested (Table 1). Because
of the level of stress needed to perform the compressions, a
k a2 force cell of 10 kN is used.
d ¼ CðEzz Þd a þ ð12Þ
1þa 2 For the compression of dry fabrics, the sample, consti-
It can be seen from Eq. (12) that such unidirectional com- tuted of several plies (Table 1), is inserted between two
pression displays both material non-linearities (modulus compression platens (Fig. 3). For the compression of
depending on the current strain) and geometric non-linear- impregnated fabrics, the setup is slightly modified. A fluid
ities. The elastoplastic tangent modulus C, which has been receptacle and a perforated compression platen are added
introduced in Eq. (11), includes both elastic and plastic (Fig. 4). The fabric, impregnated with silicone oil using a
behaviours during the loading step. syringe for each ply, is inserted between the upper compres-
In order to dissociate permanent deformations (plastic- sion platen and the lower perforated platen. The combina-
ity) from elastic deformations during loading, the strain tion of a low viscosity fluid, a low crosshead speed and a
tensor can be additively decomposed as: perforated compression platen allows to limit pressure
due to the expelled liquid flow. Therefore, with those exper-
Ezz ¼ Eezz þ Epzz ð13Þ imental setup and conditions, the measurement is limited to
Therefore, elastic (Ce) and hardening (H) moduli can be de- the mechanical response of the impregnated fabric (drained
fined, such that: compression).
The second Piola–Kirchhoff stress (Eq. (10)) and the
dRzz dRzz dRzz Green strain (Eq. (8)) are calculated using the following
dEzz ¼ ¼ dEezz þ dEpzz ¼ e þ ð14Þ
C C H relations:
with k F
Rzz ¼ ¼ ð17Þ
dRzz dRzz 1þa A0 ð1 þ aÞ
Ce ¼ e H¼ p ð15Þ
dEzz dEzz a2
Ezz ¼ a þ ð18Þ
Among the unknown functions, the plastic strain Epzz ðEzz Þ 2
has to be determined. Regarding plastic strain, two ap- where F is the force given by the force cell and a = (h h0)/
proaches can be envisaged. The first one would be to model h0. The material testing machine gives the current sample
the plastic function as a superposition of several physical ef- height h. The initial sample height h0 is measured with
fects that create plasticity such as nesting, lubrication, fric- the material testing machine, when the compression force
tion. . . This model would be a very complete tool to predict applied to the sample reaches 5 N.
the plastic behaviour of fibre reinforcements in compression
Table 1
but would also be a multi-parameter model. Generally such
Material characteristics and experimental conditions
models are not convenient to implement in simulation tools,
Areal weight 1500 g/m2
especially in industrial applications and environments, be-
Number of plies 4
cause of the large number of parameters involved and the Stacking sequence [0, 90, 90, 0]
difficulty to measure them. The second approach, the one Initial sample height 6.8 mm
retained for this study, would be a more experimental anal- Crosshead speed 0.5 mm/min
ysis using suitable compression tests to determine the plas- Silicone oil viscosity (21 C) 0.1 Pa s
Sample in-plane diameter 135 mm
ticity of the dry and lubricated fibre reinforcements.
164
Fig. 3. Experimental setup for dry fibre reinforcement compression tests.
Fig. 6. Example of compression test responses that allows to extract

compression modulus.
force of 5 N) provides the amount of permanent deforma-

tion retained in the sample Epzz . Each compression cycle
gives a data point of the plastic strain vs. total strain curve.
Secondly, whether the sample undergoes a loading or
unloading compression test, the moduli of dry and impreg-
nated fibre reinforcements are, respectively, taken as the
local slope of the stress–strain curves (Fig. 6):
Rzz ¼ f ðEzz Þ ð19Þ
Fig. 4. Experimental setup for impregnated fibre reinforcement compres-
sion tests under drained conditions. 5. Dry fabric results
Since the aim of the study is to measure the fabric com- All results presented afterwards have been obtained
pression modulus including plastic effects, the compression using a unique and low compression speed (i.e., 0.5 mm/
tests are performed following two distinct procedures. min) in order to maintain viscoelastic effects negligible.
First, for plasticity, the sample is loaded and subsequently
unloaded. As shown in Fig. 5, the loading step deforms the 5.1. Compression loading and unloading
sample up to a strain Ezz, while the unloading (down to a
The samples are submitted to loading and unloading
cycles. Fig. 7 shows the results plotted using the material
stress and the Green strain as defined in Eqs. (17) and (18).
5.2. Plasticity
From the loading and unloading compression cycles,

plastic strains can be extracted, following the procedure
detailed previously. It can be seen in Fig. 8 that the plastic
strain develops almost linearly during compression. As
Somashekar et al. mentioned [17], for some fabrics in com-
pression, most of the deformation is plastic. For the glass
twill-weave fabric studied here, the plastic strain represents
70% of the total strain.
5.3. Loading compression moduli
When the fibre reinforcement is submitted to compres-

Fig. 5. Example of compression test responses that allows to extract sion, it experiences elastic and plastic deformations. Using
plasticity. Eq. (15) and the results from Fig. 8, elastic and hardening
165
Fig. 7. Stress–strain response of the fibre reinforcement submitted to Fig. 9. Elastic (Ce) and hardening (H, plastic) compression moduli of the
loading–unloading compressions. glass twill-weave in dry state.
Fig. 8. Plasticity of the glass twill-weave in dry state.

Fig. 10. Unloading compression modulus vs. total strain of the glass twill-
moduli can be found (Fig. 9). As expected, the elastic mod- weave in dry state.
ulus is greater than the hardening one. The non-linearity
between stress and strain is confirmed here (both moduli
are not constant with respect to strain). (Fig. 11). Therefore, unloading moduli only depend on
the current elastic strain whatever the level of stress
5.4. Unloading compression modulus reached before unloading.
If we consider that fibre reinforcements in compression 5.5. Comparison between loading and unloading elastic
behave such as non-linear elastic-plastic materials, the dis- moduli
charge should be fully elastic. The unloading compression
moduli of the material obtained from different unloading If we consider that the dry fibre reinforcement behaves
stages are given in Fig. 10. Systematically, the unloading as an elasto-plastic material, the elastic moduli during
modulus decreases when the fibre reinforcement sample is loading or unloading should be equal. Fig. 12 shows that
unloaded. Using the results of compression cycles (Fig. 7) both loading and unloading elastic moduli are identical.
and the results of plasticity obtained in Fig. 8, the unload-
ing (elastic) modulus curves can be plotted with respect to 6. Impregnated fabric results
total strain. The unloading elastic modulus curves can also
be plotted with respect to elastic strain. Such graph shows The plasticity is influenced by the impregnation of the
that the unloading modulus curves create a master curve fabrics as shown in Fig. 13 where both dry and impreg-
166
nated plastic strain evolution have been plotted. For such

twill-weave fabric, plasticity is induced by the compression
of the yarns (fibre/fibre interactions) and nesting effects
(yarn/yarn interactions). The lubrication facilitates the
yarn and fibre imbrications and reorganisations.
The elastic compression modulus of the dry fibre rein-
forcement is always greater than the one of the impreg-
nated fibre reinforcement (Fig. 14). This effect can be
explained by the presence of fluid acting as lubricating
agent. The fluid helps to reduce the friction between yarns
and fibres when the compression occurs. The hardening
modulus is also lower for the impregnated fibre reinforce-
ment (Fig. 15).
Fig. 11. Unloading compression modulus vs. elastic strain of the glass
twill-weave in dry state.
Fig. 14. Comparison of elastic compression moduli with respect to total

strain for the glass twill-weave fabric in dry and impregnated states.
Fig. 12. Comparison of loading and unloading elastic compression

moduli with respect to elastic strain.
Fig. 13. Plastic strain evolution with respect to total strain for the glass Fig. 15. Comparison of hardening compression moduli with respect to
twill-weave fabric in dry and impregnated states. total strain for the glass twill-weave fabric in dry and impregnated states.
167
7. Conclusion [2] Binetruy C, Advani SG. Foam core deformation during liquid
molding of sandwich structures: modeling and experimental analysis.
J Sandwich Struct Mater 2003;5(4):351–76.
A simple compression setup based on a material testing [3] Wirth S, Gauvin R, Kendall K. Experimental analysis of
machine has been proposed to test fibre reinforcements in core crushing and core movements in RTM and SRIM foam
either dry or impregnated state. As mentioned in other cored composite parts. J Reinf Plast Comp 1998;17(11):
experimental studies [17], the results obtained for plastic 964–88.
strains characterisation show the importance of permanent [4] Al-Hamdan A, Rudd CD, Long AC. Dynamic core movement during
liquid moulding of sandwich structures. Composites Part A
deformations during unidirectional compression of fibre 1998;29(3):273–82.
reinforcements studied here. For the twill-weave fabric [5] Acheson JA, Simacek P, Advani SG. The implications of fibre
studied here, the plasticity develops linearly with respect compaction and saturation on fully coupled VARTM simulation.
to the deformation. Composites Part A 2004;35(2):159–69.
In terms of compression modulus, the fluid facilitates [6] Tackitt KD, Walsh SM. Experimental study of thickness gradient
formation in the VARTM process. Mater Manuf Process 2005;20:
the compaction leading to a lower compression modulus 607–27.
of impregnated fibre reinforcements with respect to the [7] Trevino L, Rupel K, Young WB. Analysis of resin injection molding
dry ones. This effect is due to the diminution of friction in molds with replaced fiber mats I: Permeability and compressibility
thanks to lubrication brought by the fluid. measurements. Polym Comp 1991;12(1):20–9.
A non-linear elastic-plastic constitutive law, using a [8] Toll S, Manson JAE. An analysis of the compressibility of fiber
assemblies. In: Proceedings of the FRC Conference 1994, Newcastle
large deformation formulation, whose parameters are mea- upon Tyne, UK, p. 25/1–10.
surable, has been proposed to model the behaviour of fibre [9] Rudd CD, Bulmer LJ, Morris DJ. Compaction and in-plane
reinforcements in unidirectional compression. Such model permeability characteristics of quasi-unidirectional and
would help to better simulate composite manufacturing, continuous random reinforcements. Mater Sci Tech 1996;12:
especially resin infusion processes, where a fine description 436–44.
[10] Robitaille F, Gauvin R. Compaction of textile reinforcements for
of fibre reinforcements in loading and unloading compres- composites manufacturing I: review of experimental results. Polym
sion is required. Comp 1998;19(2):198–216.
One of the assumptions made in this study was to con- [11] Saunders RA, Lekakou C, Bader MG. Compression and microstruc-
sider that the displacement field within the thickness of the ture of fibre plain woven cloths in the processing of polymer
sample was linear (i.e. the deformation is homogeneous). composites. Composites Part A 1998;29(4):443–54.
[12] Chen B, Chou TW. Compaction of woven-fabric preforms:
That assumption is acceptable for rather thin samples. In nesting and multi-layer deformation. Comp Sci Technol
the case of very thick samples, it is known that the deforma- 2000;60(12–13):2223–31.
tion is not necessarily homogeneous and that the fabrics are [13] Lekakou C, Johari MAKB, Bader MG. Compressibility
more packed in the compression platen regions than at the and flow permeability of two-dimensional woven reinforce-
centre of the sample. Further work would deal with that ments in the processing of composites. Polym Comp
1996;17(5):666–72.
inhomogeneous displacement field for thick samples. [14] Batch GL, Cumiskey S, Macosko CW. Compaction of fiber
Also, the compression involved in this study is unidirec- reinforcements. Polym Comp 2002;23(3):307–18.
tional. It is a limiting case of the more general 3D compres- [15] Kim YR, McCarthy SP, Fanucci JP. Compressibility and relaxation
sion that occurs in composite manufacturing. Further work of fibre reinforcements during composite processing. Polym Comp
would also focus on the fact that some out-of-plane defor- 1991;12(1):13–9.
[16] Kelly PA, Umer R, Bickerton S. Viscoelastic response of dry and wet
mation (shearing) occurs when double curvature is present fibrous materials during infusion processes. Composites Part A
in preforming stages, for instance. 2006;37:868–73.
[17] Somashekar AA, Bickerton S, Bhattacharyya D. An experimen-
References tal investigation of non-elastic deformation of fibrous reinforce-
ments in composites manufacturing. Composites Part A 2006;37:
[1] Han K, Trevino L, Lee LJ, Liou M. Fibre mat deformation in liquid 858–67.
composite molding. I: Experimental analysis. Polym Comp [18] Simo JC, Hughes TJR. Computational inelasticity. New York:
1993;14(2): 144–50. Springer; 1998.
168
Available online at www.sciencedirect.com
International Journal of Solids and Structures 45 (2008) 64–86

www.elsevier.com/locate/ijsolstr
Elastoplastic bifurcation and collapse of axially

loaded cylindrical shells
a,* b
Philippe Le Grognec , Anh Le van
a
Ecole des Mines de Douai, Polymers and Composites Technology and Mechanical Engineering Department,
941 rue Charles Bourseul—BP 10838, 59508 Douai Cedex, France
b
GeM (Laboratory of Civil and Mechanical Engineering), Faculty of Science, University of Nantes,
2 rue de la Houssinière—BP 92208, 44322 Nantes Cedex 3, France
Received 11 May 2006; received in revised form 10 July 2007

Available online 6 September 2007
Abstract
In this paper, a shell finite element is designed within the total Lagrangian formulation framework to deal with the plas-
tic buckling and post-buckling of thin structures, such as cylindrical shells. First, the numerical formulation is validated
using available analytical results. Then it is shown to be able to provide the bifurcation modes—possibly the secondary
ones—and describe the complex advanced post-critical state of a cylinder under axial compression, where the theory is
no longer operative.
Keywords: Bifurcation; Buckling; Elastoplasticity; Finite element method; Axially compressed cylinders
1. Introduction
Failure of slender or thin structures which is mainly due to the buckling phenomenon implies the analysis of
buckling and post-buckling behaviors for their mechanical design, namely the calculation of the critical loads,
the bifurcation modes and the post-critical equilibrium branches.
The problem is particularly difficult with thicker structures where plasticity may occur before buckling.
Shanley (1947) was the first to give the tangent modulus critical load for a discrete model. He showed that
the load increases during the post-bifurcation stage, owing to the elastic unloading in the structure. Hill
(1958) extended these results to a three-dimensional continuum by using the concept of ‘‘comparison elastic
solid’’. He examined the uniqueness and stability criteria, and pointed out the difference between bifurcation
and stability.
More precisely, the critical problem of a cylinder under axial compression was the subject of a lot of exper-
imental and numerical investigations (Bushnell, 1982, 1985). A wide variety of experimental results were
*
Corresponding author. Tel.: +33 3 27 71 23 21; fax: +33 3 27 71 29 81.
E-mail addresses: legrognec@ensm-douai.fr (P. Le Grognec), anh.le-van@univ-nantes.fr (A. Le van).
doi:10.1016/j.ijsolstr.2007.07.017
169
P. Le Grognec, A. Le van / International Journal of Solids and Structures 45 (2008) 64–86 65
obtained in terms of the boundary conditions, the geometries and the material parameters. On the other hand,
different numerical models were devised in order to predict the critical loads as well as the advanced post-
buckling behavior. Some pioneering experimental, theoretical and numerical results in the literature will be
briefly presented in the sequel.
Large discrepancies have always been observed between theoretical and experimental results. The most
important feature is that the experimental critical loads sometimes fall down to less than half of the analytical
or numerical values. These differences, due to unavoidable imperfections in experiments, were explained by
many authors. Among others, in his asymptotic analysis of plastic post-buckling, Hutchinson (1974) examined
the influence of various geometric imperfection types on the buckling and post-critical behavior of several
structures. It allows one to determine safety factors for the critical load versus the imperfection size and assess
the stability of the bifurcated branch. Despite the significant influence of the imperfections on the buckling
behavior of a cylinder under axial compression, we shall not account for any type of imperfection in this paper
and will only analyze the critical and post-critical behavior of a perfect cylinder.
From a material point of view, different critical values are obtained in the literature, depending on the plas-
ticity theory considered. Indeed, when numerically (or theoretically) solving the problem of plastic buckling,
one can choose between the deformation and the incremental theories, which have each their own advantages
and drawbacks. The critical stresses obtained by the deformation theory are fairly far from the ones corre-
sponding to the flow theory. As a matter of fact, the solutions derived from the deformation theory turn
out to compare best with the experimental results, although such a theory does not take into account the elas-
tic unloading possibility. On the contrary, the flow theory overpredicts the critical loads. However, the dis-
crepancies can be accounted for and quantified through an analysis of imperfection sensitivity.
More than 50 years ago, Batdorf (1949) and Bijlaard (1949) already tried to explain these differences
between the two theories. Later, Christoffersen and Hutchinson (1979) proposed a new ‘‘corner yield surface’’
theory which realized a compromise between both theories and was used in number of numerical develop-
ments in the literature. Dealing with axially compressed cylinders, Gellin (1979) analyzed the effects of an axi-
symmetric modal imperfection on the plastic buckling, using the deformation theory. Tvergaard analyzed the
effect of geometric and material parameters on the buckling behavior of a cylinder under axial compression
(Tvergaard, 1983a,b). He numerically solved the compressed shells using the phenomenological theory of
Christoffersen and Hutchinson, and validated his numerical approach by the experimental results obtained
by Batterman, Lee, Horton and Johnson in the sixties. He considered an initial sinusoidal axisymmetric defect
and looked for a possible bifurcation towards a non-axisymmetric mode. For relatively thin shells, he found a
bifurcation point just before the maximum load. For thicker shells, however, the maximal load occurs simul-
taneously as the axisymmetric deformed shape is localized at both ends in the form of outward axial half-
waves. Consequently, the bifurcation is delayed or may merely disappear (very thick shells display an axisym-
metric failure). When the bifurcation precedes the localization, a ‘‘diamond’’ mode is observed, displaying
both axial and circumferential waves throughout the structure. Otherwise, the localization is followed by
the loss of axisymmetry and gives rise to an ‘‘elephant foot’’ mode with regularly distributed ‘‘toes’’ at both
ends of the cylinder. According to Tvergaard, the bifurcation is delayed because of the localization which stiff-
ens the structure. Moreover, Tvergaard showed that the circumferential order of the solution decreases as the
radius-to-thickness ratio decreases. Also, he noticed that a weak hardening makes it easier for either an axi-
symmetric failure or an ‘‘elephant foot’’ mode to occur.
Recently, the analysis of this transition between the axisymmetric mode and the ‘‘diamond’’ mode was con-
ducted on a perfect cylinder by Goto and Zhang (1999) using the flow plasticity theory. It was shown therein
that the boundary conditions are a deciding factor for obtaining the previous types of buckling modes. In the
case of a moderately thick cylinder with built-in ends, Goto and Zhang found two successive bifurcation
points: the first one (on the fundamental branch) near the limit load leading to an axisymmetric mode, the
second one (on the decreasing primary bifurcated branch) leading to a localized non-axisymmetric secondary
mode, named ‘‘elephant foot’’. The thinner the shell, the earlier the secondary mode occurs. Thus, if the shell is
thin enough, the second bifurcation point supersedes the first one, prevents from any localization and gives
rise to a ‘‘diamond’’ mode. Conversely, if the shell is thick enough, the second bifurcation point disap-
pears—at least in the computed range—and the failure is very likely axisymmetric. The circumferential wave
number decreases with the thickness. Less obvious is the observed fact that an increase of the length-to-radius
170
66 P. Le Grognec, A. Le van / International Journal of Solids and Structures 45 (2008) 64–86
ratio makes the second bifurcation occur earlier too, without changing the circumferential order of the
deformed shape. In the case of simply supported ends, the wave is very near the ends, so that the localiza-
tion—if it exists—never leads to a second bifurcation. Eventually, the free end condition produces one single
non-axisymmetric bifurcation, which is immediately localized. Again, the circumferential order of the solution
decreases with the thickness.
Comparisons between the predictions of both flow and deformation theories and experimental results were
then performed in Kyriakides et al. (2005), Bardi and Kyriakides (2006) and Bardi et al. (2006), in the case of
axially loaded cylindrical shells, including the effect of anisotropy with the use of the Hill’s yield criterion. The
flow theory significantly overpredicts both the critical stresses and strains, whereas the deformation theory
gives rise to predictions much closer to the experimental values. Furthermore, better predictions were reached
with the deformation theory, when including imperfections whose amplitude was only a small fraction of the
thickness instead of the full thickness used by Gellin and Tvergaard.
Finally, Peek (2000) derived analytical solutions for the immediate post-critical behavior of a cylinder under
axial compression in the axisymmetric case only, considering finite strains.
All the above-mentioned facts show that it is necessary to use a numerical tool to analyze the influence of
the geometries, the material parameters and the boundary conditions on the complete non-linear buckling and
post-buckling response of compressed shells. All plastic bifurcations are strongly influenced by the constitutive
model used. Usually, the deformation theory of plasticity or the corner theory are preferred as they predict
more accurate critical loads than the flow theory, although some recent works have tried to show that the flow
theory is also able to give good results. In this paper, our aim is not to compare the merits of the different
theories. Rather, we adopt the J2 flow theory and aim to test the ability of numerical tools developed in pre-
dicting plastic bifurcations and post-bifurcation paths. It will be shown that (i) the proposed formulation
allows to obtain primary and secondary buckling modes and to exhibit the localization phenomenon, as
observed in experiments; (ii) and that the predicted bifurcation stresses may indeed be higher than the exper-
imental ones. Nothing will be done to compare between numerical and experimental results, since this is a tre-
mendous task which is out of the scope of the paper.
2. Outline of the paper
This paper is devoted to the numerical computations of the critical loads, the bifurcation modes and the
advanced post-critical behaviors of compressed shells in the elastoplastic range.
The principle of virtual work and the total Lagrangian formulation will be used to build a shell finite ele-
ment encompassing the geometric non-linearities (finite rotations and finite strains) and the material non-lin-
earities (elastoplastic behavior with non-linear isotropic hardening). The shell element is designed to deal with
buckling and advanced post-buckling of thin structures, it will be shown to correctly exhibit a wide variety of
bifurcation modes and possibly the secondary ones.
First, the numerical formulation will be validated using available analytical results. Then it will be seen that
the numerical computations can provide more complete results on the advanced post-critical state of a cylin-
der under axial compression, where the theory is no longer operative.
3. Shell finite element formulation
3.1. Shell kinematics with finite rotations
Let us consider a three-dimensional shell-like body with reference volume X0 at the initial time, subjected to
finite quasi-static elastoplastic transformations. In this work, the shell-like body is modeled by the Mindlin–
Reissner theory, it is represented by a middle surface and an inextensible director field which is not constrained
to be normal to the middle surface, except at the reference configuration.
The reference and current positions of a particle on the middle surface are denoted by ~ P 0 ðn; gÞ and ~
P ðn; gÞ,
respectively, where (n, g) are the surface parameters. The displacement of the middle surface particle is denoted
~ ðn; gÞ ¼ ~
U P ðn; gÞ ~
P 0 ðn; gÞ.
The reference and current positions of any particle in the shell are given respectively by:
171
~¼~
X P 0 ðn; gÞ þ f~
T ðn; gÞ
ð1Þ
x¼~
~ P ðn; gÞ þ f~tðn; gÞ
where f is the through-thickness parameter, ~

T ðn; gÞ and ~
tðn; gÞ are the reference and current director vectors,
respectively. The displacement of any particle is:
~ g; fÞ ¼ U
Dðn; ~ ðn; gÞ þ fð~
tðn; gÞ ~
T ðn; gÞÞ: ð2Þ
All the vectors will be resolved in a fixed Cartesian basis:

~ ¼ X i~
X ei ; ~
T ¼ T i~
ei ; x ¼ xi~
~ t ¼ ti~
ei ; ~ ei ð3Þ
where implicit summations are made on repeated indices.

The deformation gradient is:
gi G
F ¼~ ~i ð4Þ
where fG ~i g ~
i¼1;2;3 is the contravariant natural basis in the reference configuration, dual to fGi gi¼1;2;3 , and
f~
gi gi¼1;2;3 is the covariant natural basis in the current configuration:
~ ~ ~
~ 1 ¼ oX ; G
G ~ 2 ¼ oX ; G~ 3 ¼ oX
on og of
ð5Þ
o~x o~
x o~
x
g1 ¼ ; ~
~ g2 ¼ ; ~ g3 ¼ :
on og of
~i g
Here, the variation of the metric through the shell thickness is neglected, so that the bases fG ~i
i¼1;2;3 and fG gi¼1;2;3
are equal to those defined on the middle surface and denoted by f~ Ai g and f~ Ai g
i¼1;2;3 . Thus, the metric tensor
i¼1;2;3
is defined, respectively, in its covariant and contravariant components as Aij ¼ ~

Ai :~
Aj and Aij ¼ ~
Ai :~
Aj .
It follows from the definition of the deformation gradient:
aa ~
F ¼ ð~ t~
Aa þ ~ t;a ~
T Þ þ f~ Aa ð6Þ
where the Greek indices take the values 1 or 2.

The motion of the inextensible director is described by:
t ¼ R:~
~ T ð7Þ
where the rotation tensor R is orthogonal (R1 = RT) and can be associated with the skew-symmetric tensor H
through the exponential mapping:
X
þ1
Hk
R ¼ exp H ¼ : ð8Þ
k¼0
k!
Let us define the axial vector ~

h by H:~
h ¼~
0 and H:~v ¼~h ^~ v 2 R3 . From the equality R:~
v, 8~ h ¼~
h, it follows that
~
R is the rotation about h. Then an explicit expression for Eq. (8) is given by the Rodrigues formula:
sin h 1 cos h ~ ~
R ¼ cos hI þ Hþ hh ð9Þ
h h2
where h ¼ k~hk is the magnitude of the rotation.
~
Since h is perpendicular to ~
T , the director field ~
t can be recast in the simplified form:
sin h ~ ~
t ¼ R:~
~ T ¼ cos h~
Tþ h ^ T: ð10Þ
h
Denoting ~
h ¼ hi~
ei , one gets the components of ~
t in the fixed Cartesian basis f~
ei gi¼1;2;3 :
172
8 9 8 19 8 2 3 2
9
1 3
<t >
> = <T >
> <h T h T >
= sin h > =
t2 ¼ cos h T 2 þ h3 T 1 h1 T 3 : ð11Þ
: 3>
> ; : 3>
> ; h > : 1 2 >
;
t T h T h2 T 1
~
By defining a new local orthonormal basis f~ Ei gi¼1;2;3 at each point of the middle surface as ~
E3 ¼ ~
T, ~
E1 ¼ k~AA1 k
1
and ~
E2 ¼ ~ E3 ^ ~ h¼^
E1 , and by denoting ~ E1 þ ^h2~
h1 ~ E2 , Eq. (11) becomes:
8 9 8 9 8 9
^2
< ^t >
1
> = <0>
> = sin h >< h > =
^t2 ¼ cos h 0 þ ^
h : 1 ð12Þ
: ^3 >
> ; : >
> ; h >: >
;
t 1 0
3.2. Variations of kinematic variables
The linearization of the kinematic variables at a given configuration is performed by using the directional
derivative with respect to a small parameter e. The following relations hold for small values of e:
~
Pe ¼ ~ ~
P þ edU
ð13Þ
~ T ¼ Rð~
te ¼ Re :~ h þ ed~
hÞ:~
T
which yield:
d ~
d~
P¼ P e je¼0 ¼ dU ~
de
ð14Þ
d sin h h ~ ~ ~ i h cos h sin h
d~
t¼ ~
te je¼0 ¼ ðh:dhÞT þ d~
h^~
T ¼ ð~
h:d~
hÞð~
h^~
T Þ:
de h h3
The latter relation can be written in the local basis f~Ei gi¼1;2;3 :
8 9
1 ( )
< d^t >
> = ^ 1
d h
d^t2 ¼ ½U
: ^3 >
> ; d^h2
dt
2 2 3 2 33 ð15Þ
0 1 ^h2 ^h1 ^h2 ^h2
6 sin h 6 7 sin h h cos h 6 ^1 ^1 ^h1 ^h2 7 7
½U ¼ 4 41 0 5þ 3 4 hh 55
h h
^
h1 ^ h2 0 0
ei gi¼1;2;3 for computational purposes.
and then transformed in the Cartesian basis f~
3.3. Strains
The Green strain tensor E is written in the contravariant basis f~

Ai gi¼1;2;3 as:
1 1
E ¼ ðFT :F IÞ ¼ ð~ aj ~
ai :~ Ai :~
Aj Þ~
Ai ~ Ai ~
Aj ¼ Eij~ Aj ð16Þ
2 2
or:
Aa ~
E ¼ Eab~ Ab þ Ea3~ T þ E3a~
Aa ~ T ~
Aa þ E33~
T ~
T: ð17Þ
aa ¼ ~
Since ~ ~ ;a , the components of E can be written as functions of the through-thickness parameter f:
Aa þ U
ð0Þ ð1Þ ð2Þ
Eij ¼ Eij þ fEij þ f2 Eij ð18Þ
where:
173
ð0Þ 1
Eab ¼ ð~ ~ ;b þ ~
Aa :U Ab :U~ ;a þ U
~ ;a :U
~ ;b Þ
2
ð0Þ ð0Þ 1
Ea3 ¼ E3a ¼ ð~ t~
Aa :ð~ ~ ;a :~
TÞ þ U tÞ
2 ð19Þ
ð1Þ 1 ~ ~ ~ ~ ~ ~
~ ~ ~ ~
Eab ¼ ðAa :ðt;b T ;b Þ þ Ab :ðt;a T ;a Þ þ U ;a :t;b þ U ;b :t;a Þ
2
ð2Þ 1
Eab ¼ ð~ t; ~
t; :~ T ;a :~
T ;b Þ:
2 a b
ð0Þ ð1Þ ð2Þ
All other components, in particular E33 , E33 and E33 , are zero since the director field is assumed to be inex-
tensible. However, the expression for E33 will rather be derived from the plane stress condition, as seen further.
The variation of the Green tensor is computed by:
dEð~
P e ;~
te Þ
dE ¼ ð20Þ
de je¼0
which yields:
ð0Þ 1
dEab ¼ ðð~Aa þ U ~ ;a Þ:dU~ ;b þ ð~ ~ ;b Þ:dU
Ab þ U ~ ;a Þ
2
ð0Þ ð0Þ 1
dEa3 ¼ dE3a ¼ ðð~ Aa þ U ~ ;a Þ:d~ ~ ;a :~
t þ dU tÞ
2 ð21Þ
ð1Þ 1
dEab ¼ ðð~Aa þ U ~ ;a Þ:d~t;b þ ð~ Ab þ U~ ;b Þ:d~
t;a þ dU ~ ;a :~ ~ ;b :~
t;b þ dU t;a Þ
2
ð2Þ 1
dEab ¼ ðd~ t;a :~
t;b þ~ t;a :d~
t;b Þ:
2
3.4. Stresses
The second Piola–Kirchhoff stress tensor R is resolved in basis f~

Ai gi¼1;2;3 :
Ai ~
R ¼ Rij~ Aa ~
Aj ¼ Rab~ Ab þ Ra3~ T ~
T þ R3a~
Aa ~ Aa þ R33~
T ~
T: ð22Þ
In the sequel, use will be made of the stress resultant tensors defined as:
Z h
2
Ai ~
n ¼ nij~ Aj nij ¼ Rij df
h2
Z h
2
Ai ~
mð1Þ ¼ mijð1Þ~ Aj mijð1Þ ¼ Rij f df ð23Þ
h2
Zh
2
m ð2Þ
¼ mijð2Þ~
Ai ~
Aj m ijð2Þ
¼ Rij f2 df:
h2
3.5. Elastoplasticity formulation
The equation set for finite elastoplasticity is established on the basis of the total Lagrangian formulation
and the generalized standard materials theory introduced by Halphen and Nguyen (1975). Use is made of
the von Mises yield criterion with a non-linear isotropic hardening:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 d
f ðR; aÞ ¼ R : Rd r0 a ð24Þ
2
where Rd denotes the deviatoric part of stress tensor R, r0 the initial yield stress and a the thermodynamic
force characterizing the isotropic hardening.
174
Following Green and Naghdi (Green and Naghdi, 1965, 1971), the Green strain tensor E is split additively
into its elastic and plastic parts:
E ¼ Ee þ Ep : ð25Þ
The elastic strain is related to the second Piola–Kirchhoff stress by the three-dimensional Saint–Venant–
Kirchhoff law:
R ¼ D : Ee ¼ ktrðEe ÞI þ 2lEe ð26Þ
where k and l are the Lamé parameters, related to Young’s modulus E and Poisson’s ratio m by:
Em E
k¼ l¼ : ð27Þ
ð1 þ mÞð1 2mÞ 2ð1 þ mÞ
For computational purposes, all the equations are written in matrix representations using the covariant or
contravariant bases. For instance, the relationship (26) reads in the covariant and contravariant bases:
Rij ¼ Dijkl Eelk ¼ ðkAij Akl þ lðAik Ajl þ Ail Akj ÞÞEelk : ð28Þ
The thermodynamic force a in (24) is derived from the so-called hardening energy wa(a), where a is the hard-
ening variable:
owa ðaÞ
a¼ : ð29Þ
oa
The flow rule and the hardening law are derived from the yield function:
of Rd of
E_ p ¼ k_ ¼ k_ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a_ ¼ k_ ¼ k_ ð30Þ
oR 2 d
R : Rd oa
3
where k_ is the plastic multiplier. Use will also be made of the equivalent plastic strain P defined by its rate:
2 _p _p _
P_ ¼ E : E ¼ k ¼ a: _ ð31Þ
3
Since the normal material stress is assumed to be zero, R33 = 0, the elastic fourth-order tensor D must be re-
placed with the reduced one C for the shell model:
Dij33 D33kl
C ijkl ¼ Dijkl ði; jÞ 6¼ ð3; 3Þ; ðk; lÞ 6¼ ð3; 3Þ: ð32Þ
D3333
Substituting Eq. (18) into (23) gives the expressions for the stress resultant tensors (23):
Z h
2
eð0Þ eð1Þ eð2Þ
ij
n ¼ C ijkl Elk þ fElk þ f2 Elk df
h2
Z h
2
m ijð1Þ
¼ C ijkl Elk þ fElk þ f2 Elk f df
h2 ð33Þ
Z h
2
mijð2Þ ¼ C ijkl Elk þ fElk þ f2 Elk f2 df
h2
ði; jÞ 6¼ ð3; 3Þ; ðk; lÞ 6¼ ð3; 3Þ:

eð2Þ
According to the thin shell assumptions, the strain components Eij in Eq. (18) will be neglected in the sequel.
3.6. The discretized non-linear equations
Use is made of the principle of virtual work in order to obtain the discretized non-linear equations:
dW int þ dW ext ¼ 0 ð34Þ
175
where:
Z Z Z !
h
2
ð0Þ ð1Þ
dW int ¼ R : dE dV ¼ Rij dEji þ fdEji df dA ð35Þ
X0 M h2
and:
Z Z Z Z
dW ext ¼ ~
f P :d~
P dV þ ~
f t :d~
t dV þ s :d~
~P
P dS þ st :d~
~ t dS ð36Þ
X0 X0 oX0 oX0
where M represents the shell middle surface, and ~ f and ~

s are the volume and surface distributed loads,
respectively.
Substituting Eqs. (35) and (36) into (34) gives:
Z
ð0Þ ð1Þ
nij dEji þ mijð1Þ dEji dA þ dW ext ¼ 0 ð37Þ
M
where the components nij and mij(1) are defined in (23).

The middle surface is discretized using isoparametric shell finite elements. There are five degrees of freedom
at each node: three translations and two rotations in the tangent space. The reference position ~
P 0 , the displace-
ment U~ on the middle surface, and the director vectors ~ T and ~
t are interpolated in the same way:
N
X
~
P 0 ðn; gÞ ¼ N i ðn; gÞ~
P 0i
i¼1
N
X
~
T ðn; gÞ ¼ N i ðn; gÞ~
Ti
i¼1
ð38Þ
N
X
~¼
U i ~i
N ðn; gÞU
i¼1
N
X
~
t¼ N i ðn; gÞ~
ti
i¼1
where N is the element node number and ðN i Þi2½1;N are the shape functions.
Attention should be drawn on two types of errors due to the geometric discretization:
– The interpolation of the reference position ~

P 0 of a particle on the middle surface leads to several normals at
each node, one per contiguous element.
– The interpolated reference director vector ~
T does not coincide with the normal at the surface defined by the
~
interpolation of P 0 .
Let us mention that it is necessary to define a local orthonormal basis at each node in correspondence with the
Cartesian fixed basis in order to describe ~h in every adjoined element in a coherent way.
The tangent vectors on the middle surface are computed from the interpolation of ~ P 0:
N
X
~
A1 ¼ ~
P 0 ;1 ¼ N ;i1~
P 0i
i¼1
ð39Þ
N
X
~
A2 ¼ ~
P 0 ;2 ¼ N ;i2~
P 0i :
i¼1
The following variations and derivatives of these fields will be required:
176
N
X
~ ;a ¼
U ~i
N ;ia U
i¼1
N
X
~ ;a ¼
dU ~i
N ;ia dU
i¼1
N
X
~
t;a ¼ N ;ia~
ti ð40Þ
i¼1
N
X
d~
t¼ N i d~
ti
i¼1
N
X
d~
t;a ¼ N ;ia d~
ti :
i¼1
The second derivatives will also be required:

N
X
Dðd~
tÞ ¼ N i Dðd~
ti Þ
i¼1
ð41Þ
N
X
Dðd~
t;a Þ ¼ N ;ia Dðd~
ti Þ:
i¼1
It should be noted that DðdU~ Þ is always zero. The integrations over the middle surface and through the thick-
ness are performed using the Gaussian quadrature. Eventually, the virtual work principle (37) yields the non-
linear discretized equation system:
fRðUÞg ¼ fWðUÞg fUg ¼ f0g ð42Þ
where {U} denotes the nodal displacement vector of the whole structure, {R} the residual vector, {W} and {U}
are the internal and external force vectors, respectively.
4. Solution procedure
4.1. The Newton–Raphson procedure
The non-linear equation system (42) is solved by the iterative Newton–Raphson procedure, which requires
the computation of the structural tangent stiffness matrix:
ofRðUÞg ofWðUÞg
½KT ¼ ¼ : ð43Þ
ofUg ofUg
The geometric part of [KT] requires the computation of Dðd~ ti Þ at the middle surface nodes and entails a com-
plicated development. The interested reader is referred to Brank et al. (1997) for more details.
At each iteration, the local integration consists in solving the following problem: assuming the strain E is
known, compute the plastic strain Ep, the stress R and the reduced consistent elastoplastic tangent modulus Cp
as in (32). The implicit Euler scheme and an algorithm of radial return type are employed to integrate the state
and evolution laws as described in Simo and Hughes (1998). In particular, the von Mises criterion (24) is
rewritten as:
1 1
f ¼ RT PR ðr0 þ aÞ2 ð44Þ
2 3
where P is the deviatoric operator ðRd ¼ PRÞ. At each step n, after computing the prediction RE and the con-
sistency parameter k, the consistent elastoplastic tangent operator for computing the structural tangent stiff-
ness matrix is obtained by:
177
oRn ðCðkÞPRn ÞðCðkÞPRn ÞT

C pn ¼ ¼ CðkÞ ð45Þ
oEn RTn PT CðkÞPRn þ l
where:
1
CðkÞ ¼ ðC 1 þ kPÞ
Rn ¼ CðkÞC 1 RE
2 T ð46Þ
P n ¼ P n1 þ k R PRn
3 n
2a0 ðP n Þ T
l¼ Rn PRn c ¼ 1 23ka0 ðP n Þ :
3c
In the above, C denotes the matrix of tensor C in Eq. (32).
It should be noticed that the plane stress condition R33 = 0 is a non-trivial constraint for the non-linear
equation set which is strain driven.
4.2. Arc-length method
The nodal displacement vector {U} is split into two parts as shown in Eq. (47) below: one denoted by f Ug e
contains the unknown degrees of freedom, the other denoted by fUg contains the prescribed degrees of free-
dom. The external force and the residual vector in Eq. (42) are split in a similar way. For instance, the external
e corresponding to f Ug
force {U} is seen as the union of two parts: f Ug e contains the prescribed force compo-
nents, and fUg corresponding to fUg contains the unknown reaction force components. Likewise, the tangent
stiffness matrix [KT] is partitioned into four sub-matrices:
( ) ( ) ( ) " #
e e e e
eT K e
U R U K T
fUg ¼ fRg ¼ fUg ¼ ½KT ¼ : ð47Þ
U R U e T KT
K
Either the prescribed displacement or the external loading is assumed to be proportional:
0 ref
fUg ¼ U þ k U ð48aÞ
n o n oref
Ue ¼k U e ð48bÞ
ref e ref denote reference prescribed quantities. Vector fUg0 related
where k is the control parameter, fUg and f Ug
to zero prescribed displacements does not change the value of fUg. At each increment, the quadratic arc-
length method is used in order to proceed on the solution branches given a specified arc-length Dl, see Wemp-
ner (1971), Riks (1979) and Crisfield (1991). The constraint equation is either of the following relations,
depending on whether one has a proportional prescribed displacement or loading:
e þ Dk2 kUref k ¼ Dl2
kD Uk
2 2
ð49aÞ
e 2 þ Dk2 C 2 ¼ Dl2
kD Uk ð49bÞ
ref
where D U e ¼U e Ue n1 is the difference between the displacement U

e at the current increment (n) and that at
the previous increment (n 1). Similarly, Dk = k kn1. In Eq. (49b), the scalar Cref is a scale factor which
makes the relation consistent dimensionally.
Combining Eq. (49a,b) with the equilibrium equation leads to a quadratic equation. When solving this qua-
dratic equation, one may encounter severe computational difficulties due to complex roots which occur repeat-
edly. It is found that an efficient way to cope with these complex roots is to modify the standard solution
scheme according to Lam and Morley (1992). Let us explain the procedure in the case of prescribed loadings
(48b). The main idea is to project the residual force onto the external load vector. At a current iteration
where complex roots occur, the residual force is split into one component in the load direction and another
178
component orthogonal to this load. The last component is mainly responsible for the complex roots and
should be eliminated. The standard iteration loop is modified as follows.
Iteration loop
[ ]
In the standard arc-length method, solve the quadratic equation.
If the roots are complex, then:
(i) Compute g ¼ R e TU
e ref = U e gU
e ref and H ¼ R
e refT U e ref .
(ii) Compute:
e
e 1 H
e H ¼ K e
e Uref ¼ K
e 1 U
e ref :
dU T and U T ð50Þ
(iii) Solve a quadratic equation in g: ag2 + 2bg + c = 0, where the coefficients a, b and c depend on g, C 2ref ,
Ue Uref and d Ue H . Hopefully, this equation in g gives real roots g1 and g2. Choose a value for g at 5% of
jg2 g1j from g1 or g2.
(iv) Solve a quadratic equation in n: An2 + 2B(g)n + C(g) = 0, where the coefficient A depends on C 2ref and
Ue Uref , B(g) and C(g) are functions of g, g, C 2 , U e Uref and d U
e H . This equation in n gives real roots.
ref
Choose that root n which enables the solution point to advance in the desired direction.
(v) Make the corrections for the ith-iteration: dk = g + n and d U e ¼ gd Ue H þ nU
e Uref .
End if
[ ]
End of iteration loop.
The modified arc-length method described above is also applicable to the case of prescribed displacements
e Uref , C 2 with kUref k2 and U
e ref is replaced with K
(48a), providing that U e Uref with:
T ref
e e Uref :
e 1 K
e U ref ¼ K
U T T ð51Þ
For numerical purposes, the computer program is developed so as to handle both prescribed loads and pre-
scribed displacements in a unified way.
4.3. Branching method
The branch switching techniques are included in the numerical procedure in order to detect the bifurcation
points and bifurcate onto a given branch. Specific methods are implemented following Riks (Riks, 1979, 1991)
and Seydel (1994). The four fundamental steps are given below.
1. At the end of each increment, it must be checked whether one has gone across one or several critical points.
The detection of critical points is based on the singularity of the tangent stiffness matrix, which is factorized
following the Crout formula KT = LDLT, where L is a lower triangular matrix with unit diagonal elements
and D is a diagonal matrix. Since the number of negative eigenvalues of KT is equal to the number of neg-
ative diagonal elements (pivots) of D, the critical points are determined by counting the negative pivot
number.
2. Each critical point detected has to be isolated in order to determine its nature: limit point or bifurcation
point. To do this, the current arc-length Dl is re-estimated several times using a dichotomy-like method.
In the case of prescribed loadings, a simple way to distinguish a limit point from a bifurcation point is
to determine the sign of the current stiffness parameter introduced by Bergan et al. (1978):
n oT n o
e ref
U e Uref
U
k¼n oT n o ð52Þ
e Uref
U e Uref
U
179
e Uref g in (50). The sign of parameter k changes when passing a limit

e ref g is defined in (48b) and f U
where f U
point, whereas it remains unchanged when passing a bifurcation point. In the case of prescribed displace-
ments, parameter k is computed by:
n o n o
e Uref T U
K e U ref
T
k¼ n oT n o ð53Þ
e U ref
U e U ref
U
where fUref g is defined in (48a) and f U e U ref g in (51).

3. If it is a bifurcation point, the step increment is renewed so as to reach a point just behind the bifurcation
point.
4. Finally, the switching on a bifurcated branch is performed using the mode injection method (Wagner and
Wriggers, 1988; Seydel, 1994): at the first step of a bifurcating branch, the eigenvector Z, e solution of
e
e TZ e ¼ 0, is computed and the following predictions are used:
K
e
Z
dk ¼ 0 e ¼
dU : ð54Þ
e
k Zk
5. Numerical results
Let us consider a cylindrical tube defined in the reference configuration by the length l along the x-axis, the
average radius R and the thickness t R. One end of the cylinder lies in the plane X = 0 (more details will be
given according to the boundary condition type) whereas the other end at X = l is loaded by a uniformly dis-
tributed nominal axial compressive stress PXX = k < 0.
First, comparisons will be made between the numerical results and available analytical solutions, in the
case of axisymmetric modes with a linear isotropic hardening. Next, still considering axisymmetric modes,
we will compare our numerical results with those obtained by Brank et al. (1997) using the flow plasticity
theory with a non-linear isotropic hardening. Eventually, we will go further in the description of the post-
buckling behavior and obtain both axisymmetric and non-axisymmetric deformed shapes, especially the
so-called ‘‘diamond’’ modes or the ‘‘elephant foot’’ secondary modes. It will be shown how the geometric
and material parameters, together with the boundary conditions, affect the buckling mode type. For sim-
plicity sake, the numerical results will be presented assuming a linear isotropic hardening. The obtained
results show that the proposed formulation is able to provide complex experimental modes which are hard
to predict in theory.
In all examples, the Newton–Raphson method is used for solving the matrix non-linear equations of the
problem. The path-following is carried out either by displacement (49a) or by force control (49b), use is also
made of the extended version of the arc-length method in Lam and Morley (1992) as explained in Section 4 to
deal with the complex roots in the solution scheme.
A few preliminary mesh convergence tests lead us to use about 200 quadratic eight-node elements for the
considered portion of the cylinder and four Gaussian points for the numerical quadrature through the shell
thickness.
5.1. Comparison between analytical and numerical results
The comparison between analytical and numerical results is possible within the context of a linear isotropic
hardening and axisymmetric bifurcation modes. In order to simplify the discussion, let us take the yield stress
r0 small enough so that the plastification occurs before the tangent modulus critical load is reached. Further-
more, it is assumed that the hardening energy wa(a) in Eq. (29) is a quadratic function, i.e. wa ¼ 12 ha2 , where
constant h is the hardening modulus. The numerical values are shown in Tables 1 and 2. The hardening mod-
ulus h is given a very high value just to make sure of obtaining an axisymmetric mode, so that the numerical
results can be compared with the analytical solution.
180
Table 1
Geometry used in Section 5.1
Length l Radius R Thickness t
2m 1m 1 cm
Table 2
Material properties used in Section 5.1
Young’s modulus E Poisson’s ratio m Yield stress r0 Isotropic hardening modulus h
11 8
2.1 · 10 Pa 0.3 Free edges: 1 · 10 Pa 2 · 1010 Pa
Simply supported edges: 3.5 · 108 Pa
5.1.1. Free edges

In the case of free edge cylinders, the yield stress r0 is supposed to be small enough for the buckling to occur
in the plastic range.
The analytical results for a cylinder of infinite length can be obtained by describing the shell kinematics with
the Koiter model, which is equivalent to the Love–Kirchhoff plate model. The critical stress was given by Bat-
terman (1968):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u ET
free edges Et u
kc ¼ t h E i ð55Þ
R 3 5 4m ð1 2mÞ2 ET
E
where ET is the tangent modulus, related to the Young’s modulus E and the isotropic hardening modulus h by
1
ET
¼ E1 þ 1h. Batterman showed that the critical stress is a little bit smaller for a finite shell and it hardly depends
on the length. The eigenmode is:
8 qffiffiffi ET
>
> U ¼ 2t 3t 1 1ð12mÞ E
sin pX
p ffiffiffi exp p ffiffiffiffi
pX ffi
>
< Rp E
1þ3 ET Rt 3Rt
V ¼ 0 ð56Þ
>
> pffiffiffi
>
: W ¼ t 3 cos pffiffiffi sin pffiffiffi exp pffiffiffiffiffi
pX pX pX
Rt Rt 3Rt
where U, V and W are axial, circumferential and radial displacements, respectively, and:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i ffi
u h
u27 5 4m ð1 2mÞ2 ET ET
u4 E E
p¼ t 2
: ð57Þ
1 þ 3 EET
pffiffiffi
In the above, the mode has been normalized so that Wð0Þ ¼ t 3.
The numerical computations show that an axisymmetric bifurcation under increasing load is obtained at
the critical value of 184 MPa, which is identical with the analytical critical stress kfree
c
edges
derived from Eq.
(55), within 1%. The bifurcation shape is a sinusoid weighted by a decreasing exponential function, so that
its magnitude is minimal at the middle of the cylinder and maximal at the ends. This result is in full accordance
with Batterman’s results (Batterman, 1968). Fig. 1 shows that the numerical modal shape is very close to the
Batterman’s analytical one corresponding to the semi-infinite cylinder.
5.1.2. Simply supported edges

The case of simply supported edges is simpler and does not require the infinite length assumption. Let the
radial displacements be fixed and the rotations be free on both edges.
Batterman (1965) showed that, if the cylinder is long enough, a good approximate expression for the critical
value is:
181
Numerical axisymmetric mode shape
λ λ
λ λ
Analytical mode shape
Zoom of numerical mode shape
Fig. 1. Analytical and numerical axisymmetric bifurcation modes of an elastoplastic cylinder under axial compression with free edges.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u ET
2Et u
ksupported
c
edges
¼ t h E i ¼ 2kfree
c
edges
: ð58Þ
R 3 5 4m ð1 2mÞ E 2 ET
In the case of simply supported edges, it is also possible to obtain the analytical slope to the post-critical curve
at the bifurcation point. Since the calculation is somewhat intricate, we give the final result without proof:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h i
5 þ ð8m 1Þ E þ 4 3 5 4m ð1 2mÞ2 EET EET
ET
t
ksupported
1
edges
¼ 2ET
: ð59Þ
R 1 þ 3 EET 5 ð1 2mÞ EET
In the case of simply supported edges, the numerical critical value is found to be twice as large as in the free
edge case, which is in agreement with the analytical solution ksupported
c
edges
obtained from Eq. (58). From a
qualitative point of view, the bifurcation comes just before the maximum load value, as commonly observed
in other numerical applications in the literature.
400 Theoretical tangent
350
Bifurcation point
300
Bifurcated branch
250
λ (N/mm2)
200
Fundamental path
150
100 l/R=2, R/t=100
50
0
0 0.001 0.002 0.003
|u|/l
Fig. 2. Elastoplastic cylinder under axial compression with simply supported edges: compressive stress versus axial shortening.
182
Fig. 2 shows the load versus the axial shortening in the simply supported edge case. The analytical tangent
to the post-critical curve at the bifurcation point is also plotted using Eq. (59) and fits well with the numerical
curve.
5.2. Comparison with a numerical result of the literature
We still consider axisymmetric modes and now compare the numerical results with those obtained by Brank
et al. (1997) using the flow plasticity theory with a non-linear isotropic hardening. The computations are
carried out with the same geometry and material as in Brank et al. (1997), see Tables 3 and 4. The coefficients
in the non-linear isotropic hardening law are determined by fitting the data given in Brank et al. (1997), see
Table 5.
The computations are performed for two boundary conditions: simply supported edges and built-in edges.
The bifurcation loads are 280 MPa in the case of simply supported edges and 300 MPa in the case of built-in
edges, which agree very well with the solutions of Brank et al. (1997). As shown in Fig. 3, the bifurcation loads
are close to the maximum loads. Both boundary conditions lead to similar bifurcation modes, see Fig. 4. In
addition, Fig. 3 shows that although the bifurcation loads are similar, the corresponding axial shortenings are
significantly distinct.
Fig. 5 shows the maximum equivalent plastic strain P in the shell thickness. Fig. 5a is related to the case of
simply supported edges when the axial shortening equals 1.6%; Fig. 5b to the case of built-in edges when the
axial shortening equals 2.3%. As regards the maximum plastic strain in the whole shell, it is 5.6% in the case of
simply supported edges and 6.7% in the case of built-in edges. The points where the plastic strain P reaches the
maximum are also indicated in Fig. 5. For both boundary condition types, it is found that P reaches its max-
imum value near the ends of the cylinder where the axisymmetric localization takes place.
5.3. Influence of thickness, hardening modulus and boundary conditions
In the previous Sections 5.1 and 5.2, the numerical formulation has been validated by comparing analytical
versus numerical results and numerical versus numerical results. Attention is next focused on more complex
modes. Several thicknesses, isotropic hardening moduli and boundary conditions (free, simply supported or
built-in edges, prescribed displacements or stresses) are considered in order to exhibit the variety of post-crit-
ical responses of the cylinder under axial compression. It is expected that the boundary conditions have a
strong influence on the deformed shapes, especially for non-axisymmetric modes. Indeed, in the particular case
of free edges, prescribing the axial displacements at the ends of the cylinder leads more naturally to an axisym-
Table 3
Geometry used in Section 5.2
1 2
1.271 · 10 m 4.333 · 10 m 2.36 · 103 m
Table 4
Young’s modulus E Poisson’s ratio m Yield stress r0 Isotropic hardening coefficients
11 8
2.16962 · 10 Pa 0.274 1.62722 · 10 Pa Obtained by fitting values in Brank et al. (1997)
Table 5
Non-linear isotropic hardening curve
P (103) 0 1 2.55 4.25 6.75 9.25 14.25 24.25 54.25
K(P) (MPa) 0 68.95 86.19 106.2 123.4 133.1 151 166.9 211
183
Built-in edges
300
Simply supported edges

250
Bifurcation points
200
λ (N/mm2)
150
100 l/R=2.93, R/t=18.36
50
0
0 0.005 0.01 0.015 0.02
|u|/l
Fig. 3. Elastoplastic cylinder under axial compression: compressive stress versus axial shortening. Case of non-linear isotropic hardening.
Simply supported edges Built-in edges
Fig. 4. Axisymmetric bifurcation modes of an elastoplastic cylinder under axial compression. Case of non-linear isotropic hardening.
(a) Simply supported edges (b) Built-in edges
P P
0.056 0.067
0.052 0.062
0.048 0.057
0.044 0.053
0.040 0.048
0.036 0.043
0.032 0.038
0.028 0.034
0.024 0.029
0.020 0.024
Pmax in the whole shell (5.6 %) 0.016 Pmax in the whole shell (6.7 %) 0.019
0.012 0.014
reached about this parallel 0.008 reached about this parallel 0.010
0.004 0.005
0.000 0.000
Fig. 5. Maximum equivalent plastic strain in the thickness of the cylinder under axial compression. Case of non-linear isotropic
hardening: (a) simply supported edges, axial shortening = 1.6%; (b) built-in edges, axial shortening = 2.3%.
184
metric mode, whereas prescribing the stresses leads to non-axisymmetric modes as the ends are free to move
now.
The numerical results in this section are obtained using the geometry and the material properties shown in
Tables 6 and 7. The length-to-radius ratio is constant and equal to l/R = 3.33, while several radius-to-thick-
ness ratios will be used.
Remark. Why the unrealistically high values of the hardening modulus h in Table 7? It is found that with
usual h values, the numerical computations in plasticity give axisymmetric modes. Only very high moduli give
rise to non-axisymmetric modes, similar to those encountered in elasticity. The very high moduli are
admittedly unrealistic, but they are useful to check how and when non-axisymmetric modes can be obtained
using the proposed plasticity formulation.
5.3.1. Free edges

Consider a rather large thickness t = 7 cm, a medium hardening modulus h = 2 · 1010 Pa, and prescribed
stresses at the ends of the cylinder. Fig. 6 shows that the first bifurcated solution has three circumferential
waves, which is in agreement with experimental and numerical results obtained by Goto and Zhang (1999).
Fig. 7 depicts the distribution of the equivalent plastic strain P in the cylinder when the nominal axial com-
pressive stress is k = 1.617 · 108 Pa. The maximum value of 8.4% is reached at the junction of two circumfer-
ential waves which are localized at the ends of the cylinder.
Let us maintain the boundary conditions and reduce the thickness to t = 5 cm so as to see the influence of
the radius-to-thickness ratio R/t on the circumferential wave number. Fig. 8 shows that the wave number
increases with the ratio R/t—now there are four circumferential waves with the new thickness—as was already
observed in Goto and Zhang (1999) and Tvergaard (1983a). The bifurcation mode is localized at both ends
and displays a series of periodic inward dimples.
Fig. 9 describes the load–displacement curves for t = 5 cm and t = 7 cm. In both cases, the (complete) plas-
tification of the cylinder just precedes the bifurcation, which itself precedes the maximum stress value.
Table 6
Geometries used in Section 5.3
5m 1.5 m 7 cm, 5 cm, 3 cm, 2 cm, 12 mm, 6 mm according to the considered case
Table 7
Young’s modulus E Poisson’s ratio m Yield stress r0 Isotropic hardening modulus h
11 8
2.1 · 10 Pa 0.3 4 · 10 Pa 2 · 1010 Pa, 5 · 1010 Pa, 5 · 108 Pa according to the considered case
Fig. 6. Non-axisymmetric bifurcation mode of an elastoplastic cylinder under axial compression (free edges, prescribed stresses, t = 7 cm).
185
P
0.084
0.078
0.072
0.066
0.060
0.054
0.048
0.042
0.036
0.030
0.024
0.018
Pmax in the whole shell (8.4%) 0.012
0.006
reached here 0.000
Fig. 7. Maximum equivalent plastic strain in the thickness of the cylinder under axial compression (free edges, prescribed stresses,
t = 7 cm, nominal axial stress k = 1.617 · 108 Pa).
Fig. 8. Non-axisymmetric bifurcation mode of an elastoplastic cylinder under axial compression (free edges, prescribed stresses, t = 5 cm).
1.4
Free edges
Prescribed stress
1.2
Bifurcation points
1
Plastification onsets
0.8
λ/σ0
0.6 R/t=21.43
0.4 R/t=30
0.2
0
0 0.01 0.02 0.03 0.04 0.05 0.06
|u|/l
Fig. 9. Elastoplastic cylinder under axial compression (free edges, prescribed stresses): compressive stress versus axial shortening.
Remark. About the mesh: it is necessary to mesh a quarter of the cylinder in order to get three circumferential
waves as shown in Fig. 6. From now on only one eighth of the cylinder will be meshed so as to reduce the
computational cost. Consequently, one necessarily obtains even circumferential wave numbers, as shown in
Fig. 8.
186
5.3.2. Built-in edges

In the sequel, we consider the cases of built-in and simply supported edges, which are closer to experimental
conditions. Here, the linear isotropic hardening modulus is given by the new value h = 5.1010 Pa in order to
obtain non-axisymmetric modes with dimples spreading over the entire surface of the cylinder (‘‘diamond’’
modes) or limited in regions near the ends.
Under prescribed stresses, with built-in edges, two ‘‘diamond’’ modes are obtained with the thicknesses
t = 5 cm and t = 3 cm, as shown in Fig. 10. These ‘‘diamond’’ modes display both axial and circumferential
waves, whose numbers increase with the ratio R/t, as in the case of free edges.
Again, the load–displacement curves in Fig. 11 show that the bifurcation point occurs just before the limit
load. It is noteworthy that the ratio between the two stresses at the bifurcation points is almost equal to the
ratio of thicknesses, which reminds us the proportionality between the analytical stress in Eq. (58) and the
thickness of the cylinder.
Fig. 10. Primary modes (‘‘diamond’’ modes) of an elastoplastic cylinder under axial compression (built-in edges, prescribed stresses).
3.5
3 Bifurcation points
2.5 R/t=30
2
λ/σ 0
R/t=50
1.5
1 Plastification onsets
0.5 Built-in edges

Prescribed stress
0
0 0.01 0.02 0.03 0.04 0.05
|u|/l
Fig. 11. Elastoplastic cylinder under axial compression (built-in edges, prescribed stresses): load–displacement curves for primary modes.
187
Remark. Prescribed displacements give rise to similar results.

The secondary modes are obtained with lower hardening moduli, and only in the case of prescribed dis-
placements, because the localization of the axisymmetric mode seems to appear more easily in such conditions.
As in the case of the primary modes, the elastoplastic response is not homogeneous in the cylinder since the
edges are clamped. The cylinder is partly plastified before the bifurcation occurs, so that the bifurcation does
occur in plasticity. But unlike the primary mode case, here the plastification onset and the bifurcation are very
close. It is more difficult to computationally detect the bifurcation points for secondary modes. The step size
should be smaller than usual and some numerical devices have to be added in order to correctly deal with this
bifurcation type. Fig. 12 shows two ‘‘elephant foot’’ transitional secondary modes with h = 5.108 Pa and med-
ium thicknesses. As in the primary mode case, the circumferential wave number mainly depends on the radius-
to-thickness ratio.
Fig. 12. Secondary modes (‘‘elephant foot’’ modes) for an elastoplastic cylinder under axial compression (built-in edges, prescribed
displacements).
1
Primary bifurcation points (just after the plastification onsets)
0.9
Secondary bifurcation points
0.8
0.7
Built-in edges
0.6 Prescribed displacements
λ/σ0
0.5 Secondary modes
0.4
0.3
0.2
R/t=125
0.1 R/t=250
0
0 0.005 0.01 0.015 0.02
|u|/l
Fig. 13. Elastoplastic cylinder under axial compression (built-in edges, prescribed displacements): load–displacement curves for secondary
modes.
188
The load–displacement curves for secondary modes are shown in Fig. 13. It is worth mentioning the pres-
ence of a so-called snap-back (a limit point in displacement) in the case of the thinner shell.
5.3.3. Simply supported edges

Consider now a cylinder with simply supported edges and thickness t = 2 cm. In the case of prescribed
stresses on the ends, a non-axisymmetric mode is obtained with dimples localized at the ends of the cylinder,
as shown in Fig. 14a. As shown in Eq. (48a,b), the axial stress is prescribed on the edges through the control
parameter k. Here, since the numerical computations show that there can be some limit load, one has to use
the arc-length method (see Section 4.2) to control the prescribed stress.
Fig. 14b shows that in the case of prescribed displacements on the ends, the deformation is more significant
and there are more dimples in the middle of the cylinder than in the case of prescribed stresses. This fact could
be accounted for as follows: since the lower and upper circumferences are constrained to remain plane under
prescribed displacements, the global stiffness increases as compared with the case of prescribed stresses, thus
increasing the dimples number too.
Fig. 15 shows that the load–displacement curves are very close before the critical point. Contrary to the case
of prescribed stresses, the post-critical branch displays several successive limit points in the case of prescribed
displacements, which must appear in the same time as additional dimples.
5.4. Summary
Table 8 summarizes the encountered bifurcation modes and the post-bifurcation behaviors versus the cyl-
inder thickness and the boundary conditions. The major influence of the shell thickness on the axial and cir-
cumferential wave numbers has been highlighted for all encountered mode types.
(a) In the case of sufficiently thick shells and for some combinations of R/t and stress–strain properties, the
axisymmetric mode degenerates into a non-axisymmetric one after the limit load of the axisymmetric
response and the second bifurcation corresponds to the so-called ‘‘elephant foot’’ mode. Note that in
the cases of simply supported or built-in edges, the pre-critical deformations are found to be negligible
and an axisymmetric mode is still possible.
Fig. 14. Primary modes (‘‘diamond’’ modes) of an elastoplastic cylinder under axial compression with simply supported edges.
189
Bifurcation points
1.5
Prescribed displacement
Plastification onsets
λ/σ0
Prescribed stress
0.5 Simply supported edges

R/t=75
0
0 0.005 0.01 0.015 0.02
|u|/l
Fig. 15. Elastoplastic cylinder under axial compression (simply supported edges, t = 2 cm): compressive stress versus axial shortening.
Table 8
Bifurcation modes and post-bifurcation behaviors of a cylinder under axial compression versus the thickness and the boundary conditions
Cylinders Simply supported edges Built-in edges Free edges
Thick Sinusoidal axisymmetric mode Sinusoidal axisymmetric mode Damped sinusoidal axisymmetric
Post-bif: localization at the ends Post-bif: localization somewhat far from the mode (Fig. 1)
(Fig. 4) ends (Fig. 4)
Moderately (Results between those for thick Axisymmetric mode (+localization) (Results between those for thick and
thick and thin cylinders) ‘‘Elephant foot’’ secondary mode (Fig. 12) thin cylinders)
Thin ‘‘Diamond’’ mode (Fig. 14) ‘‘Diamond’’ mode (Fig. 10) Damped sinusoidal non-
axisymmetric mode (Figs. 6 and 8)
(b) For thin shells, it has been found that there are no more axisymmetric modes (as if the two bifurcation
points coalesce into one) and one gets straight to the so-called ‘‘diamond’’ mode, before the limit load of
the ‘‘diamond’’ response.
The hardening modulus value has a great influence on the solution, as it may give rise to secondary bifurcation
modes or not. Only quite high hardening modulus values (and the thicknesses considered) lead to interesting
post-critical deformed shapes as presented in this work.
6. Conclusions
This work deals with the numerical solution of the elastoplastic thin shell buckling and post-buckling
problem.
The proposed shell finite element has been validated by showing that the numerical results on the plastic
buckling of axially compressed cylinders compare well with the analytical solutions derived from the plastic
bifurcation theory.
The numerical computations have been carried out considering the influence of the geometry, the material
properties and the boundary conditions. For the cylinder under axial compression, it has been shown that a
large range of primary and secondary buckling modes can be computed in a unified way without using imper-
fections, and that the post-critical behavior can be described in an advanced deformation state, exhibiting the
localization phenomenon as observed in experiments.
190
Acknowledgements
The authors thank the reviewers for their instructive comments which have helped to clarify many state-
ments in the manuscript.
References
Bardi, F.C., Kyriakides, S., 2006. Plastic buckling of circular tubes under axial compression—part I: Experiments. International Journal of
Mechanical Sciences 48, 830–841.
Bardi, F.C., Kyriakides, S., Yun, H.D., 2006. Plastic buckling of circular tubes under axial compression—part II: Analysis. International
Journal of Mechanical Sciences 48, 842–854.
Batdorf, S.B., 1949. Theories of plastic buckling. Journal of Aeronautical Sciences 16, 405–408.
Batterman, S.C., 1965. Plastic buckling of axially compressed cylindrical shells. AIAA Journal 3 (2), 316–325.
Batterman, S.C., 1968. Free-edge plastic buckling of axially compressed cylindrical shells. Journal of Applied Mechanics 35, 73–79.
Bergan, P.G., Horrigmoe, G., Krakeland, B., Soreide, B., 1978. Solution techniques for non-linear finite element problem. International
Journal for Numerical Methods in Engineering 12, 1677–1696.
Bijlaard, P.P., 1949. Theory and tests on the plastic stability of plates and shells. Journal of Aeronautical Sciences 16, 529–541.
Brank, B., Peric, D., Damjanic, F.B., 1997. On large deformations of thin elastoplastic shells: implementation of a finite rotation model for
quadrilateral shell element. International Journal for Numerical Methods in Engineering 40, 689–726.
Bushnell, D. 1982. Plastic buckling, Pressure vessels and piping: Design technology. In: Zamrik, S.Y., Dietrich, D. (Eds.), A decade of
Progress, Book No. G00213, 47–117.
Bushnell, D., 1985. Computerized Buckling Analysis of Shells. M. Nijhoff.
Christoffersen, J., Hutchinson, J.W., 1979. A class of phenomenological corner theories of plasticity. Journal of the Mechanics and Physics
of Solids 27 (5–6), 465–487.
Crisfield, M.A., 1991. Non-linear finite element analysis of solids and structures. In: Essentials, 1. John Wiley & Sons.
Gellin, S., 1979. Effect of an axisymmetric imperfection on the plastic buckling of an axially compressed cylindrical shell. Journal of
Applied Mechanics 46, 125–131.
Goto, Y., Zhang, C., 1999. Plastic buckling transition modes in moderately thick cylindrical shells. Journal of Engineering Mechanics 125
(4), 426–434.
Green, A.E., Naghdi, P.M., 1965. A general theory of an elastic–plastic continuum. Archive for Rational Mechanics and Analysis 18, 251–
281.
Green, A.E., Naghdi, P.M., 1971. Some remarks on elastic–plastic deformation at finite strain. International Journal of Engineering
Sciences 9, 1219–1229.
Halphen, B., Nguyen, Q.S., 1975. Sur les matériaux standard généralisés (On the generalized standard materials). Journal de Mécanique 14
(1), 39–63.
Hill, R., 1958. A general theory of uniqueness and stability in elastic–plastic solids. Journal of Mechanics and Physics of Solids 6, 236–249.
Hutchinson, J.W., 1974. Plastic buckling. Advances in Applied Mechanics 14, 67–144.
Kyriakides, S., Bardi, F.C., Paquette, J.A., 2005. Wrinkling of circular tubes under axial compression: effect of anisotropy. Journal of
Applied Mechanics 72, 301–305.
Lam, W.F., Morley, C.T., 1992. Arc-length method for passing limit points in structural calculation. Journal of Structural Engineering 118
(1), 169–185.
Peek, R., 2000. Axisymmetric wrinkling of cylinders with finite strain. Journal of Engineering Mechanics 126 (5), 455–461.
Riks, E., 1979. An incremental approach to the solution of snapping and buckling problems. International Journal of Solids and
Structures 15, 524–551.
Riks, E., 1991. On formulations on path-following techniques for structural stability analysis. In: New Advances in Computational
Structural Mechanics. Elsevier, 65–80.
Seydel, R., 1994. Practical Bifurcation and Stability Analysis. From Equilibrium to Chaos. Springer-Verlag.
Shanley, F.R., 1947. Inelastic column theory. International Journal of Aeronautical Sciences 14, 261–267.
Simo, J.C., Hughes, T.J.R., 1998. Computational Inelasticity. Springer.
Tvergaard, V., 1983a. On the transition from a diamond mode to an axisymmetric mode of collapse in cylindrical shells. International
Journal of Solids and Structures 19 (10), 845–856.
Tvergaard, V., 1983b. Plastic buckling of axially compressed circular cylindrical shells. Thin-Walled Structures 1 (2), 139–163.
Wagner, W., Wriggers, P., 1988. A simple method for the calculation of post-critical branches. Engineering and Computations 5, 103–109.
Wempner, G.A., 1971. Discrete approximations related to non-linear theories of solids. International Journal of Solids and Structures 7,
1581–1599.
191
Available online at www.sciencedirect.com
Engineering Fracture Mechanics 75 (2008) 1027–1044

www.elsevier.com/locate/engfracmech
Numerical and analytical study of severity of cracks

in cylindrical and spherical shells
Abdelhadi El Hakimi, Philippe Le Grognec *, Said Hariri
Ecole des Mines de Douai, Polymers and Composites Technology & Mechanical Engineering Department,
941 rue Charles Bourseul – BP 10838, 59508 Douai Cedex, France
Received 9 October 2006; received in revised form 3 April 2007; accepted 29 April 2007
Available online 13 May 2007
Abstract
The present study deals with the severity of cracks in pressure equipments, where such defects are often involved. Our
work is particularly concerned with the problem of cylindrical shells and also the little well-known problem of spherical
shells, including all sorts of practical defects, namely axisymmetric or semi-elliptic, both internal and external cracks.
The stress intensity factor in the linear elastic domain and the J integral in the elastoplastic range are performed using
the finite element method and compared to the results provided by the application of the semi-analytical A16 or R6 sim-
plified criteria, depending on a limit load calculation. The nocivity of the defects depends on the crack shape and size and
other structural geometrical parameters. Use is made of a polynomial decomposition of the stress field in the vicinity of the
crack in order to cover all industrial loadings. All the numerical results, for a wide range of shell and crack geometries, are
depicted using appropriate tables and curves in order to check the fracture criteria more easily.
Keywords: Stress intensity factor; J integral; A16-R6 methods; Cylindrical and spherical shells
1. Introduction
Cylindrical and spherical thin-walled structures are often encountered in industrial or household appli-
ances, such as pressure vessels. Most of the time, one or several internal or external cracks, either axisymmetric
or semi-elliptic, can be observed in this type of structures. It is thus essential to be able to measure the nocivity
of these defects in order to avoid any accident. Lots of numerical and experimental works about cylindrical
shells under pressure have already been carried out [1]. On the contrary, the problem of severity of cracks
in spherical shells has only been examined quite recently. For instance, Folias [2] derived an approximate
expression of the stress intensity factor in a spherical shell displaying a semi-elliptic defect from finite element
calculations on an infinite plate, multiplying the corresponding factor KI by an appropriate coefficient taking
*
Corresponding author.
E-mail addresses: elhakimi@ensm-douai.fr (A. El Hakimi), legrognec@ensm-douai.fr (P. Le Grognec), hariri@ensm-douai.fr
(S. Hariri).
doi:10.1016/j.engfracmech.2007.04.027
192
1028 A. El Hakimi et al. / Engineering Fracture Mechanics 75 (2008) 1027–1044
Nomenclature
rn elementary stress distributions

in influence parameters
KI stress intensity factor
C integration path for the J integral evaluation
M current point in the integration path C
x, y coordinates of the current point M
W strain energy density
~; ~
U T displacement and stress vectors
Je elastic value of the J integral
E Young’s modulus
m Poisson’s ratio
ry yield stress
H hardening modulus
eref, rref reference strain and stress
PL limit load of the cracked structure
P applied load
t shell thickness
R internal radius
a crack depth
c semi crack length
into account the shell curvature. In another way, Wang and Hu [3] used the so-called ‘‘line-spring’’ model,
referring to the theoretical developments of Sih and Hagendorf [4], in order to determine the stress intensity
factor in a spherical shell with an external semi-elliptic crack.
In this study, the problem is solved using the finite element method, possibly coupled with analytical esti-
mation methods, considering both elastic and elastoplastic behaviors. The main objective is to dispose of a
measure of the severity of cracks in both types of structures, for a wide range of geometrical parameters,
dimensions of the defect and loading cases, in a rather simple way. First, a set of numerical results in elasticity,
covering all the shell and crack geometries considered, allows us to perform the stress intensity factor for any
loading type using the superposition principle. Then, the use of simplified estimation methods of the J integral
in plasticity, such as A16 or R6 rules, validated by elastoplastic finite element calculations, will provide good
approximations of the severity of cracks in cylindrical and spherical shells for all the geometries and loadings
considered, in an efficient way.
2. Elasticity: stress intensity factor
Assuming a linear elastic constitutive law, the severity of a crack is given using stress intensity factors. For
computational purposes, the loading on the boundary of the cracked structure is replaced by the equivalent
stresses applied onto the crack lips, what leads to a localized determination of the stress intensity factors
standing for the nocivity degree of the defect. The superposition principle then allows us to resolve any com-
plicated loading as the sum of simple loads, just developing the local stress field [5–7]. A polynomial expression
is generally retained to describe the stress distribution in the vicinity of the crack, knowing that a polynomial
of the third degree can represent most of the practical loadings encountered in industrial applications [1,8–10].
The stress field will thus take the following form:
u u u2 u3
r ¼ r0 þ r1 þ r2 þ r3 ð1Þ
t t t t
193
A. El Hakimi et al. / Engineering Fracture Mechanics 75 (2008) 1027–1044 1029
Fig. 1. Distance u in the cracked structure.
where u stands for the radial distance between a point inside the inner wall and its projection on the crack
emergence surface, independently of the crack shape and size (Fig. 1). The components r0, r1, r2 and r3 cor-
respond to constant, linear, quadratic and cubic stress distributions, respectively, in the thickness of the
cracked structure.
A stress intensity factor is associated to each elementary stress distribution, which writes
pffiffiffiffiffiffian
K nI ¼ rn pa in ð2Þ
t
thus defining the corresponding function of influence in, the integer n varying only from 0 to 3 in practice.
Following the superposition principle, the stress intensity factor corresponding to a complex loading is
obtained by adding up the contributions of the different stress components, that is to say
X
3 X
3
pffiffiffiffiffiffian
K tot
I ¼ K nI ¼ rn pa in ð3Þ
n¼0 n¼0
t
3. Plasticity: J integral
The J integral is the most commonly used parameter for characterizing the nocivity of a defect in a ductile
material, in the framework of elastoplastic fracture mechanics. Rice and Rosengren [11] express the J integral
for a non-linear constitutive law as the variation of potential energy during the crack propagation. They show
that this non-linear strain energy release rate is independent of the integration path encircling the crack tip and
of the external contour of the structure. Its formulation, which can apply in the tridimensional case, so long as
one considers a path in a plane orthogonal to the crack front at the current point, reads as follows:
Z !
~
dU
J¼ W dy ~
T: dC ð4Þ
C dx
where x and y represent the coordinates of the current point M in the integration path C, defined with respect
to the crack front. Let us mention that the use of the J integral in plasticity is rigorously restricted to the cases
of monotonous radial loadings, because of the irreversible nature of the elastoplastic behavior. Nevertheless,
its principle can reasonably apply for any monotonous loadings, just assuming there is no unloading.
The so-called simplified estimation methods, in the context of the calculation of the J integral in plasticity,
produce approximations of J using the elastic stress intensity factors, and the constitutive law together with a
limit analysis calculation. The most commonly used methods in the area of pressure equipments are the R6
rule [12] and the A16 one [13,14].
Both approaches provide an evaluation of the elastoplastic J integral using:
• the true stress–strain curve;

• the limit load;
• and a corrective formula for the J integral connecting the elastic and elastoplastic values.
194
In our study, the following equations are used, depending on the method:
• R6 rule:
8 2
>
> J e ¼ 1m
E
K 2I
<
Lr ¼ rrrefy ¼ PPL ð5Þ
>
>
: K R6 ¼ JJR6e ¼ Ee ref L3 r
þ 2Eer refy
Lr ry
• A16 rule:
8 2
>
> J e ¼ 1m
E
K 2I
<
Lr ¼ rrrefy ¼ PPL ð6Þ
>
>
:K J A16 Eeref 1 L2r
A16 ¼ J e ¼ Lr ry þ 2 1þL2r
where the reference strain eref can be deduced from the reference stress rref, for the considered loading P, using
the stress–strain curve, and where Je is the elastic value of the J integral in the same loading conditions, in
plane strains.
Remark. Analytical solutions for the limit load of a cracked cylinder under pressure are available in the
literature. For other loading types or in the case of a spherical shell, the same approaches can be transformed
or, in a simpler way, the yield stress can be used, empirically modified by an appropriate factor. Fortunately,
all these solutions give rise to almost similar limit loads, very close to the numerical results obtained by the
finite element method. Thus it shows the efficiency of such estimation methods proposed in this work for the
calculation of the elastoplastic J integral. As a consequence, we will not develop here the different limit load
calculation methods and the finite element method will only be retained in the following applications.
Fig. 2. Calculation process using A16 and R6 rules.
195
In Fig. 2, the flowchart describes the calculation process of the elastoplastic J integral, using one of the two
previous approximation methods.
4. Numerical simulation
4.1. Geometrical features
The J integrals, or the stress intensity factors depending on the functions of influence in elasticity, are firstly
determined by the finite element method. Cylindrical and spherical hollow structures are considered, contain-
ing axisymmetric or semi-elliptic cracks emerging on the internal or external surface (Figs. 3–8). The geometry
is described by a single adimensional parameter t/R where t represents the shell thickness and R its internal
radius. The semi-elliptic defects are characterized by the two ratios a/t and a/c where a is the crack depth
and 2c its length. The case of an axisymmetric defect can be considered as the limit case where c tends towards
infinity.
Fig. 3. Cylindrical shell with an external longitudinal semi-elliptic crack.
Fig. 4. Cylindrical shell with an internal circumferential semi-elliptic crack.
Fig. 5. Cylindrical shell with an external axisymmetric crack.
196
Fig. 6. Cylindrical shell with an internal axisymmetric crack.
Fig. 7. Spherical shell with an internal or external semi-elliptic crack.
Fig. 8. Spherical shell with an internal or external axisymmetric crack.
In the case of semi-elliptic defects, the following geometrical ratios have been considered:
8
< t=R ¼ 1=2; 1=10; 1=80
>
a=c ¼ 1=2; 1=4; 1=8
>
:
a=t ¼ 0:2; 0:4; 0:6; 0:8
For axisymmetric cracks, ratios are slightly different:

t=R ¼ 1=2; 1=10; 1=80
a=t ¼ 0:2; 0:4; 0:5 ðonly for spherical shellsÞ; 0:6; 0:7 ðonly for spherical shellsÞ; 0:75
All these values of the geometrical parameters provide an amount of 144 and 72 configurations for cylindrical
and spherical shells, respectively, in the case of semi-elliptic cracks, and 24 and 36 configurations with an axi-
symmetric crack, considering internal or external, axial or circumferential defects.
197
4.2. Meshing
The meshing, as well as all the numerical calculations, are performed with the help of Castem software [15].
In the case of axisymmetric defects, a 2D axisymmetric modeling is retained, taking into account that both
geometry and loading are axisymmetric (Fig. 9).
A 3D model is required in the more complicated case of semi-elliptic defects. Use is made of a specific
adjusted crack block so as to mesh the vicinity of the crack front (Fig. 10). A plain crack mesh is extruded
along the elliptic crack front, so as to obtain a reference volume surrounding the whole crack, which is com-
pleted by transitional volumes in order to get finally a plane-parallel volume. This crack block is then wrapped
in order to obtain the exact radius of curvature, before being incorporated in the global mesh of the structure
(only one quarter by symmetry) generated by suitable translations and rotations (Figs. 11 and 12). For more
details, see [8], for instance.
Use is made of quadratic finite elements. The meshing in the vicinity of the crack front is particularly
refined (most of the elements are concentrated in the crack block) in order to ensure a good description of
the stress field where it is singular and accordingly correct values for stress intensity factors and J integrals.
The developed crack block has been optimized and validated in the range of crack sizes presented above,
in comparison with other previous numerical results of the literature, by some of the authors in [8].
4.3. Loading
Stresses are directly prescribed onto the crack lips, involving constant, linear, quadratic and cubic distribu-
tions. This procedure enables us to perform stress intensity factors for arbitrary global loading conditions,
according to the superposition principle in linear elasticity.
Fig. 9. Mesh of a spherical shell with an axisymmetric crack.
Fig. 10. Crack block.
198
Fig. 11. Mesh of a cylindrical shell with a semi-elliptic crack.
Fig. 12. Mesh of a spherical shell with a semi-elliptic crack.
Fig. 13. Definition of the angle u ().
Fig. 14. Function of influence i0 in a cylindrical shell with an external circumferential semi-elliptic crack (t/R = 1/10; a/c = 1/4; a/t = 0.4).
199
Fig. 15. Function of influence i0 in a spherical shell with an external semi-elliptic crack (t/R = 1/10; a/c = 1/5; a/t = 0.2).
4.4. Validation
Before presenting the whole results in the following section, let us validate our numerical model by com-
parison with other previous numerical solutions. In the case of semi-elliptic cracks, the current point on the
front is defined using the parameter u as seen in Fig. 13. This angle is equal to 0 at the surface and 90 at
the bottom. It allows us to study the variation of the stress intensity factor along the crack front.
In elasticity, with a constant stress distribution, the function of influence i0 obtained here for a cylindrical
shell is compared to that found by Chapuliot [1] for the same geometrical configuration. In the case of a spher-
ical shell, the same function is compared to that derived by Brighenti [16], which is in good accordance with
the earlier solution of Joseph and Erdogan [17], in another configuration.
Figs. 14 and 15 show a very good agreement between our numerical results and the previous ones.
5. Numerical results
5.1. Elasticity
The main objective of our work, in elasticity, is the tabulation of a great amount of values for each function
of influence considered, in a wide range of geometrical configurations. From this point, it will be easy to derive
the stress intensity factor in any loading case, using Eq. (3).
5.1.1. Axisymmetric cracks

All the results, that is to say the functions of influence i0, i1, i2 and i3 for various t/R and a/t ratios, are
depicted in the tables of Appendix A. Tables A.1 and A.2 correspond to external and internal cracks, respec-
tively, in cylindrical shells and Tables A.3 and A.4 give similar results in the spherical case.
In a systematic way, one can notice that the four parameters in (with n = 0,1,2,3) become greater when the
a/t ratio increases or the t/R ratio decreases. Indeed, for any loading case, the nocivity degree of an axisym-
metric defect grows with its depth. Moreover, the thinner shells represent the less favorable cases, other
parameters being constant. Actually, one can mention that the first parameter i0 overcomes the other ones
(the values of in decrease with respect to n), that is to say the constant component of a possible complex load-
ing is the most damaging.
5.1.2. Semi-elliptic cracks

In the case of semi-elliptic cracks, the functions of influence in depend on the current point on the crack
front. The maximum values of all these functions are always reached at the deepest point of the front
(u = 90), as observed in Figs. 14 and 15, given that all the cracks considered verify a < c. In the following,
we will only focus on these maximum values.
All the numerical results, for various t/R, a/c and a/t ratios, are depicted in the tables of Appendix A.
Tables A.5 and A.6 correspond to external and internal circumferential cracks, respectively, in cylindrical
200
shells. Tables A.7 and A.8 concern the cases of axial cracks, respectively external and internal. Finally, Tables
A.9 and A.10 give similar results in the spherical case.
5.2. Plasticity
Let us now consider the particular case of a cylindrical or spherical shell under pressure. It implies here that
only i0 is non zero (the stress distribution is constant on the crack lips), but the method described above can be
applied generally to more complex loading cases. Firstly, the elastoplastic J integral is estimated using the
semi-analytical R6 and A16 criteria, which depend on the corresponding elastic values, easily obtained from
the previous tables. Secondly, in order to validate these simplified approaches, the elastoplastic J integral is
also calculated using the finite element method.
The influence factors of the previous section in elasticity did not depend on the material properties. On the
contrary, in elastoplasticity, it is necessary to specify all the material parameters. In further applications, the
following values will be considered: Young’s modulus E = 200 000 MPa and Poisson’s ratio m = 0.3. The yield
stress is ry = 217 MPa and a linear kinematic hardening is involved with hardening modulus H = 40 000 MPa.
Let us illustrate first the general methodology presented above with the help of an example. Then, the
results from R6 or A16 methods will be validated by comparison with the finite element method.
5.2.1. Example of a spherical shell with an internal semi-elliptic crack

Let us consider a spherical shell containing an internal semi-elliptic defect with the following dimensions:
t = 60 mm; t/R = 1/80; a/c = 1/4; a/t = 0.4. The applied load consists of an internal uniform pressure
(4 MPa), which provides a constant stress distribution onto the lips of the crack. In elasticity, the influence
parameter i0 corresponding to the given geometry is extracted from the associated table and the stress intensity
factor KI together with the J integral are deduced by Eq. (2). Then, using the applied and limit loads, one can
assess the analytical plastic corrections for the J integral, based on both R6 and A16 methods, particularly
involving the reference stress and strain. Relations (5) and (6) lead to the proper evaluations of the elastoplas-
tic J integral. In the following Table 1, all the intermediate quantities involved in the simplified estimation
methods are depicted, including the final semi-analytical values of the J integral. The numerical value of
the J integral, obtained by the finite element method and denoted by JFE, is given too for comparison
purposes.
This approach (with either R6 or A16 criteria) has already been applied to a wide set of configurations.
Moreover, direct finite element calculations have been performed for most of the cases. A few results are plot-
ted in the next section, in order to compare semi-analytical and numerical solutions.
5.2.2. Validation and analysis of the results

First, in the following curves (Figs. 16 and 17), the J integrals obtained by both analytical and numerical
methods are plotted versus the applied load (divided by the pressure at which the shell would plastify without
defect) for two examples of cylindrical and spherical shells and two crack geometries.
In each case, semi-analytical results following from the R6 or A16 methods are very similar. Moreover,
both simplified solutions are in very good accordance with finite element calculations. Some noticeable differ-
Table 1
Results of elastoplastic calculations for an internal semi-elliptic crack in a spherical shell (t = 60 mm; t/R = 1/80; a/c = 1/4; a/t = 0.4)
pffiffiffiffiffiffiffiffi
i0 K I ðMPa mmÞ Je (N/mm) P (MPa) PL (MPa)
1.109995889 1542.133282 10.82069651 160 201.81
rref (MPa) eref Lr

172.0430108 0.000860215 0.792824934
KR6 JR6 (N/mm) KA16 JA16 (N/mm) JFE (N/mm)
1.314285688 14.22148657 1.192982446 12.908901 12.174252
201
Fig. 16. Elastoplastic J integral vs. the applied load for a cylindrical shell with an external longitudinal semi-elliptic crack (t/R = 1/10;
a/c = 1/8; a/t = 0.4).
Fig. 17. Elastoplastic J integral vs. the applied load for a spherical shell with an external semi-elliptic crack (t/R = 1/10; a/c = 1/4;
a/t = 0.4).
Fig. 18. Elastoplastic J integral vs. the a/t ratio for a spherical shell with an external semi-elliptic crack (t/R = 1/10; a/c = 1/2; P/Pe = 0.8).
ences only appear either when the applied pressure becomes largely greater than the reference pressure (plas-
ticity is not confined yet) or when the crack depth reaches more than 75% of the shell thickness. It could be
mentioned that analytical values are always greater than numerical ones, what makes the former more con-
servative. Thus the R6 and A16 criteria provide satisfactory approximations of the elastoplastic J integral
without any more numerical computations, in the sense of safety.
In the next Fig. 18, the same results are plotted versus the a/t ratio, for a given pressure in a spherical shell.
It can naturally be noticed that a deeper crack is more severe.
These are only a few samples among all the configurations studied, varying the crack size and the shell
geometry. The whole results show the same agreement between analytical and numerical values. So the R6
and A16 methods are proved to be effective in solving easily the problem of severity of cracks in cylindrical
and spherical shells.
202
6. Conclusion
This study deals with the severity of cracks in cylindrical and spherical shells. Two crack shapes were retained,
either axisymmetric or semi-elliptic, in order to take into account the whole defects detected in practice. The
stress intensity factors in elasticity could be derived from the functions of influence corresponding to polynomial
stress distributions in the vicinity of the crack front, in order to consider any complicated loading, according to
the superposition principle. Calculations were performed using the finite element method and all the results were
put together in appropriate tables (Tables A.1–A.10) describing many configurations for a wide range of geomet-
rical parameters and all the elementary stress distributions. In plasticity, the common case of internal pressure
was particularly retained. The J integral was performed as a correction of the elastic value, depending on the limit
pressure of the cracked structure. The semi-analytical R6 and A16 methods were then validated in comparison
with purely numerical solutions obtained by the finite element method. Our numerous results, in plasticity as
much as in elasticity, give lots of information about the influence of geometrical features on the nocivity degree
of a given defect. The advocated methodology, involving the use of elastic values coming from the previous tables
and the analytical plastic correction methods described above, provides a noticeable and efficient help in the ser-
vice support of cylindrical and spherical shells under pressure.
Appendix A. Influence factors for axisymmetric and semi-elliptic cracks
Tables A.1–A.10.
Table A.1
Influence factors for an external axisymmetric crack in a cylindrical shell
t/R a/t i0 i1 i2 i3
1/2 0.2 1.229 0.712 0.534 0.44
1/2 0.4 1.457 0.799 0.589 0.488
1/2 0.6 1.802 0.93 0.659 0.529
1/2 0.75 2.422 1.185 0.8074 0.627
1/10 0.2 1.269 0.729 0.544 0.447
1/10 0.4 1.637 0.84 0.629 0.515
1/10 0.6 2.172 1.067 0.735 0.578
1/10 0.75 2.706 1.287 0.8615 0.6612
1/80 0.2 1.328 0.761 0.568 0.468
1/80 0.4 1.896 0.978 0.689 0.547
1/80 0.6 2.963 1.37 0.901 0.684
1/80 0.75 4.392 1.878 1.179 0.8617
Table A.2
Influence factors for an internal axisymmetric crack in a cylindrical shell
t/R a/t i0 i1 i2 i3
1/2 0.2 1.218 0.708 0.532 0.444
1/2 0.4 1.426 0.794 0.574 0.483
1/2 0.6 1.712 0.902 0.642 0.516
1/2 0.75 2.115 1.174 0.804 0.602
1/10 0.2 1.256 0.715 0.538 0.446
1/10 0.4 1.565 0.835 0.623 0.511
1/10 0.6 1.982 0.981 0.732 0.57
1/10 0.75 2.516 1.233 0.836 0.646
1/80 0.2 1.316 0.757 0.565 0.465
1/80 0.4 1.824 0.972 0.684 0.543
1/80 0.6 2.566 1.293 0.897 0.681
1/80 0.75 4.007 1.745 1.172 0.86
203
Table A.3
Influence factors for an external axisymmetric crack in a spherical shell
t/R a/t i0 i1 i2 i3
1/2 0.2 1.3446 0.75514 0.56442 0.46532
1/2 0.4 1.7643 0.92771 0.66067 0.52852
1/2 0.5 2.0909 1.0493 0.72725 0.57169
1/2 0.6 2.4881 1.1952 0.8064 0.62263
1/2 0.7 2.9402 1.3616 0.89678 0.68086
1/2 0.75 3.1753 1.4507 0.94614 0.71313
1/10 0.2 1.3484 0.76898 0.57242 0.47071
1/10 0.4 2.0065 1.0208 0.71261 0.56262
1/10 0.5 2.5872 1.2361 0.83008 0.63852
1/10 0.6 3.463 1.5545 1.0014 0.7485
1/10 0.7 4.8054 2.0341 1.2561 0.90924
1/10 0.75 5.7106 2.3543 1.4248 1.0154
1/80 0.2 1.3598 0.7736 0.57516 0.47255
1/80 0.4 2.0932 1.054 0.7317 0.57512
1/80 0.5 2.7871 1.3111 0.87127 0.66612
1/80 0.6 3.9425 1.7306 1.0967 0.80926
1/80 0.7 6.0827 2.4929 1.5006 1.0643
1/80 0.75 7.9204 3.139 1.8393 1.2765
Table A.4
Influence factors for an internal axisymmetric crack in a spherical shell
t/R a/t i0 i1 i2 i3
1/2 0.2 1.2821 0.74345 0.55812 0.46131
1/2 0.4 1.7456 0.9215 0.65756 0.52665
1/2 0.5 2.0797 1.0465 0.72619 0.57123
1/2 0.6 2.4883 1.198 0.80896 0.62475
1/2 0.7 2.9636 1.3761 0.90672 0.68818
1/2 0.75 3.2188 1.4748 0.96205 0.72461
1/10 0.2 1.3435 0.76731 0.57155 0.47017
1/10 0.4 2.0095 1.022 0.71332 0.56309
1/10 0.5 2.5936 1.2385 0.83137 0.63935
1/10 0.6 3.471 1.5574 1.003 0.74906
1/10 0.7 4.8118 2.0368 1.2576 0.91028
1/10 0.75 5.7146 2.3566 1.4263 1.0164
1/80 0.2 1.3592 0.77343 0.57499 0.47244
1/80 0.4 2.094 1.0543 0.73129 0.57486
1/80 0.5 2.7887 1.3117 0.87158 0.66544
1/80 0.6 3.9449 1.7314 1.0972 0.80954
1/80 0.7 6.0859 2.494 1.5011 1.0647
1/80 0.75 7.9239 3.1402 1.8399 1.2769
Table A.5
Influence factors for an external circumferential semi-elliptic crack in a cylindrical shell
t/R a/c a/t i0 i1 i2 i3
1/2 1/2 0.2 0.9086168 0.55923 0.43554684 0.3674667
1/2 1/2 0.4 0.9707519 0.5883312 0.45356589 0.3814232
1/2 1/2 0.6 1.05350 0.625274 0.47302 0.39632
1/2 1/2 0.8 1.2079899 0.65152222 0.50683309 0.4170637
1/2 1/4 0.2 1.0754754 0.64042622 0.48736905 0.4059111
1/2 1/4 0.4 1.26324 0.69822952 0.51900869 0.4268091
1/2 1/4 0.6 1.557203 0.8094532 0.579374 0.463281
1/2 1/4 0.8 1.999365 0.98053374 0.67005992 0.5228508
1/2 1/8 0.2 1.1758348 0.68119666 0.51376287 0.4368066
(continued on next page)
204
Table A.5 (continued)

1/2 1/8 0.4 1.4724616 0.79609906 0.57671713 0.4665349
1/2 1/8 0.6 1.894534 0.95013427 0.6574923 0.5234177
1/2 1/8 0.8 3.0513516 1.43435488 0.82327786 0.6266641
1/10 1/2 0.2 0.88636 0.55740 0.43507794 0.3674564
1/10 1/2 0.4 0.9565709 0.58573632 0.45330716 0.3818889
1/10 1/2 0.6 1.0505422 0.6201343 0.475291 0.3974521
1/10 1/2 0.8 1.1118042 0.65152222 0.49579706 0.4129901
1/10 1/4 0.2 1.0568871 0.6342894 0.48434937 0.4041444
1/10 1/4 0.4 1.2430695 0.69822952 0.51900869 0.4268091
1/10 1/4 0.6 1.4823807 0.78836123 0.5664278 0.456896
1/10 1/4 0.8 1.73094 0.88843085 0.62532686 0.4968856
1/10 1/8 0.2 1.1758348 0.68119666 0.51376287 0.4368066
1/10 1/8 0.4 1.4566936 0.78884282 0.57219852 0.4633717
1/10 1/8 0.6 1.9435723 0.95554376 0.6605623 0.5043262
1/10 1/8 0.8 2.4622705 1.20934213 0.71648962 0.5583252
1/80 1/2 0.2 0.8980772 0.55396 0.433449 0.3665416
1/80 1/2 0.4 0.9258241 0.5883312 0.45356589 0.3814232
1/80 1/2 0.6 1.0323156 0.614528 0.4725689 0.397554
1/80 1/2 0.8 1.072221 0.6203171 0.49249078 0.4170637
1/80 1/4 0.2 1.0352474 0.62670415 0.48043401 0.4017502
1/80 1/4 0.4 1.2278066 0.69822952 0.51900869 0.4268091
1/80 1/4 0.6 1.4523987 0.77645635 0.5609887 0.4507875
1/80 1/4 0.8 1.55885 0.83304467 0.59851664 0.4812903
1/80 1/8 0.2 1.1758348 0.68119666 0.51376287 0.4368066
1/80 1/8 0.4 1.4152124 0.77310568 0.56348362 0.4576993
1/80 1/8 0.6 1.954378 0.9612805 0.663789 0.518056
1/80 1/8 0.8 2.6596263 1.2702422 0.74109557 0.5722313
Table A.6
Influence factors for an internal circumferential semi-elliptic crack in a cylindrical shell
1/2 1/2 0.2 0.87496 0.5429 0.426722 0.3618129
1/2 1/2 0.4 0.8971001 0.564324 0.412884 0.3746133
1/2 1/2 0.6 0.90348 0.6000343 0.467215 0.394267
1/2 1/2 0.8 0.992975 0.6594027 0.456828 0.3893471
1/2 1/4 0.2 0.9258442 0.5822104 0.454334 0.3838792
1/2 1/4 0.4 0.95675 0.6626068 0.465062 0.3924361
1/2 1/4 0.6 1.179434 0.6934503 0.523018 0.4325623
1/2 1/4 0.8 1.12082 0.688570 0.52664 0.4382121
1/2 1/8 0.2 1.0282473 0.6297836 0.484995 0.4177781
1/2 1/8 0.4 1.0782346 0.6524608 0.49956 0.417432
1/2 1/8 0.6 1.342768 0.7624393 0.570033 0.4764821
1/2 1/8 0.8 1.5077579 0.8904494 0.651365 0.4752007
1/10 1/2 0.2 0.89313 0.54291 0.426722 0.3618129
1/10 1/2 0.4 0.8971001 0.564324 0.453307 0.3746133
1/10 1/2 0.6 1.0050542 0.610374 0.470035 0.3974536
1/10 1/2 0.8 1.2079899 0.6515222 0.506833 0.4170637
1/10 1/4 0.2 1.0194565 0.6199079 0.476263 0.3987906
1/10 1/4 0.4 1.1372582 0.6626068 0.500324 0.4151314
1/10 1/4 0.6 1.3008352 0.7365832 0.544396 0.4455036
1/10 1/4 0.8 1.36853 0.7728132 0.569553 0.4644269
1/10 1/8 0.2 1.11911 0.6648325 0.504763 0.4309493
1/10 1/8 0.4 1.3195398 0.7398836 0.546312 0.4471055
1/10 1/8 0.6 1.45746 0.8600431 0.610343 0.4875843
1/10 1/8 0.8 1.9992073 1.0572885 0.651365 0.51962
205

1/80 1/2 0.2 0.8967051 0.55209 0.432371 0.3658147
1/80 1/2 0.4 0.9175788 0.5728923 0.447013 0.3782089
1/80 1/2 0.6 1.0004545 0.615334 0.472875 0.3846452
1/80 1/2 0.8 1.0377694 0.6368079 0.489952 0.4103021
1/80 1/4 0.2 1.0301411 0.6247546 0.479345 0.4010325
1/80 1/4 0.4 1.1540093 0.6700233 0.504416 0.4178494
1/80 1/4 0.6 1.4004626 0.7654234 0.557433 0.4523478
1/80 1/4 0.8 1.48790 0.8107461 0.587891 0.476824
1/80 1/8 0.2 1.11277 0.6736875 0.503451 0.430175
1/80 1/8 0.4 1.3845562 0.7622876 0.557838 0.4541956
1/80 1/8 0.6 1.8302849 0.92835 0.465239 0.5124472
1/80 1/8 0.8 2.0605688 1.0213364 0.702644 0.5496262
Table A.7
Influence factors for an external longitudinal semi-elliptic crack in a cylindrical shell
1/2 1/2 0.2 0.930064 0.568604 0.4411166 0.371291
1/2 1/2 0.4 1.031625 0.613697 0.4683843 0.391462
1/2 1/2 0.6 1.2550566 0.681732 0.405627 0.416728
1/2 1/2 0.8 1.442269 0.771694 0.5571016 0.425291
1/2 1/4 0.2 1.098907 0.650441 0.4932625 0.420685
1/2 1/4 0.4 1.335855 0.738045 0.5412023 0.441218
1/2 1/4 0.6 1.7136042 0.871953 0.6172893 0.490248
1/2 1/4 0.8 2.315145 1.102251 0.7393429 0.568594
1/2 1/8 0.2 1.203554 0.697928 0.523467 0.432026
1/2 1/8 0.4 1.543115 0.826571 0.5947983 0.478852
1/2 1/8 0.6 2.091533 1.034192 0.902819 0.568201
1/2 1/8 0.8 2.91222 1.345387 0.8816915 0.665798
1/10 1/2 0.2 0.893575 0.560363 0.4368008 0.368619
1/10 1/2 0.4 0.979982 0.594874 0.4584224 0.385247
1/10 1/2 0.6 1.109147 0.634012 0.483721 0.403563
1/10 1/2 0.8 1.204806 0.688807 0.5153971 0.425291
1/10 1/4 0.2 1.067557 0.638527 0.4867338 0.405707
1/10 1/4 0.4 1.288058 0.714675 0.5279527 0.432517
1/10 1/4 0.6 1.623064 0.832917 0.582781 0.472917
1/10 1/4 0.8 2.003146 0.982401 0.6733723 0.52618
1/10 1/8 0.2 1.177273 0.686993 0.5170568 0.427686
1/10 1/8 0.4 1.55252 0.824562 0.5914518 0.47558
1/10 1/8 0.6 2.235689 1.058193 0.715829 0.547924
1/10 1/8 0.8 3.236192 1.425135 0.9087894 0.675194
1/80 1/2 0.2 0.899519 0.553839 0.4334014 0.366523
1/80 1/2 0.4 0.931032 0.577619 0.4494902 0.379762
1/80 1/2 0.6 1.051673 0.618294 0.4781937 0.395742
1/80 1/2 0.8 1.068653 0.64607 0.4941964 0.425291
1/80 1/4 0.2 1.032863 0.625807 0.4799493 0.412252
1/80 1/4 0.4 1.190832 0.683082 0.5112161 0.422048
1/80 1/4 0.6 1.46451 0.773519 0.56283 0.456917
1/80 1/4 0.8 1.629376 0.855315 0.6091175 0.487381
1/80 1/8 0.2 1.105236 0.658942 0.501379 0.417454
1/80 1/8 0.4 1.409451 0.769582 0.5610168 0.45585
1/80 1/8 0.6 2.014529 0.947183 0.6735298 0.527463
1/80 1/8 0.8 2.497652 1.163473 0.7720915 0.590418
206
Table A.8
Influence factors for an internal longitudinal semi-elliptic crack in a cylindrical shell
1/2 1/2 0.2 0.8584641 0.54356 0.4280845 0.36324
1/2 1/2 0.4 0.887368 0.5628368 0.4420725 0.3753087
1/2 1/2 0.6 0.9820160 0.6029103 0.467281 0.493172
1/2 1/2 0.8 1.1652454 0.6921609 0.522616 0.432295
1/2 1/4 0.2 0.9946399 0.6115675 0.4722209 0.3964899
1/2 1/4 0.4 1.1174586 0.6572803 0.4978125 0.4137567
1/2 1/4 0.6 1.3529120 0.7472931 0.548392 0.447209
1/2 1/4 0.8 1.7267513 0.9115809 0.6467848 0.5146758
1/2 1/8 0.2 1.0949536 0.6554061 0.4995192 0.416254
1/2 1/8 0.4 1.3299371 0.7430871 0.5478306 0.447969
1/2 1/8 0.6 1.7017328 0.883672 0.6253819 0.4990483
1/2 1/8 0.8 2.2246379 1.1072125 0.7586977 0.5899972
1/10 1/2 0.2 0.8884913 0.5696409 0.4498775 0.3835263
1/10 1/2 0.4 0.9417307 0.5887928 0.4609169 0.3913888
1/10 1/2 0.6 1.008201 0.6103628 0.468162 0.3924518
1/10 1/2 0.8 1.0957825 0.6655925 0.5100001 0.4263965
1/10 1/4 0.2 1.0517727 0.6428779 0.4968957 0.4185731
1/10 1/4 0.4 1.1956846 0.6920436 0.521394 0.4328328
1/10 1/4 0.6 1.405829 0.7603722 0.553728 0.4501342
1/10 1/4 0.8 1.6555456 0.8773342 0.6273353 0.5031388
1/10 1/8 0.2 1.152554 0.6865632 0.5170568 0.4276858
1/10 1/8 0.4 1.4375241 0.7880246 0.5766621 0.4702785
1/10 1/8 0.6 1.879345 0.939029 0.642719 0.5103426
1/10 1/8 0.8 2.5246789 1.1820381 0.7862742 0.6014373
1/80 1/2 0.2 0.869568 0.5518285 0.4323559 0.36324
1/80 1/2 0.4 0.9137593 0.571606 0.4464056 0.37788
1/80 1/2 0.6 1.025391 0.612918 0.4701821 0.392617
1/80 1/2 0.8 1.0331684 0.6353432 0.4892625 0.4099177
1/80 1/4 0.2 1.0214092 0.6216495 0.4777475 0.4000674
1/80 1/4 0.4 1.1540093 0.6700233 0.504416 0.4178494
1/80 1/4 0.6 1.4300261 0.75361 0.54267 0.451025
1/80 1/4 0.8 1.5077118 0.8169238 0.5907997 0.476824
1/80 1/8 0.2 1.0813685 0.6612623 0.4964497 0.4255976
1/80 1/8 0.4 1.3299371 0.7880246 0.5766621 0.4702785
1/80 1/8 0.6 1.832182 0.941839 0.652341 0.51394
1/80 1/8 0.8 2.1277306 1.0425317 0.7586977 0.5899972
Table A.9
Influence factors for an external semi-elliptic crack in a spherical shell
1/2 1/2 0.2 0.83850847 0.57120543 0.43315326 0.36636021
1/2 1/2 0.4 0.89276768 0.57645227 0.44888128 0.37937947
1/2 1/2 0.6 0.91100257 0.60745886 0.46652615 0.39150102
1/2 1/2 0.8 1.07577889 0.64706336 0.49456452 0.41166266
1/2 1/4 0.2 1.03111897 0.62522417 0.47965903 0.40127084
1/2 1/4 0.4 1.18506526 0.6812726 0.51036169 0.4215604
1/2 1/4 0.6 1.42667077 0.76647305 0.55532037 0.45002088
1/2 1/4 0.8 1.63422367 0.85079405 0.60416282 0.48317227
1/2 1/8 0.2 1.10488031 0.65917428 0.51297883 0.42900142
1/2 1/8 0.4 1.40910217 0.77031108 0.56174861 0.45647844
1/2 1/8 0.6 1.91259958 0.94768944 0.65473331 0.51470884
1/2 1/8 0.8 2.49795418 1.16581543 0.77412936 0.59205116
1/10 1/2 0.2 0.88981071 0.56964087 0.43689923 0.36858034
1/10 1/2 0.4 0.98239534 0.59487827 0.4580539 0.3848098
1/10 1/2 0.6 1.10230214 0.63772838 0.48133048 0.40015677
1/10 1/2 0.8 1.2266883 0.70040624 0.52373338 0.43273831
207

1/10 1/4 0.2 1.0730674 0.64029699 0.48758686 0.40619666
1/10 1/4 0.4 1.28994571 0.72580192 0.53913818 0.44387802
1/10 1/4 0.6 1.60778041 0.82977532 0.58801555 0.4700899
1/10 1/4 0.8 2.01480512 0.98576783 0.67483663 0.52690741
1/10 1/8 0.2 1.23194 0.70798502 0.54008855 0.44665809
1/10 1/8 0.4 1.54008942 0.82594988 0.59486998 0.47913404
1/10 1/8 0.6 1.96131617 1.01523929 0.69328006 0.53918241
1/10 1/8 0.8 2.67223722 1.27613839 0.85198638 0.6504581
1/80 1/2 0.2 0.92851 0.57120543 0.44175951 0.3712884
1/80 1/2 0.4 1.04881041 0.61577938 0.46776823 0.39014582
1/80 1/2 0.6 1.23398845 0.67859825 0.50023806 0.41056292
1/80 1/2 0.8 1.49753834 0.77407379 0.55160612 0.4435023
1/80 1/4 0.2 1.12559409 0.65907866 0.49745244 0.41232117
1/80 1/4 0.4 1.37378547 0.749013 0.54602529 0.44375102
1/80 1/4 0.6 1.60778041 0.88941854 0.62001853 0.49028435
1/80 1/4 0.8 2.40697035 1.12407189 0.74671381 0.57112154
1/80 1/8 0.2 1.18249867 0.68887204 0.52938694 0.43962989
1/80 1/8 0.4 1.52724417 0.81604309 0.58725399 0.473117
1/80 1/8 0.6 2.08043767 1.01523929 0.69328006 0.54021209
1/80 1/8 0.8 2.8474644 1.30496957 0.85267268 0.64347885
Table A.10
Influence factors for an internal semi-elliptic crack in a spherical shell
1/2 1/2 0.2 0.72427025 0.49142776 0.3964915 0.34110532
1/2 1/2 0.4 0.74506363 0.50867164 0.41196593 0.35545807
1/2 1/2 0.6 0.79660094 0.53267681 0.42713805 0.36659229
1/2 1/2 0.8 0.95789547 0.61771571 0.48297983 0.40715464
1/2 1/4 0.2 0.88724554 0.64029699 0.44680686 0.37913624
1/2 1/4 0.4 0.92239527 0.58531013 0.45860262 0.38841786
1/2 1/4 0.6 1.04720536 0.63983071 0.49146608 0.41122062
1/2 1/4 0.8 1.35884 0.78850978 0.58483291 0.47702154
1/2 1/8 0.2 0.985314 0.61313137 0.48694403 0.41167071
1/2 1/8 0.4 1.06891354 0.64847782 0.49712171 0.41569767
1/2 1/8 0.6 1.26575289 0.73267661 0.54715007 0.44997839
1/2 1/8 0.8 1.68487563 0.92941774 0.66988996 0.53649766
1/10 1/2 0.2 0.8413814 0.53986372 0.42514233 0.36084163
1/10 1/2 0.4 0.88167503 0.55900546 0.4580539 0.3848098
1/10 1/2 0.6 0.94042495 0.59274979 0.45450957 0.38408596
1/10 1/2 0.8 1.00398396 0.6250561 0.48391178 0.40660908
1/10 1/4 0.2 1.00223243 0.61472254 0.47407334 0.3977586
1/10 1/4 0.4 1.1002755 0.65145731 0.49490517 0.41203926
1/10 1/4 0.6 1.25752369 0.71267497 0.52818935 0.43364842
1/10 1/4 0.8 1.42294981 0.78946419 0.57735885 0.46889452
1/10 1/8 0.2 1.03971918 0.61313137 0.49981055 0.42060578
1/10 1/8 0.4 1.22262491 0.70308519 0.52618286 0.41569767
1/10 1/8 0.6 1.51676332 0.81113423 0.58470884 0.47185366
1/10 1/8 0.8 1.8474644 0.93049696 0.65267268 0.54347885
1/80 1/2 0.2 0.85836197 0.54762951 0.43005527 0.36439102
1/80 1/2 0.4 0.88632694 0.56201058 0.44142869 0.37480144
1/80 1/2 0.6 0.945352 0.45450957 0.45566402 0.38498244
1/80 1/2 0.8 1.2266883 0.70040624 0.52373338 0.43273831
1/80 1/4 0.2 1.0730674 0.64029699 0.48758686 0.40619666
1/80 1/4 0.4 1.10999589 0.65417789 0.49587662 0.41235205
1/80 1/4 0.6 1.26553193 0.71321328 0.5287123 0.43401867
1/80 1/4 0.8 1.49309562 0.88576783 0.67483663 0.52690741
(continued on next page)
208

1/80 1/8 0.2 1.14885979 0.6841697 0.52938694 0.43962989
1/80 1/8 0.4 1.52724417 0.81604309 0.58725399 0.473117
1/80 1/8 0.6 1.70106858 0.94427723 0.7031352 0.58377975
1/80 1/8 0.8 1.85115989 0.95445264 0.77017705 0.63052617
References
[1] Chapuliot S. Formulaire de KI pour les tubes avec un défaut de surface semi-elliptique longitudinal ou circonférentiel, interne ou
externe. Rapport de recherche – CEA 2000; R-5900.
[2] Rice JR, Levy N. The part-through surface cracks in elastic plates. J Appl Mech 1972;39:185–94.
[3] Wang B, Hu N. Study of a spherical shell with a surface crack by the line-spring model. Engng Struct 2000;22(8):1006–12.
[4] Sih GC, Hagendorf HC. On cracks in shells with shear deformation. Mech Fract 1977;3:201–29.
[5] Brighenti R. Axially-cracked pipes under pulsating internal pressure. Int J Fatigue 2000;22:559–67.
[6] Joseph PF, Erdogan F. A surface crack in shells under mixed-mode loading conditions. J Appl Mech 1988;55:795–804.
[7] McGowan JJ, Raymond M. Stress intensity factor for internal longitudinal semi-elliptical surface flaws in a cylinder under arbitrary
loadings. Am Soc Test Mater 1979:365–80.
[8] Saffih A. Etude de la nocivité des fissures dans un cylindre comportant une transition d’épaisseur: comparaison avec un cylindre droit.
PhD Thesis 2003; University of Sciences and Technologies of Lille.
[9] Saffih A, Hariri S. Numerical study of elliptical cracks in cylinders with a thickness transition. Int J Pres Ves Pip 2006;83(1):35–41.
[10] Saffih A, Hariri S. Comparison of semi-elliptical cracks in cylinders with a thickness transition and in straight cylinders – Elastic–
plastic behaviour. Engng Fract Mech 2006;73(17):2685–97.
[11] Rice JR, Rosengren GF. A path-independent integral and the approximate analysis of strain concentration by notches and cracks. J
Appl Mech 1980;35:379–86.
[12] Ainsworth RA. R6: assessment of the integrity of structures containing defects. British Energy Generation Ltd. 1999; R/H/R6 -
Revision 3.
[13] Kumar V, German MD, Shih CF. Elastic–plastic fracture analysis. EPRI 1981; NP-3607.
[14] Drubay B, Marie S, Chapuliot S, Lacire MH, Michel B, Deschanels H. A16: guide for defect assessment at elevated temperature. Int J
Pres Ves Pip 2003;80(7–8):499–516.
[15] CASTEM. <http://www-cast3m.cea.fr2006>.
[16] Brighenti R. Surface cracks in shells under different hoop stress distributions. Int J Pres Ves Pip 2000;77(9):503–9.
[17] Joseph PF, Erdogan F. Surface cracks in toroidal shells. Nucl Engng Des 1995;158:263–76.
209
ARTICLE IN PRESS
Thin-Walled Structures 47 (2009) 879–889
Contents lists available at ScienceDirect
Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws
Some new analytical results for plastic buckling and initial post-buckling
of plates and cylinders under uniform compression
Philippe Le Grognec a,, Anh Le van b
a
Ecole des Mines de Douai, Polymers and Composites Technology and Mechanical Engineering Department, 941 rue Charles Bourseul, BP 10838, 59508 Douai Cedex, France
b
GeM (Laboratory of Civil and Mechanical Engineering), Faculty of Science, University of Nantes, 2 rue de la Houssinière, BP 92208, 44322 Nantes Cedex 3, France
a r t i c l e in fo abstract
Article history: This paper is devoted to the theoretical aspects of the elastoplastic buckling and initial post-buckling of
Received 30 July 2008 plates and cylinders under uniform compression. The analysis is based on the 3D plastic bifurcation
Received in revised form theory assuming the J 2 flow theory of plasticity with the von Mises yield criterion and a linear isotropic
24 February 2009
hardening. The proposed method is shown to be a systematic and unified way to obtain the critical
Accepted 24 February 2009
loads, the buckling modes and the initial slope of the bifurcated branch for rectangular plates under
Available online 10 April 2009
uniaxial or biaxial compression(-tension) and cylinders under axial compression, with various
Keywords: boundary conditions.
Bifurcation & 2009 Elsevier Ltd. All rights reserved.
Buckling
Elastoplasticity
Axially compressed plates and cylinders
1. Introduction yields the same expansion with fractional exponents as obtained

by Hutchinson. Much later, Cimetière et al. [6–9] thoroughly
Failure of thin structures which is mainly due to the buckling solved the bifurcation problem for the Shanley discrete model and
phenomenon implies the analysis of buckling and post-buckling a compressed beam, and first provided the necessary theoretical
behaviors for their mechanical design, namely the calculation ingredients such as the validity, the convergence of the previous
of the critical loads, the bifurcation modes and the post-critical expansion and the existence of the post-critical branches. Another
equilibrium branches. significant result is the existence of continua of bifurcation points
In elasticity, the bifurcation is related to the structural in plastic buckling problems, which was discovered by Cimetière
instability as shown in Koiter’s theory. The problem is much [10] when dealing with the case of compressed rectangular plates.
more difficult with thicker structures where plasticity may occur These continua enable the structure to bifurcate within intervals
before buckling. Shanley [1] was the first to give the tangent of critical loads, by continuously modifying the unloaded zone
modulus critical load for a discrete model. He showed that and the structural stiffness.
the load increases during the post-bifurcation stage, owing to the The main results were obtained for the plastic bifurcation at
elastic unloading in the structure. Hill [2] extended these results the tangent modulus critical value. The load l acts as a bifurcation
to a 3D continuum by using the concept of ‘‘comparison elastic parameter; the perturbation parameter denoted by x (xX0) can be
solid’’. He examined the uniqueness and stability criteria, and viewed as a measure of the difference between the fundamental
pointed out the difference between bifurcation and stability. In and post-buckling solutions. The asymptotic expansion proposed
order to study the plastic post-buckling, Hutchinson [3] extended by Hutchinson [3] for a general structure reads
Koiter’s results in elasticity by using a series expansion with
fractional exponents, assuming an homogeneous pre-bifurcation l ¼ lT þ l1 x þ l2 x1þb þ (1)
state. In the 1980s, Nguyen and Stolz [4] and Akel [5] reformulated
the problem within the framework of the generalized standard where l1 is positive, thus verifying the hypothesis of an increasing
materials and expanded both the load and the perturbation bifurcation load, l2 is negative and the exponent b is a rational
(i.e. the bifurcation mode amplitude) parameters in power series number such that 0obo1. The l2 o0 condition leads one to
of a time-like parameter. They showed that their power expansion expect a load limit point on the bifurcated branch, i.e. an upper
bound for the strength of the elastoplastic structure in the post-
Corresponding author. critical behavior.
E-mail addresses: legrognec@ensm-douai.fr (P. Le Grognec), From that time, the main theoretical results available in the
anh.le-van@univ-nantes.fr (A. Le van). literature concern the calculation of the critical loads lT and the
0263-8231/$ - see front matter & 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2009.02.010
210
ARTICLE IN PRESS
880 P. Le Grognec, A. Le van / Thin-Walled Structures 47 (2009) 879–889
corresponding buckling modes for different geometries and buckling state, the obtained solutions were found to be in very
loading conditions. good agreement with experimental results [18].
The case of a rectangular plate was investigated by a few The interested reader can find more details on the theoretical
authors. Among others, Durban and Zuckerman [11] derived developments and more complete states-of-art in the quoted
semi-analytical solutions for the plastic buckling of rectangular references.
plates under uniform biaxial compression/tension with various
boundary conditions. They particularly showed the possible
discrepancy between the results provided by the J2 flow and 2. Outline of the paper
deformation theories, thus illustrating the well-known plastic
buckling paradox. The aim of this work is to obtain analytical results for the
One of the most considered cases in plastic buckling is the axial elastoplastic buckling and initial post-buckling of compressed
compression of circular cylindrical shells. In this context, the plates and cylinders, namely (i) the critical load, the bifurcation
two following aspects are usually analyzed: the influence of the mode and (ii) the initial slope of the bifurcated branch. The
constitutive law and the effect of the boundary conditions. proposed method provides a systematic and unified way to obtain
Except for very long cylinders which buckle like beams, the sought quantities since they will all be derived from the 3D
two types of analytical deformed shapes can occur in the post- elastoplastic bifurcation framework.
buckling behavior of cylinders under axial compression. In the Section 3 summarizes the main known results for the 3D
case of thin shells, the buckling mode may usually be non- plastic bifurcation assuming the J2 flow theory with the von Mises
axisymmetric, whereas for moderately thick shells, when plastic yield criterion and a linear isotropic hardening. In Section 4, the
strains appear before buckling, axisymmetric modes are most 3D results are particularized to the case of a uniform biaxial stress
often obtained. state with the purpose of treating the plastic bifurcation of plates
In the latter case, Batterman [12] derived the critical stress and cylinders.
from the rate equilibrium equations, using the flow theory of Using adequate assumptions for plate and shell models then
plasticity: enables one to apply the previous developments to a rectangular
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi plate under uniaxial or biaxial compression(-tension) in Section 5
Ee 4H and a cylinder under axial compression in Section 6, with various
spc ¼ (2)
boundary conditions in both geometries.
R 3½ð5 4nÞE þ 4Hð1 n2 Þ
It should be noted that the initial slope of the bifurcated
where e denotes the shell thickness, R the radius of the cylinder, branch gives information about the load variation in the vicinity
and E, n and H represent the Young’s modulus, the Poisson’s ratio of the critical point and is sufficient for the initial post-buckling
and the linear hardening modulus, respectively. The critical stress analysis. The study of advanced post-buckling necessitates higher-
(2) was obtained for a cylinder with simply supported ends. It can order terms like l2 in expansion (1) and is beyond the scope of
be shown that built-in edges almost yield the same critical value this work.
and mode shape. In order to study the influence of the boundary
conditions, Batterman [13] analytically solved a semi-infinite
cylindrical shell with free ends, which should be the most 3. The 3D plastic bifurcation theory
disadvantageous case. With a constant tangent modulus, he
obtained a critical stress equal to half of the value for simply It is essential in the bifurcation analysis to distinguish between
supported edges. Batterman also showed that the critical stress is the reference and the current configuration so as to take into
a little bit smaller for finite shells, yet it hardly depends on the account the geometric changes and correctly derive the terms
cylinder length. Furthermore, the axisymmetric buckling mode responsible for the bifurcation phenomenon. For this purpose, the
looks like an exponential wave, the radial displacement being total Lagrangian formulation is usually adopted since it proves to
maximal at the ends of the cylinder and almost zero in the middle. be a convenient framework in which the bifurcation equation can
Ore and Durban [14] also derived semi-analytical values for the easily be linearized.
critical load of a cylinder under axial compression and various In the 3D context, let us consider a 3D body with reference
boundary conditions, with a special emphasis on axisymmetric volume O0 ; the position of the current particle in the reference
modes. They showed again the discrepancy between the results configuration will be denoted by X 2 O0. The kinematic decom-
provided by the flow and deformation theories of plasticity. More position is the first step to be defined in a finite plasticity analysis.
recently, Mao and Lu [15] considered the more general case of a In general, one has to choose among the different kinematic
cylindrical shell under combined axial and circumferential descriptions proposed in the literature, the most widely used
compressive loading stresses. The transverse shear is taken into being the Lee multiplicative decomposition of the deformation
account in order to analyze properly the interactive effect of gradient [19,20]. However, as the structural elements of interest
stresses in both directions on the plastic buckling. here are plates and cylinders, it is assumed that the pre-critical
Let us finally mention other analytical approaches for this deformations are small—see assumption (20) below—so that all
post-buckling problem, involving geometric considerations. From the kinematic decompositions are equivalent in the present
a mathematical point of view, the post-critical deformed shape— bifurcation analysis and the choice of a kinematic decomposition
whether the buckling is local or global—can be seen as an has no significant consequences. In order to deal with relations
isometric transformation of the initial cylindrical geometry. analogous to those in small strains, use will be made of the Green
Accordingly, it is possible to reach the final deformed shape and and Naghdi additive decomposition [21,22], which splits the
evaluate the limit load without describing the whole equilibrium Green strain tensor E additively into its elastic and plastic parts:
curve. In a slightly different way, some authors—among which
E ¼ Ee þ Ep (3)
Singace [16] and Gupta and Abbas [17]—represented the post-
e p
buckling of the cylinder by a cyclic formation of folds giving rise to E (resp. E ) is the so-called elastic (resp. plastic) strain tensor. As
plastic hinges and investigated the location and the shape of these mentioned by Green and Naghdi [21], here the ‘‘elastic’’ tensor Ee
folds in both axisymmetric and non-axisymmetric cases. Within is not the classical elastic tensor, it is named so just by analogy
the scope of the limit analysis, which means an advanced post- with the case of small deformations. In this approach, the total
211
ARTICLE IN PRESS
P. Le Grognec, A. Le van / Thin-Walled Structures 47 (2009) 879–889 881
strain E and the plastic one Ep are considered as independent where the symmetric tensor N is
variables, whereas the elastic strain Ee is viewed merely as the
@f
difference between E and Ep [23–25]. The plastic strain Ep is D:
assumed to be a symmetric and objective tensor; and so is Ee .
@ R
N ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (11)
@f @f
The description in the sequel is based on the generalized Hþ :D:
@R @R
standard materials theory [26,27]. Let us assume that the free
energy by unit volume w in the reference configuration can One then derives the nominal tangent elastoplastic tensor:
be written as the sum of the elastic energy we, depending on the
@P @R T
elastic strain Ee , and the hardening energy wa, depending on Kp ¼ ¼F F þ ðI RÞT (12)
@F @E
the hardening parameter a. Since we confine ourselves to small
elastic strains during the whole process, the elastic energy can be In the previous equation, F stands for the deformation gradient,
expressed as a quadratic function of the elastic strain. The free P ¼ F R the first Kirchhoff (non-symmetric) stress tensor, I
energy by unit volume then writes represents the fourth-order unit tensor (Iijkl ¼ dil dkj ) and the
superscript T the major transposition of a fourth-order tensor
wðEe ; aÞ ¼ we ðEe Þ þ wa ðaÞ ¼ 12Ee : D : Ee þ wa ðaÞ defined by ðDT Þijkl ¼ Dklij. Like (10), Eq. (12) can be rewritten as
¼ 12Eeji Dijkl Eelk þ wa ðaÞ (4)
Kp ¼ Ke MT M (13)
where D represents the fourth-order elasticity tensor which is with
assumed to be constant. The material is isotropic so that the
elastic coefficients in an orthonormal basis are Dijkl ¼ ldij dkl þ Ke ¼ F D FT þ ðI RÞT ; M ¼ N FT (14)
mðdik djl þ dil dkj Þ, where dij is the Kronecker symbol, l and m are
Now, let us assume that there exists a fundamental equilibrium
the Lamé constants. Use will also be made of Young’s modulus E
path l/Uf ðlÞ, which is the displacement solution of the
and Poisson’s ratio n which are related to l and m by l ¼
elastoplastic problem under an increasing load factor l (l_ 40),
En=ð1 þ nÞð1 2nÞ and m ¼ E=2ð1 þ nÞ.
and that there exists a bifurcated solution U at a critical time tc
With the additive decomposition (3), the state laws and the
corresponding to the critical load factor lc ¼ lðt c Þ.
evolution laws are analogous to those in small strains. The state
The bifurcated branch is described by an asymptotic expansion
laws express the second Kirchhoff stress tensor R (symmetric) and
with respect to the perturbation parameter xX0:
the thermodynamical force A (conjugate of the hardening
(
parameter a) as the derivatives of the free energy with respect l ¼ lc þ l1 x þ oðxÞ
to the elastic strain and the hardening variable, respectively: (15)
U ¼ Uf ðlÞ þ xX þ oðxÞ
@w @w
R¼ ¼ D : Ee ; A¼ (5) Assumption 1. At critical time tc , the plastic zone Op ðt c Þ
@Ee @a
corresponding to the bifurcated solution is equal to that of the
We assume that the elastic region is defined by the von Mises fundamental solution Opf ðt c Þ, i.e. the bifurcation takes place at
yield function with an isotropic hardening: the tangent modulus critical load with incipient unloading.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 d With the above assumption, the critical load lc ¼ lT and
f ðR; AÞ ¼ R : Rd s0 A (6)
2 the bifurcation mode X are obtained by solving the following
where Rd means the deviatoric part of the stress tensor R and s0 bifurcation equation [28,29]:
is the yield stress. Z
In the framework of generalized standard materials [26], the 8dU; rT dU : Ki ðUf ðlc ÞÞ : rX dO0 ¼ 0 (16)
O0
normality rules state that the plastic strain Ep as well as the
hardening variable a derive from a dissipative potential, which is where Ki stands for Ke or Kp (see Eqs. (13) and (14)), depending on
equal to the yield function: whether the current point belongs to the elastic or plastic zone,
respectively.
p
_ @f ; a_ ¼ L
E_ ¼ L _ @f (7) The coefficient l1 in expansion (15) represents the initial slope
@R @A of the bifurcated branch at the critical point. Its expression
where L _ is the plastic multiplier. Note that the rates of the involves some more assumptions, which are listed below.
Lagrangian variables Ep and a are ordinary derivatives with
Assumption 2. lc is a simple eigenvalue of Eq. (16).
respect to time. The consistency condition leads to the following
expression for the plastic multiplier: Assumption 3. The so-called transversality coefficient Tr is
@f positive:
: D : E_
@ R Z
_
L¼ (8) : dKi ðUf ðlÞÞ
Hþ
@f
:D:
@f Tr¼ rT X : : rX dO0 40 (17)
O0 dl
@R @R l¼lc
where H ¼ ð@f =@AÞ2 @2 w=@a2 is the hardening modulus. The Assumption 4. The plastic zone corresponding to the funda-
material tangent elastoplastic tensor is mental solution at the critical time Opf ðt c Þ is active (i.e. there is
loading and no neutral loading).
@f @f
@R D: :D
Dp ¼ ¼D @R @R (9) Assumption 5. 9D40; 8X 2 Opf ðt c Þ; MðX; lc Þ : rUf ;l ðX; lc ÞXD40.
@E @f @f
Hþ :D:
@R @R Under these assumptions, one has an angular bifurcation and
where the tensor product of two second-order tensors S and T is the initial slope l1 is given by
defined by ðS TÞijkl ¼ Sij T kl. Relation (9) can be recast as
MðX; lc Þ : rX
l1 ¼ max (18)
Dp ¼ D N N (10) X2Opf ðtc Þ MðX; lc Þ : rUf ;l ðX; lc Þ
212
ARTICLE IN PRESS
4. Biaxial stress state The nominal tangent elastoplastic tensor in Eq. (13) becomes
@R
In this section, we shall derive the expressions for the Kp þ ðI RÞT ¼ Dp lei e1 e1 ei Zlei e2 e2 ei
@E
elastoplastic tangent moduli Dp and Kp (Eqs. (10) and (13)) in
(25)
the particular case of biaxial stress states. In the next sections,
these expressions will be applied to the problem of rectangular which is independent of the spatial coordinates. Furthermore,
plates under uniaxial or biaxial compression(-tension) and the when dealing with two-dimensional models like plates or shells,
problem of cylinders under axial compression. an ad hoc assumption has to be added in order to enforce some
Let ðe1 ; e2 ; e3 Þ be an orthonormal basis and the first Kirchhoff specific stress state in the body. Namely, the normal material
stress tensor P be expressed in this basis as follows: stress is assumed to be zero: S33 ¼ 0 (although this contradicts
2 3 the equality E33 ¼ 0 induced by the classical kinematics). Taking
l 0 0 into account this assumption leads one to recast the 3D
P ¼ le1 e1 Zle2 e2 ¼ 4 0 Zl 0 7
6
5 ðl40; Z 2 RÞ constitutive law and to replace tensor Dp in (25) with a suitable
0 0 0 one denoted by Cp, which will be computed in the next section.
(19) Eventually, let us make the following assumption which is
satisfied by the plates and cylinders considered in this work:
where the tensor expression and its matrix representation in the
basis ðe1 ; e2 ; e3 Þ have been put together for brevity. The scalar Z Assumption 7. The whole solid is plastified on the fundamental
represents the ratio between the load in the e2 -direction and that branch:
in the e1 -direction. The load is compressive in the e1 -direction, it
Opf ðtc Þ ¼ O0 (26)
is a tension or compression load in the e2 -direction, depending on
i p
the sign of Z. In order to get explicit expressions for the tangent which implies that K ¼ K throughout the solid.
elastoplastic tensors, let us make the following assumption, which
is satisfied in practice: Then, the bifurcation equation (16) writes in the biaxial stress
case:
Assumption 6. The pre-critical deformations are small: Z
8dU; rT dU : ðCp lei e1 e1 ei
krUk51 (20) O0
Zlei e2 e2 ei Þ : rX dO0 ¼ 0 (27)
Thus, the stress tensor R writes
Furthermore, from Eq. (14) and assumption (20), one has
R ¼ F1 P P (21) M ¼ N FT N. Hence:
The tensor N in Eq. (11) simplifies as follows: m

M : rX ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi½I 3e1 e1
ðH þ 3mÞð1 Z þ Z2 Þ
m
N ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi½I 3e1 e1 þ ZðI 3e2 e2 Þ þ ZðI 3e2 e2 Þ : rX
2 3 m
2 þ Z 0 0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðdiv X 3ðrXÞ11
m 6 7
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi6
4
0 1 2Z 0 7
5 (22) þ Zðdiv X 3ðrXÞ22 ÞÞ (28)
ðH þ 3mÞð1 Z þ Z Þ 2
0 0 1þZ
Accordingly, the initial slope of the bifurcated branch (18)
becomes
Hence, the material tangent elastoplastic tensor in Eq. (10)
!
reads 3ðrXÞ11 div X þ Zð3ðrXÞ22 div XÞ
l1 ¼ max
m2 X2O0 3ðrUcf;l Þ11 divðUcf;l Þ þ Zð3ðrUcf;l Þ22 divðUcf;l ÞÞ
Dp ¼ D ½I 3e1 e1 þ ZðI 3e2 e2 Þ
ðH þ 3mÞð1 Z þ Z2 Þ (29)
½I 3e1 e1 þ ZðI 3e2 e2 Þ (23) where Ucf;l denotes the differentiation Uf ;l ðX; lÞ evaluated at
The components of Dp in the orthonormal basis ðe1 ; e2 ; e3 Þ are l ¼ lc .
Eventually, let us check that the transversality condition (17) is
ð2 ZÞ2
m2 satisfied in the biaxial stress state. From expression (25) for the
Dp1111 ¼ l þ 2m nominal tangent elastoplastic tensor and assumption (26), it can
H þ 3m 1 Z þ Z2
be readily shown that the transversality coefficient Tr in (17) is
m2 ð1 2ZÞ2
Dp2222 ¼ l þ 2m Z
H þ 3m 1 Z þ Z2
Tr ¼ ðjX;1 j2 þ ZjX;2 j2 Þ dO0 (30)
m2 ð1 þ ZÞ2 O0
Dp3333 ¼ l þ 2m
H þ 3m 1 Z þ Z2 It will be seen in the next sections that the eigenmode X does
m2 ð2 ZÞð1 2ZÞ depend on abscissa X 1 , so that Tr is strictly positive as long as Z is
Dp1122 ¼ l þ
H þ 3m 1 Z þ Z2 not too negative and the transversality condition (17) is always
m2 ð2 ZÞð1 þ ZÞ satisfied.
Dp1133 ¼ l þ
H þ 3m 1 Z þ Z2
Remark 1. (i) All the previous developments apply to the special
m2 ð1 2ZÞð1 þ ZÞ
Dp2233 ¼ l case of a uniaxial stress state by taking Z ¼ 0.
H þ 3m 1 Z þ Z2
(ii) Whereas the above results have been obtained using the J 2
Dp1212 ¼ Dp1313 ¼ Dp2323 ¼ m (24) flow theory of plasticity, similar expressions can straightforwardly
The other components are either zero or derived from be derived for the deformation theory. The same methodology is
Eqs. (24) using both major and minor symmetries of tensor Dp valid, provided the elastoplastic tangent moduli in Eq. (24) are
(Dpijkl ¼ Dpklij ¼ Dpjikl ¼ Dpijlk ). replaced with those corresponding to the deformation theory.
213
ARTICLE IN PRESS
5. Plastic bifurcation of a rectangular plate under uniaxial or b ¼ C p1122

biaxial compression(-tension) ET
2 5Z þ 2Z2 ½2ð1 2nÞð1 þ Z2 Þ ð5 4nÞZ
¼E E
ET
Let us consider a rectangular plate with reference dimensions a ð5 4nÞð1 þ Z2 Þ 2ð4 5nÞZ ð1 2nÞ½ð1 2nÞð1 þ Z2 Þ 2ð2 nÞZ
E
along x ¼ e1 , b along y ¼ e2 (aXb or aob) and thickness e along
g ¼ C p2222
z ¼ e3 (e5a; b). The edges are subjected to a compressive nominal
ET
stress PXX ¼ lo0 normal to the sides of length b and another ð2 ZÞ2 þ 3Z2
¼E E
stress PYY ¼ Zl normal to the sides of length a (which is ET
ð5 4nÞð1 þ Z2 Þ 2ð4 5nÞZ ð1 2nÞ½ð1 2nÞð1 þ Z2 Þ 2ð2 nÞZ
compressive when Z40). The biaxial stress state can be dealt with E
by applying the results established in Section 4. (36)
The plate is made of an elastoplastic material with a linear
Eq. (35) can be analytically solved in the classical case where
isotropic hardening (make A ¼ Ha and H ¼constant in (5) and (8)). the four edges are simply supported on the Oxy plane. The mode
Use will also be made of the tangent modulus ET related to the
is assumed to be sinusoidal along the x and y-axes in order to
Young’s modulus E and the isotropic hardening modulus H by automatically satisfy the boundary conditions:
1=ET ¼ 1=E þ 1=H. The yield stress s0 is assumed small enough for 8
the plastic strains to appear before buckling. > U¼0
>
<
The kinematics is described by the Love–Kirchhoff model, V¼0 (37)
hence the displacement field is computed from the three >
> e ppX qpY
: W ¼ sin sin ðp; q 2 N Þ
displacement components UðX; Y; tÞ, VðX; Y; tÞ and WðX; Y; tÞ of 2 a b
the middle surface: By substituting for W from Eq. (37) into Eq. (35), one gets the
expression for the critical stress:
U ZW;X

p4 q4 p2 q2
U ¼ V ZW;Y (31) a þ g 4 þ ð2b þ 4mÞ 2
p2 e2 a4
W lc ¼ b a2 b (38)
12 p2 q2
2
þZ 2
a b
The integers p and q minimizing lc are obtained after solving
5.1. Critical load and bifurcation mode
the following equations:
8 4
The bifurcation mode X and the displacement variation dU are >
> p p2 q2 q4
expressed according to the Love–Kirchhoff kinematics: < aa4 þ 2aZa2 2 þ ðð2b þ 4mÞZ gÞ 4 ¼ 0
>
b b
(39)
>
> p4 p2 q2 q4
U Z W; X ; dU Z dW;X >
: ð2b þ 4m aZÞ 4 þ 2g 2 2 þ gZ 4 ¼ 0
a a b b

X ¼ V Z W;Y ; dU ¼ dV Z dW;Y (32)
Since the expressions for p and q are intricate in the case of
W; dW
arbitrary a=b, let us confine ourselves to the case of a square plate
Whence the bifurcation mode gradient: (a ¼ b) in the equi-biaxial stress state (Z ¼ 1). The problem is
symmetric with respect to x and y, so that integers p and q in (37)
2 3 must be equal and Eq. (38) becomes
U;X Z W;XX U;Y Z W;XY W;X
rX ¼ 6
4 V;X Z W;YX V;Y Z W;YY W;Y 7
5 (33) p2 e2
lc ¼ ða þ 2b þ g þ 4mÞp2 (40)
W;X W;Y 0 24a2
Clearly, the integer p minimizing lc is p ¼ 1. Relations (36) and
In the case of a thin plate, the normal stress SZZ is assumed to (40) then yield the critical stress:
be zero (the subscripts Z and 3 are used alike). Then, the reduced
tensor Cp involved in (27) can be recast as 3ET
p2 e2 1þ
lc ¼ E E (41)
Dpij33 Dp33kl 12a2 ET
ð1 þ nÞ 1 þ ð1 2nÞ
C pijkl ¼ Dpijkl ði; jÞað3; 3Þ; ðk; lÞað3; 3Þ (34) E
Dp3333
In the case of a square plate, with Z 2 ½0:1; 1, the sought mode
The tensor Cp has the major and both minor symmetries. After is obtained by taking p ¼ q ¼ 1. Table 1 displays the numerical
integrating the bifurcation equation (27) through the thickness, values which are used throughout. Fig. 1 plots the plastic buckling
integrating by parts with respect to X and Y, and eliminating stress of a square plate under biaxial compression(-tension) with
negligible higher-order terms, one obtains the following local simply supported edges versus the load ratio Z, including the
equation for the out-of-plane component W of the eigenmode: particular case of a uniaxial loading (Z ¼ 0) which will be studied
in details in the sequel. As indicated in the figure, adding a small
e2 ½aW;XXXX þ gW;YYYY þ ð2b þ 4mÞW;XXYY tensile stress in the y-direction has a stabilizing effect on the
þ 12lc ðW;XX þ ZW;YY Þ ¼ 0 (35) behavior of the plate under uniaxial compression. On the contrary,
where the notations (a, b, g) defined as follows have been used for
brevity: Table 1
Geometric and material parameters.
a ¼ C p1111
Dimension Thickness Young’s Poisson’s Tangent
ET
1 þ 4ZðZ 1Þ þ 3 b e modulus E ratio n modulus ET
¼E E
ET
ð5 4nÞð1 þ Z2 Þ 2ð4 5nÞZ ð1 2nÞ½ð1 2nÞð1 þ Z2 Þ 2ð2 nÞZ 100 mm 3 mm 70 000 MPa 0:3 30 000 MPa
E
214
ARTICLE IN PRESS
Fig. 1. Plastic buckling stress for a square plate under biaxial compression(-tension) with simply supported edges.
adding a compressive stress in the y-direction lowers the critical 5.1.1.2. Simply supported and built-in edge solution. Let us consider
load. The minimum value for the critical load corresponds to the now the little-known problem of a rectangular plate with
equi-biaxial compression case (Z ¼ 1) and is given by Eq. (41). simply supported and built-in edges. The plate is the same as
before, except that the simply supported edge at X ¼ 0 is now
5.1.1. Particular case of a uniaxial loading replaced with a clamped edge. The edges X ¼ 0 and X ¼ a remain
The particular case of a uniaxial stress state is readily obtained subjected to a compressive nominal stress PXX ¼ lo0 along the
from the biaxial case by taking Z ¼ 0. The entailed simplifications x-axis.
will enable one to discuss further the critical value and consider The mode is still assumed to be sinusoidal along the y-axis in
another boundary condition type. order to automatically satisfy the boundary conditions on edges
Y ¼ 0 and Y ¼ b:
5.1.1.1. Simply supported edge solution. The critical stress is derived qpY
W ¼ FðXÞ sin ðq 2 N Þ (46)
from Eq. (38): b
" #
p2 e2 p2 q4 a2 q2 where function F acting as a magnitude varying with abscissa
lc ¼ a 2 þ g 4 þ ð2b þ 4mÞ 2 (42) X is to be determined. By substituting for W from Eq. (46)
12 a p2 b b
into Eq. (35), it leads to the following differential equation
It can readily be checked that the wave numbers corresponding to in F:
the smallest critical load are qp 2 qp 4
8 0 1 e2 aF;XXXX ð2b þ 4mÞ F;XX þ g F þ 12lc F;XX ¼ 0
>
> rffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi b b
>
> B u
< p ¼ Int 4 g a ¼ Int Bu
> 4 aC
C
@t 4 (47)
ab ET bA (43)
>
> 1 þ 3
>
> E Taking into account the order of magnitude of the geometric
>
:q ¼ 1
and material parameters, one finds that the general solution of
Eq. (47) can be written as
where IntðrÞ with r 2 R means the one of the two integers closest
to r and minimizing Eq. (42) with q ¼ 1. FðXÞ ¼ k1 cosðrXÞ þ k2 sinðrXÞ þ k3 cosðsXÞ þ k4 sinðsXÞ (48)
With the so-defined integer p, the critical stress reads
2 with
2 2 3
ET ET pb a 8 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 26 4 1 ð1 2nÞ þ 1þ3 þ4 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
pb 7
u
p e 6 2 E E a 7 >
>
> u eqp 2 eqp 2 2 eqp 4
lc ¼ E > u
26 7
þ >
> u 12 l ð2 b þ 4 m Þ 12 l ð2 b þ 4 m Þ 4 ag
12b 41 þ n 2 ET 5 >
> t b b b
5 4n ð1 2nÞ >
>
E <r ¼
>
2e2 a
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
(44) > u
>
>
> u eqp 2 eqp 2 2 eqp 4
> u
>
> u12l ð2b þ 4mÞ þ 12l ð2b þ 4mÞ 4ag
In the particular case of a square plate (a ¼ b), since ET =E lies >
>
> t b b b
>
:s ¼
between 0 and 1 in practice, Eq. (43) leads to choosing between 2e2 a
p ¼ 1 and p ¼ 2, i.e. choosing the greater number of a þ g and (49)
4a þ g=4. It can readily be checked that for all ET =E 2 ½0; 1,
a þ go4a þ g=4, so that p ¼ 1 is the proper value. One eventually
The boundary conditions at the clamped edge X ¼ 0 (W ¼ 0
finds the critical stress obtained in [10] by a different way:
and W;X ¼ 0) enable one to eliminate k3 and k4 according to
2 3
ET 8
p2 e2 6 2 9 þ ð8n 1Þ < k3 ¼ k1
E 7
lc ¼ E 4 þ 5 (45) r (50)
12a2 1 þ n 2 ET : k4 ¼ k2
5 4n ð1 2nÞ s
E
215
ARTICLE IN PRESS
With expression (46), it is not a trivial matter to satisfy the details, one finds the fundamental displacement field:
remaining boundary conditions at X ¼ a: " ! #
l 1 s0
( Uf ¼ ð1 ZnÞ þ l pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2 þ ZÞ Xx
E 2H 1 Z þ Z2
W¼0 " ! #
(51)
aW;XX þ bW;YY ¼ 0 l 1 s0
þ ðZ nÞ þ l pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2Z þ 1Þ Yy (54)
E 2H 1 Z þ Z2
These boundary conditions lead to the following linear equation In the case of simply supported edges, substituting Eqs. (37) and
system: (54) into Eq. (29) yields the initial slope of the bifurcated branch:
8 p2 q2
< k1 ðcosðraÞ cosðsaÞÞ þ k2 sinðraÞ r sinðsaÞ ¼ 0 ð2 ZÞ þ 2 ð2Z 1Þ
p2 e2 a2 b
s (52) l1 ¼ ET (55)
: k ðs2 cosðsaÞ r 2 cosðraÞÞ þ k ðrs sinðsaÞ r 2 sinðraÞÞ ¼ 0 2 ET
1 2 5 8Z þ 5Z2 ð1 2nÞð1 4Z þ Z2 Þ
E
where p and q are the integers minimizing lc in (38), as explained
The critical loads lc are such that the determinant of the in (39). In the case of a square plate (a ¼ b) with Z 2 ½0:1; 1,
previous system vanishes, in order to ensure a non-zero solution Eq. (55) yields
for ðk1 ; k2 Þ. Hence the equation for the critical loads lc :
p2 e2 1þZ
l1 ¼ ET (56)
2a2 ET
2 2 2 2
rðs r Þ cosðraÞ sinðsaÞ þ sðr s Þ cosðsaÞ sinðraÞ ¼ 0 (53) 5 8Z þ 5Z ð1 2nÞð1 4Z þ Z2 Þ
2
E
The initial slope l1 is plotted versus the load ratio Z in Fig. 3,
The minimum critical load is then obtained by numerically which shows that its maximum value is reached at Z 0:88.
solving Eq. (53). A single half-wave mode along the y-axis is In the case of a square plate in the equi-biaxial stress state, the
obtained (q ¼ 1) when minimizing the critical load with respect to initial slope is
parameter q. The critical load for a large range of the aspect ratio
a=b is depicted in Fig. 2 together with the one given by Eq. (44) p2 e2 1
l1 ¼ ET (57)
corresponding to the previous case of simply supported edges, for 2a2 ET
1 þ ð1 2nÞ
comparison purposes. E
As expected, the two boundary condition types considered In the case of a uniaxial stress state, the initial slope of the
yield the same critical load for large a=b values. On the other hand, bifurcated branch is
the critical load lc is identical for the two boundary condition 2
p 1
types, when the aspect ratio a=b takes some particular values, e.g. p2 e2 2 2
a 2b
a=b 1:25; 2:2; 3, as shown in Fig. 2. This fact is noteworthy, l1 ¼ ET (58)
ET
although the authors have no explanations for it. 5 ð1 2nÞ
E
where p is given by (43).
5.2. Initial slope of the bifurcated branch

6. Plastic bifurcation of a cylinder under axial compression
The initial slope l1 of the bifurcated branch is given by Eq. (18)
and depends on the fundamental solution. The fundamental The last section of this paper is devoted to thin cylinders under
solution for the plate under biaxial loading is obtained by solving axial compression. Let us consider a cylindrical tube defined in the
the elastoplasticity equation set, Eqs. (3)–(7). Without giving reference configuration by the length l along the x-axis, the
Fig. 2. Plastic buckling stress for a rectangular plate under uniaxial compression.
216
ARTICLE IN PRESS
Fig. 3. Initial slope of the bifurcated branch for a square plate under biaxial compression(-tension) with simply supported edges.
average radius R and the thickness e5R. One end of the cylinder X and y. Without going into details, we give the resulting local
lies in the plane X ¼ 0 where the radial displacements and equations:
the rotations are free; whereas the other end at X ¼ l is loaded 8
by a uniformly distributed nominal axial compressive stress > m bþm b
>
> ða lc ÞU;XX þ 2 U;yy þ V;X y þ W;X ¼ 0
>
> R R R
PXX ¼ lo0. The material properties are the same as those of the >
>
>
> b þ m e 2
e2
plate in the previous section. To simplify the discussion, we still >
>
>
> U;X y m V þ ðm l c Þ 1 þ V;XX
assume that the yield stress s0 is chosen small enough so that the >
> R 12R 2
12R 2
>
>
plastification occurs before the tangent modulus critical load is >
> g e2
>
>
> þ 2 1þ
> V;yy
reached. >
> R 12R 2
>
>
The kinematics is described by the Koiter model, equivalent to >
> e2 e2 e2
>
>
the Love–Kirchhoff plate model. The displacement field of the < þm
> 2
W;y ðb þ 2m lc Þ 2
W;XX y g W;yyy ¼ 0
12R 12R 12R4
middle surface is defined by the three components: UðX; y; tÞ
>
> b g e2 e2
(axial), VðX; y; tÞ (circumferential) and WðX; y; tÞ (radial). Consider > U;X þ
> þm V;y ðb þ 2m lc Þ V;XX y
>R
> R 2
12R 4
12R2
a current particle in the cylinder, whose position in the reference >
>
>
>
>
> e2 g e2
configuration is defined by ðX; yÞ and the thickness parameter r >
> g V;yyy þ 2 W þ lc W;XX m W;yy
>
> 12R 4
R 12R 4
(e=2prpe=2). Its displacement can be expressed in the local >
>
>
> e2
e2
cylindrical basis ðx; ey ; er Þ as: >
>
> þða lc Þ12W;XXXX þ ð2b þ 4m lc Þ
>
> 2
W;XX yy
>
> 12R
U rW;X >
>
>
>
> e2
r : þg W;yyyy ¼ 0
U ¼ V ðW;y VÞ (59) 12R4
R
W (62)
where coefficients a, b and g are defined as in (36) by making
Z ¼ 0.
6.1. Critical load and bifurcation mode In the axisymmetric case, the circumferential displacements V
and V are zero and all fields are independent of the circumfer-
The bifurcation mode X and the displacement variation dU are ential parameter y. Only two equations remain among Eqs. (62),
expressed according to the kinematics (59): which after omitting negligible terms can be written as
8
U rW;X ; dU rdW;X b
>
>
r r < aU;X þ RW ¼ 0
>
X ¼ V ðW;y VÞ; dU ¼ dV ðdW;y dVÞ (60)
R R (63)

W;

dW > e2
> g b
: a12W;XXXX þ lc W;XX þ 2 W þ RU;X ¼ 0
>
R
Whence the bifurcation mode gradient:
2 3 Combining these two relations leads to a single equation
1
U;X rW;XX RðU;y rW;X y Þ W;X involving the radial component of the eigenmode:
6 r 1 r 1 7
rX ¼ 6
4 V;X RðW;yX V;X Þ RðV;y R ðW;yy V;y Þ þ WÞ
7
RðV W;y Þ 5 !
W;X 1
ð W; y V þ rRðW;y VÞÞ 0 e2 1 b2
R
a W;XXXX þ lc W;XX þ 2 g W¼0 (64)
(61)
12 R a
As in the case of plates, the normal stress Srr is assumed to be Expression (64), valid for plastic buckling, is similar to the
zero and the reduced tensor Cp in Eq. (34) can be used again. One simplified form (in the axisymmetric case) of the Donnell
has to integrate the bifurcation equation (27) in the shell equation [30] for the well-known elastic buckling of a cylinder
thickness and then adequately integrate by parts with respect to under axial compression (see also [31]).
217
ARTICLE IN PRESS
6.1.1. Free edge solution The complete eigenmode is obtained by solving Eqs. (63) with
Since it is difficult to determine the bifurcation mode in the boundary conditions Uð0Þ ¼ 0 and (67):
the case of a free edge cylinder with finite length, we limit 8 rffiffiffiffiffiffi
consideration to a cylinder of infinite length. The symmetry >
> 3e q pX pX
>
> U ¼ e sin p ffiffiffiffiffi
ffi exp p ffiffiffiffiffiffiffiffi
makes it possible to study one half of the cylinder only, >
< R p Re 3Re
with the free edge at X ¼ 0 and the plane of symmetry put at V ¼ 0 (71)
>
>
infinity. >
> pffiffiffi pX pX pX
>
: W ¼ e 3 cos pffiffiffiffiffiffi sin pffiffiffiffiffiffi exp pffiffiffiffiffiffiffiffi
The axisymmetric buckling mode is assumed to be a sinusoidal Re Re 3Re
function weighted by a decreasing exponential function:
where
W ¼ ðk1 cosðxXÞ þ k2 sinðxXÞÞ expðzXÞ (65)
ET
1 ð1 2nÞ
: E
where k1 , k2 , x and z are to be determined. Relation (65) expresses q¼2 (72)
ET
the fact that the radial displacement is significant at the end of the 1þ3
E
cylinder and is zero in the plane of symmetry.
By inserting Eq. (65) in (64), one obtains In thepffiffiffi
above, the mode has been normalized so that
Wð0Þ ¼ e 3.
8 2
> e
>
>
> a ðk1 x4 þ 4k2 x3 z 6k1 x2 z2 4k2 xz3 þ k1 z4 Þ
>
> 12 !
>
>
>
> b 2
k1 6.1.2. Simply supported edge solution
>
> 2 2
þlc ðk1 x 2k2 xz þ k1 z Þ þ g ¼0
>
> Consider now the case of simply supported edges, by assuming
< a R2
(66) that the radial displacements are fixed and the rotations free on
>
> e2
>
>
> a ðk2 x4 4k1 x3 z 6k2 x2 z2 þ 4k1 xz3 þ k2 z4 Þ both edges. This case is simpler than the free edge case and does
> 12
> !
>
> not require the infinite length assumption. The same bifurcation
>
>
> 2 2 b2 k2 equation (64) can be used in order to get the critical load and the
>
> þlc ðk2 x þ 2k1 xz þ k2 z Þ þ g ¼0
: a R2 bifurcation mode, providing that the free edges fundamental
solution is used. This approximation is justified as long as the pre-
The boundary conditions at infinity—namely the displacement critical deformations are small.
W and its derivatives with respect to X are zero—result in the The bifurcation mode is assumed to be sinusoidal so as to
single condition z40. On the edge X ¼ 0, the boundary conditions identically satisfy the kinematic boundary conditions:
are less trivial:
npX
W ¼ e sin ðn 2 N Þ (73)
8 l
< W;XX ¼ 0
e2 (67) By inserting Eq. (73) in Eq. (64), we obtain the critical stress as a
: a W;XXX þ lc W;X ¼ 0
12 function of integer n:
!
2
that is to say 2 2 e2 b2 1 l
lc ðnÞ ¼ an p þ g (74)
12l
2 a n2 p R2
2
8
2 2
>
< k1 x 2k2 xz þ k1 z ¼ 0
2 (68) The n value minimizing lc ðnÞ is
> e 3 2 2 3
: a ðk2 x þ 3k1 x z þ 3k2 xz k1 z Þ þ lc ðk2 x k1 zÞ ¼ 0
12 2lp
n ¼ pffiffiffiffiffiffiffiffi (75)
p 3Re
The bifurcation mode is defined within a factor which can be
determined by enforcing k2 ¼ e. The four Eqs. (66)–(68) then In general, the previous relation provides a real value for n.
give However, if the cylinder is long enough this real value is large
enough to be a good approximation of the sought integer n.
pffiffiffi
k1 ¼ e 3 Inserting Eq. (75) in Eq. (74) then yields the critical value:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u E E
u27 5 4n ð1 2nÞ2 T T u ET
p : u
4 E E 2Eeu
u E
x ¼ pffiffiffiffiffiffi; p¼uu lsupported edges
¼ u ¼ 2lfree edges
(76)
Re t ET 2 c
R t 2 ET
c
1þ3 3 5 4n ð1 2nÞ
E E
p
z ¼ pffiffiffiffiffiffiffiffi (69) Again, this result is in good agreement with Batterman’s
3Re
solution [12] obtained for the same boundary conditions, yet in
and the critical stress: a different way.
The eigenmode is obtained by solving Eqs. (63):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 8
u
u ET > qle npX
Eeu >
> U ¼ cos 1
u E >
lfree
c
edges
¼ u (70) >
< npR l
Rt ET
3 5 4n ð1 2nÞ2 >
V ¼0 (77)
E >
> W ¼ e sin npX
>
>
:
l
This relation compares very well with Batterman’s ones [13],
which show that the critical stress is a little bit smaller for a finite where n is given by Eq. (75). The mode has been normalized so
cylinder and it hardly depends on the cylinder length. that Wð0Þ ¼ e.
218
ARTICLE IN PRESS
6.2. Initial slope of the bifurcated branch the inequality Co4 to be fulfilled. One then finds
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
6.2.1. Free edge solution maxðf Þ ¼ minðf Þ ¼ 3ð16 þ 4C þ C 2 Þ
4 pffiffiffi
The initial slope l1 is given by (29). Here, the symbol O0 stands 1 4þC
for the reference volume of the cylinder and the fundamental exp pffiffiffi tan1 3 (82)
3 4C
solution is
The extremum value of f is reached
pffiffiffi at r ¼ 12 and the smaller

t40 value satisfying tan t ¼ 3ð4 þ CÞ=ð4 CÞ.
l þ s0 1 1 X
U f ðXÞ ET ET E So far, the initial slope l1 has been computed using the

Uf ¼ 0 ¼ 0 (78) eigenmode X given by Eqs. (71), such that the radial displacement

Wf R 1 1 2n 1 1 Wð0Þ at X ¼ 0 is positive, i.e. the free edge is bent outwards. Now
l s0
2 ET E ET E let us change the signs of the modal displacements U and W in
(71), which implies that the radial displacement Wð0Þ at X ¼ 0 is
In the case of an infinite cylinder, Eq. (60) entails negative and the free edge is bent inwards. In that case, the same
div X ¼ ðW=rÞ þ U;X rW;XX , where r ¼ R þ r is the radius at a calculation as above leads to another expression for the initial
current position. From Eq. (78), one has slope:

pX e e pffiffiffi pX pX 16 r pX
exp pffiffiffiffiffiffiffiffi 2q þ 3 cos pffiffiffiffiffiffi sin pffiffiffiffiffiffi þ p2 sin pffiffiffiffiffiffi
3Re R r Re 3 R
l1 ¼ max Re Re (79)
XX0 1 R R ET
r2½Re=2;Rþe=2 4 þ ð1 2nÞ
2ET r r E
Since the cylinder is a thin shell, 1=r can be replaced by 1=R in

ET
Eq. (79). Hence: 2ET e 5 þ ð8n 1Þ
linward free edges
¼ E max f ðt; rÞ
1 ET ET r2½1=2;1=2
ET R tX0
5 þ ð8n 1Þ 1þ3 5 ð1 2nÞ
2ET e E E E
lfree
1
edges
¼ max ½f ðt; rÞ
R ET ET tX0 (83)
1þ3 5 ð1 2nÞ r2½1=2;1=2
E E
where
(80) pffiffiffi
maxðf Þ ¼ 3amaxðf Þ (84)
where
The extremum value for f is reached at t ¼ 0, i.e. on the whole
t pffiffiffi interval r 2 ½1=2; 1=2.
f ðt; rÞ ¼ exp pffiffiffi ½ 3 cos t ð1 C rÞ sin t
3 In the case of compressed plates, Section 5, one has to find the
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ET ET maximum value of a sinusoidal function f ðtÞ, thus satisfying
16 3 5 4n ð1 2nÞ2 maxðf Þ ¼ maxðf Þ. One thereby gets the same value for l1
E E
C¼ 40 (81) whatever the sign chosen for the bifurcation mode X. For the free
ET
5 þ ð8n 1Þ edge cylinder in hand, the extremal values of the damped
E
sinusoidal function f ðt; rÞ are such that maxðf Þamaxðf Þ. De-
and r stands for r=e for brevity. pending on the sign retained for the eigenmode X, i.e. depending
Assume for definiteness that the ratio ET =E is small enough to on whether the free edge moves outwards or inwards, one obtains
have Co4. A thorough analysis of coefficient C defined by (81) two different values for the initial slope l1 . More precisely, the l1
shows that ET =E must be bounded by ET =Eo5=49 0:10204 for value corresponding to Wð0Þo0 is greater than that corresponding
Fig. 4. Initial slope of bifurcated branches for a cylinder under axial compression.
219
ARTICLE IN PRESS
to Wð0Þ40. As regards the load–displacement curves, this fact (2) The initial slopes of the bifurcated branches. These are the
gives rise to two bifurcating (half-)branches with different initial most significant results of the proposed elastoplastic bifurca-
slopes, stemming at the same critical load lc . tion analysis.
6.2.2. Simply supported edge solution References

Again, the case of simply supported edges is simpler than
the free edge case and does not require the infinite length [1] Shanley FR. Inelastic column theory. International Journal of Aeronautical
assumption. From the fundamental solution (78) and Eq. (29), one Sciences 1947;14:261–7.
obtains [2] Hill R. A general theory of uniqueness and stability in elastic–plastic solids.
Journal of the Mechanics and Physics of Solids 1958;6:236–49.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi [3] Hutchinson JW. Plastic buckling. Advances in Applied Mechanics 1974;14:

ET ET ET 67–144.
5 þ ð8n 1Þ þ 4 3 5 4n ð1 2nÞ2 [4] Nguyen QS, Stolz C. Sur la méthode de développement asymptotique en
2ET e E E E
lsupported
1
edges
¼ flambage plastique (On the asymptotic development method in plastic
R ET ET buckling). Comptes Rendus de l’Académie des Sciences 1985;Paris t. 300(7):
1þ3 5 ð1 2nÞ 235–8.
E E
[5] Akel S. Sur le flambage des structures élastoplastiques (On the buckling of
(85) elastoplastic structures). PhD thesis, Ecole Nationale des Ponts et Chaussées,
France; 1987.
[6] Cimetière A, El Koulani A, Léger A. Flambage naissant et post-flambage d’un
Contrary to the free edge case, there is only one single value for
modèle simple en élastoplasticité (Buckling and post-buckling of an
the initial slope, whatever the sign of the radial displacements. elastic–plastic simple model). Comptes Rendus de l’Académie des Sciences
The three values of initial slope (two for the free edges and one for 1994;Paris t. 319(11):1263–9.
the simply supported edges) are compared in Fig. 4 versus the [7] El Koulani A. Continuation dans les problèmes à frontières libres de type
bifurcations plastiques (Continuation in moving boundary problems of the
moduli ratio ET =E. The numerical values are the same as in plastic bifurcation type). PhD thesis, University of Metz, France; 1996.
the plate problem (Section 5), in addition the mean radius of the [8] El Koulani A, Léger A. Solutions bifurquées du problème en vitesses initiales
cylinder is R ¼ 100 mm. pour une poutre élastoplastique (Bifurcated solutions of the initial velocity
problem for an elastic–plastic beam). Comptes Rendus de l’Académie des
Fig. 4 shows that the initial slope when the free edge is Sciences 1996;Paris t. 322(10):1007–13.
bent inwards is always greater than in the case when the free edge [9] El Koulani A, Léger A. Comportement post-critique des poutres élastoplas-
is bent outwards. This stiffening effect on the post-critical tiques : existence et régularité des branches bifurquées (Post-critical behavior
of elastoplastic beams: existence and regularity of the bifurcated branches).
behavior is mainly due to the compressive hoop stress in the Comptes Rendus de l’Académie des Sciences 1997;Paris t. 324(11):1307-13.
inward free edge case, giving rise to incipient elastic unloading at [10] Cimetière A. Sur la modélisation et le flambage des plaques élastoplastiques
the edge X ¼ 0 where the exponential displacement field is (On the modelling and buckling of elastoplastic plates). PhD thesis, University
of Poitiers, France; 1987.
maximal.
[11] Durban D, Zuckerman Z. Elastoplastic buckling of rectangular plates in biaxial
One particular cylinder geometry has been chosen to be compression/tension. International Journal of Mechanical Sciences 1999;41:
numerically computed in [32] and the numerical initial slope 751–65.
[12] Batterman SC. Plastic buckling of axially compressed cylindrical shells. AIAA
obtained compares very well with the analytical one given by
Journal 1965;3(2):316–25.
Eq. (85). Eventually, it should be mentioned that the case of built- [13] Batterman SC. Free-edge plastic buckling of axially compressed cylindrical
in ends, which is not studied here, provides results similar to shells. Journal of Applied Mechanics 1968:73–9.
those with simply supported ends. [14] Ore E, Durban D. Elastoplastic buckling of axially compressed circular
cylindrical shells. International Journal of Mechanical Sciences 1992;34(9):
727–42.
[15] Mao R, Lu G. Plastic buckling of circular cylindrical shells under combined in-
7. Conclusion plane loads. International Journal of Solids and Structures 2001;38:741–57.
[16] Singace AA. Axial crushing analysis of tubes deforming in the multi-lobe
mode. International Journal of Mechanical Sciences 1999;41:865–90.
An analytical approach has been proposed to investigate the [17] Gupta NK, Abbas H. Mathematical modeling of axial crushing of cylindrical
elastoplastic buckling and initial post-buckling of plates under tubes. Thin-Walled Structures 2000;38:355–75.
uni- or bi-axial loadings and cylinders under axial compression. [18] Jones N. Structural impact. Cambridge: Cambridge University Press; 1997.
[19] Lee EH, Liu DT. Finite-strain elastic–plastic theory with application to plane-
First the analysis has been carried out in the unified framework of wave analysis. Journal of Applied Physics 1967;38:19–27.
the 3D elastoplastic bifurcation theory, and then the resulting [20] Lee EH. Elastic–plastic deformations at finite strains. Journal of Applied
equations have been solved under ad hoc assumptions for plate Mechanics 1969;36:1–6.
[21] Green AE, Naghdi PM. A general theory of an elastic–plastic continuum.
and shell models.
Archive for Rational Mechanics and Analysis 1965;18:251–81.
The following geometries and boundary conditions have been [22] Green AE, Naghdi PM. Some remarks on elastic–plastic deformation at finite
considered: strain. International Journal of Engineering Science 1971;9:1219–29.
[23] Naghdi PM, Trapp JA. Restrictions on constitutive equations for finitely
deformed elastic–plastic materials. Quarterly Journal of Mechanics and
(a) Simply supported plates under biaxial compression. Applied Mathematics 1975;28(1):25–46.
(b) Simply supported (and possibly built-in) plates under uniaxial [24] Casey J, Naghdi PM. A remark on the use of the decomposition F ¼ FeFp in
compression. plasticity. Journal of Applied Mechanics 1980;47:672–5.
[25] Naghdi PM. A critical review of the state of finite plasticity. Journal of Applied
(c) Free edge cylinders under axial compression. Mathematics and Physics 1990;41:315–94.
(d) Simply supported edge cylinders under axial compression. [26] Halphen B, Nguyen QS. Sur les matériaux standard généralisés (On the
generalized standard materials). Journal de Mécanique 1975;14(1):39–63.
[27] Germain P, Nguyen QS, Suquet P. Continuum thermodynamics. Journal of
The following results have been obtained: Applied Mechanics 1983;50:1010–20.
[28] Hutchinson JW. On the post-buckling behavior of imperfection-sensitive
structures in the plastic range. Journal of Applied Mechanics 1972:155–62.
(1) The critical loads and the bifurcation modes. Some of the
[29] Nguyen QS. Stability and non-linear solid mechanics. New York: Wiley; 2000.
results are in good agreement with those existing in the [30] Donnell LH. Beams, plates and shells. New York: McGraw-Hill; 1976.
literature. The results obtained for square simply supported [31] Timoshenko SP, Gere JM. Theory of elastic stability. New York: McGraw-Hill;
plates under biaxial loading and rectangular simply supported 1961.
[32] Le Grognec P, Le van A. Elastoplastic bifurcation and collapse of axially loaded
(and possibly built-in) plates under uniaxial loading are to the cylindrical shells. International Journal of Solids and Structures 2008;
best of the authors’ knowledge not known. 45(1):64–86.
220
Marine Structures 22 (2009) 836–854
Marine Structures
journal homepage: www.elsevier.com/locate/
marstruc
Influence of residual stresses and geometric imperfections

on the elastoplastic collapse of cylindrical tubes under
external pressure
Philippe Le Grognec a, *, Pascal Casari b, Dominique Choqueuse c
a
Ecole des Mines de Douai, Polymers and Composites Technology & Mechanical Engineering Department,
941 rue Charles Bourseul – BP 10838, 59508 Douai Cedex, France
b
GeM (Laboratory of Civil and Mechanical Engineering), Institut Universitaire de Technologie de Saint-Nazaire,
58 rue Michel-Ange – BP 420, 44606 Saint-Nazaire Cedex, France
c
Ifremer, Materials and Structures Department, Technopôle de Brest-Iroise – BP 70, 29280 Plouzané, France
a r t i c l e i n f o a b s t r a c t
Article history:
The buckling problem of a circular cylindrical shell has long been
Received 23 January 2009
widely investigated due to its great importance in the design of
Received in revised form 24 July 2009
Accepted 5 September 2009 aerospace and marine structures. Geometric imperfections and
residual stresses are inevitable in practice and have been so far
Keywords:
frequently considered in analytical and numerical predictions. But
Buckling little attention has been paid until now on the combined influence
Elastoplasticity of such initial defects on the critical and often unstable response of
Imperfection sensitivity such elastoplastic structures. In this paper, a shell finite element is
Residual stresses designed within the total Lagrangian formulation framework to
Cylindrical containers under deal with the elastoplastic buckling and post-buckling of thin
external pressure cylindrical tubes under external pressure and axial compression.
A specific experimental process will be introduced in order to
measure residual stresses in the shell very accurately, so as to
include them in the numerical calculations. The present formula-
tion will enable us to describe the complete non-linear solutions,
namely the critical pressures (bifurcation and limit (collapse)
loads), the bifurcation modes and the bifurcated equilibrium
branches up to advanced post-critical states. Comparisons will be
* Corresponding author. Tel.: þ33 (0)3 27 71 23 21; fax: þ33 (0)3 27 71 29 81.
E-mail addresses: legrognec@ensm-douai.fr (P. Le Grognec), pascal.casari@univ-nantes.fr (P. Casari), dominique.choqueuse@
ifremer.fr (D. Choqueuse).
0951-8339/$ – see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marstruc.2009.09.003
221
P. Le Grognec et al. / Marine Structures 22 (2009) 836–854 837
made between numerical results and the experimental critical

value and deformation patterns of a new generation profiler.
Furthermore, the combined effects of geometric imperfections,
residual stresses and plasticity will be analyzed.
1. Introduction
Failure of thin structures submitted to compressive loadings is mainly due to the buckling
phenomenon. Their mechanical design implies thus the analysis of their buckling and post-buckling
behaviors, namely the calculation of the critical loads, the bifurcation modes and possibly the post-
critical equilibrium branches.
In elasticity, the bifurcation is related to the structural instability as shown in Koiter’s theory. The
problem is much more difficult with thicker structures where plasticity may occur before buckling.
From a theoretical point of view, Shanley [1] was the first to give the tangent modulus critical load for
a discrete model. Then, Hill [2] extended these results to a three-dimensional continuum by using the
concept of ‘‘comparison elastic solid’’. He examined the uniqueness and stability criteria, and pointed
out the difference between bifurcation and stability.
A comprehensive work on the buckling of structures was presented by Brush and Almroth [3]. They
examined the effect of initial imperfections, inherent in many real structures, on the critical loads.
When dealing with the buckling phenomenon, cylindrical shells are certainly among the most sensitive
structures to all imperfection types (in comparison with beams and plates) and, as a matter of fact, very
useful in industrial applications. For example, the bifurcation of a cylinder under axial compression,
like a spherical shell under external pressure, corresponds to a sub-critical point (a load limit point on
the bifurcated branch), namely an upper bound for the strength of the shell in the post-critical
behavior. For that reason, the critical problem of a cylinder under axial compression was the subject of
a lot of experimental and numerical investigations [4,5]. A wide variety of experimental results were
obtained in terms of the boundary conditions, the geometries and the material parameters. Further-
more, different numerical models were devised in order to predict the critical loads as well as the
advanced post-buckling behavior. Among others, Tvergaard [6] analyzed the effect of geometric and
material parameters on the buckling behavior of a cylinder under axial compression. It was also shown
that the boundary conditions are a deciding factor for obtaining various types of buckling modes. From
a material point of view, different critical values are obtained in the literature, depending on the
plasticity theory considered. In the case of a cylinder under axial compression, the critical stresses
obtained by the deformation theory are fairly far from the ones corresponding to the flow theory. As a
matter of fact, the solutions derived from the deformation theory turn out to compare best with the
experimental results, although such a theory does not take into account the elastic unloading possi-
bility. On the contrary, the flow theory overpredicts the critical loads. Comparisons between the
predictions of both flow and deformation theories and experimental results were recently performed
in Bardi and Kyriakides [7] and Bardi, Kyriakides and Yun [8], in the case of axially loaded cylindrical
shells, including the effect of anisotropy with the use of Hill’s yield criterion. The flow theory is shown
to significantly overpredict both the critical stresses and strains, whereas the deformation theory gives
rise to predictions much closer to the experimental values.
For this loading case, large discrepancies have always been observed between numerical and
experimental results. The most significant feature is that the experimental critical loads sometimes fall
down to less than half of the numerical values, see for instance Singer [9] for a review on the impor-
tance of shell buckling experiments. These differences, due to unavoidable imperfections in experi-
ments, were explained by many authors. If the major influence of geometric imperfections on the
buckling behavior of cylindrical shells is universally recognized, the importance of residual stresses has
also been enhanced in the case of axial compressive stresses. Just let us mention Mandal and Calladine
[10] and Lancaster, Calladine and Palmer [11], who investigated the buckling under self-weight of
cylinders with free ends by means of an adequate experimental set-up. Since the ends were able to
222
838 P. Le Grognec et al. / Marine Structures 22 (2009) 836–854
move freely in the axial direction, there were no longer imperfections due to localized stresses and the
critical value was found to increase in comparison to the classical experimental result obtained with
enforced displacements at the ends. Therefore, these authors concluded that localized stresses were
strongly responsible for the drop of the critical value.
In some cases, the structure may be quite insensitive to imperfections in elasticity and yet markedly
sensitive in plasticity. Cylindrical shells under external pressure are quite less imperfection-sensitive
than axially compressed ones but their sensitivity is more noticeable in the plastic domain than in
elasticity. As a matter of fact, this problem has more rarely been studied in the literature. Let us mention
the buckling of offshore pipelines under external pressure as one of the main application exhaustively
studied by Kyriakides and Corona [12]. These authors examined the occurrence of elastic or plastic
buckling/collapse depending on the thinness of the tubes and also the influence of several types of
imperfections on the critical loads. The combination of bending and external pressure had previously
been addressed from both experimental and numerical points of view by the same authors [13]. They
came to the conclusion that practical factors that mainly affect the buckling of cylindrical shells under
external pressure are geometric imperfections (initial ovality or thickness variations), residual stresses
and yield anisotropy. Other discrepancies between experimental and simulation results could also
come from the experimental imperfect boundary conditions or loadings; they could finally result from
material heterogeneities, which are often difficult to be measured.
In elasticity, just few authors investigated the imperfection sensitivity of a cylinder under external
pressure. For example, Gusic, Combescure and Jullien [14] analyzed the influence of thickness
imperfections on the buckling behavior. But in many practical applications, the cylindrical shell may
buckle plastically, or at least, collapse plastically as soon as plasticity occurs in the post-buckling stage.
In this case, one deals with a limit load instability due to interactions between geometric and material
non-linearities, which leads to a more or less localized collapse [15].
In that context, many studies deal with the unstable collapse of cylinders, considering all sorts of
imperfections. Until now, most attention has been paid to geometric imperfections. The effect of
thickness variations on plastic buckling and collapse of pipelines under external pressure were
analyzed in Yeh and Kyriakides [16,17], where it was shown that this type of geometric imperfection, as
well as wall eccentricity, has a secondary influence on the critical behavior, compared with an initial
ovality, as long as realistic amplitudes are considered.
Imperfections in the form of residual stresses have more rarely been investigated. Most of the time,
the residual stresses just enhance the yielding plastic zone in the structure and therefore decrease the
buckling load. In elastoplastic column buckling problems, for which residual stresses are important and
widely recognized, initial stresses have never shown any effect on the buckling loads, as far as
a homogeneous purely elastic or plastic behavior is assumed. The effect of residual stresses is thus
mostly significant for members of mid-slenderness, when non-homogeneous (partly elastic and
plastic) pre-bifurcation conditions are involved (see Triantafyllidis [18] for an analysis of general non-
homogeneous pre-bifurcation conditions). This property is due to the simple residual stress patterns
and the one-dimensional stress state encountered in such wire structures, and is no longer valid for
plates or shells. The effect of residual stresses on the buckling of shell-type structures has recently been
examined by a few authors. The problem of collapse under combined bending and external pressure
was revisited by Corona and Kyriakides [19] in order to take into account asymmetric imperfections
together with residual stresses and better explain the scatter between experimental and numerical
solutions under high pressures. See also Kyriakides and Corona [12] for a complete review on the
subject of elastoplastic buckling and collapse of cylindrical shells under various loads (among which
external pressure) with geometric and material imperfections, including residual stresses.
Analytical solutions can be derived for the critical loads and bifurcation modes of beams, plates or
possibly shells with simple boundary conditions and loadings, under homogeneous pre-bifurcation
conditions or even in the presence of simple initial imperfections, both in elasticity and plasticity (see
the pioneering works of Timoshenko and Gere [20] in elasticity and Dubey [21] in plasticity, for
instance). Intermediate semi-analytical approaches involving asymptotic expansions can be derived in
order to estimate not only the buckling behavior but also the post-bifurcated solution. For example, in
his asymptotic analysis of plastic post-buckling, Hutchinson [22] examined the influence of various
geometric imperfection types on the buckling and post-critical behavior of several structures, among
223
which cylindrical shells under various loads. It allowed him to determine safety factors for the critical
load vs. the imperfection size and particularly to assess the stability of the bifurcated branch. Likewise,
in order to include a non-linear pre-buckling behavior, Kheyrkhahan and Peek [23] proposed an
asymptotic analysis using a Lyapunov–Schmidt–Koiter decomposition and an asymptotic expansion
technique in the framework of the finite element method. But as far as the complete non-linear solution
is required or complex initial conditions are assumed, exact solutions obtained by numerically tracking
the bifurcated equilibrium branches and giving rise, if needed, to advanced post-critical states are
preferred. In the case of compressed shells such as cylinders, which are usually very imperfection-
sensitive, numerical tools are thus most often employed to analyze their buckling behavior and collapse.
The present work deals with the buckling behavior and collapse of cylindrical tubes submitted to
both hydrostatic external pressure and axial compression (end effect) in seabeds. Some experiments
performed by one of the authors seem to be in disagreement with classical linearized elastic buckling
pressures. All the above-mentioned facts show that it is preferable to use a numerical tool to analyze
the influence of geometric imperfections and residual stresses, among others, on the complete
non-linear buckling and post-buckling response of such compressed shells. It will be shown that the
occurrence of plasticity is mainly responsible for the early collapse of the shell. Yielding is caused by
the combination of stresses induced by the applied loads with the residual stresses which remain after
the manufacturing process, so that residual stresses are of great importance. Moreover, it will be shown
that it is difficult to separate the effects of residual stresses and geometric imperfections, such as the
ovality of the initially perfect circular cylindrical shell.
2. Outline of the paper
This paper is devoted to the understanding of the buckling and post-buckling behaviors of
a cylindrical tube under combined external pressure and axial compression in the elastoplastic range,
taking into consideration of geometric imperfections together with residual stresses. A specific
numerical tool has been developed to perform the numerical computation of the critical loads, the
bifurcation modes and the advanced post-critical states.
The virtual work principle and the total Lagrangian formulation are used to build a shell finite
element encompassing geometric non-linearities (finite rotations) and material non-linearities
(elastoplastic behavior with the von Mises yield criterion and a linear isotropic hardening). We shall
only focus on the physically more correct flow plasticity theory, observing that both flow and defor-
mation theories of plasticity give rise to nearly similar results in the case of paramount external
pressure. The shell element is designed to deal with buckling and advanced post-buckling of thin
structures. A specific arc-length method and branch-switching procedures are included in order to
resolve snap-through (or even snap-back) and bifurcation phenomena, encountered when dealing
with cylindrical shells under external pressure.
First, the numerical formulation will be concisely presented. Then a specific experimental process
will be introduced in order to measure residual stresses in the shell very accurately. Numerical results
will finally be presented, relating to a new generation profiler, emphasizing the relative and total effect
of residual stresses and geometric imperfections on the elastic or plastic buckling loads and collapse
behavior. The finite element formulation and the solution procedure will be validated by comparison
with related experimental data.
3. Shell finite element formulation
The elastoplastic thin shell finite element formulation is briefly reviewed in the following
paragraphs. A more detailed description, which is outside the scope of this paper, may be found in Le
Grognec and Le van [24].
3.1. Shell kinematics with finite rotations
The shell-like body is represented by a middle surface and an inextensible director field (in the
direction of the shell thickness) which is not constrained to be normal to the middle surface, except at
the reference configuration, thus following a Mindlin–Reissner type formulation.
224
The motion of the inextensible director is described by:
! !
t ¼ R$ T (1)
where the rotation tensor R is orthogonal (R1 ¼ RT) and can be associated with the skew-symmetric
tensor Q through the exponential mapping:
þN k
X Q
R ¼ exp Q ¼ (2)
k!
k¼0
! ! ! !
Let us define the axial vector q by Q$! v ¼ q ^! v , c!
v ˛R3 . From the equality R$ q ¼ q , it follows
!
that R is the rotation about q . Then an explicit expression for Equation (2) is given by the Rodrigues
formula:
sin q 1 cos q! !
R ¼ cos qI þ Qþ 2
q5q (3)
q q
!
where q ¼ k q k is the magnitude of the rotation.
! ! !
Since q is perpendicular to T , the director field t can be recast in the simplified form:
! ! ! sin q! !
t ¼ R$ T ¼ cosq T þ q ^T (4)
q
3.2. Elastoplasticity formulation
The equation set for finite elastoplasticity is established on the basis of the total Lagrangian
formulation and the generalized standard materials theory introduced by Halphen and Nguyen [25].
The elastic strain is related to the second Piola–Kirchhoff stress by the three-dimensional Saint–
Venant–Kirchhoff law, including the residual stress state at the reference configuration. Since the
normal material stress is assumed to be zero, the elastic fourth-order tensor D must be replaced with
a reduced one C for the shell model. Following Green and Naghdi [26], the Green strain tensor E is split
additively into its elastic and plastic parts. Use is made of the von Mises yield criterion with a linear
isotropic hardening.
3.3. The discretized non-linear equations
Use is made of the virtual work principle in order to obtain the discretized non-linear equations. The
middle surface is discretized using eight-node isoparametric shell finite elements. There are five
degrees of freedom at each node: three translations and two rotations in the tangent space. The
! !
reference position, the displacement on the middle surface, and the director vectors T and t are
interpolated in the same way.
The integration over the middle surface and through the thickness is performed using the Gaussian
quadrature. Eventually, the virtual work principle yields the non-linear discretized equation system:
fRðUÞg ¼ fJg fFg ¼ f0g (5)

where {U} denotes the nodal displacement vector of the whole structure, {R} the residual vector, {J}
and {F} are the internal and external force vectors, respectively.
4. Solution procedure
4.1. Local integration
The non-linear equation system (5) is solved by the iterative Newton–Raphson procedure, which
requires the computation of the structural tangent stiffness matrix:
225
vfRðUÞg
½KT ¼ (6)
vfUg
At each iteration, the local integration consists of solving the following problem: assuming the strain
E is known, compute the plastic strain Ep, the stress S and the reduced consistent elastoplastic tangent
modulus Cp. The implicit Euler scheme and an algorithm of radial return type are employed to integrate
the state and evolution laws as described in Simo and Hughes [27].
4.2. Arc-length method
e contains the unknown

The nodal displacement vector is split into two parts: one denoted by fUg
degrees of freedom, the other denoted by fUg contains the prescribed degrees of freedom. The external
force and the tangent stiffness matrix are split in a similar way:
( ) " #
e e e
eT e
U ¼ U fFg ¼ F ½KT ¼ K K T (7)
U F eT
K KT
At each increment, the quadratic arc-length method is used in order to proceed on the equilibrium
branches given a specified arc-length Dl, see Riks [28] and Crisfield [29]. The constraint equation writes,
assuming a proportional prescribed loading:
e k2 þ Dl2 C 2 ¼ Dl2
kDU (8)
ref
where l represents the load factor and Cref is a scale parameter which makes the relation consistent
dimensionally.
Combining Equation (8) with the equilibrium equation leads to a quadratic equation. When solving
this quadratic equation, one may encounter severe computational difficulties due to complex roots
which occur repeatedly. It is found that an efficient way to cope with these complex roots is to modify
the standard solution scheme according to Lam and Morley [30]. The main idea is to project the
residual force onto the external load vector. At a current iteration where complex roots occur, the
residual force is split into one component in the load direction and another component orthogonal to
this load. The last component is mainly responsible for the complex roots and should be eliminated.
4.3. Branching method
Branch-switching techniques are included in the numerical procedure in order to detect the
bifurcation points and bifurcate onto a given branch. Specific methods are implemented following Riks
[28,31] and Seydel [32]. The four fundamental steps are given below.
1. At the end of each increment, it must be checked whether one has gone across one or several
critical points. The detection of critical points is based on the singularity of the tangent stiffness
matrix, which is factorized following the Crout formula. The critical points are determined by
counting the negative pivot number.
2. Each critical point detected has to be isolated in order to determine its nature: limit point or bifur-
cation point. To do this, the current arc-length Dl is re-estimated several times using a dichotomy-like
method. A simple way to distinguish a limit point from a bifurcation point is to determine the sign of
the current stiffness parameter introduced by Bergan, Horrigmoe, Krakeland and Soreide [33]:
n oT n o
e
F e
U
ref Fref
k ¼ n oT n o (9)
e
U e
U
Fref Fref
e g is the reference force vector and fU e g ¼ ½K e e g. The sign of parameter k
e T 1 fF
where fF ref Fref ref
changes when passing a limit point, whereas it remains unchanged when passing a bifurcation point.
226
3. If it is a bifurcation point, the step increment is renewed so as to reach a point just behind the
bifurcation point.
4. Finally, the switching on a bifurcated branch is performed using the mode injection method
e
e solution of ½K e g ¼ f0g, is
e T fZ
[34,32]: at the first step of a bifurcating branch, the eigenvector fZg,
computed and the following predictions are used:
n o
n o e
Z
dl ¼ 0 e ¼
dU (10)
ek
kZ
5. Experimental data
5.1. Context
Unexpected collapse of cylindrical aluminum shells subjected to hydrostatic compression has been
recently observed during qualification of equipments (Ifremer). These cylindrical shells were designed
as electronic containers for subsurface Lagrangian drift buoys used in worldwide program of ocean
monitoring (Coriolis). Analytical models are used to design the containers. These models were
previously validated by experimental tests on structures and finite element modeling in the framework
of linearized elastic buckling, and were supposed to guarantee a conservative design.
After analysis, the unexpected collapse was attributed to the presence of initial geometric imper-
fections, but also non-negligible residual stresses which were not taken into account in the models,
induced during the manufacturing process (aluminum T6 instead of T6511 grade), and then to an
overestimation of the capacity of the container to withstand the pressure. Indeed, a new grade of
aluminum has been selected for a new application which can only be provided in T6 grade, so that
residual stresses are still present in the material and their effect on the capability of the structures to
withstand the pressure must be evaluated.
The following section aims at estimating the influence of such residual stresses, in addition to
plasticity and geometric imperfections, on the collapse behavior of one cylindrical container, using
a complete non-linear buckling and post-buckling analysis based on the formulation presented above.
5.2. Geometric and material data
Let us consider a cylindrical tube defined in the reference configuration by the length l, the average
radius R and the thickness h R. The shell is made of an elastoplastic material and a linear isotropic
hardening is assumed (what can properly stand for the real stress–strain curve). The numerical values
are listed in Tables 1 and 2.
This cylindrical tube is submitted to a uniform hydrostatic external pressure P, and axial
compression due to the head and bottom of the tube, whose geometry will not be modeled. The ends of
the cylinder are clamped for the same reason.
5.3. Collapse experiment
The cylindrical shell has been equipped with strain gauges, as seen in Fig. 1, in order to record the
local deformations of the structure during the pressure loading.
A set of six biaxial strain gauges (axial and circumferential) has been bonded on the structure to
cover the whole external circumference of the central part of the tube. The position of the gauges is
Table 1
Geometric parameters.
Length l Radius R Thickness h

970 mm 53.375 mm 6.75 mm
227
Table 2
Material properties.
Young’s modulus E Poisson’s ratio n Yield stress s0 Isotropic hardening

modulus H
71,000 MPa 0.32 365 MPa 625 MPa
described in Fig. 2. This configuration has been chosen taking into account the expected mode shape of
the structure.
End closures are fixed at the ends of the tube in order to ensure a perfect water tightness and the
structure is tested in a hyperbaric tank. A speed of 1 MPa/min for pressure increase has been retained,
following the loading history depicted in Fig. 3.
The complete experimental set-up is presented in Fig. 4. The circumferential microstrains are
recorded during the test and are reported in Fig. 5. The failure of the structure occurred at P ¼ 366 bar
by unstable buckling, as shown in Fig. 6.
A linear response of the strain gauges is observed until about 300 bar, where divergence between
the different sensors seems to appear. In order to highlight this divergence phenomenon, the following
expression could be proposed as an indicator of the ‘‘ovalization’’ of the initially circular central
circumference:
1
3oval ¼ 3gauge Min 3J1 ; 3J3 ; 3J5 ; 3J7 ; 3J9 ; 3J11 þ Max 3J1 ; 3J3 ; 3J5 ; 3J7 ; 3J9 ; 3J11 (11)
2
Fig. 7 indicates that the gauges placed at 0 and 180 display values of 3oval very close to zero and thus
are located on nodes of ovalization. Gauges J3, J5, J7 and J9, placed respectively at 45 , 90 , 135 and
270 , have a significant 3oval and can be used to reveal the ovalization process. As an example, in Fig. 8
are plotted the ovalization parameters for J3 and J5 during the whole test. These results clearly indicate
the divergence of the microstrains around 200–250 bar, long before the buckling phenomenon.
Therefore, an initial ovality of the shell should be introduced in the geometry, in order to better
describe the collapse of the tube, which probably depends on this initial geometric imperfection.
5.4. Measurement of residual stresses
The main objective of the following section is to better evaluate numerically the collapse load of the
containers, taking into account initial geometric imperfections but also initial stresses. Indeed, the
Fig. 1. Strain gauges.
228
Fig. 2. Circumferential position of the strain gauges.
manufacturing process of such aluminum tubes involves both axial and circumferential non-negligible
residual stresses related to the cycle of temperature submitted to the material during manufacture.
Therefore, accurate measurements of these initial stress fields are needed to complete the model data.
5.4.1. Experimental procedure

The residual stresses are assumed to depend on the radial position only. Axial and circumferential
gauges are bonded on the inner wall of the tube. Then, stresses are progressively released, using chip
removal machining (turning) on the outer wall surface. Machining parameters are principally the
depth of cut (0.5 mm) and the sampling frequency (2 Hz). The response curves (longitudinal and
circumferential strains during the turning process) are presented in Fig. 9. Each zone with an oscillatory
behavior corresponds to a material removal.
Fig. 3. Applied external pressure during the experimental test.
229
Fig. 4. Cylindrical tube at the initial state.
5.4.2. Data reduction

Some specific assumptions must be made in order to derive the internal stresses from the strains
plotted in Fig. 9. First, the plane stress condition is assumed, due to the thinness of the tube. Stresses are
supposed to be piecewise constant (one value for each layer). During deformation, the tube remains
cylindrical: its diameter and radius are changing but there is neither ovalization nor local or global
bending curvature in the tube.
Fig. 5. Circumferential microstrains around the tube vs. applied external pressure.
230
Fig. 6. Collapse of the cylindrical tube.
231
Fig. 7. Ovalization indicator around the circumference at a high pressure level.
Considering all the hypotheses above, the stresses in the tube are simply derived from the corre-
sponding strains using the classical linear elastic Hooke’s law in small strains. Then, the stresses
released on the opposite side are deduced, following the iterative procedure illustrated by Fig. 10 and
explained below.
Once a layer of material is removed, stresses are released that give rise to particular strains on the
opposite wall, where gauges are located. It allows one to calculate the variation of stresses in the
remaining part of the cylinder and the amplitude of the released stresses coming from the layer
removed (see Fig. 10b).
Before a second layer is removed, the internal equilibrium is achieved, ensuring consistent residual
stresses in the thickness (see Fig. 10c).
Removing a second layer gives rise to a new stress release related to the deformation history. With
such a sequential proceeding, the longitudinal and circumferential stresses are iteratively evaluated in
each successive layer, and then smoothed to make easier the numerical implementation (Figs.11 and 12).
Jiao and Zhao [35] also measured the through-thickness variation of residual stresses in circular
tubes manufactured by several methods. They always found similar results, namely compressive
stresses on the inside surface and tensile ones on the outside surface, for both longitudinal and
transverse directions. Moreover, the residual stress patterns obtained experimentally (especially the
small decrease at both internal and external radii) are very likely for such extruded tubes that are
currently straightened afterwards.
At this stage, it can be mentioned that the maximum amplitude of both axial and circumferential
residual stresses is large enough to modify significantly the carrying capacity of the tube. The most
harmful stress values are the compressive ones, which are located on the inside surface and will be
added to the compressive stresses due to the external pressure.
6. Numerical results: validation and analysis
6.1. Numerical calculations
In the following calculations, symmetries are taken into account in such a way that only one-eighth
of the cylinder is meshed, what considerably reduces the computational cost.
A few preliminary mesh convergence tests lead us to use about 200 quadratic elements for the
considered portion of the cylinder and 4 or 7 Gaussian points for the numerical quadrature through the
shell thickness, depending whether the behavior is assumed to be elastic or elastoplastic.
6.1.1. Perfect cylindrical shell

A first calculation is performed, considering a perfectly circular cylindrical shell without any
residual stresses.
232
Fig. 8. Ovalization growth.
In elasticity, a bifurcation point is obtained at the critical pressure Pc ¼ 39.55 MPa. This bifurcation
point is super-critical, namely the pressure increases in the post-buckling range. It corroborates
the little imperfection sensitivity of a cylindrical shell under external pressure in elasticity. Indeed, the
axial compression has very little effect on the buckling load and the global behavior is governed by the
external pressure. As a consequence, no limit load has been found and no unstable collapse is expected.
When plasticity
pffiffiffi is considered, the critical pressure remains identical. Indeed, plasticity may occur at
P0 ¼ 2hs0 = 3R ¼ 53:3 MPa if free ends and thus homogeneous strains are assumed. Actually,
plasticity starts to occur in the middle of the shell on the outer wall, due to the built-in ends, at a lower
pressure level. Nevertheless, without any defect or residual stress, bifurcation occurs in elasticity at the
previous critical pressure Pc, even if plasticity is considered. In the post-buckling range, contrary to
the elastic case, the critical point is followed by a limit point which is hardly above the bifurcation load.
The maximum pressure Pmax ¼ 39.9 MPa defines the shell load carrying capacity and leads to unstable
failure.
6.1.2. Geometric imperfections

Geometric imperfections are very often encountered in such cylindrical shells, due to the
manufacturing process. In many cases, the inner and outer walls of the cylinder are not coaxial, what
leads to heterogeneous strains and stresses in the shell. Moreover, the radius of the cylinder is
known to be no longer constant, even if it is most often difficult to dispose of accurate
Fig. 9. Evolution of the longitudinal and circumferential strains during the turning process.
233
Fig. 10. Iterative procedure for the calculation of residual stresses.
measurements. In that study, geometric imperfections are taken into account in the form of an
ovality of the initially perfect circular cylinder. This imperfection shape represents the prevalent and
most damaging one (as far as buckling is concerned). The shell is supposed to have an elliptic shape
and the defect amplitude is given by the radius difference DR ¼ Rmax Rmin between the two
principal axes of the ellipse.
Next calculations are performed in order to show the influence of such an imperfection on the
buckling and/or limit loads of the structure. Because precise measurements are not available,
the imperfection amplitude is supposed to vary arbitrarily from DR ¼ 0 to 0.4 mm, in accordance with
the dimensional tolerances of such extruded tubes.
Fig. 11. Variation of longitudinal residual stresses in the shell thickness.
234
Fig. 12. Variation of circumferential residual stresses in the shell thickness.
Due to the built-in ends and most of all to this geometric imperfection, no bifurcation point is
obtained, except for the smallest amplitude DR ¼ 0.05 mm. Indeed, the buckling problem singularity is
replaced by a degenerated regular equilibrium curve, just displaying large displacements and possibly
limit points. Interesting results are thus naturally obtained when considering plasticity effects. Fig. 13
depicts the maximum supported pressure vs. the imperfection amplitude. It shows a non-negligible
loss of the load carrying capacity of the shell when the ovality increases (the elastic buckling pressure is
also added for comparison purposes).
Fig. 13. Collapse external pressure of an elastoplastic cylindrical shell vs. geometric imperfection size, with or without residual
stresses.
235
Fig. 14. External pressure vs. maximum radial displacement in the cylindrical tube: elasticity, ovality DR ¼ 0.4 mm.
6.1.3. Residual stresses

Finally, the axial and circumferential residual stress states presented above are included in the
model in order to see the influence of initial stresses on the buckling behavior of a cylindrical shell
under external pressure.
First, purely elastic calculations are performed and allow us to conclude that residual stresses do not
play any role in the elastic buckling of such shells. In the problem in hand, residual stresses will just
increase more or less quickly the spreading plastic zone, due to non-homogeneous strains, and thus
reduce the buckling stress until all the shells are plastified.
Fig. 15. External pressure vs. maximum radial displacement in the cylindrical tube: plasticity, ovality DR ¼ 0.4 mm, residual stresses.
236
Fig. 16. Post-critical deformation shape of the cylindrical tube in the elastic case: maximum radial displacement 32.8 mm, external
pressure 48.1 MPa.
The results obtained in plasticity with residual stresses, for a perfect cylindrical shell or including
the previous geometric imperfections of various amplitudes, are also plotted in Fig. 13. Without any
imperfection, the residual stresses seem to have no particular effect on the maximum pressure. This
influence appears and grows with increasing imperfection amplitudes towards a constant value.
6.2. Validation and analysis
First, the purely elastic and elastoplastic calculations have shown that elasticity and plasticity
yielded totally different post-buckling behaviors in the case of a cylindrical shell under external
pressure. In elasticity, the shell can support loadings above the critical pressure, even if large
displacements are observed on the post-bifurcated equilibrium branch, what fatally leads to failure. On
the contrary, when plasticity occurs in the vicinity of the buckling phenomenon, the bifurcation
Fig. 17. Post-critical deformation shape of the cylindrical tube in the plastic case: maximum radial displacement 50.9 mm, external
pressure 6.19 MPa.
237
pressure is just followed by a maximum pressure on the post-critical curve, what clearly leads to an
unstable collapse of the shell, as already observed by Dyau and Kyriakides [15] and Kyriakides and
Corona [12], among others. Figs. 14 and 15 show the applied external pressure vs. the maximum radial
displacement in the shell in elasticity and plasticity, respectively (the maximum imperfection ampli-
tude DR ¼ 0.4 mm is considered, and residual stresses are included in plasticity).
Furthermore, the advanced deformed shape of the collapsed cylinder is depicted in Figs. 16 and 17,
in the elastic stable and plastic unstable cases, respectively. Despite identical buckling modes in both
cases, the plasticity occurrence, just localized in the middle of the shell, gives rise to the localization of
buckling patterns at the same place, which is in better accordance with experimental results.
In a second stage, the imperfection sensitivity of such a shell has been illustrated, including some
geometric imperfections (ovality of the initially circular shell) of various amplitudes. When the
imperfection reaches the maximum value encountered in practice (only 0.75% of the average radius),
the loss of carrying capacity attains 22.5% of the maximum pressure in the plastic perfect shell.
Finally, residual stresses are considered in the model. The stronger influence of such initial
manufacturing axial and particularly circumferential stresses occurs at the maximum imperfection
amplitude. In the most detrimental case, the limit pressure does not exceed 75% of the elastic buckling
pressure.
All the results show that neither plasticity effects, nor ovality and residual stresses can be neglected
in such a buckling calculation. These three modeling components are really influential when consid-
ered together.
The experimental collapse load (366 bar) has also been plotted in Fig. 13 for validation purposes.
With the combined effects of plasticity, ovality and residual stresses, it is shown that, for the smallest
imperfection amplitude considered (DR ¼ 0.05 mm) which is inevitable in practice, the critical pres-
sure, initially expected at a value of about 40 MPa perfectly coincides with the experimental value.
7. Conclusion
In this work, a shell finite element model has been developed, encompassing finite rotations and
plasticity, and involving appropriate arc-length and branching methods, in order to resolve the
buckling and post-buckling of a cylindrical shell under combined axial and circumferential loading
stresses. As mentioned by Schafer and Pekoz [36], the primary impediment to success is neither the
formulation nor the solution procedure but the lack of knowledge about the initial state, including
residual stresses. Therefore, axial and circumferential residual stresses coming from the shell
manufacturing process were accurately measured and included in the model. Geometric imperfections
were also taken into account, considering the ovality of the initially perfect circular cylindrical shell.
Whereas plasticity and all imperfection types related to the initial state have a rather limited
influence when considered separately, the combination of these several factors has a more significant
effect. All these initial parameters, when carefully measured, enable us to enhance the numerical
predictions of buckling or collapse pressure and thus improve the safety level of such structures.
References
[1] Shanley FR. Inelastic column theory. International Journal of Aeronautical Sciences 1947;14:261–7.
[2] Hill R. A general theory of uniqueness and stability in elastic–plastic solids. Journal of the Mechanics and Physics of Solids
1958;6(3):236–49.
[3] Brush DO, Almroth BO. Buckling of bars, plates and shells. Mc Graw-Hill; 1975.
[4] Bushnell D. Plastic buckling. In: Pressure vessels and piping: design technology – a decade of progress. S.Y. Zamrik and D.
Dietrich; 1982. Book no. G00213, pp. 47–117.
[5] Bushnell D. Computerized buckling analysis of shells. M. Nijhoff; 1985.
[6] Tvergaard V. Plastic buckling of axially compressed circular cylindrical shells. Thin-Walled Structures 1983;1(2):139–63.
[7] Bardi FC, Kyriakides S. Plastic buckling of circular tubes under axial compression – part I: experiments. International
Journal of Mechanical Sciences 2006;48:830–41.
[8] Bardi FC, Kyriakides S, Yun HD. Plastic buckling of circular tubes under axial compression – part II: analysis. International
Journal of Mechanical Sciences 2006;48:842–54.
[9] Singer J. On the importance of shell buckling experiments. Applied Mechanics Review 1999;52(6):17–25.
[10] Mandal P, Calladine CR. Buckling of thin cylindrical shells under axial compression. International Journal of Solids and
Structures 2000;37(33):4509–25.
238
[11] Lancaster ER, Calladine CR, Palmer SC. Paradoxical buckling behavior of a thin cylindrical shell under axial compression.
International Journal of Mechanical Sciences 2000;42:843–65.
[12] Kyriakides S, Corona E. Mechanics of offshore pipelines: buckling and collapse, vol. 1. Elsevier; 2007.
[13] Corona E, Kyriakides S. On the collapse of inelastic tubes under combined bending and pressure. International Journal of
Solids and Structures 1988;24(5):505–35.
[14] Gusic G, Combescure A, Jullien JF. The influence of circumferential thickness variations on the buckling of cylindrical shells
under external pressure. Computers and Structures 2000;74:461–77.
[15] Dyau JY, Kyriakides S. On the localization of collapse in cylindrical shells under external pressure. International Journal of
[16] Yeh MK, Kyriakides S. On the collapse of inelastic thick-walled tubes under external pressure. Journal of Energy Resources
Technology 1986;108:35–47.
[17] Yeh MK, Kyriakides S. Collapse of deepwater pipelines. Journal of Energy Resources Technology 1988;110:1–11.
[18] Triantafyllidis N. On the bifurcation and post-bifurcation analysis of elastic–plastic solids under general pre-bifurcation
conditions. Journal of the Mechanics and Physics of Solids 1983;31(6):499–510.
[19] Corona E, Kyriakides S. Asymmetric collapse modes of pipes under combined bending and external pressure. Journal of
Engineering Mechanics 2000;126:1232–9.
[20] Timoshenko SP, Gere JM. Theory of elastic stability. Mc Graw-Hill; 1961.
[21] Dubey RN. Instabilities in thin elastic–plastic tubes. International Journal of Solids and Structures 1969;5(7):699–711.
[22] Hutchinson JW. Plastic buckling. Advances in Applied Mechanics 1974;14:67–144.
[23] Kheyrkhahan M, Peek R. Post-buckling analysis and imperfection sensitivity of general shells by the finite element
method. International Journal of Solids and Structures 1999;36(18):2641–81.
[24] Le Grognec P, Le van A. Elastoplastic bifurcation and collapse of axially loaded cylindrical shells. International Journal of
[25] Halphen B, Nguyen QS. Sur les matériaux standard généralisés. [On the generalized standard materials]. Journal de
Mécanique 1975;14(1):39–63.
[26] Green AE, Naghdi PM. A general theory of an elastic–plastic continuum. Archive for Rational Mechanics and Analysis 1965;
18:251–81.
[27] Simo JC, Hughes TJR. Computational inelasticity. Springer; 1998.
[28] Riks E. An incremental approach to the solution of snapping and buckling problems. International Journal of Solids and
Structures 1979;15(7):529–51.
[29] Crisfield MA. Non-linear finite element analysis of solids and structures: essentials, vol. 1. John Wiley & Sons; 1991.
[30] Lam WF, Morley CT. Arc-length method for passing limit points in structural calculation. Journal of Structural Engineering
1992;118(1):169–85.
[31] Riks E. On formulations on path-following techniques for structural stability analysis. In: New advances in computational
structural mechanics. Elsevier; 1991, pp. 65–80.
[32] Seydel R. Practical bifurcation and stability analysis: from equilibrium to chaos. Springer-Verlag; 1994.
[33] Bergan PG, Horrigmoe G, Krakeland B, Soreide B. Solution techniques for non-linear finite element problem. International
Journal for Numerical Methods in Engineering 1978;12:1677–96.
[34] Wagner W, Wriggers P. A simple method for the calculation of post-critical branches. Engineering and Computations 1988;
5:103–9.
[35] Jiao H, Zhao XL. Imperfection, residual stress and yield slenderness limit of very high strength (VHS) circular steel tubes.
Journal of Constructional Steel Research 2003;59:233–49.
[36] Schafer BW, Pekoz T. Computational modeling of cold-formed steel: characterizing geometric imperfections and residual
stresses. Journal of Constructional Steel Research 1998;47:193–210.
239
International Journal of Solids and Structures 47 (2010) 34–50
International Journal of Solids and Structures

journal homepage: www.elsevier.com/locate/ijsolstr
GBT-based semi-analytical solutions for the plastic bifurcation of thin-walled

members
Rodrigo Gonçalves a, Philippe Le Grognec b,*, Dinar Camotim c
a
Departamento de Engenharia Civil, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, 2829-516 Caparica, Portugal
b
Polymers and Composites Technology and Mechanical Engineering Department, Ecole des Mines de Douai, 941 rue Charles Bourseul BP 10838, 59508 Douai Cedex, France
c
Department of Civil Engineering and Architecture, ICIST/IST, Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
Article history: This paper presents the development and illustrates the application of semi-analytical solutions for plas-
Received 18 July 2008 tic buckling (bifurcation) problems involving perfectly straight and uniformly compressed thin-walled
Received in revised form 5 September 2009 metal members. These solutions are derived on the basis of a non-linear Generalised Beam Theory
Available online 18 September 2009
(GBT) formulation, recently proposed by two of the authors, that resorts to a linearised buckling analysis
that adopts Hill’s hypo-elastic comparison solid method to obtain the plastic bifurcation loads and asso-
Keywords: ciated buckling mode shapes. Moreover, both J2 small-strain incremental and deformation plasticity the-
Thin-walled members
ories are employed. Several numerical illustrative examples are presented and discussed throughout the
Plastic bifurcation
Generalised Beam Theory (GBT)
paper and closed-form analytical formulae are also derived. The accuracy of the GBT-based semi-analyt-
Local-plate buckling ical plastic bifurcation predictions is assessed through the comparison with (i) available theoretical solu-
Distortional buckling tions and/or (ii) results yielded by a special shell finite element model developed by one of the authors.
Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction Hutchinson (1974)). This approach is commonly viewed as the

generalisation of the tangent modulus theory developed by Enges-
Plastic buckling (bifurcation) remains a domain in which ana- ser (1889) for uniformly compressed columns, later validated by
lytical results are rather scarce, particularly in comparison with Shanley (1946, 1947).
the situation concerning the buckling behaviour of elastic mem- It is well-known that bifurcation loads determined on the basis
bers and structures (Nguyen, 2000). The incremental and irre- of the incremental theory of plasticity overestimate, sometimes
versible nature of the elastic–plastic constitutive equations considerably, the experimental results, which in turn exhibit a
makes it generally necessary to resort to numerical methods in better agreement with those provided by the deformation theory
order to detect (exactly or approximately) bifurcations and fol- of plasticity (e.g., Hutchinson, 1974; Bazant and Cedolin, 1991 or
low up the corresponding bifurcated branches (e.g., Bazant and Nguyen, 2000). In this context, the experimental results reported
Cedolin, 1991; Cimetière and Léger, 1996 or Le Grognec and Le by Gerard and Becker (1957), concerning uniformly compressed
van, 2008). Concerning this matter, Hill (1956, 1958) introduced simply supported aluminium cruciform columns, play a historical
the ‘‘hypo-elastic comparison solid method”,1 which was subse- role. The paradox involving the plastic bifurcation analyses based
quently generalised by Raniecki for non-associated flow rules on deformation and incremental theories of plasticity attracted
(Raniecki, 1979; Raniecki and Bruhns, 1981 or Bigoni, 2000). The the attention of several researchers (e.g., Drucker, 1949; Cicala,
bifurcation of the comparison solid constitutes a necessary condi- 1950; Bijlaard, 1950; Onat and Drucker, 1953; Hutchinson,
tion for the bifurcation of the real solid, which becomes also a suf- 1974; Hutchinson and Budiansky, 1976; Needleman and Tverg-
ficient condition whenever the whole plastic zone is active in the aard, 1982 or Gjelsvik and Lin, 1987), with the discussion mostly
fundamental solution (the so-called ‘‘total loading” condition – see focused on the influence of the initial geometrical imperfections
and the possible development of corners on the yield surface.
However, today it seems fair to say that most researchers agree
* Corresponding author. Tel.: +33 (0) 3 27 71 23 21; fax: +33 (0) 3 27 71 29 81. that the differences between the experimental and incremental
E-mail addresses: rodrigo.goncalves@fct.unl.pt (R. Gonçalves), legrognec@ theory results stem mainly from a high sensitivity to initial geo-
ensm-douai.fr (P. Le Grognec), dcamotim@civil.ist.utl.pt (D. Camotim). metrical imperfections (e.g., Brochard and Combescure, 1987;
1
In this method, the ‘‘real” solid is replaced by a so-called ‘‘hypo-elastic
Tuǧcu, 1998; Chakrabarty, 2002 or Nguyen, 2000). In fact, defor-
comparison solid”, whose tangent constitutive equations differ from the ‘‘real” solid
ones in the fact that no elastic unloading from the yield surface is allowed. mation theory provides results that are in better agreement with
0020-7683/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijsolstr.2009.09.013
240
R. Gonçalves et al. / International Journal of Solids and Structures 47 (2010) 34–50 35
the experimental ones simply because its ‘‘total” nature works S ¼ Ce Ee ¼ Ce ðE Ep Þ; ð2Þ
like a vertex in the yield surface (Needleman and Tvergaard,
where Ce is the 4th order elastic constitutive tensor. One now intro-
1982).
duces an alternative notation, according to which the previous
Generalised Beam Theory (GBT), which was originally devel-
equation is rewritten using ‘‘vectorial forms” of 2nd order tensors,
oped by Schardt (e.g., 1966, 1989, 1994a), may be viewed as an
identified by the subscript ‘‘v”. Therefore, (2) becomes
extension of Vlasov’s prismatic thin-walled bar theory that takes
into account cross-section distortion and wall transverse bending. S v ¼ C e Eev ¼ C e ðEv Epv Þ ð3Þ
Due to its unique modal decomposition features, GBT has been
in the 6-dimensional space
established as an extremely versatile and computationally efficient
approach to analyse the local and global structural response of ½Ev t ¼ ½ Exx Eyy Ezz 2Exy 2Exz 2Eyz ð4Þ
thin-walled members, particularly when compared with the more t
½S v ¼ ½ Sxx Syy Szz Sxy Sxz Syz ð5Þ
traditional finite strip (FS) and shell finite element (SFE) methods
(e.g., Camotim et al., 2004, 2006). and an orthogonal coordinate system (x, y, z) is adopted. In the case
Two of the authors have recently developed and implemented of a St. Venant–Kirchhoff material law, one has
numerically a GBT formulation that is computationally very effi- 2 3
cient and makes it possible to calculate plastic bifurcation loads C exxxx C exxyy C exxzz 0 0 0
6 7
in thin-walled members (Gonçalves and Camotim, 2004, 2007), 6 C eyyyy C eyyzz 0 0 0 7
6 7
through FEM-based linear stability analyses, i.e., buckling analyses 6
6 C ezzzz 0 0 0 7
7
that neglect the geometrically non-linear effects associated with ½C e ¼ 6 7 ð6Þ
6
6 C exyxy 0 0 7
7
the pre-buckling displacements. However, due to (i) the extreme 7
6
scarcity of analytical solutions for plastic buckling problems 4 C exzxz 0 5
involving thin-walled members and (ii) the fact that the currently Sym: C eyzyz
available SFE or FS commercial software packages do not offer the
C exxxx ¼ C eyyyy ¼ C ezzzz ¼ k þ 2l
possibility of calculating plastic bifurcation loads, it was not possi-
ble to validate and assess the accuracy of most GBT-based numer- C exxyy ¼ C exxzz ¼ C eyyzz ¼k ð7Þ
ical results. This situation was recently altered through the C exyxy ¼ C exzxz ¼ C eyzyz ¼l
cooperation with the remaining author, who developed and imple-
mented a SFE model capable of (i) handling finite rotations, (ii) fol- where k and l are the Lamé constants, related to E, G and m (Young’s
lowing non-linear equilibrium paths (also non-monotonic ones) modulus, shear modulus and Poisson’s ratio) by
with specific arc-length methods, (iii) detecting the occurrence of Em E
elastic–plastic bifurcations and (iv) examining the corresponding k¼ l¼G¼ : ð8Þ
ð1 þ mÞð1 2mÞ 2ð1 þ mÞ
bifurcated equilibrium paths by means of branch-switching tech-
niques (Le Grognec, 2001). The first (preliminary) results stemming According to J 2 (von Mises) theory with isotropic strain harden-
from this cooperation were already published (Gonçalves et al., ing, the yield function f is given by
2006). f ðJ 2 ; ep Þ ¼ re ðJ 2 Þ ry ðep Þ ð9Þ
On the other hand, GBT offers remarkable capabilities concern-
ing the development of analytical or semi-analytical formulae for where the effective stress re depends on J2 (second invariant of the
a wide range of problems. Such capabilities have been already ex- deviatoric stresses) and the yield stress ry is a function of the equiv-
plored in the context of the elastic bifurcation of thin-walled alent plastic strain ep ,
members (e.g., Schardt, 1989, 1994b or Silvestre and Camotim, pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi
2004a,b,c). The aim of this paper is to do the same for plastic
re ðJ2 Þ ¼ 3J 2 ¼ 3=2kdevðSÞk ð10Þ
Z t pffiffiffiffiffiffiffiffi
bifurcation problems, namely by deriving GBT-based semi-analyt- ep ¼ 2=3kdEp k: ð11Þ
ical buckling formulae for uniformly compressed and perfectly 0
straight thin-walled members. The plastic bifurcation loads and
For an associated flow rule, the plastic strain evolves perpendic-
associated buckling mode shapes are obtained from linear stabil-
ularly to the yield surface, i.e.,
ity analyses (the evolution and stability of the bifurcated equilib-
rium paths is not addressed) based on both J 2 small-strain dEp ¼ dcN; ð12Þ
incremental and deformation theories of plasticity. In this paper,
where dc ¼ dep is the plastic multiplier and the normal N is given by
besides presenting and discussing new numerical results, one also
the expression
revisits published ones, either because semi-analytical solutions
have now been derived or to show comparisons with Le Grognec’s df pffiffiffiffiffiffiffiffi devðSÞ
N¼ ¼ 3=2 : ð13Þ
SFE results. dS kdevðSÞk
The consistency condition df = 0 yields
2. Elastic–plastic constitutive equations
N v C e dEv
dc ¼ ð14Þ
2.1. Incremental theory N v C e N v þ H0
where
Since this work is restricted to small strains, the Green–La-
grange strain tensor can be decomposed into an elastic and a plas- ½N v t ¼ ½ Nxx Nyy Nzz 2Nxy 2N xz 2Nyz ð15Þ
tic parts, i.e., one has
and the hardening modulus is a function of the uniaxial tangent
E ¼ E e þ Ep : ð1Þ modulus Et ,
dry Et
Assuming a St. Venant–Kirchhoff material law, the elastic H0 ¼ ¼ : ð16Þ
dep 1 Et =E
strains Ee are related to their work-conjugate second Piola–Kirch-
hoff stresses S by Using the vectorial forms, the differentiation of (3) leads to
241
36 R. Gonçalves et al. / International Journal of Solids and Structures 47 (2010) 34–50
dS v ¼ C e ðdEv dcN v Þ ¼ C t dEv ; ð17Þ dS v ¼ C t dEv ð30Þ

!
with the tangent elastic–plastic constitutive operator given by Nv Nv C
Ct ¼ C 1 : ð31Þ
N v CN v þ H0
Nv Nv Ce
Ct ¼ Ce 1 0 : ð18Þ
Nv CeNv þ H The tangent constitutive operators (31) and (18) differ in the
presence of either C or C e . Taking into account (24), one concludes
that they coincide ðC ¼ C e Þ whenever there is (i) no variation of the
2.2. Deformation theory normal ðdN ¼ 0Þ or (ii) a null equivalent plastic strain ð ep ¼ 0Þ,
which implies that Es ¼ E.
According to J 2 deformation theory (Hencky, 1924), the incre- It is important to emphasise that, unlike incremental theory,
mental relation (12) is replaced by one involving the total plastic deformation theory does not take into consideration the true
deformation ep , namely deformation history. In fact, (19) and (20) imply that both the
equivalent plastic strain and direction of plastic deformation de-
Ep ¼ ep N; ð19Þ
pend exclusively on the ‘‘point of arrival”, regardless of the path
where the normal N is again given by (13) and ep , obtained from an followed to reach it. It is precisely due to this feature that the term
uniaxial stress–strain curve, is given by associated with the variation of the normal ðdNÞ appears in the
incremental relation (24), thus leading to different C and C e .

1 1 re Finally, it is worth finding under which conditions the two
ep ¼ e ee ¼ re ¼ ; ð20Þ
Es E H0s above tangent constitutive operators coincide with the elastic
one (i.e., C t ¼ C e Þ. In the incremental theory, this occurs if and only
where Es ¼ r=e is the uniaxial secant modulus and H0s is the secant if N v C e dEv ¼ 0, i.e., the elastic prediction C e dEv is tangent to the
hardening modulus, provided by the expression yield surface. In the deformation theory, one must have
N v C dEv ¼ 0, i.e., the ‘‘pseudo elastic prediction” C dEv tangent
Es
H0s ¼ : ð21Þ to the yield surface. The two conditions obviously coincide if
1 Es =E C ¼ Ce.
Incorporating (20), (21) and (10) into (19), one obtains Next, one obtains the tangent constitutive operator components
for the particular case of initial uniaxial stress states.
3
Ep ¼ devðSÞ ð22Þ
2H0s
and the ‘‘total” constitutive relation becomes 2.3. Uniaxial initial stress states

3 For uniaxial initial stress states along the fundamental path (say
S ¼ Ce ðE Ep Þ ¼ Ce ðE ep NÞ ¼ Ce E 0 devðSÞ : ð23Þ
2Hs Sxx – 0), one has
In order to obtain the corresponding incremental relation, one

must differentiate this equation, which leads to (adopting the vec- ½S v t ¼ ½ Sxx 0 0 0 0 0 ð32Þ
torial forms) 1
½N v t ¼ ½ 2 1 1 0 0 0 : ð33Þ
2
dS v ¼ C e ðdEv dep N v ep dN v Þ: ð24Þ
The following notation is adopted for the tangent constitutive
Taking into account that operator components:

3 Nv 3 dS v
dN v ¼ PdS v ðN v dS v Þ ¼ P Nv Nv ; ð25Þ 2 3
2re re 2 re C xx
t C xy
t C xz
t 0 0 0
6 C yy C yz 0 0 07
6 t t 7
where 6 7
6 C zz 0 0 07
2 3 ½C t ¼ 6
6
t 7: ð34Þ
2=3 1=3 1=3 0 0 0 6 Gt 0 07
7
6 1=3 2=3 1=3 0 0 6 7
6 07
7 4 Gt 05
6 7
6 1=3 1=3 2=3 0 0 07 Sym: Gt
½P ¼ 6
6 0
7; ð26Þ
6 0 0 2 0 07
7
6 7 For the incremental and deformation theories, applying (18)
4 0 0 0 0 2 05 and (31) leads respectively to
0 0 0 0 0 2
the above incremental relation may be simplified to 3kl þ 2l2 þ ð2l þ kÞH0
C xx
t ¼
3l þ H0
dS v ¼ CðdEv dep N v Þ ð27Þ 3kl þ 5l2 þ ð2l þ kÞH0
1 C yy zz
t ¼ Ct ¼
3 1 3l þ H 0
C ¼ 1 þ Ce P Nv Nv Ce: ð28Þ
2 H0s 3kl þ 2l2 þ kH0 ð35Þ
C xy xz
t ¼ Ct ¼
As for the incremental theory, the consistency condition leads 3l þ H 0
again to the derivation of an expression for dep , given by 3kl l þ kH0
2
C yz
t ¼
3l þ H0
N v C dEv Gt ¼ G
dep ¼ ; ð29Þ
N v CN v þ H0
which, upon substitution in (27), yields and
242
3kl þ 2l2 þ ð2l þ kÞH0 E2 þ 3E2 H0 =Es þ EH0

C xx
t ¼ C xx
t ¼
3l þ H 0 ð3E=Es þ 2 4mÞE ð4m2 3E=Es 1ÞH0
lH0s ð2l þ H0 Þ 3l3 4E2 þ 4EH0
C yy zz
t ¼ Ct ¼ k þ l þ
C yy
t ¼
ð3l þ H0 Þð3l þ H0s Þ ð3E=Es þ 2 4mÞE ð4m2 3E=Es 1ÞH0
ð41Þ
3kl þ 2l2 þ kH0 2E2 þ 4mEH0
C xy ¼ C xz ð36Þ C xy
¼ ;
t ¼ : ð3E=Es þ 2 4mÞE ð4m2 3E=Es 1ÞH0
t
t
3l þ H 0
E
lH0s ð4l þ H0 Þ þ 3l3 Gt ¼
2m 1 þ 3E=Es
C yz
t ¼ kþl
ð3l þ H0 Þð3l þ H0s Þ
for the deformation theory. In the (further reduced) 2-dimensional
E space
Gt ¼
2m 1 þ 3E=Es
½S v t ¼ ½ Sxx Sxy ð42Þ
Future developments include the investigation of situations in
which some incremental stress ðdSÞ components are null. For an ½Ev t ¼ ½ Exx 2Exy ð43Þ
xx
incremental plane stress (xy plane), the constraint equations Ct 0
½C t ¼ ð44Þ
dSzz ¼ dSxz ¼ dSyz ¼ 0 make it possible to work in the reduced 3- 0 Gt
dimensional space
corresponding to the additional assumption dSyy ¼ 0, one obtains
½S v t ¼ ½ Sxx Syy Sxy ð37Þ
C xx
t ¼ Et Gt ¼ G; ð45Þ
½Ev t ¼ ½ Exx Eyy 2Exy ð38Þ
2 xx 3 for the incremental theory, and
Ct C xy
t 0
6 7 E
½C t ¼ 6 C xx
t ¼ Et Gt ¼ ð46Þ
4 C yy
t 07 5: ð39Þ 2m 1 þ 3E=Es
Sym: Gt for the deformation theory.
One then has In order to provide a better quantification of the differences be-
tween the two plasticity theories, Fig. 1 shows how the compo-
E2 þ 4EH0 nents of the plane stress tangent constitutive operators (40)
C xx
t ¼
ð5 4mÞE ðm2 1Þ4H0 (incremental theory) and (41) (deformation theory), normalised
with respect to the corresponding elastic values, vary with the total
4E2 þ 4EH0 uniaxial strain e, for an uniaxial constitutive relation of the Ram-
C yy
t ¼
ð5 4mÞE ðm2 1Þ4H0 berg–Osgood type (Ramberg and Osgood, 1943)
a
2E2 þ 4mEH0 r r
C xy
t ¼ ; ð40Þ e ¼ þ 0:002 ð47Þ
ð5 4mÞE ðm2 1Þ4H0 E r0:2
Gt ¼ G with E = 70 GPa, r0:2 ¼ 200 MPa; m ¼ 0:3 and a ¼ 5, values corre-
sponding to a typical aluminium alloy (e.g., Mazzolani and Valtinat,
for the incremental theory, and
1992). These results clearly indicate that, with the exception of C xx
t ,
(i) there are significant differences between the two theories and
Fig. 1. Variation of the normalised values of the components of the constitutive operators (40) (incremental theory) and (41) (deformation theory) with the total uniaxial
strain e.
243
(ii) deformation theory predicts a much more ‘‘flexible” material ‘‘member analysis” (first-order, buckling, vibration, etc.), yielding
behaviour. the solution of the appropriate differential equilibrium equations.
Complete and in-depth descriptions of the concepts and proce-
3. Bifurcation equation – Linearised buckling analysis dures engaged in each task can be found in the literature (e.g.,
Schardt, 1989; Camotim et al., 2004). The purpose of this subsec-
At a bifurcation point, at least two solutions satisfy the incre- tion is to provide a brief account of the GBT kinematic description,
mental virtual work equation whereas the following subsection addresses the semi-analytical
plastic bifurcation solutions (the ‘‘member analysis” output).
DdW int ð/Þ½D/ð1Þ þ DdW ext ð/;kÞ½D/ð1Þ þ DdW ext ð/;kÞ½Dk ¼ 0 ð48Þ Consider the arbitrary thin-walled prismatic member depicted
DdW int ð/Þ½D/ð2Þ þ DdW ext ð/;kÞ½D/ð2Þ þ DdW ext ð/;kÞ½Dk ¼ 0; ð49Þ in Fig. 2, which shows also the mid-surface local axes (x, y, z).
According to the standard GBT approach, the wall displacement
where D ½a is the standard directional derivative along direction a components, measured along the local axes, are obtained by means
(e.g., Bonet and Wood, 1997), k is the loading parameter and / of Kirchhoff’s plate theory
groups all kinematic parameters required to define the solid config-
uration (a total Lagrangian description is assumed). Subtracting the U x ¼ u zw;x U y ¼ v zw;y U z ¼ w; ð59Þ
two equations, one is led to the eigenvalue problem defined by
where u, v and w are the wall mid-surface displacement compo-
DdWð/; kÞ½D/ ¼ 0 ð50Þ nents along x, y and z. These displacement components are ex-
pressed as
where D/ ¼ D/ð1Þ D/ð2Þ . For bifurcations under neutral equilib-
rium, one has Dk ¼ 0 for both solutions and, thus, the eigenvalue ðyÞ /;x ðxÞ;
uðx; yÞ ¼ u v ðx; yÞ ¼ v ðyÞ /ðxÞ;
problem may be directly obtained from (48) or (49).

wðx; yÞ ¼ wðyÞ /ðxÞ; ð60Þ
In linearised buckling analyses (also termed linear stability analy-
ses), the fundamental path is determined neglecting all geometrical where u ; v
and w are N-dimensional vectors, containing the com-
non-linear effects. The bifurcation problem is then ‘‘reduced” to the ponents of the N cross-section deformation modes (obtained from
simpler initial stress problem (i.e., / ¼ 0 but S – 0) the ‘‘cross-section analysis”), and /ðxÞ is the N-dimensional vector
DdWð0; kÞ½D/ ¼ 0: ð51Þ of the associated modal amplitude functions /ð1Þ ðxÞ /ðNÞ ðxÞ, which
constitute the problem unknowns.
In the following, it is assumed that the displacement vector U is The small-strain tensor e follows directly from (59), (60). In
a linear function of the parameters contained in / (which is true in component form, separating membrane and bending terms, one
GBT analyses – see Section 4). Then, DdU ¼ 0 and DdD ¼ 0, where D has
is the displacement derivative tensor, and one has
2 3 2 3 2 3
DdW ext ¼ 0: ð52Þ
exx eMxx eBxx
6 e 7 6 eM 7 6 eB 7
½ev ¼ 4 yy 5 ¼ 4 yy 5 þ 4 yy 5 ¼ ½eM B
v þ ½ev ð61Þ
Using the second Piola–Kirchhoff stress tensor S and its work-
cxy cMxy cBxy
conjugate Green–Lagrange strain tensor E, the directional deriva- 2 3 2 3
tive of the internal virtual work reads /;xx
u /;xx
w
Z 6 v ;y / 7 6 / 7
½eM
v¼4 5; ½eBv ¼ z4 w ;yy 5: ð62Þ
DdW int ½D/ ¼ ðdE : Ct DE þ S : DdEÞ dV 0 ; ð53Þ ;y þ v
ðu Þ /;x ;y /;x
2w
V0
where V 0 is the initial volume of the solid and Ct is the tangent con- According to the standard GBT procedure, followed in this work,
stitutive operator of the associated hypo-elastic comparison solid. null membrane transverse extensions and shear strains are en-
Since this directional derivative is calculated in the initial configu- forced on the deformation modes, by means of the constraint
ration ð/ ¼ 0Þ, one has equations
dE ¼ de ð54Þ
DE ¼ De ð55Þ
t
DdE ¼ symðDdD þ DD dDÞ; ð56Þ
t
where e ¼ 1=2ðD þ D Þ is the small-strain tensor. Then, recalling
that DdD ¼ 0, one may write the bifurcation equation as
Z
DdWð0; kÞ½D/ ¼ ðde : Ct De þ S : DDt dDÞ dV 0 ¼ 0: ð57Þ
V0
Finally, for initial uniaxial stress states ðSxx – 0Þ and using vec-
torial forms, the bifurcation equation may be further simplified to
Z
ðdev C t Dev þ Sxx ½DDt dDxx Þ dV 0 ¼ 0: ð58Þ
V0
4. GBT-based semi-analytical solutions
4.1. Overview of the GBT kinematic description
The performance of a GBT analysis involves two main tasks,

namely (i) a ‘‘cross-section analysis”, which leads to the identifica-
tion and characterisation of its deformation modes, and (ii) a Fig. 2. Arbitrary thin-walled member geometry and local coordinate systems.
244
Z
eMyy ¼ 0 ) v;y ¼ 0 ð63Þ C yy 3
t t
B¼ ;yy dy
w;yy w
cMxy ¼ 0 ) u ;y ¼ v ; ð64Þ C 12
C ¼ Cr þ Cs
which means that the longitudinal extension eM xx is the only non-null Z Z xx 3
Ct t
membrane strain term. Cr ¼ Et t u dy C s ¼
u w
w dy
The calculation of the deformation modes (i.e., the cross-section C C 12
t
analysis, not addressed in this paper) takes into account the previ- D ¼ Dr Ds Ds
ous constraint equations and leads to an hierarchical set of modes, Z Z xy 3
Gt t3 Ct t
where (i) the first four are the classic prismatic bar theory modes Dr ¼ w ;y dy Ds ¼
;y w w
w ;yy dy
C 3 C 12
(extension, major and minor axis bending and torsion about the Z Z
shear centre) and (ii) the remaining ones involve cross-section dis- Xs ¼ tSM
xx u u dy Xr ¼ tSM
xx ðv v þ w wÞ dy;
ð73Þ
tortion and local-plate bending (see Figs. 6 and 10, which concern C C
the illustrative examples). with t being the wall thickness and C the cross-section mid-line.
The bifurcation equation (70) and associated boundary condi-
4.2. Semi-analytical solutions for plastic uniformly compressed tions may be employed to obtain semi-analytical solutions for arbi-
members trary boundary conditions (see Section 5.2). However, far-reaching
conclusions can be drawn if these equations are specialised for
For an elastic–plastic material behaviour, the incremental stres- members simply supported in the longitudinal direction (end sec-
ses are calculated from the incremental strains using the tangent tions pinned locally and globally, and free to warp). In fact, semi-
constitutive operators derived in Section 2.3. For the (only non- analytical solutions cannot be reached for most boundary condi-
null) membrane term, one has tions, which must then be handled by means of a beam finite ele-
ment implementation, such as the one developed by Gonçalves and
dSxx ¼ Et deM
xx ð65Þ
Camotim (2007). Moreover, it should be pointed out that (i) simply
and, following the conventional GBT approach, a plane stress state supported members constitute the most typical and widely used
is assumed for the bending terms, thus yielding benchmark problems in the buckling literature and (ii) the local-
2 B
3 2 B 3 plate and/or distortional buckling behaviour of sufficiently long
dSxx dexx members is virtually unaffected by the end support conditions,
6 B 7
6 dSyy 7 ¼ ½C 6B
4 d eByy 7
5 ð66Þ which means that the semi-analytical solutions obtained for sim-
4 5 t
B dcxyB ply supported members also apply to ‘‘long enough” members with
dSxy
2 xx 3 other support conditions.
Ct C xy
t 0 For members simply supported in the longitudinal direction,
6 7
½C Bt ¼ 4 C yy
t 0 5; ð67Þ sinusoidal modal amplitude functions provide exact solutions
Sym: Gt 2 ð1Þ 3

/
mpx6 7
where the tangent constitutive matrix components are given in (40) ½/ ¼ sin 6 .. 7 ð74Þ
(incremental theory) and (41) (deformation theory). L 4 . 5
ðNÞ
/
In uniformly compressed and perfectly straight thin-walled
members, the membrane longitudinal normal stresses SM xx are the where m is the number of longitudinal half-waves of the buckling
only non-null stress components along the fundamental path, con- ðNÞ represent the contributions of the
ð1Þ /
mode and the scalars /
stant within the whole member volume at a given load level. In
corresponding deformation mode to the member buckling mode.
this particular case, the fundamental solution satisfies the total
Since the number of degrees of freedom is equal to the number of
loading condition and the analysis of the comparison solid pro-
deformation modes included in the analysis (N), this formulation
vides true plastic bifurcation loads (Hutchinson, 1974). Replacing
leads to an obvious computational economy with respect to SFE
SM
xx by the bifurcation stress rb and noting that the displacement analyses.
vector U is a linear function of /, the bifurcation equation (58)
Incorporating (74) into (70), the eigenvalue problem becomes
becomes !
Z m2 p2 L2
B C þ D þ 2 2 B þ rb X / ¼ 0; ð75Þ
Et deM M B B
xx Dexx þ dev C t Dev dV 0 L2 m p
V0
Z
þ rb ½ðDDM Þt dDM xx dV 0 ¼ 0: ð68Þ where the new modal matrix X is given by
V0 Z
m2 p2
In order to proceed, it is necessary to calculate ½DD dDxx . After t
X¼ t v v þw w þ uu dy
C L2
some algebraic manipulation, one obtains Z
m2 p2 r
½DDt dDxx ¼ d/;xx ðu ÞD/;xx þ d/;x ðv
u v wÞ
þw D/;x : ð69Þ ¼ tðv
v þw wÞ
dy þ C : ð76Þ
C Et L 2
Then, the use of the fundamental lemma of the calculus of vari- The last term in this expression reflects the influence of the lon-
ations leads to the bifurcation differential equation system and gitudinal displacements and is routinely discarded (e.g., Prola, 2001
associated boundary conditions given by or Silvestre, 2005) – this term will also be omitted in this work. The
bifurcation stresses may then be determined from the characteris-
ðC þ X s Þ/;xxxx ðD þ X r Þ/;xx þ B/ ¼ 0 ð70Þ
tic equation
s
d/;x ððC þ X Þ/;xx þ D s
/ÞjL0 ¼0 ð71Þ !
s s
d/ ððC þ X Þ/;xxx þ ðD D X r r
Þ/;x ÞjL0 ¼ 0; ð72Þ m2 p2 L2
det C þ D þ 2 2 B þ rb X ¼0 ð77Þ
L2 m p
where L is the member length, the prefix D of / was dropped to alle-
viate the notation and the GBT modal matrices (B, C, D and X) read and the corresponding buckling modes from (75).
245
2 3
The bifurcation stress associated with buckling in a single defor- sinðpy=bÞ
mation mode (say mode kÞ can be readily obtained by means of the 6 .. 7
¼ ½v
½u ¼ ½0; ¼6
½w 4 .
7;
5 ð89Þ
expressions
! sinðN py=bÞ
ðkÞ 1 m2 p2 L2 thus yielding the GBT modal matrices (1 is the N N identity
r ¼ ½Ckk þ ½Dkk þ 2 2 ½Bkk ð78Þ
b
½Xkk L 2 m p matrix)
mpx 3
/ðkÞ ¼ sin ; ð79Þ C xx
t bt
L Cr ¼ 0 Cs ¼ 1
24
G t p2 t 3 C xy p2 t 3
which constitute the generalisation, for thin-walled members made Dr ¼ diagf12 ; . . . ; N2 g Ds ¼ t diagf12 ; . . . ; N2 g:
of elastic–plastic materials, of expressions derived earlier by 6b 24b
C yy p4 t 3 bt
Schardt (1994a,b), in the context of linear elastic members. B ¼ t 3 diagf14 ; . . . ; N4 g X ¼ 1 ð90Þ
Although (78) is non-linear (the tangent constitutive operator com- 24b 2
ponents appearing in matrices C, D and B also depend on the ap- Since all these matrices are diagonal, no mode coupling occurs
plied stress), note that while the first term decreases with L, the and the critical buckling mode always exhibits a single half-wave
last one (associated with transverse plate bending) increases with along y (i.e., mode k = 1). From (78), the bifurcation stress is then
L. As recognized by Schardt, the minimum bifurcation stress occurs given by
for !
p2 b2 m2 L2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rb ¼ 2
K xx þ 2K xy þ K yy ; ð91Þ
Lmin ¼ mp 4 ½Ckk =½Bkk ð80Þ tb L2 2
b m2
where
and corresponds to
C xx
t t
3
C yy
t t
3
t 3 xy
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi K xx ¼ ; K yy ¼ ; K xy ¼ ðC þ 2Gt Þ; ð92Þ
ðkÞ 1 12 12 12 t
r b;min ¼ ð½Dkk þ 2 ½Bkk ½Ckk Þ: ð81Þ
½Xkk and coincides with the classical expression obtained by Handelman
and Prager (1948). Expression (91) may be also written as
For the four Vlasov cross-section deformation modes, namely !
axial extension, major and minor axis bending and torsion (about p2 m2 C xx
t C xy
t þ 2Gt C yy
t L
2
the shear centre), the GBT matrix components read rb ¼ 2 þ þ ; ð93Þ

b 12L2 6 12m2
½C11 ¼ Et X ½D11 ¼ 0 ½B11 ¼ 0 ½X11 ¼ 0 ð82Þ where L ¼ L=b and b ¼ b=t. For a given L ratio, the m value corre-
½C22 ¼ Et II ½D22 ¼ 0 ½B22 ¼ 0 ½X22 ¼ X ð83Þ sponding to the lowest (critical) stress is given by
½C33 ¼ Et III ½D33 ¼ 0 ½B33 ¼ 0 ½X33 ¼ X ð84Þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m ¼ IntðL C yy xx
4
2 t =C t Þ ð94Þ
½C44 ¼ Et Ix ½D44 ¼ Gt J ½B44 ¼ 0 ½X44 ¼ X; ic ð85Þ
where Int(a) is the integer adjacent to a leading to the true critical
where X is the cross-section area, II and III are the second moments stress. Finally, (81) provides the lowest bifurcation stress, given by
of area about the centroidal principal axes, Ix is the warping con- qffiffiffiffiffiffiffiffiffiffiffiffiffi
p2
stant, J is St. Venant’s torsional constant and iC is the polar radius rb;min ¼ 2 ðC xy
t þ 2G t þ C xx yy
t C t Þ: ð95Þ
6b
of gyration (also about the shear centre). For modes 2–4, the use
of (78) immediately leads to the classical tangent modulus formulae
5.2. Simply supported and built-in rectangular plates
1 m 2
p Et II
2
rbð2Þ ¼ ð86Þ If the x = 0 edge is built-in (clamped), the critical deformation
X L2 mode still exhibits a single half-wave sinusoid along the y direction
1 m2 p2 Et III and, therefore, only the first term of the GBT modal matrices (90)
rbð3Þ ¼ ð87Þ
X L2 applies. However, due to the non-simply supported edge, one must
2 2
1 m p Et Ix resort to the more general equation system (70), which now re-
rbð4Þ ¼ 2
þ Gt J : ð88Þ
iC X L2 duces to the single equation
For the higher order distortional modes, analytical formulae for ½C11 /;xxxx ð½D11 þ rb ½X11 Þ/;xx þ ½B11 / ¼ 0: ð96Þ
the associated matrix C, D, B and X components are available for In general, one has ½D11 þ rb ½X11 << 0 and the solution reads
lipped channel, hat, zed and rack-section members (Schardt,
1989, 1994b or Silvestre and Camotim, 2004a,b,c). Although these / ¼ A1 cosðaxÞ þ A2 sinðaxÞ þ A3 cosðbxÞ þ A4 sinðbxÞ
formulae were developed in the context of linear elastic thin- xÞ þ A2 sinðatbL
¼ A1 cosðat bL xÞ þ A3 cosðbt bL
xÞ þ A4
walled members, they can also be applied to elastic–plastic ones, xÞ;
just by using the appropriate tangent constitutive operator sinðbtbL ð97Þ
components. with x ¼ x=L and
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u½D r ½X þ ð½D þ r ½X Þ2 4½B ½C
5. Illustrative examples t 11 b 11 11 b 11 11 11
a¼ ð98Þ
2½C11
5.1. Fully simply supported rectangular plates vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uð2G þ C xy Þðp=bÞ 2 6r þ ðð2G þ C xy Þðp=bÞ 2 þ 6r Þ2 C xx C yy ðp=bÞ 4
t t t b t t b t t
For rectangular plates simply supported along all edges, with at ¼ xx
Ct
length L and width b, and subjected to uniform longitudinal com-
pression, the exact solutions are given by (74) and ð99Þ
246
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u esting to observe that, for each constitutive model, the local
u½D r ½X ð½D þ r ½X Þ2 4½B ½C
t 11 b 11 11 b 11 11 11 maxima of the simply supported curve lie on top of the corre-
b¼ ð100Þ
2½C11 sponding simply supported and built-in curve.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
uð2G þ C xy Þðp=bÞ
5.3. Local-plate buckling of rectangular hollow sections
2 6r þ ðð2Gt þ C xy Þðp=bÞ 2 þ 6r Þ2 C xx C yy ðp=bÞ 4
t t t b t b t t
bt ¼ xx :
Ct After assuming a particular buckling mode shape, it is a quite
ð101Þ straightforward task to derive GBT-based closed-form semi-analyt-
ical solutions (formulae). In order to illustrate this statement, let us
The longitudinal boundary conditions are
examine the local-plate buckling of uniformly compressed rectan-
gular hollow section (RHS) stub-columns. As shown in Fig. 4, the
/ð0Þ ¼ 0 ) A3 ¼ A1 ð102Þ
a
/;x ð0Þ ¼ 0 ) A4 ¼ A2 ð103Þ
b
/ðLÞ ¼ 0 ) A1 ðcosðaLÞ cosðbLÞÞ

a
þ A2 sinðaLÞ sinðbLÞ ¼ 0 ð104Þ
b
/;xx ðLÞ ¼ 0 ) A1 ðb2 cosðbLÞ a2 cosðaLÞÞ
þ A2 ðab sinðbLÞ a2 sinðaLÞÞ ¼ 0 ð105Þ
and, in order to ensure a non-trivial solution for A1 and A2 , the bifur-

cation loads are obtained by solving (numerically) the characteristic
equation associated with (104) and (105), i.e.,
aðb2 a2 ÞcosðaLÞsinðbLÞ þ bða2 b2 ÞcosðbLÞsinðaLÞ ¼ 0

() aðb2 a2 ÞcosðatbLÞsinðbt þ bða2 b2 ÞcosðbtbLÞsinð
bLÞ ¼ 0:
atbLÞ
ð106Þ
For illustrative purposes, Fig. 3 shows the variation of the bifur-

cation stress (positive in compression) with the length/width ratio
¼ 33:3 and a uniaxial bilinear stress–strain law character-
L, for b
ised by E = 70 GPa, Et ¼ 30 GPa; ry ð0Þ ¼ 100 MPa and m ¼ 0:3. As
expected, the bifurcation loads associated with deformation theory
are significantly lower than those stemming from incremental the- Fig. 4. Local-plate buckling mode shape for a rectangular hollow section stub-
ory (which obviously fall below the elastic ones). It is quite inter- column.
Fig. 3. Rectangular plates: variation of rb with L.
247

buckling pattern exhibits double symmetry and, therefore, only a p2 g 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
quarter of the cross-section needs to be analysed. This local-plate rb ¼ 2 ð wC xxt C yy 2 2 xy
t ðK þ 1=K Þ þ 4Gt þ 2C t Þ;
3h w g þ 1
mode shape may then be closely approximated by the sinusoidal
ð112Þ
function

2p y ph where K is a normalised length ratio defined by
¼ cos
w cos
hþb hþb
K ¼ L=Lcr : ð113Þ
2p y pg
¼ cos cos ; ð107Þ
bðg þ 1Þ gþ1 The minimum bifurcation stress corresponds to K ¼ 1 and,
therefore, is given by
where g ¼ h=b P 1, which (i) ensures the satisfaction of conditions

g 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
w ;y ðb=2 þ h=2Þ ¼ 0 (symmetry) and wðh=2Þ
;y ð0Þ ¼ w ¼ 0 (see Fig. 4)
2p2
and (ii) corresponds to the exact solution for g ¼ 1 (square hollow rb;min ¼ 2 ð wC xx yy xy
t C t þ 2Gt þ C t Þ: ð114Þ
section). By adopting this buckling mode shape, one obtains 3h w g þ 1
3
C xx
t bt 2p2 t 3
½C s 11 ¼ ðg þ 1Þw ½D11 ¼ ð2Gt þ C xy
t Þ For linear elastic materials, the above expressions may be writ-
12 3bðg þ 1Þ
ten in the standard plate buckling notation
4C yy 4 3
t p t
½B11 ¼ 3
½X11 ¼ btðg þ 1Þw ð108Þ 2
3b ðg þ 1Þ3 kp2 E t
0 !2 rb ¼ ð115Þ
2 12ð1 m2 Þ h
p
2
g
@ mðg þ 1Þ wC xx
rb ¼ t
12h2 w g þ 1 gL where k and kmin are the plate buckling coefficients, given by
!2 1
4gL
4 g 2 pffiffiffiffi 2
xy A k¼ ð wðK þ 1=K2 Þ þ 2Þ ð116Þ
þ16Gt þ 8C t þ C yy
t ð109Þ w gþ1
mðg þ 1Þ
8 g 2 pffiffiffiffi
kmin ¼ ð w þ 1Þ: ð117Þ
where w gþ1
g ¼ h=b P 1; L ¼ L=h; h ¼ h=t; Fig. 5 makes it possible to compare the k values provided by
(116), (117) and yielded by finite strip (FS) analyses carried out
pg
w ¼ 1 þ 2 cos2 : ð110Þ in the code CUFSM (Schafer, 2003), for 1 6 g 6 2 and 1 6 K 6 2. As
1þg
mentioned before, for g ¼ 1 (square cross-section) (107) consti-
The minimum bifurcation stress occurs for a member length gi- tutes the exact solution and, therefore, the k values obtained from
ven by (80), i.e., (116) are exact. However, the FS results for g ¼ 1 are 1.3% below
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the exact solution, a difference that must be taken into account
mðg þ 1Þ
wC xx yy
4
Lcr ¼ Lcr =h ¼ t =C t : ð111Þ when comparing the values corresponding to g – 1. The curves
2g
presented in Fig. 5 show that the difference between the analytical
This result makes it possible to rewrite (109) in the more com- and FS kmin values increases gradually as g grows, which means
pact form that (107) progressively departs from the exact solution. In
Fig. 5. Elastic local-plate buckling coefficients for RHS members.
248
(a)
(b)
Mode 1 Mode 2 Mode 3 Mode 4
Fig. 6. Lipped channel (a) cross-section geometry and (b) shapes of the first eight GBT deformation modes.
percentage terms,2 these differences amount to 2.4% ðg ¼ 1:2Þ, 3.9% lytical formula (78), for single-mode buckling (using the data given
ðg ¼ 1:5Þ and 3.5% ðg ¼ 2Þ – however, they decrease as the distance in Table 1), and (ii) the numerical solution of (77), for multi-mode
to the curve minimum grows. Moreover, it is also interesting to ob- buckling (all deformation modes included in the analysis). The
serve that, as predicted, the numerical minima occur for K ¼ 1. modal participation diagrams displayed in the bottom of Fig. 7
make it possible to assess how important is the contribution of
each GBT deformation mode to the column buckling mode. More-
5.4. Aluminium lipped channel columns over, the proximity between the thicker (all modes) and thinner
(single mode) curves provides clear evidence that formula (78),
One now analyses simply supported lipped channel columns concerning single-mode buckling, may be used to obtain accurate
with the cross-section geometry shown in Fig. 6(a) (Gonçalves estimates of the column true single half-wave bifurcation stresses:
and Camotim, 2004), adopting once again the constitutive relation (i) mode v for short-to-intermediate columns (or half-wave
(47), with E ¼ 70 GPa; r0:2 ¼ 200 MPa; m ¼ 0:33 and a ¼ 5 (typical lengths), (ii) mode u for intermediate-to-long columns and (iii)
values for an aluminium alloy). The cross-section is discretised into mode t for very long columns – in addition, mode x also provides
six natural and seven auxiliary nodes (three web intermediate accurate estimates for very short columns (not shown in Fig. 7).
nodes, one intermediate node per flange and one end node per Fig. 8, on the other hand, provides a comparison between the
lip), which leads to 13 deformation modes: (i) four Vlasov modes column critical bifurcation stresses obtained through (i) ‘‘all m”
(axial extension r, major s and minor t axis bending and torsion GBT-based analyses (with all modes included), (ii) FS analyses
u), (ii) two distortional modes ðv; wÞ and (iii) seven local-plate (CUFSM – linear elastic material behaviour only) and (iii) SFE analy-
modes. The shapes of the first eight deformation modes are de- ses based on the model developed by Le Grognec (2001).3 Also in-
picted in Fig. 6(b) and Table 1 shows the corresponding diagonal cluded is the GBT-based single half-wave (m = 1) buckling curve, i.e.,
components of the GBT matrices, calculated for an elastic material the thicker curve already presented in Fig. 7. As for Fig. 9, it shows
behaviour ð ep ¼ 0Þ.
Fig. 7 displays the GBT-based bifurcation stresses concerning 3
Concerning this model, the column length and cross-section mid-line are
single half-wave buckling (m = 1), obtained by means of (i) the ana- discretised into 20 and 18 8-node shell elements, respectively – in the latter case,
the discretisation includes one element per corner, to compensate for the absence of a
2
These percentages are calculated with respect to the values provided by (117). drilling degree-of-freedom in the shell element formulation.
249
Table 1
Elastic modal cross-section properties for the lipped channel cross-section.
i ½C r ii =E ½C s ii =C xx ½Dr ii =G ½Ds ii =C xy ½Bii =C yy ½Xii
1 600 0 0 0 0 0
2 1:0600 106 156.25 0 0 0 600
3 300:00 103 156.25 0 0 0 600
4 602:59 106 737:55 103 1250 0 0 2:7538 106
5 18.037 0.81892 3:5054 103 27:466 106 151:66 109 2.3109
6 19.736 0.69691 3:9783 103 103:02 106 365:91 109 2.1525
7 1.1152 66.155 0.33657 78:151 10 3
144:50 106 127.06
8 1.8279 94.915 1.5054 0.28009 939:44 106 182.31
Fig. 7. Lipped channel column GBT single mode and multi-mode bifurcation stresses and participation factors for single half-wave buckling (m = 1).
Fig. 8. Lipped channel column critical bifurcation stresses.
250
Fig. 9. Lipped channel column buckling mode shapes yielded by GBT-based and shell finite element analyses (elastic–plastic material, incremental theory).
the buckling mode shapes yielded by the GBT4 and SFE analyses for (i) There is an excellent agreement between the elastic results
columns with different lengths (incremental theory only5). The yielded by the three formulations, thus demonstrating that
observation of all these results prompts the following remarks: GBT-based linear stability analyses provide extremely accu-
rate results with a very small computational cost.
(ii) The plastic results obtained with the SFE model fall slightly
4
Note that the GBT-based buckling mode shapes are 3-dimensional representa- below the GBT incremental theory curve (the difference
tions of results provided by a 1-dimensional (beam) formulation. never exceeds 7%) and practically coincide with the GBT
5
Since there is a perfect coincidence between the GBT-based buckling mode deformation theory curve. This correlation was to be
shapes yielded by incremental and deformation theories, both may be deemed expected, since the GBT analysis allows for a quite straight-
represented in Fig. 9. It is also worth mentioning that the SFE model does not provide
true buckling mode shapes – instead, they are post-critical deformed configurations
forward modelling of the uniaxial stress state and boundary
adjacent to the bifurcation point (suitably scaled and having all the pre-buckling conditions (pinned and free-to-warp end sections) - no such
deflections removed). simplicity exists with the SFE model, for which localised
251
(a)
(b)
Fig. 10. RHS (a) cross-section geometry and (b) eight first GBT deformation mode shapes.
stress concentrations are bound to occur, acting as imperfec- bending, torsion u, distortion v and 10 local-plate modes.
tions and lowering the bifurcation stress values. Moreover, Fig. 10(b) shows the shapes of the first eight of these 15 modes.
the SFE analyses take into account the pre-buckling deflec- Like Fig. 7, Fig. 11 displays (i) single half-wave (m = 1) bifurca-
tions (unlike the GBT linearised buckling ones). tion stresses obtained by means of the analytical formula (78) (sin-
(iii) The plastic buckling mode shapes yielded by the GBT and gle-mode buckling) and the numerical solution of (77) (multi-
SFE analyses are also in excellent agreement. Depending mode buckling), and (ii) modal participation diagrams. These re-
on their length, the columns may buckle in ðiii1 Þ a mode sults clearly show that the column buckling behaviour is mostly
combining local-plate and symmetric distortional contribu- governed by the GBT cross-section deformation modes s (minor
tions (v þ x, for L = 15.8 cm), ðiii2 Þ a symmetric distortional axis bending) and w (local-plate). Obviously, the minor axis flex-
mode with one, two or three half-waves (v, for L = 33.6 cm, ural buckling stresses may be straightforwardly obtained from
L = 55 cm and L = 92 cm) or ðiii3 Þ a distortional–flexural–tor- (87). It is interesting to compare the present numerical results (ob-
sional mode (s þ u þ w, for L = 104.3 cm). tained from a genuine GBT cross-section analysis) with the values
yielded by the approximate formulae derived in Section 5.3 – this
comparison is presented in Fig. 12, where one readily observes that
5.5. Stainless steel rectangular hollow section columns
a virtually perfect match has been achieved.
Finally, Figs. 13 and 14 are similar to Figs. 8 and 9: while the for-
Finally, one addresses the buckling behaviour of rectangular
mer provides a comparison between the critical buckling stresses
hollow section (RHS) columns with the cross-section dimensions
yielded by the GBT-based (all m), FS and Le Grognec’s SFE analy-
shown in Fig. 10(a) and the constitutive law defined by the Ram-
ses,6 the latter compares the GBT and SFE plastic buckling mode
berg–Osgood parameters E ¼ 200 GPa; r0:2 ¼ 300 MPa; m ¼ 0:3
shapes (obtained with incremental theory). One may draw the fol-
and a ¼ 5, which correspond to a stainless steel alloy. The GBT
lowing conclusions:
cross-section analysis is based on a discretisation involving four
natural and ten (three per web and two per flange) intermediate
nodes, thus leading to 15 deformation modes (Gonçalves and 6
The columns cross-section mid-line and length are now discretised into 12 and 20
Camotim, 2004): axial extension r, minor s and major t axis elements, respectively (as before, one element per cross-section corner is included).
252
Fig. 11. RHS column GBT single mode and multi-mode bifurcation stresses and participation factors for single half-wave buckling (m = 1).
Fig. 12. RHS column GBT-based local-plate bifurcation stresses (m = 1).
Fig. 13. RHS column critical bifurcation stresses.
253
Fig. 14. RHS column buckling mode shapes yielded by GBT-based and shell finite element analyses (elastic–plastic material model, incremental theory).
Fig. 15. RHS column with L = 15 cm: GBT bifurcation stress and SFE equilibrium path.
254
(i) The elastic results yielded by the three formulations7 are tion stresses. In addition, the SFE analyses include the influence of
again in excellent agreement, thus confirming the accuracy the pre-buckling deflections, which is completely neglected by the
of the GBT-based linear stability analyses. GBT linear stability analyses.
(ii) The SFE local-plate plastic bifurcation stresses underesti- It should be stressed that the developed GBT-based formulae
mate the GBT-based results associated with the incremental are found to be quite accurate whenever the simplifying assump-
theory by a fair amount (the difference reaches 19%) but tions underlying their derivation remain valid. In particular, they
continue to lie above the GBT deformation theory curve may be very useful to generate benchmark results for future stud-
(note that the two GBT plastic buckling curves are now con- ies involving the plastic bifurcation behaviour of compressed thin-
siderably more far apart than in the case of the lipped chan- walled members.
nel columns). A possible explanation for this discrepancy
resides in the well-known high imperfection sensitivity of References
structural systems bifurcating in the plastic range, which is
Bazant, Z.P., Cedolin, L., 1991. Stability of Structures – Elastic, Inelastic, Fracture and
properly captured by the SFE analyses. In order to illustrate
Damage Theories. Oxford University Press, Oxford.
this statement, Fig. 15 shows the GBT bifurcation stress8 and Bigoni, D., 2000. Bifurcation and instability of non-associative elastoplastic solids.
the SFE model equilibrium path concerning a column with In: Petryk, H. (Ed.), Material Instabilities in Elastic and Plastic Solids (CISM
L = 15 cm. One observes a typical imperfection-sensitive Course 414). Springer-Verlag, Vienna, pp. 1–52.
Bijlaard, P.P., 1950. On the plastic buckling of plates. Journal of the Aeronautical
behaviour: soon after the bifurcation point, the bifurcated Sciences 17, 742–743.
branch attains a limit point (subcritical bifurcation). As Bonet, J., Wood, R., 1997. Nonlinear Continuum Mechanics for Finite Element
mentioned earlier, geometrical imperfections are often gen- Analysis. Cambridge University Press, Cambridge.
Brochard, J., Combescure, A., 1987. Dix ans de comparaison calculs-essais dans le
erated in the SFE model by the occurrence of stress concen- domaine du flambage et du post-flambage des structures. Calcul des Structures
trations associated with the modelling of the boundary et Intelligence Artificielle. Pluralis, Paris, vol. 1.
conditions. Camotim, D., Silvestre, N., Gonçalves, R., Dinis, P.B., 2004. GBT analysis of thin-
walled members: new formulations and applications. In: Loughlan, J. (Ed.),
(iii) The GBT and SFE plastic buckling mode shapes again corre- Thin-walled Structures: Recent Advances and Future Trends in Thin-walled
late extremely well. Depending on their length, the columns Structures Technology. Canopus Publishing Ltd., Bath, pp. 137–168.
may buckle in a local-plate mode with one or three half- Camotim, D., Silvestre, N., Gonçalves, R., Dinis, P.B., 2006. GBT-based structural
analysis of thin-walled members: overview, recent progress and future
waves (w, for L = 10 cm and L = 20 cm) or a minor axis flex-
developments. In: Pandey, M., Xie, W.-C., Xu, L. (Eds.), Advances in
ural mode (s, for L = 100 cm). The GBT buckling mode shapes Engineering Structures, Mechanics and Construction (SMCD 2006-Waterloo,
stemming from incremental and deformation theory are still 14-17/5). Springer, pp. 187–204.
Chakrabarty, J., 2002. Influence of anisotropy on the plastic buckling of rectangular
qualitatively similar (the latter are not shown) – however,
plates. In: Wang, C.M., Liu, G.R., Ang, K.K. (Eds.), Proceedings of 2nd
note that, according to deformation theory, the L = 100 cm International Conference on Structural Stability and Dynamics (ICSSD 2002–
column bifurcates in a local-plate mode (see Fig. 13). Singapore, 16-18/12). World Scientific, Singapore, pp. 448–452.
Cicala, P., 1950. On plastic buckling of a compressed strip. Journal of the
Aeronautical Sciences 17, 378–379.
Cimetière, A., Léger, A., 1996. Some problems about elastic–plastic post-buckling.
International Journal of Solids and Structures 33 (10), 1519–1533.
6. Conclusion Drucker, D.C., 1949. A discussion of the theories of plasticity. Journal of the
Aeronautical Sciences 16, 567–568.
Engesser, F., 1889. Über die Knickfestigkeit gerader Stäbe. Zeitschrift für Architektur
This paper presented the derivation and illustrated the applica- und Ingenieurwesen 35, 455–462 (German).
tion of GBT-based semi-analytical formulae to determine plastic Gerard, G., Becker, H., 1957. Handbook of Structural Stability: Part 1 – Buckling of
Flat Plates. NACA Technical Note 3781.
bifurcation loads of uniformly compressed and perfectly straight Gjelsvik, A., Lin, G.-S., 1987. Plastic buckling of plates with edge frictional shear
thin-walled plates and columns. The formulae were obtained on effects. Journal of Engineering Mechanics (ASCE) 113 (7), 953–964.
the basis of computationally efficient GBT-based linear stability Gonçalves, R., Camotim, D., 2004. GBT local and global buckling analysis of
aluminium and stainless steel columns. Computers and Structures 82 (17–19),
analyses, using Hill’s comparison solid method and adopting J 2
1473–1484.
small-strain incremental and deformation plasticity theories. Gonçalves, R., Camotim, D., 2007. Thin-walled member plastic bifurcation analysis
Several illustrative examples were presented and discussed, using generalized beam theory. Advances in Engineering Software 38 (8–9),
637–646.
concerning (i) the buckling of rectangular plates with simply sup-
Gonçalves, R., Le Grognec, P., Camotim, D., 2006. Plastic bifurcation of thin-walled
ported and built-in edges, (ii) the local-plate buckling of simply members: thin shell elements vs. GBT-based beam elements. In: Soares, C.A.M.
supported RHS stub-columns and (iii) the local and global buckling et al. (Eds.), Book of Abstracts of III European Conference on Computational
of simply supported lipped channel and RHS columns. While for Mechanics: Solids, Structures and Coupled Problems in Engineering (III ECCM –
Lisboa, 5-9/6). Springer (full paper in CD-ROM Proceedings).
the rectangular plates, the derived analytical formulae correspond Handelman, G.H., Prager, W., 1948. Plastic Buckling of a Rectangular Plate under
to exact solutions, for the columns, their accuracy was assessed by Edge Thrusts. NACA Technical Note 1530.
means of (i) CUFSM finite strip analyses (elastic materials) and (ii) Hencky, H., 1924. Zur Theorie plastischer Deformationen und der hierdurch im
Material hervorgerufenen Nachspannungen. Zeitschrift für angewandte
shell finite element analyses based on a model developed by one Mathematik und Mechanik (ZAMM) 4, 323–334 (German).
of the authors (elastic and plastic materials). Although a virtually Hill, R., 1956. On the problem of uniqueness in the theory of a rigid/plastic solid.
perfect agreement was found between all the elastic results ob- Journal of the Mechanics and Physics of Solids 4, 247–255.
Hill, R., 1958. A general theory of uniqueness and stability in elastic–plastic solids.
tained, the plastic results yielded by the SFE model fall slightly be- Journal of the Mechanics and Physics of Solids 6, 236–249.
low the GBT incremental theory ones and practically coincide with Hutchinson, J.W., 1974. Plastic buckling. Advances in Applied Mechanics 14, 67–
the GBT deformation theory values. A logical explanation for this 144.
Hutchinson, J.W., Budiansky, B., 1976. Analytical and numerical study of the effects
discrepancy resides in the fact that the GBT analysis allows for a of initial imperfections on the inelastic buckling of a cruciform column. In:
quite straightforward modelling of the uniaxial stress state and Budiansky, B. (Ed.), Buckling of Structures. Springer-Verlag, Berlin, pp. 98–
boundary conditions, whereas the use of a SFE model entails the 105.
Le Grognec, P., 2001. Contribution à la Modélisation du Flambement et du Post-
occurrence of localised stress concentrations that act as geometri-
Flambement des Structures Élastoplastiques. Ph.D. thesis, École Centrale de
cal imperfections, thus lowering the corresponding plastic bifurca- Nantes (French).
7
Le Grognec, P., Le van, A., 2008. Elastoplastic bifurcation and collapse of axially
Note that only three elastic SFE results are presented (L/h = 20, 32 and 60). loaded cylindrical shells. International Journal of Solids and Structures 45 (1),
8
Since the GBT linear stability analyses neglect the pre-buckling deflections and 64–86.
correspond to an eigenvalue solution, they are not capable of providing post-buckling Mazzolani, F.M., Valtinat, G., 1992. Bars, beams and beam-columns. In: Bulson, P.S.
equilibrium paths. (Ed.), Aluminium Structural Analysis. Elsevier, New York, pp. 35–192.
255
Needleman, A., Tvergaard, V., 1982. Aspects of plastic postbuckling behaviour. In: Schardt, R., 1989. Verallgemeinerte Technische Biegetheorie. Springer-Verlag, Berlin
Hopkins, H. (Ed.), Mechanics of Solids (R. Hill Anniversary Volume), pp. 453– (German).
498. Schardt, R., 1994a. Generalized beam theory – an adequate method for coupled
Nguyen, Q.S., 2000. Stability and Non-Linear Solid Mechanics. Wiley, stability problems. Thin-Walled Structures 19 (2–4), 161–180.
Chichester. Schardt, R., 1994b. Lateral torsional and distortional buckling of channel and hat-
Onat, E.T., Drucker, D.C., 1953. Inelastic instability and incremental theories of sections. Journal of Constructional Steel Research 31 (2–3), 243–265.
plasticity. Journal of the Aeronautical Sciences 20, 181–186. Shanley, F.R., 1946. The column paradox. Journal of the Aeronautical Sciences 13 (5),
Prola, L.C., 2001. Local and Global Stability Behaviour of Cold-Formed Steel 678.
Members. Ph.D. thesis, Instituto Superior Técnico, Technical University of Shanley, F.R., 1947. Inelastic column theory. Journal of the Aeronautical Sciences 14
Lisbon (Portuguese). (5), 261–267.
Ramberg, W., Osgood, W.R., 1943. Description of Stress–Strain Curves by Three Silvestre, N., 2005. Generalized Beam Theory: Formulations, Numerical
Parameters. NACA Technical Note 902. Implementation and Applications, Ph.D. thesis, Instituto Superior Técnico,
Raniecki, B., 1979. Uniqueness criteria in solids with non-associated plastic flow Technical University of Lisbon (Portuguese).
laws at finite deformations. Bulletin de l’Académie Polonaise des Sciences, Série Silvestre, N., Camotim, D., 2004a. Distortional buckling formulae for cold-formed
Science et Technique 27, 391–399. steel C and Z-section members: part I – derivation. Thin-Walled Structures 42
Raniecki, B., Bruhns, O.T., 1981. Bounds to bifurcation stresses in solids with non- (11), 1567–1597.
associated plastic flow law at finite strain. Journal of the Mechanics and Physics Silvestre, N., Camotim, D., 2004b. Distortional buckling formulae for cold-formed
of Solids 29, 153–171. steel C and Z-section members: part II – validation and application. Thin-
Schafer, B.W., 2003. CUFSM – Finite Strip Buckling Analysis of Thin-Walled Members Walled Structures 42 (11), 1599–1629.
(version 2.6), Department of Civil Engineering, Johns Hopkins University, Silvestre, N., Camotim, D., 2004c. Distortional buckling formulae for cold-formed
Baltimore (http://www.ce.jhu.edu/bschafer/cufsm). steel rack-section members. Steel and Composite Structures 4 (1), 49–75.
Schardt, R., 1966. Eine Erweiterung der technische Biegetheorie zur Berechnung Tuǧcu, P., 1998. Effect of axial loading on plastic buckling of long strips under pure
prismatischer Faltwerke. Stahlbau 35, 161–171 (German). shear. Computers and Structures 66 (2–3), 155–161.
256
Mechanics of Materials 42 (2010) 686–697
Mechanics of Materials
journal homepage: www.elsevier.com/locate/mechmat
A constitutive model for semi-crystalline polymers at high temperature

and finite plastic strain: Application to PA6 and PE biaxial stretching
Fanfei Zeng, Philippe Le Grognec *, Marie-France Lacrampe, Patricia Krawczak
Univ. Lille Nord de France, F-59000 Lille, France
Ecole des Mines de Douai, Polymers and Composites Technology and Mechanical Engineering Department, 941 rue Charles Bourseul, BP 10838,
F-59508 Douai, France
Article history: This paper deals with the mechanical behavior of semi-crystalline polymer films at high
Received 15 May 2009 temperature (90 °C) in finite plastic strains. Due to the complex morphology of such mate-
Received in revised form 6 April 2010 rials, standard elastoplastic formulations are not suitable for describing in a unified way
their uniaxial and biaxial tensile behaviors which involve specific physical phenomena.
According to physical considerations concerning the mesoscopic semi-crystalline structure
Keywords: of such polymers, a constitutive model has been specially developed, with the idea of a
Semi-crystalline polymers
three-phase morphology in relation to the average distance between crystalline blocks. A
Finite elastoplastic strains
Three-phase morphology
quite classical constitutive law is used to describe the mechanical (hyper)-elastic and/or
Eight-chain model plastic behavior of each phase (an eight-chain ‘‘network” model is employed). As a key
Entanglements of the work, a ‘‘multi-axial factor” is introduced, in the context of perfect plasticity, so as
Biaxial stretching to relate the threshold stress and the deformation mode, considering the effect of entangle-
ments. Uniaxial and equi-biaxial tensile experimental tests have been performed using two
materials: polyamide 6 and polyethylene. Model parameters are calibrated using uniaxial
stress–strain experimental curves. Then, the biaxial simulation curves are shown to be in
very good agreement with the corresponding experimental results.
1. Introduction which bringing an interesting property to the combination

thus formed. For example, the combination of polyamide 6
Polymer films for food packaging are very convenient (PA6) and polyethylene (PE) has shown a particular inter-
and widely applied in everyday’s life. However, these con- est for combining the good mechanical strength and
sumer products must respond simultaneously to so many impermeability to gas (e.g. O2) of PA6 with the high ductil-
performance criteria (impermeability to liquids and gas, ity and impermeability to liquids (e.g. H2O) of PE.
mechanical strength, transparency) that the use of a single Modeling the mechanical behavior of such multi-layer
polymer does not generally meet the specifications re- films, especially in uni- and biaxial stretching under high
quired for this type of application. Therefore, in most cases, temperatures which is supposed to improve the mechani-
polymer films are made from several polymers, each of cal properties, is therefore an important objective for
industrial applications. This knowledge is related to the
* Corresponding author at: Ecole des Mines de Douai, Polymers and establishment of a constitutive model for each polymer
Composites Technology and Mechanical Engineering Department, 941 material that can account for large irreversible deforma-
rue Charles Bourseul, BP 10838, F-59508 Douai, France. Tel.: +33 3 27 71 tions and reproduce in a unified way the uni- and biaxial
23 21; fax: +33 3 27 71 29 81. behaviors.
E-mail addresses: fanfei.zeng@mines-douai.fr (F. Zeng), philippe.le.
In recent years, many authors investigated the mechan-
grognec@mines-douai.fr (P. Le Grognec), marie-france.lacrampe@
mines-douai.fr (M.-F. Lacrampe), patricia.krawczak@mines-douai.fr ical behavior of semi-crystalline polymers, following
(P. Krawczak). different points of view, from the physical description of
doi:10.1016/j.mechmat.2010.04.006
257
F. Zeng et al. / Mechanics of Materials 42 (2010) 686–697 687
the morphology at various scales to the numerical simula- uda and Boyce, 1993), the p-chain models (James and Guth,
tion of the macroscopic behavior possibly relying on phys- 1943) and the full-chain model.
ical considerations. More recently, network models were modified and
On one hand, several studies were based on micro- widely applied in numerical simulations involving amor-
graphic observations and aimed at depicting the structure phous or semi-crystalline polymers (Diani et al., 2004;
of such polymers, in order to better explain their overall Richeton et al., 2007; van Dommelen et al., 2003). Let us
behavior. For example, Boni et al. (1982) and Aboulfaraj mention Dupaix and Boyce (2007) who developed a new
et al. (1995) discovered a variety of spherulitic morpholo- constitutive model in order to capture the rate-dependent
gies composed by lamellae of crystalline phase in polybu- stress–strain behavior of an amorphous polymer (poly-
tene and polypropylene, respectively. In high-density (ethylene terephthalate)-glycol) at and above the glass
polyethylene, ringed spherulites were also surveyed using transition temperature (Tg). These authors particularly
an optical microscope, as mentioned in Peacock (2000). considered the effect of intermolecular interactions be-
Furthermore, Heck et al. (2000) observed several semi- tween neighboring polymer chain segments. Richeton
crystalline polymers, among which poly-(ethylene-co-oc- et al. (2007) modified the classical eight-chain model so
tene)s and polypropylenes, and measured the average size as to simulate the behavior of poly-(methyl methacrylate)
of crystal blocks which was found to be the minimum va- and polycarbonate, taking into account the influence of
lue ensuring the stability of the crystalline structure. The relaxation in polymer chains. Dealing now with semi-
experimental achievement of micro-structural parameters crystalline polymers, the ‘‘Arruda-Boyce”, ‘‘Hasan-Boyce”,
such as the crystallinity, the orientation and size factors of ‘‘Bergström-Boyce” models, and the so-called ‘‘Hybrid”
each phase, allowed Fatahi et al. (2007) to correlate the model, were developed as a generalization of the eight-
tensile modulus of polyethylene blown films with their chain model for predicting the large strains, time- and
physical characteristics. temperature-dependent elastoplastic response of ultra-
Referring to experimental stress–strain tensile curves, it high molecular weight polyethylene, as reviewed in
was often reported in the literature that semi-crystalline Bergström et al. (2002). Gueguen et al. (2008), as well as
polymers sometimes exhibit a double yielding point. The Lee et al. (1993) and van Dommelen et al. (2003) in a dif-
influence of structural morphology such as the degree of ferent way, investigated the elasto-(visco)-plastic behavior
crystallinity or the lamellae thickness, and external factors, of semi-crystalline polymers using a micromechanically-
such as temperature or strain rate, on the occurrence of a based formulation, considering two phases (crystalline
double yield point was analyzed by many authors (among and amorphous) with independent yield processes. Hong
which Balsamo et al. (1993) and Feijoo et al. (1997) dealing (2005) constructed a specific model for the simulation of
with various polyethylenes and their blends), who tried to various kinds of polyethylene (e.g. linear polyethylene,
understand this particular feature in different ways. Sev- branched polyethylene, poly-(ethylene-co-vinylacetate))
eral micromechanical processes were suggested. Among under uniaxial tension. Two models were used in parallel
others, Séguéla and Rietsch (1990) proposed that the two to represent the combined effect of amorphous chains
yield points were due to the slip of the crystal blocks past and crystalline skeleton (see also Hong et al. (2004)). A
each other in the mosaic crystalline structure (heteroge- homogenization procedure was also applied in Drozdov
neous slip) and the homogeneous shear of the crystal and Christiansen (2008) in order to predict the thermo-
blocks (homogeneous slip). Considering polyamide 6 com- viscoelastic and viscoplastic behavior of high-density poly-
posites, Shan et al. (2005) showed that the change of crys- ethylene, but considering not only two but three different
talline structure was one of the main mechanisms involved phases, with an inter-phase at the interface between crys-
in the formation of a double yield point. They proved that a talline and amorphous phases.
double yielding phenomenon is more likely to occur with Although many models have been proposed so far and
low crystallinity ratios and low strain rates. validated with uniaxial experimental tests, their use in
On the other hand, many constitutive models were pro- the case of a biaxial tensile behavior has rarely been inves-
posed to simulate the mechanical behavior of semi-crystal- tigated until now. In this paper, a special model is intro-
line polymers. A preliminary approach was based on the duced, partly based on the constitutive models discussed
so-called ‘‘network” models, initially developed for elasto- above, which is intended to catch both uniaxial and biaxial
mers, emphasizing the rubber-like properties of the amor- behaviors of semi-crystalline polymers, where standard
phous phase of semi-crystalline polymers. Rubber-like elastoplastic models naturally failed. The general idea of
materials are predominantly defined by an extreme defor- the present model results from the three-phase description
mability combined with an almost full recovery in unload- of the morphology of the extruded films considered, in
ing procedures. Following these ideas, many network relation to the average distance between crystalline blocks,
models have been proposed. For instance, one can mention and of the physical description of their mechanical behav-
the classical models based on Gaussian statistics or non- ior. On one hand, somewhat classical hyperelastic and
Gaussian statistics (e.g. using Langevin distributions) plastic models are employed for the two phases corre-
(Kuhn and Grün, 1942; James and Guth, 1943), developed sponding to the extreme distances between crystalline
in the context of the statistical mechanics of polymer blocks. On the other hand, the third model related to the
molecular chains. According to the number of chains em- intermediate phase includes, in the context of perfect plas-
ployed in the network model with non-Gaussian statistics, ticity, a ‘‘multi-axial factor”, so as to relate the threshold
one distinguishes the single-chain model (Kuhn and Grün, stress and the deformation mode, considering the effect
1942), the three-chain model, the eight-chain model (Arr- of entanglements.
258
688 F. Zeng et al. / Mechanics of Materials 42 (2010) 686–697
Eventually, the objective is to consider explicitly the As shown in Fig. 1a, which describes a morphological
crystalline structure and the global morphology of the ex- model of polypropylene (Peacock, 2000), semi-crystalline
truded films. Nonetheless, it is convenient to confirm first polymers can be viewed as three-phase materials, includ-
the capability of such a model to reproduce the macro- ing interfacial regions between successive crystalline and
scopic response of the films in uniaxial and biaxial draw- non-crystalline zones. This simple model is still used by
ing. This is the framework of the present work, which is many authors, among which Drozdov and Christiansen
not concerned with the correlation between the model (2008) and Hong et al. (2004), and it will be retained for
parameters and the crystallinity of the materials (whose our materials. Interfacial regions are about 10–30 Å thick,
identification is out of the scope of this study and will be depending on the polymer and crystallization conditions,
performed later). Thereby, for the two materials consid- and can comprise up to 30% of the matter for medium to
ered (polyamide 6 and polyethylene), uniaxial and equi- high molecular weight materials (Mandelkern, 2006).
biaxial tensile experimental tests will be performed. The Moreover, the thickness of the non-crystalline regions is
uniaxial stress–strain experimental curves are used to cal- about 50–200 Å, whereas the average extended length of
ibrate the model parameters. Then, the biaxial simulation a typical polyethylene molecule is 1 lm (Peacock, 2000).
curves will be compared with the corresponding experi- As reviewed in Hong (2005), when the volume fraction
mental results, in order to evaluate the efficiency of the of crystallites is higher than 10% (which is the case for the
present model. materials used in this study, see Table 1), these crystallites
form a lamellar structure (the crystallization of homo-
2. Morphology of semi-crystalline polymers polymers such as polyethylene from a dilute solution is
lamellar-like) and experimental observations show that
The mechanical behavior of a semi-crystalline polymer the lamellar crystals are arranged in arrays of blocks, as
is strongly influenced by its morphology. Thus, a quick pre- shown in Fig. 1b. The thickness of lamellae is of the order
sentation on the physical structure of semi-crystalline of 80–200 Å.
polymers is first carried out in order to better explain the It can thus be assumed that semi-crystalline polymers
macroscopic properties of these materials observed in are merely composed of many crystalline blocks which
experimental testing and justify the subsequent choice of are almost randomly distributed in an amorphous med-
the different components of the future model. ium, so that the material appears to be almost homoge-
Polymers are macro-molecules which are composed of neous at the macroscopic scale. However, the links
giant chains in repeated patterns of molecular sequences between crystalline blocks and the surrounding amor-
connected by covalent bonds. The molecular shape and phous region may greatly influence the mechanical behav-
the way by which molecules are arranged in the solid state ior. Three scenarios will be distinguished in the sequel: a
are important factors for the determination of the physical first case where crystalline blocks are inter-connected by
and/or mechanical properties of polymer products. In crys- very short intra-lamellar chains and two other cases
talline materials, molecules are perfectly arranged in re- involving the inter-lamellar distance which is assumed to
peated patterns. Such materials display an highly ordered be either middle-sized or rather long.
and regular structure. In amorphous materials, by contrast,
molecules are randomly arranged in long chains which 3. Experimental procedure and results
twist and curve around one another, so that large regions
of highly structured morphology are very unlikely. In be- Two materials are considered in this study. The first one
tween, semi-crystalline polymers (like PA6, PE, and most is a non-nucleated grade of polyamide 6 (PA6) provided by
polymers) consist in mixtures of small crystals in an amor- DSM (Geleen, Netherlands, reference AKULON F236C).
phous medium. The second one is a blend of 30 wt% of low-density
Fig. 1. Morphology of semi-crystalline polymers (Peacock, 2000).
259
Table 1 hardly differ from the Kirchhoff stresses, as it will be ex-

Thermal properties of PA6 and PE (Sallem-Idrissi, 2008). plained in the next section. As shown in Fig. 2, all the speci-
Polymer Glass transition Melting Crystal weight mens (PA6 or PE) behave similarly for the three following
temperature (°C) temperature (°C) fraction (%) speeds into consideration: 10 mm/min (i.e. a strain rate of
PA6 48 220 25 ± 1 5.55 103 s1), 20 mm/min (1.11 102 s1) and
2 1
PE 78 120 40 ± 5 50 mm/min (2.77 10 s ). Viscous effects will thus be
neglected in the sequel, since the loading speed does not
exceed 50 mm/min, what is verified in practice for our
polyethylene (LDPE, reference 2101TN00) and 70 wt% of future applications (the minimum value of 10 mm/min will
linear low-density polyethylene (LLDPE, reference 118 N) be retained for the following tests).
both provided by SABIC. The films were manufactured by On one hand, the previous tests prove that these semi-
extrusion blowing with a constant thickness of 100 lm. crystalline polymers display an extreme deformability and
FTIR experiments have indicated a nearly perfect isotropy a strong stress hardening in large strains like elastomers
of the PA6 films, while the PE films exhibit a substantial (see Fig. 2). On the other hand, they differ from elastomers
planar chain orientation induced by the process. Moreover, because of their elastoplastic irreversible behavior, as
the crystalline structure of the PA6 films is mainly in the shown in Fig. 3 involving cyclic loading curves.
b-form (85%) with a little amount of c-form (15%). This The volume of the specimens are measured at the
particular crystalline distribution is induced by rapid cool- beginning and the end of each test. It is found to change
ing during processing. More details on the manufacturing only very slightly, what agrees with the conclusions of
conditions and micro-structure of the films are presented Hong (2005).
elsewhere (Sallem-Idrissi et al., 2009). Furthermore, some thermal properties of PA6 and PE
Experimental uniaxial tests are performed using an In- are listed in Table 1, which may mainly explain the few
stron tensile testing machine together with a conditioning differences encountered in our experiments between the
chamber in order to heat the specimens up to the desired behaviors of PA6 and PE. Despite these differences, the
temperature (90 °C) and keep it constant during the tests. two materials will be modeled using the same ideas and
Considering only isothermal deformations, the tempera- the same methodology.
ture will not explicitly appear in the description of our Finally, the monotonic uniaxial and equi-biaxial curves
model. The same formulation is however suitable for any are compared for each material in Fig. 4. The double yield-
other value between the glass transition and the melting ing points under uniaxial and biaxial stress states seem to
temperatures, with elastic or plastic parameters depending be very different, what can not be explained by using only
implicitly on the temperature. Strains are precisely mea- the classical plasticity theories such as the generalized
sured in the dumbbell specimens thanks to an Apollor vi- standard materials theory involving the von Mises crite-
deo-traction system. The biaxial curves used in the rion, as the corresponding equivalent stress gives the same
sequel come from an earlier experimental work carried value in both loading cases.
out by Sallem-Idrissi (2008). Biaxial stretching experi-
ments were performed on a Cellier tenter frame consisting 4. Constitutive model
of four pantographs driven by two hydraulic jacks at right
angle, and operated in a temperature-regulated oven. In this section, a new constitutive model is designed,
Square specimens were submitted to simultaneous equi- based on the previous physical and experimental observa-
biaxial drawing at a constant jack speed of 60 mm/min tions. It will then be applied to both cases of PA6 and PE. In
(i.e. a strain rate of 9.52 103 s1). this study, due to the very large strains encountered, true
Preliminary uniaxial tests are carried out in order to esti- logarithmic strains are employed and additively decom-
mate the effect of the loading rate on the stress–strain law. posed into elastic and plastic parts, according to the classi-
In all the following stress–strain curves, the true logarithmic cal Lee multiplicative decomposition of the deformation
strains are plotted versus the true Cauchy stresses, which gradient. Moreover, as volumetric changes are proved to
Fig. 2. Uniaxial stretch curves of the semi-crystalline polymers at 90 °C under various strain rates.
260
Fig. 3. Loading and unloading curves of the semi-crystalline polymers at 90 °C.
Fig. 4. Comparisons between uniaxial and biaxial experimental curves of the semi-crystalline polymers at 90 °C.
be negligible, the more classical true Cauchy stresses can tween these blocks are connected or even embedded
be employed in place of Kirchhoff stresses (they practically thereto. When the sample is stretched under a uniaxial
coincide with one another) which should be preferred as a stress state, the chains within the amorphous region may
more reliable counterpart of logarithmic strains. turn tensile for even a small strain. Once the stress be-
comes large enough, the chains begin to glide and the
4.1. Basic components of the three-phase model material hardens, due to the strong connecting links with
the crystalline blocks. The typical uniaxial stress–strain
As shown in Fig. 5, one assumes that a random sample behavior in this case, usually observed with crystalline
of a semi-crystalline polymer comprises quite a few crys- materials, is depicted in Fig. 6, where E1 represents Young’s
talline blocks connected by molecular chains (standing modulus, ET1 the tangent elastoplastic modulus, and r1 is
for the amorphous phase and represented by curved lines). the yield stress of the considered sample.
First, the uniaxial drawing of a selected sample typically The generalization to multi-axial stress states is
involves two crystalline blocks linked together with a straightforward, in the framework of generalized standard
qffiffiffiffiffiffiffiffiffiffiffiffi
non-crystalline region. According to the distance between materials, using the von Mises criterion (rVM ¼ 32 S : S
the two blocks, three cases are considered, leading to three
with the deviatoric stress tensor S ¼ R 13 trðRÞI where R
particular behaviors.
When the specimen considered is stretched under mul- represents the true stress tensor) and an isotropic harden-
ti-axial conditions, the representative sample might in- ing type. The coupling effects between the elastic and
volve more blocks. The multi-axial behavior will be also plastic parts of the behavior, clearly observed during
discussed in the sequel: it will derive from the uniaxial re- unloading (see Fig. 3), will naturally appear with the global
sults by using various criteria partly based on a ‘‘multi-ax- model, due to the specific combination of the elementary
ial factor”. models employed for each phase.
4.1.1. Case 1 4.1.2. Case 2

The distance between the two considered crystalline The distance between crystalline blocks is medium-
blocks is very short. Therefore, most chains located be- sized. Many chains in the amorphous region are not
261
Fig. 5. Distinction of three behaviors according to the distance between crystalline blocks.
ments then naturally appear and a new sliding phenomenon

may occur, as the stress increases. All the entanglements are
supposed to provide the same contribution to the global
resistance against sliding, which is represented by a limit
force s corresponding to the incipient slippage. The macro-
scopic uniaxial threshold stress ru2 (where superscript .u
stands for ‘‘uniaxial”) resulting from this limit force is
therefore proportional to the uniform surface density nu of
entanglements involved in the uniaxial behavior, as follows:
ru2 ¼ nu s ð1Þ
Fig. 6. Stress–strain curve under uniaxial drawing for case 1.

In this case, the typical mechanical behavior under uniaxial
drawing is expressed as shown in Fig. 7, displaying the
yield stress r2 defined above (which is also the maximum
admissible stress in the context of perfect plasticity) and
directly connected to all the surrounding crystalline blocks
the elastic modulus E2 (encompassing the elastic response
and thus less constrained by their mechanical behavior as
of both the crystalline blocks and the amorphous phase)
compared to case 1. Here, the mechanical response of such
a sample is conversely governed by the presence of numer-
ous entanglements between couples of chains.
Isolating two entangled chains at the mesoscopic scale,
it is possible to explain their joined mechanical response
by the spatial configuration of the couple of chains, namely
their relative position and motion. In a macroscopic point
of view, due to the huge number of molecular chains in-
volved and their random space distribution, the sample is
assumed to behave uniformly. Only the deformation mode
(e.g., uniaxial or biaxial) will affect the relative displace-
ments of the chains and consequently the number of
entanglements brought into play.
When the sample is stretched uniaxially, the chains in
the non-crystalline region tighten rather quickly. Entangle- Fig. 7. Stress–strain curve under uniaxial drawing for case 2.
262
whose influence will further be considered together with be involved in the calculation of the ‘‘multi-axial factor”.
the elastic behavior of case 1 in the global model. Nevertheless, let us consider the particular case of a triaxial
Under general loading conditions, one has to replace the isotropic stress/strain state. The von Mises type criterion
uniaxial stress by an equivalent scalar stress function of leads to a null equivalent stress r0VM , independently of
the multi-axial stress tensor, to be compared to the the value of K. Indeed, one can easily imagine that both
previous threshold stress. On one hand, the von Mises ends of all chains lie in the vicinity of a crystalline block
equivalent stress is used, as the sliding phenomenon is and are equally loaded in such a way that no sliding may
supposed to be governed by shear effects and distortion. occur. Therefore, plasticity is not observed at all in the
On the other hand, a new limit stress is defined, in corre- material.
spondence with the number of entanglements involved
in the multi-axial behavior, which clearly depends on the 4.1.3. Case 3
deformation mode. Let us consider an arbitrary multi-axial The distance between crystalline blocks is long enough
strain state and focus on the positive principal strains eI to consider that the chains in the amorphous region are al-
which are the ones and only strains in the principal axes most unconstrained. In these conditions, a network model
that contribute to the interlocking of entanglements. A such as the eight-chain model is a satisfactory solution to
‘‘multi-axial factor” K is introduced, namely the scaled simulate the macroscopic mechanical behavior.
sum of these strains: In the 1940’s, Kuhn and Grün (1942) and James and
P Guth (1943) introduced the so-called Langevin model (in
e
I I
K¼ ðeI P 0Þ ð2Þ the context of non-Gaussian statistics) to deal with large
maxI eI
stretches up to the limiting end-to-end distance given by
The number of entanglements nm (where superscript .m the length of the chains. Considering a molecular chain
stands for ‘‘multi-axial”) involved in the current multi-ax- composed of N monomer segments pffiffiffiffi with length l, its aver-
ial behavior is then supposed to depend linearly on the age unstretched length is l N according to statistical
parameter K. The associated threshold stress rm 2 follows mechanics of freely jointed chains (Elías-Zúñiga and
the same rules, considering that all the limit forces s corre- Beatty, 2002), and its total stretched length is lN.
sponding to each entanglement can be added, regardless of First, a probability density for the stretch distribution in
their orientation. In the uniaxial case, Eq. (2) entails K = 1, the material is established as follows:
since the strain measured in the loading direction is the
k b
single positive value. The following relations are thus PðkÞ ¼ P0 exp N pffiffiffiffi b þ ln ð5Þ
N sinh b
obtained:

nm ¼ Knu with b ¼ L1 pkffiffiffi
N
, where the inverse of the Langevin func-
ð3Þ
rm2 ¼ K ru2 tion LðxÞ ¼ cothðxÞ 1=x cannot be explicitly obtained
and is conveniently evaluated by a Padé approximation
For the sake of simplicity, a specific modified von Mises cri- as mentioned in Miehe et al. (2004):
terion is defined, based on the following modified von
Mises equivalent stress: ð3 x2 Þx
L1 ðxÞ ð6Þ
1 x2
rVM
r0VM ¼ ð4Þ Using the Boltzmann’s equation (with the so-called Boltz-
K
mann’s constant k), the strain energy function W of a given
and with a single yield stress ru2 which will be further de-
chain writes:
noted by r2. The same modified function is retained for the
evolution law, within the framework of associated k b
W ¼ kT ln PðkÞ ¼ NkT pffiffiffiffi b þ ln kT ln P 0
plasticity. N sinhðbÞ
When dealing with the equi-biaxial tensile test, the ð7Þ
‘‘multi-axial factor” takes the value K = 2, what means that
Then, the principle of minimum potential energy leads to
the number of entanglements involved in the sliding phe-
the following expression for the stress in a single chain:
nomenon is twice as important as in the uniaxial case. Any
other biaxial case, when the two relevant positive strains @W pffiffiffiffi 1 k
are different one from each other, gives rise to an interme-
r¼k ¼ kkT NL pffiffiffiffi ð8Þ
@k N
diate value of K. The ‘‘multi-axial factor” evolves continu-
ously and even linearly between 1 and 2 corresponding The more recent model based on this approach is the eight-
to the two limit cases of uniaxial and equi-biaxial states, chain model, the eight chains being the half-diagonals of a
respectively. unit cube. Under the assumption of a constant volume dur-
In practice, dealing with 2D polymer films without any ing the loading procedure, namely kIkIIkIII = 1 (where ki
compressive loading in the plane, at least one of the (i = I, II, III) are the principal stretches), an equivalent
principal strains is strictly positive and the parameter K stretch is introduced (Miehe et al., 2004; Elías-Zúñiga
is always properly defined, except at the reference and Beatty, 2002):
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
configuration (the corresponding value can be arbitrarily
k2I þ k2II þ k2III
selected). Moreover, the out-of-plane strain will always k¼ ð9Þ
be negative, so that only the in-plane deformations will 3
263
The eight chains being stretched by this same extension ra- of case 3 and case 1-2 in the initial and advanced deforma-
tio, the principal stresses can thus be simply expressed as tion stages, respectively. Moreover, due to the relative
(Arruda and Boyce, 1993): lengths of the different specimens (cases 1, 2 and 3), one
pffiffiffiffi k2 may surely encounter a few samples of short and med-
1 k
ri ¼ p þ C R N i
L1 pffiffiffiffi ð10Þ ium-sized chains aligned along each long chain of purely
3 k N amorphous phase, in any arbitrary direction.
In Eq. (10), CR and N stand for the model parameters, Based on the above remarks, the following relations can
accounting for the density and the limit of extensibility be deduced for the full model:
of a chain, respectively. These parameters will be further :
Rcase1 ¼ Rcase2 ¼ Rcase1—2
derived from the uniaxial experimental curves (see Section
4.3). Rtotal ¼ Rcase12 þ Rcase3
ð11Þ
ecase1—2 ¼ ecase1 þ ecase2 ¼ eecase1—2 þ epcase1—2
4.2. Full model etotal ¼ ecase1—2 ¼ ecase3
At the macroscopic scale, the semi-crystalline polymers where Rtotal and etotal are the global stress and strain ten-
considered seem to be homogeneous and roughly isotro- sors for the whole model, and Rcase and ecase correspond
pic: the amorphous phase is assumed to be randomly dis- to the stresses and strains supported by one or two associ-
tributed around the crystalline blocks and no preferential ated cases. The superscripts .e and .p stand for the elastic
direction is observed. The structural and mechanical and plastic parts of the strain tensor. As previously ex-
properties of these materials were previously analyzed in plained, the elastic behaviors of cases 1 and 2 will not be
Sallem-Idrissi et al. (2009) using methods such as Differen- particularized and a unique Young’s modulus E1–2 = E1E2/
tial Scanning Calorimetry (DSC), Wide-Angle X-ray (E1 + E2) will be used for the serial arrangement, in the
Scattering (WAXS) and infrared spectroscopy. A great sim- framework of the Saint–Venant–Kirchhoff linear elastic
ilarity of the absorption spectra of the films in the main theory. Let us mention that this model will be imple-
directions was observed, thus revealing a nearly perfect mented within the context of a plane stress formulation,
isotropy of the initial semi-crystalline structures. The PE for thin films applications. It means that Rtotal must verify
films displayed a slight orthotropy as compared to PA6, the plane stress condition. For simplicity purposes, we as-
but no particular difference has been pointed out between sume that both Rcase 1–2 and Rcase 3 have a null transverse
the two in-plane directions. Moreover, the experimental normal component. This approximation is not so detri-
results show a similar uniaxial behavior in any stretching mental since case 1-2 plays first a major role in the range
direction for both PA6 and PE. of small deformations (case 3 is then insignificant), and
The global model encompassing all the behaviors re- case 3 is in turn the most predominant for larger strains.
vealed for each case previously presented is depicted in It can be noticed that the proportion of each phase,
Fig. 8, in the style of a rheological model. Models for cases which is related to the crystallinity of the material, does
1 and 2 are assumed to bear the same stress so that a series not explicitly appear in the model, but it is indirectly taken
arrangement is retained. The model for case 3 initially cor- into account in the respective model parameters. For
responds to a much softer behavior which is paralleled example, a Young’s modulus is only defined for the combi-
with the others. The tangent modulus of case 3 is very nation of cases 1 and 2 and not for each single case, and it
weak in the first deformation stage as compared to the takes implicitly into consideration the volume fraction of
other two cases, whereas it is conversely predominant for each phase. Only such global parameters are needed to
larger strains. The configuration in Fig. 8 is the most ade- simulate the macroscopic behavior and they can be easily
quate one, being able to predict the neglecting influence derived from experimental tests, contrary to the intrinsic
Fig. 8. Global representation of the mechanical model for semi-crystalline polymers.
264
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2 0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi131
moduli of each phase. Moreover, it allows us to define a
req k2eq þ k2eq k2 þ 2
simpler expression of the model with very few parameters, CR ¼ 4L1 @ eq keq A5 ð14Þ
which is valid for any crystallization conditions (only the k2eq k1eq 3N 3N
numerical values of the material parameters are changing).
4.3. Computation of the model parameters from uniaxial tests
The procedure involved for the determination of the Remark. The global behavior is assumed to be almost
model parameters is explained throughout the example incompressible. Indeed, when large elastoplastic strains
of PA6. The corresponding uniaxial experimental curve is are involved, both models in parallel (1–2 and 3) display
plotted in Fig. 9. such a behavior. On one hand (case 1–2), the elastic strains
As shown in Fig. 8, cases 1 and 2 may be formally put are negligible, and the von Mises type criteria used imply
together, giving rise to a unified set of elastoplastic coeffi- an incompressible plastic behavior. On the other hand, the
cients, namely E1–2, ET1, r1 and r2. eight-chain model is consistent with this hypothesis. On
In Fig. 9, the first two parts of the curve (for 0 6 e 6 eA the contrary, for particularly small strains, the global
and eA 6 e 6 eB, respectively) are assumed to be linear, behavior is governed by the elastic part of case 1-2
and their respective slopes E and H are obtained by linear (considering that stresses in the amorphous phase are
regression with a reasonable correlation coefficient. For relatively small, as compared to those concerning the two
e P eB in the vicinity of eB (while strains remain moderate), other cases in series). A Poisson coefficient m can thus be
the response curve is also quite linear, so that a third slope simply derived for this model as the ratio between the
E0 can be deduced again by linear regression, which corre- transverse strain and the longitudinal one in the onset of
sponds to the initial elastic modulus of case 3. the uniaxial test:
First, the four parameters of case 1-2 are derived from
the experimental curve using the following relations: etrans
m¼ ð15Þ
elong
E12 ¼ E E0
The numerical value obtained is not so far from 0.5 (elastic
ET1 ¼ H E0
ð12Þ strains are also nearly incompressible), and most of all, it
r1 ¼ rA E0 eA only concerns very small strains so that the effect of the
r2 ¼ rB E0 eB Poisson’s ratio can be neglected in most of the strain range
considered. The apparent contradiction between a slightly
Then, CR and N are determined as in Bergström and Boyce
compressible material and a perfectly incompressible
(1998). From the maximum strain elim deduced from the
model in parallel is not an issue since only bi-dimensional
uniaxial experimental curve by extrapolation, the limit
strains and stresses are further considered within the plane
stretch value is straightforward klim = exp(elim), and one
stress formulation retained.
obtains:

1 2 2
N¼ klim þ ð13Þ 5. Simulation results: validation and analysis
3 klim
Using an arbitrary point (et, rt) on the stress–strain curve, The procedure discussed above is applied to the uniax-
with keq = exp (et) and req = rt r2, one finally gets: ial experimental curves obtained for PA6 and PE. Once the
Fig. 9. Calculation of the model parameters for PA6 using the uniaxial curve.
265
model parameters are known, it is possible to simulate the r2 = 1.6 MPa, E12 = 18.1 MPa, N = 49.5, CR = 1.18 MPa,
mechanical behavior of PA6 and PE films under uniaxial or ET1 = 9.2 MPa, m = 0.39.
equi-biaxial drawing, and compare the numerical solutions Fig. 13 displays both numerical and experimental
to the corresponding experimental results (Sallem-Idrissi curves of PE in uniaxial and equi-biaxial drawing. From
et al., 2009). On one hand, a very good agreement between small up to moderate strains, the simulation results fit
both experimental and numerical curves is expected in the the experimental ones very well. When the strains become
uniaxial case, since material parameters were chosen in or- much larger, the two curves slightly differ due to the strain
der to closely fit the uniaxial test response. On the other softening phenomena.
hand, the biaxial results are supposed to validate the pre- Finally, it is interesting to remark that, despite an ini-
vious formulation used for this type of material. tially isotropic behavior, induced anisotropy may occur,
For PA6, the following parameters are identified accord- due to the large elastoplastic strains and to the micro-
ing to Fig. 9: r1 = 11.8 MPa, r2 = 21.7 MPa, E12 = 431 MPa, structure of the materials, especially after successive elas-
N = 9.17, CR = 7.34 MPa, ET1 = 121 MPa, m = 0.44. tic unloading and plastic reloading. Our model naturally
As expected, in Fig. 10, the uniaxial simulation results involves such a phenomenon, because of the global
fit the experimental results very well. In the biaxial case, arrangement of the three phases. Moreover, to some ex-
simulation results are also in good agreement with the cor- tent, the adequacy of the experimental and simulation
responding experimental results. The slightly higher values unloading curves in Fig. 14 corroborates the fact that the
obtained with the model may be due to the slight anisot- evolution of anisotropy is well-depicted.
ropy of PA6 which is not taken into account. The mechan-
ical properties of PA6 used in the biaxial simulation are 6. Conclusion
derived from uniaxial tests performed in the strongest
direction of the material sheet. In this paper, a specific elastoplastic constitutive model
In the case of PE, the testing temperature (90 °C) is clo- is developed for semi-crystalline polymers in the frame-
ser to the melting temperature of the material (see Table work of isothermal transformations (between the glass
1). As a consequence, the stress–strain curve of PE (see transition and the melting temperatures) under low-level
Fig. 11) is slightly different compared to that of PA6. Under strain rates, so that viscosity effects are not considered.
such a temperature, the yielding stresses of cases 1 and 2 Based on the analysis of the structural morphology of
are very near one from each other, and the ‘‘viscous” effects semi-crystalline polymers from a mesoscopic point of
may be not negligible yet. It may be induced by the slip- view, three cases are considered, depending on the
ping of chains in case 1, that makes the curve not so fluent distance between crystalline blocks. A particular model is
than in Fig. 9. As shown in Fig. 11, each continuous part of proposed for each configuration, namely small (1),
the curve which presents almost the same hardening slope medium-sized (2) and large (3) amorphous regions. For
ET1 (see Section 4.1.1) is followed by a stress softening case 1, a classical elastoplastic model is retained. For case
phenomenon. 2, a perfect plastic model is suggested with a new thresh-
For comparison purposes, Fig. 12 depicts the uniaxial old stress, defined using a ‘‘multi-axial factor” as a function
experimental stress–strain behavior of PE at a much lower of the deformation state. For case 3, the so-called ‘‘eight-
temperature, which seems to present a similar tendency as chain” model is employed. The intermediate case mainly
for PA6 (the two yielding points are more distinguishable). describes the sliding phenomenon between entangled
In spite of these differences, one can obtain the follow- chains and is capable to explain the most significant fea-
ing parameters in a quite similar way: r1 = 0.88 MPa, tures of the experimental behavior.
Fig. 10. Comparisons between simulation and experimental results for PA6 at 90 °C in both uniaxial and equi-biaxial cases.
266
Fig. 11. Calculation of the model parameters for PE using the uniaxial curve.
seven material parameters. A methodology for the calibra-

tion of these parameters from uniaxial tests is presented
and applied in the case of PA6 and PE.
Once the models defined, simulations are performed
also in equi-biaxial stretching and numerical results agree
very well with the experimental ones. Large discrepancies
only appear during unloading, when it takes place at high
stress values, in the vicinity of null stresses (when the load
vanishes). Therefore, this model describes properly the
arbitrary behavior of such a polymer film in plane stresses,
once strain rates are not too high and unloading is not too
strong or just restricted to a small area.
The constitutive model developed in this study has thus
proved to be able to reproduce the biaxial behavior
according to parameters calibrated from uniaxial tests,
Fig. 12. Experimental uniaxial stress–strain curve of PE at 20 °C.
for two different semi-crystalline polymers. The next
A global constitutive model is then built by properly step can now be undertaken, namely the identification
assembling the three components above, displaying only of the correlation between the model parameters and
Fig. 13. Comparisons between simulation and experimental results for PE at 90 °C in both uniaxial and equi-biaxial cases.
267
Fig. 14. Comparisons between simulation and experimental results during unloading in the uniaxial case.
characteristic quantities describing the crystalline mor- Feijoo, J.L., Sanchez, J.J., Müller, A.J., 1997. The phenomenon of double
yielding in oriented high density polyethylene films. Journal of
phology of the considered polymers.
Materials Science Letters 16 (21), 1721–1724.
Gueguen, O., Richeton, J., Ahzi, S., Makradi, A., 2008.
Acknowledgements Micromechanically based formulation of the cooperative model
for the yield behavior of semi-crystalline polymers. Acta Materialia
56 (7), 1650–1655.
Thanks are due to Conseil Régional (Regional Council) Heck, B., Hugel, T., Iijima, M., Strobl, G., 2000. Steps in the formation of the
Nord-Pas-de-Calais, France, and the European Union (Euro- partially crystalline state. Polymer 41 (25), 8839–8848.
Hong, K., 2005. A Model Treating Tensile Deformation of Semi-crystalline
pean Funds for Regional Development, FEDER) for having Polymers. Ph.D. Thesis, Albert-Ludwigs-Universität Freiburg im
partly funded the video-traction equipment. DSM and SA- Breisgau.
BIC are also deeply acknowledged for the PA6 and PE raw Hong, K., Rastogi, A., Strobl, G., 2004. A model treating tensile deformation
of semi-crystalline polymers: quasi-static stress–strain relationship
materials. The authors also wish to thank Naïma Sallem- and viscous stress determined for a sample of polyethylene.
Idrissi for her biaxial stretching experiments. Macromolecules 37 (26), 10165–10173.
James, H.M., Guth, E., 1943. Theory of the elastic properties of rubber.
Journal of Chemical Physics 11 (10), 455–481.
References Kuhn, W., Grün, F., 1942. Beziehungen zwischen elastischen Konstanten
und Dehnungsdoppelbrechung hochelastischer Stoffe. Kolloid-
Aboulfaraj, M., G’Sell, C., Ulrich, B., Dahoun, A., 1995. In situ observation of Zeitschrift 101 (3), 248–271.
the plastic deformation of polypropylene spherulites under uniaxial Lee, B.J., Argon, A.S., Parks, D.M., Ahzi, S., Bartczak, Z., 1993. Simulation of
tension and simple shear in the scanning electron microscope. large strain plastic deformation and texture evolution in high density
Polymer 36 (4), 731–742. polyethylene. Polymer 34 (17), 3555–3575.
Arruda, E.M., Boyce, M.C., 1993. A three-dimensional constitutive model Mandelkern, L., 2006. Crystalline polymer: some reminiscences over the
for the large stretch behavior of rubber elastic materials. Journal of years. Thermochimica Acta 442 (1–2), 31–34.
the Mechanics and Physics of Solids 41 (2), 389–412. Miehe, C., Göktepe, S., Lulei, F., 2004. A micro–macro approach to rubber-
Balsamo, V., Müller, A.J., 1993. The phenomenon of double yielding under like materials – Part I: The non-affine micro-sphere model of rubber
tension in low-density polyethylene, linear low-density polyethylene elasticity. Journal of the Mechanics and Physics of Solids 52 (11),
and their blends. Journal of Materials Science Letters 12 (18), 1457–1459. 2617–2660.
Bergström, J.S., Boyce, M.C., 1998. Constitutive modeling of the large Peacock, A.J., 2000. Handbook of Polyethylene: Structures Properties and
strain time-dependent behavior of elastomers. Journal of the Applications. Marcel Dekker.
Mechanics and Physics of Solids 46 (5), 931–954. Richeton, J., Ahzi, S., Vecchio, K.S., Jiang, F.C., Makradi, A., 2007. Modeling
Bergström, J.S., Kurtz, S.M., Rimnac, C.M., Edidin, A.A., 2002. Constitutive and validation of the large deformation inelastic response of
modeling of ultra-high molecular weight polyethylene under large- amorphous polymers over a wide range of temperatures and strain
deformation and cyclic loading conditions. Biomaterials 23 (11), rates. International Journal of Solids and Structures 44 (24), 7938–
2329–2343. 7954.
Boni, S., G’Sell, C., Weynant, E., Haudin, J.M., 1982. Microscopic in situ Sallem-Idrissi, N., 2008. Comportement mécanique et évolution
observation of the plastic deformation of polybutene-1 films under structurale induite dans les films multi-couches à base de
simple shear. Polymer Testing 3 (1), 3–24. polyamide 6 et de polyéthylène. Ph.D. Thesis, Université des
Diani, J., Brieu, M., Vacherand, J.M., Rezgui, A., 2004. Directional model for Sciences et Technologies de Lille/Ecole des Mines de Douai.
isotropic and anisotropic hyperelastic rubber-like materials. Sallem-Idrissi, N., Miri, V., Elkoun, S., Krawczak, P., Lacrampe, M.F.,
Mechanics of Materials 36 (4), 313–321. Lefebvre, J.M., Séguéla, R., 2009. Trichroic infrared analysis of the
Drozdov, A.D., Christiansen, J. de C., 2008. Thermo-viscoelastic and strain-induced structural changes in the PA6 layer of PA6/PE
viscoplastic behavior of high-density polyethylene. International multilayer films under biaxial drawing. Polymer 50 (24), 5812–
Journal of Solids and Structures 45 (14–15), 4274–4288. 5823.
Dupaix, R.B., Boyce, M.C., 2007. Constitutive modeling of the finite strain Séguéla, R., Rietsch, F., 1990. Double yield point in polyethylene
behavior of amorphous polymers in and above the glass transition. under tensile loading. Journal of Materials Science Letters 9 (1),
Mechanics of Materials 39 (1), 39–52. 46–47.
Elías-Zúñiga, A., Beatty, M.F., 2002. Constitutive equations for amended Shan, G.F., Yang, W., Xie, B.H., Li, Z.M., Chen, J., Yang, M.B., 2005. Double
non-Gaussian network models of rubber elasticity. International yielding behaviors of polyamide 6 and glass bead filled polyamide 6
Journal of Engineering Science 40 (20), 2265–2294. composites. Polymer Testing 24 (6), 704–711.
Fatahi, S., Ajji, A., Lafleur, P.G., 2007. Correlation between different van Dommelen, J.A.W., Parks, D.M., Boyce, M.C., Brekelmans, W.A.M.,
microstructural parameters and tensile modulus of various Baaijens, F.P.T., 2003. Micromechanical modeling of the elasto-
polyethylene blown films. Polymer Engineering and Science 47 (9), viscoplastic behavior of semi-crystalline polymers. Journal of the
1430–1440. Mechanics and Physics of Solids 51 (3), 519–541.
268
International Journal of Non-Linear Mechanics 46 (2011) 693–702
International Journal of Non-Linear Mechanics

journal homepage: www.elsevier.com/locate/nlm
On the plastic bifurcation and post-bifurcation of axially compressed beams

Philippe Le Grognec a,, Anh Le van b
a
Ecole des Mines de Douai, Polymers and Composites Technology & Mechanical Engineering Department, 941 rue Charles Bourseul - BP 10838, 59508 Douai Cedex, France
b
GeM (Laboratory of Civil and Mechanical Engineering), Faculty of Science - University of Nantes, 2 rue de la Houssinie re - BP 92208, 44322 Nantes Cedex 3, France
a r t i c l e i n f o abstract
Article history: The paper presents two new results in the domain of the elastoplastic buckling and post-buckling of
Received 30 March 2010 beams under axial compression. (i) First, the tangent modulus critical load, the buckling mode and the
Received in revised form initial slope of the bifurcated branch are given for a Timoshenko beam (with the transverse shear
10 January 2011
effects). The result is derived from the 3D J2 flow plastic bifurcation theory with the von Mises yield
Accepted 13 February 2011
criterion and a linear isotropic hardening. (ii) Second, use is made of a specific method in order to
Available online 21 February 2011
provide the asymptotic expansion of the post-critical branch for a Euler–Bernoulli beam, exhibiting one
Keywords: new non-linear fractional term. All the analytical results are validated by finite element computations.
Axially compressed beams & 2011 Elsevier Ltd All rights reserved.
Bifurcation
Buckling
Elastoplasticity
Asymptotic developments
1. Introduction The pioneering works on the critical load of an elastoplastic

beam under axial compression dated back to the end of the nine-
Failure of slender or thin structures is mainly due to the teenth century and were conducted by Engesser, Conside re, and
buckling phenomenon and necessitates the analysis of the buck- later by von Karman. Yet, the early results before the 1940s were not
ling and post-buckling behaviors for their mechanical design, quite correct or properly justified. Some authors derived the tangent
namely the calculation of the critical loads, the bifurcation modes modulus critical load by discarding the unloading possibility in the
and the post-critical equilibrium branches. structure; others obtained the reduced modulus critical load by
The buckling of a straight beam under axial compression is the assuming that the bifurcation takes place at constant load as in
most commonly studied in the literature, either in elasticity or in elasticity. The first significant result is due to Shanley [2], who
plasticity. The first critical load of a compressed beam in elasticity provided a rational explanation for the plastic buckling of the
was found as the solution of an eigenvalue problem by Euler at so-called Shanley’s column, which had been introduced by von
the beginning of the eighteenth century. Much later on, in 1945, Karman in elasticity. The considered model is a rigid rod with two
Koiter succeeded in describing the elastic post-bifurcation beha- degrees of freedom; it is supported by two elastoplastic springs at
vior using an asymptotic power expansion. At the same time, he the bottom and subjected to an axial compressive force at the tip.
also achieved the imperfection sensitivity analysis for an elastic This discrete model supposedly able to reproduce the behavior of a
buckled beam. On the other hand, the plastic bifurcation analysis, beam cross-section did lead to results which are qualitatively similar
even on a simple model such as a beam, makes much less to those of a continuum structure under plastic buckling. Shanley
progress than in elasticity. It still is the subject of a lot of recent thus provided a satisfactory answer to the value and the nature of
theoretical as well as experimental investigations. the first critical load. In discrete or continuous structures, the first
In the sequel, we will briefly recall the main results available bifurcation occurs at the tangent modulus critical load, giving rise to
in the literature on the plastic bifurcation, with no attempt to an incipient unloaded zone and an increasing load during the initial
make an exhaustive review (see, for instance, the survey paper of post-bifurcation. Hill [3] extended these results to a 3D continuum
Sewell [1] for a general review on plastic buckling). The interested by using the concept of ‘‘comparison elastic solid’’. He examined the
reader can find more details on the theoretical developments and uniqueness and stability criteria, and pointed out the difference
more complete states-of-art in the quoted references below. between bifurcation and stability.
As regards the plasticity theory to be used for solving the plastic
buckling problem, one can choose between the J2 deformation and
Corresponding author. Tel.: + 33 3 27 71 23 21; fax: + 33 3 27 71 29 81. flow theories. These two theories may yield different critical values
E-mail addresses: philippe.le.grognec@mines-douai.fr (P. Le Grognec), and they coexist as they have each their own advantages and
anh.le-van@univ-nantes.fr (A. Le van). drawbacks. The deformation theory, although it does not take into
0020-7462/$ - see front matter & 2011 Elsevier Ltd All rights reserved.
doi:10.1016/j.ijnonlinmec.2011.02.001
269
694 P. Le Grognec, A. Le van / International Journal of Non-Linear Mechanics 46 (2011) 693–702
account the elastic unloading possibility, yields critical loads that More recently, Cheng et al. [14] proposed a simpler approach in
compare best with the experiments, whereas the flow theory gen- order to obtain the second fractional term in the beam case with a
erally overpredicts the critical values. The discrepancy—known as the rectangular cross-section (the method can be applied to plates and
plastic buckling paradox—can be accounted for and quantified shells as well). By means of a new coordinate system which is not
through an analysis of imperfection sensitivity, as can be found in directly related to the geometry of the unloaded zone, they found
Durban [4]. the known first three terms in relation (1) again, determined the
In order to study the plastic post-buckling, Hutchinson [5] fourth term, and showed that one cannot go further without
extended Koiter’s results in elasticity by using a series expansion breaking the continuity of the solution. Writing the virtual work
with fractional exponents. He improved the Shanley model by principle at different ‘‘time’’ orders enabled them to obtain a new
replacing the above-mentioned two springs system with a uniform equilibrium branch, bifurcating at the tangent modulus critical
distribution of springs. He also studied compressed beams and some value, with a new maximum load, much greater than that found
more general cases by assuming a homogeneous pre-bifurcation in earlier works. Later on, the same authors extended their analysis
state. The series expansion with fractional exponents was derived by adding a geometric imperfection [15]. They emphasized that the
laboriously and is only valid in initial post-bifurcation. Much later, elastic unloading prevails over geometric non-linearities, at least in a
Cimetie re et al. [6–9] thoroughly solved the bifurcation problem for first deformation stage. Moreover, they showed that a growing
the Shanley column and the compressed beam, and first provided the imperfection modifies the previous maximum load value and finally
necessary theoretical ingredients such as the validity, the convergence makes it completely disappear.
of the previous expansion and the existence of the post-critical Whilst the bifurcation occurs at the tangent modulus critical load
branches. In the beam case, the Hutchinson asymptotic expansion in practice, it can theoretically take place at any other point of the
giving the load parameter versus the perturbation parameter (i.e. the continuous range. Yet, only a few qualitative results are available for
bifurcation mode amplitude) was limited to the first non-linear term, the bifurcated branches arising from the critical loads other than the
because of the complexity of the coordinate system used to describe tangent modulus one. During initial post-bifurcation, the load
the evolution of the unloaded zone. In the 1980s, Nguyen and increases when bifurcation occurs between lT and lR and decreases
Stolz [10] and Akel [11] reformulated the problem within the frame- between lR and lE . By using a stability criterion based on a plastic
work of the generalized standard materials and expanded both the potential, Nguyen [16] showed that the bifurcated branches with an
load and the perturbation parameter in power series of a ‘‘time’’ increasing load (lc A ½lT , lR ) have a stable behavior and the post-
parameter. They showed that these simple power series yield the buckling behavior far from the fundamental solution depends on the
same series with fractional exponents as obtained by Hutchinson. geometric non-linearities taken into account.
Another significant result in plastic buckling problems is the
existence of continua of bifurcation points, which was discovered by
Cimetie re [12] when dealing with the case of compressed rectan- 2. Scope and outline of the paper
gular plates. These continua enable the structure to bifurcate within
intervals of critical loads, by continuously modifying the unloaded This paper is devoted to two new theoretical issues of the
zone and the structural stiffness. Each continuous range of bifurca- plastic bifurcation and post-bifurcation of axially compressed
tion points (one per mode) observed in plasticity spreads from the beams. The beams have no geometric imperfection and the
tangent modulus critical value lT to the elastic one lE . It contains a bifurcation is assumed to occur at the tangent modulus value.
particular intermediate value which is the reduced modulus critical (i) Section 3 deals with the plastic bifurcation of a Timoshenko
load lR corresponding to a constant loading bifurcation. In general, beam, where the transverse shear effects are taken into account. The
the plastic buckling may appear in three different ways, depending analysis is based on the 3D plastic bifurcation theory, using the J2
on the yield strength value ly . If the yield force precedes the first flow theory assuming the von Mises yield criterion and a linear
continuous range (ly r lT ), the plastic buckling occurs at the isotropic hardening. This 3D approach is a general and efficient way
tangent modulus critical load. If ly Z lE , the buckling is purely to cope with structures subjected to a uniform uniaxial pre-critical
elastic. Eventually, in the intermediate case when lT r ly r lE , stress state. It was successfully applied to the case of compressed
referred to as the plastic breakdown [11], the plastic buckling occurs plates and cylinders in an earlier study [17]. Here, the same
at the plastification load ly precisely. approach will provide the tangent modulus critical load, the buck-
Most results were obtained for the plastic bifurcation at the ling mode and the initial slope of the bifurcated branch for a
tangent modulus critical value. The load l acts as a bifurcation Timoshenko beam. The analytical values will be displayed together
parameter; the perturbation parameter denoted by x ðx Z 0Þ can with finite element results for a beam with a hollow rectangular
be viewed as a measure of the difference between the funda- section, where the transverse shear effects are significant.
mental and post-buckling solutions. The asymptotic expansion (ii) In Section 4, we will limit ourselves to Euler–Bernoulli
proposed by Hutchinson [5] for a general structure reads beams and aim to go further in the post-bifurcation analysis
stemming from the tangent modulus bifurcation. The goal will be
l ¼ lT þ l1 x þ l2 x1 þ b þ ð1Þ
achieved by using a specific method which provides both the
where l1 is generally positive, thus verifying the hypothesis of an fractional expansion of the post-critical branch containing one
increasing bifurcation load, l2 is negative, and the exponent b is a new non-linear term and the shape of the elastic unloading region
rational number such that 0 o b o1. The l2 o 0 condition leads one in the beam during the bifurcation process. All the analytical
to predict a load limit point on the bifurcated branch, i.e. an upper solutions presented will be compared to finite element results.
bound for the strength of the elastoplastic structure in the post-
critical behavior. In the case of a beam with a circular cross-section,
the unloading initiates at a single point and b is 1/3. For a 3. Plastic bifurcation analysis of a Timoshenko beam
rectangular cross-section, a whole line is simultaneously unloaded
and b is 2/5. It is noteworthy that in the case of a symmetric section, 3.1. Critical load and bifurcation mode
the previous expansion can be used with x o 0 by replacing x with
x in relation (1), thus making lðxÞ an even function; conversely, Let us consider a straight Timoshenko beam subjected to an
this procedure does not work for a non-symmetric section, as axial compressive force which leads to the buckling in the plastic
mentioned by Tvergaard and Needleman [13]. regime. The critical load and the bifurcation mode of the beam
270
P. Le Grognec, A. Le van / International Journal of Non-Linear Mechanics 46 (2011) 693–702 695
will be derived from a 3D approach which is in essence similar to Let us make the following assumption:
that presented in Le Grognec and Le van [17] for the plastic
buckling of plates and cylinders. The ingredients in the quoted Assumption 1.
reference will be summarized below and it will be shown how the
3D approach can also be applied to the beam case. At critical time tc, the plastic zone Op ðtc Þ corresponding to the
The theory is developed using a total Lagrangian formulation bifurcated solution is equal to that of the fundamental solution
where the beam is seen as a 3D body. In the reference configura- Opf ðtc Þ, i.e. the bifurcation takes place at the tangent modulus
tion, the beam occupies a cylindrical volume Oo of constant cross- critical load with incipient unloading.
section area S and length L. In the elastic regime, the material is The whole solid is plastified on the fundamental branch:
assumed to be isotropic, defined by the fourth-order elasticity Opf ðtc Þ ¼ Oo ð10Þ
tensor D whose components in an orthonormal basis are
Dijkl ¼ ldij dkl þ mðdik djl þ dil dkj Þ, where dij is the Kronecker symbol, which implies that the nominal tangent elastoplastic tensor is
l and m are the Lamé constants. Use is also made of Young’s equal to Kp throughout the solid. &
modulus E and Poisson’s ratio n related to l and m by standard
relations l ¼ En=ð1 þ nÞð12nÞ and m ¼ E=2ð1 þ nÞ. We assume that
With the above assumption, the critical load lc ¼ lT and the
the plastic threshold is defined by the von Mises yield function
bifurcation mode X are obtained by solving the following bifurca-
with a linear isotropic hardening:
tion equation [18,19]:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z
f ðR,AÞ ¼ 32Rd : Rd s0 A, A ¼ Hp ð2Þ 8dU, rT dU : Kp ðUf ðlc ÞÞ : rX dOo ¼ 0 ð11Þ
Oo
where R denotes the second Kirchhoff stress tensor (symmetric),
Rd its deviatoric part, s0 the yield stress, H the isotropic hard- We shall now derive more explicit expressions of the above
ening modulus (constant) and p the equivalent plastic strain. results by exploiting the uniaxial stress state in the compressed
The material tangent elastoplastic tensor is beam at hand. Let the beam be subjected in the pre-critical state
to a nominal axial compressive stress PXX ¼ l o0, so that the
@R D : @@fR @@fR : D first Kirchhoff stress tensor P is expressed in an orthonormal
Dp ¼ ¼ D ð3Þ basis (e1, e2, e3) as (we indifferently denote (X,Y,Z) and (1,2,3))
@E H þ @@fR : D : @@fR
2 3
l 0 0
where E denotes the Green strain tensor and the tensor product
P ¼ le1 e1 ¼ 6 7
4 0 0 0 5, ðl 4 0Þ ð12Þ
of two second-order tensors S and T is defined by ðS TÞijkl ¼ Sij Tkl .
Relation (3) can be recast as 0 0 0
Dp ¼ DN N ð4Þ Let us make the following assumption, which is usually

satisfied in practice.
where the symmetric tensor N is
Assumption 2. The pre-critical deformations are small:
D : @@fR
N ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð5Þ
H þ @@fR : D : @@fR JrUJ 5 1 ð13Þ
One then derives the nominal tangent elastoplastic tensor by Thus, the stress tensor R writes
Kp ¼
@P
¼F
@R T
F þ ðI:RÞT ð6Þ
R ¼ F1 P P ð14Þ
@F @E
The tensor N in Eq. (5) then simplifies as follows:
In the previous equation, F is the deformation gradient, P ¼ F R 2 3
the first Kirchhoff stress tensor (non-symmetric), I represents the 2 0 0
m m 6 7
fourth-order unit tensor ðIijkl ¼ dil dkj Þ and the superscript T the N ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðI3e1 e1 Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4 0 1 0 5 ð15Þ
H þ 3m H þ3m
major transposition of a fourth-order tensor ððDT Þijkl ¼ Dklij Þ. 0 0 1
Like (4), Eq. (6) can be rewritten as
Hence, the material tangent elastoplastic tensor in Eq. (4) reads
Kp ¼ Ke MT M ð7Þ m2
Dp ¼ D ðI3e1 e1 Þ ðI3e1 e1 Þ ð16Þ
with H þ 3m
The components of Dp in the orthonormal basis (e1, e2, e3) are

Ke ¼ F D FT þðI RÞT , M ¼ N FT ð8Þ
4m2 m2
Now, let us assume that there exists a fundamental equili- Dp1111 ¼ l þ 2m , Dp2233 ¼ l
H þ 3m H þ 3m
brium path l/Uf ðlÞ, which is the displacement solution of the
elastoplastic problem under an increasing load factor l ðl_ 4 0Þ,
m2 2m2
and that there exists a bifurcated solution U at a critical time tc Dp2222 ¼ l þ 2m , Dp1133 ¼ l þ
H þ 3m H þ 3m
corresponding to lc ¼ lðtc Þ. Furthermore, the yield stress s0 is
assumed to be small enough for the plastic strains to appear
m2 2m2
before the tangent modulus critical load is reached. Dp3333 ¼ l þ 2m , Dp1122 ¼ l þ
H þ 3m H þ 3m
The bifurcated branch is described by an asymptotic expansion
with respect to the perturbation parameter x ðx Z 0Þ:
Dp1212 ¼ Dp1313 ¼ Dp2323 ¼ m ð17Þ
l ¼ lc þ l1 x þoðxÞ
The other components are either zero or derived from Eq. (17)
using both major and minor symmetries of tensor Dp (Dpijkl ¼Dpklij ¼
U ¼ Uf ðlÞ þ xX þoðxÞ ð9Þ Dpjikl ¼Dpijlk).
271
The nominal tangent elastoplastic tensor in Eq. (7) becomes k is introduced as the transverse shear correction factor depend-
@R ing on the cross-section of the beam.
Kp þ ðI RÞT ¼ Dp lei e1 e1 ei ð18Þ Taking into account Uð0Þ ¼ Vð0Þ ¼ Yð0Þ ¼ 0, one obtains the
@E
first critical load and the corresponding bifurcation mode [20]:
which is independent of the spatial coordinates.
p2 ET I
Furthermore, when dealing with 1D models like beams, ad hoc 4L2
lc S ¼ lT S ¼ ð25Þ
assumptions are usually added in order to enforce some specific 2
1þ p ET I2
4kmSL
stress state in the body. Namely, the transverse normal material
stresses are assumed to be zero: S22 ¼ S33 ¼ 0. Taking into 8
>
> U ¼0 !
account these assumptions leads one to replace tensor Dp >
>
> 2
in (16) with the reduced tensor Cp defined as < V ¼ C 1 þ p ET I
>
1cos
pX
ðC is an arbitrary constantÞ
4kmSL2 2L
Dpij22 ðDp2233 Dp33kl Dp3333 Dp22kl Þ þ Dpij33 ðDp3322 Dp22kl Dp2222 Dp33kl Þ >
>
p >
> pC pX
Cijkl ¼ Dpijkl þ >
>Y¼ sin
Dp2222 Dp3333 Dp2233 Dp3322 :
2L 2L
ði,jÞ a ð2,2Þ,ð3,3Þ, ðk,lÞ að2,2Þ,ð3,3Þ ð19Þ ð26Þ
It can be readily checked that tensor Cp has the major and both Similar differential equations and solutions for the J2 deforma-
minor symmetries. In the sequel, we only need the following tion theory can be obtained from the above results, provided the
reduced moduli (and their equivalents obtained by major or tangent moduli are replaced by the equivalent secant moduli.
minor symmetries):
p
C1111 ¼ ET , p
C1212 p
¼ C1313 p
¼ C2323 ¼m ð20Þ 3.2. Initial slope of the bifurcated branch
where ET is the tangent elastoplastic modulus related to Young’s The initial slope of the bifurcated branch at the critical point,
modulus E and the isotropic hardening modulus H by 1=ET ¼ equal to coefficient l1 in the expansion (9), can be expressed
1=E þ1=H. under the following assumption:
Eventually, the bifurcation equation (11) writes in the uniaxial
stress case Assumption 3.
Z
8dU, rT dU : ðCp lc ei e1 e1 ei Þ : rX dOo ¼ 0 ð21Þ lc is a simple eigenvalue.
Oo
The so-called transversality coefficient Tr is positive:
Let us consider the bending problem in the xy-plane of a beam Z
dKp ðUf ðlÞÞ
built-in at end X ¼0 and compressed by load l at end X ¼L. The Tr6 rT X : : rXdOo 4 0 ð27Þ
Oo dl jl ¼ lc
Timoshenko kinematics is defined by two scalar displacement
fields U(X,t) and V(X,t), respectively the axial and transverse The plastic zone corresponding to the fundamental solution at
displacements of the centroid axis of the beam, and the cross- the critical time Opf ðtc Þð ¼ Oo Þ is active (i.e. there is loading and
section rotation yðX,tÞ, independent of deflection V. When the no neutral loading).
beam buckles from the straight position (the fundamental solu- (D 4 0,8 X A Opf ðtc Þ,MðX, lc Þ : rUf , l ðX, lc Þ Z D 4 0, where X A Oo
tion) to a bent shape, the expressions for the bifurcation mode X denotes the position of the current particle in the reference
and the displacement variation dU are chosen according to the configuration. &
Timoshenko kinematics:

UY Y dUY dy
Under this assumption, one has an angular bifurcation and the

X¼V , dU ¼ dV ð22Þ initial slope l1 is given by

0 0
MðX, lc Þ : rX
l1 ¼ max ð28Þ
The bifurcation equation (21) becomes X A Opf ðtc Þ MðX, lc Þ : rUf , l ðX, lc Þ
8dU, dV, dy satisfying dUð0Þ ¼ dVð0Þ ¼ dyð0Þ ¼ 0,
In fact, it can be readily checked that the transversality condi-
Z tion (27) is always satisfied in the uniaxial stress state. From
½ET ðU, X Y Y, X ÞðdU, X Y dy, X Þ expression (18) for the nominal tangent elastoplastic tensor, the
Oo
transversality coefficient Tr in (27) is reduced to
mV, X dymYdV, X þ mV, X dV, X þ mYdy Z
Tr ¼ jX, 1 j2 dOo ð29Þ
Oo
lc ðU, X Y Y, X ÞðdU, X Y dy, X Þlc V, X dV, X dOo ¼ 0 ð23Þ
whence, by (26), Tr is strictly positive.
First, integrating over the cross-section, then integrating by
Furthermore, from Eq. (8) and assumption (13), one has
parts with respect to X and eliminating negligible higher order
M ¼ N FT N. Hence
terms yields three local equations for the components U, V and Y
of the eigenmode: m
M : rX ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðI3e1 e1 Þ : rX
H þ 3m
ET SU, XX ¼ 0
m
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðdivX3ðrXÞ11 Þ ð30Þ
kmSðY, X V, XX Þ þ lc SV, XX ¼ 0 H þ 3m
Accordingly, the initial slope of the bifurcated branch (28)
ET IY, XX þ kmSðV, X YÞ ¼ 0 ð24Þ becomes
together with the stress boundary conditions at the end X¼ L: !
3ðrXÞ11 divX
U, X ðLÞ ¼ 0, kmSðYðLÞV, X ðLÞÞ þ lc SV, X ðLÞ ¼ 0 and Y, X ðLÞ ¼ 0, where l1 ¼ max ð31Þ
X A Oo 3ðrUcf, l Þ11 divðUcf, l Þ
S denotes the cross-section area, I the second moment of area and
272
where Ucf, l denotes the differentiation Uf , l ðX, lÞ evaluated at

l ¼ lc .
The fundamental solution for the entirely plastified beam
subjected to load l Z s0 has the same form as in small deforma-
tions:

ls0 s0
Uf ¼ þ Xx ð32Þ
ET E
From Eq. (31), one then can derive the initial slope for a beam
with a rectangular cross-section of height e in the y-direction,
choosing the value of C equal to e/2 in expression (26) for V:
!
3p2 ET I p2 ET I
l1 S ¼ ¼ 3 1þ lT S ð33Þ
4L2 4kmSL2
Remark. Making k tend to infinity in relations (25), (26) and (33)

yields the well-known expressions for the first critical load, the
Fig. 1. Plastic buckling mode of a compressed Timoshenko beam with a rectan-
corresponding bifurcation mode and the initial slope in the case gular hollow cross-section, obtained by the finite element method.
of a Euler–Bernoulli beam:
p2 ET I
lc S ¼ lT S ¼ ð34Þ
4L2
Table 1
8 Critical loads of a rectangular hollow beam under axial compression with various
< U ¼ 0
pX thicknesses.
ð35Þ
: V ¼ C 1cos 2L ðC is an arbitrary constantÞ
Thickness lc S ðNÞ lc S ðNÞ lc S ðNÞ Error (%)
(mm) (Euler– (Timoshenko) (FE) (Timoshenko
3p2 ET I Bernoulli) versus FE)
l1 S ¼ ¼ 3lT S ð36Þ
4L2
2 3.823 106 3.596 106 3.596 106 0.0005
The Timoshenko beam model provides a slightly smaller critical 3 5.735 106 5.395 106 5.444 106 0.9
5 9.558 106 8.991 106 9.11 106 1.3
load than the Euler–Bernoulli model, whereas the initial slope is
the same.
3.3. Finite element validation

1E+07
In order to check the plastic buckling load expression (25) for a
Timoshenko beam, a numerical computation is performed by
9E+06
means of a home-made finite element program. Developed within
a total Lagrangian framework, the program encompasses finite
plasticity and geometric non-linearities (finite displacements and 8E+06
rotations) in order to deal with the plastic buckling and advanced
Bernoulli
Critical load (N)
post-critical behavior of general structures. For the computation Finite Element

of interest, the beam is discretized using a finite rotation shell 7E+06
finite element which is described in full details in Le Grognec and
Timoshenko
Le van [21] in the context of cylindrical shells under axial 6E+06
compression.
Consider a beam of length L¼1 m in the x-direction, with a
rectangular hollow cross-section of height e¼ 250 mm in the 5E+06
y-direction and width b¼ 200 mm in the z-direction. The wall
thickness is 2, 3 or 5 mm. Young’s modulus is E¼2.1 1011 Pa,
4E+06
Poisson’s ratio n ¼ 0:3 and the yield stress s0 ¼ 5 107 Pa. The
isotropic hardening modulus is H¼1.5 1011 Pa, hence the tan-
gent elastoplastic modulus ET ¼ EH=ðEþ HÞ ¼ 8:75 1010 Pa (the
1.5 2 2.5 3 3.5 4 4.5 5 5.5
hardening modulus H has been given a high value in order to
Thickness (mm)
separate the different continua of bifurcation points and
thus help us to analyze the numerical results more easily). Fig. 2. Comparison between numerical and analytical critical values with or
The shear correction factor introduced in Eq. (24) takes the value without transverse shear effect.
k ¼ 5=12 0:417 usually chosen for square hollow sections.
Some rotational degrees of freedom of the shell elements are
prescribed to zero in order to prevent a local buckling mode and thicknesses. The term p2 ET I=4kmSL2 in (25) responsible for the
enforce the global mode to take place in the xy-plane. Fig. 1 difference between the two kinematical models does not depend
shows the buckling mode obtained by the finite element on the thickness of the thin-walled beam and is about 6.3%. It is
computation. found that for cross-section geometries such as the one consid-
The analytical critical loads obtained from the Timoshenko ered here—for which the transverse effects are significant—the
model (25) and the Euler–Bernoulli model (34) are compared Timoshenko model gives better results and is more relevant than
with the finite element values in Table 1 and Fig. 2 for various the Euler–Bernoulli one.
273
4. Plastic post-bifurcation analysis of a Euler–Bernoulli beam The quantities introduced above depend on s and t, and
the following notations are used for the sake of brevity:
The aim of this section is to go further in the expansion (9) and v(X,t) ¼v(s(X, a(t)),t)¼v(s,t), r(X,t)¼r(s(X,a(t)),t) ¼r(s,t), d(X,t)¼
investigate the post-bifurcation behavior of the beam. d(s(X,a(t)),t) ¼d(s,t).
The post-bifurcation analysis will be carried out for the Euler– It is shown in Nguyen [19] that functions lðtÞ, v(s,t), r(s,t) and
Bernoulli beam model only; although the Timoshenko model d(s,t) are governed by the following equation system:
could be treated using the same line of argument, the correspond-
ing equations would be much more intricate. Also, for the sake of for s A 0,1½:
definiteness, we will focus on the first bifurcation mode arising
PðdÞr 00 þt 2 L2 ¼ 0 ðaÞ
from the tangent modulus critical load of a beam, built-in at
Q ðdÞr 00 þ t 2 L2 ðlr þvÞ ¼ 0 ðbÞ
end X ¼0, with a rectangular cross-section of height e in the
s
y-direction and width b in the z-direction, see Fig. 3. l_ r ¼ vv
_ 0 ðcÞ
t
A generic approach will be presented, which enables one to
obtain one more term in comparison with Hutchinson’s asymp- where the dot denotes the derivative with respect to time, v0
totic expansion of the bifurcated branch. and r 00 stand for the first and second derivatives with respect
Let the compressed beam buckle into a bent position with to the space variable s, respectively. Functions P(d) and Q(d)
negative Y -values. The elastic unloaded zone, reduced to a single are polynomials defined as
point in the xy-plane at the critical time, is assumed to spread out Z
b
into a connected region separated from the plastic zone by a PðdÞ ¼ ðdYÞEi dS ¼ ½4ðET EÞd2 þ 4ðET þEÞed þ ðET EÞe2
S 8
smooth boundary defined by Y ¼ dðX,tÞ, X A ½0,a, as shown in Fig. 3.
Z
b
Q ðdÞ ¼ ðdYÞYEi dS ¼ ½4ðET EÞd3 þ 3ðET EÞe2 d þ ðET þ EÞe3
S 24
4.1. Power series expansions for the load and the deflection
ð38Þ
According to the procedure described in Nguyen [19], the four where S is the cross-section of the beam and Ei ¼E or ET,
following operations are performed: depending on whether the considered point in the cross-
section belongs to the elastic unloading zone or the plastic
(a) To define v(X,t)¼V (X,t) V (L,t). loading zone;
(b) To assume a non-constant load factor: l_ a 0 (in this section, for s A 1,2½:
l denotes the axial force instead of the axial stress) and
ET Ir00 þ ð1tÞ2 L2 ðlr þ vÞ ¼ 0 ðdÞ
introduce the function r(X,t) defined by
2s
l_ r ¼ vv
_ 0 ðeÞ
_
vðX,tÞ ¼ l_ ðtÞrðX,tÞ ð37Þ 1t
(c) To use the following change of variables, which enables one to furthermore, the boundary and connecting conditions write
eliminate parameter a and write the equations on fixed space e
dð1,tÞ ¼ ðfÞ r 0 ð0,tÞ ¼ 0 ðgÞ rð2,tÞ ¼ 0 ðhÞ
intervals: 2
for X A 0,a½: X ¼ as2s ¼ X=a with s A 0,1½ v0 ð0,tÞ ¼ 0 ðiÞ vð2,tÞ ¼ 0 ðjÞ
for X A a,L½: X ¼ a þ ðLaÞðs1Þ2s ¼ ððXaÞ=ðLaÞÞ þ1 with rð1 ,tÞ ¼ rð1 þ ,tÞ ðkÞ vð1 ,tÞ ¼ vð1 þ ,tÞ ðlÞ
s A 1,2½ ð1tÞr 0 ð1 ,tÞ ¼ tr 0 ð1 þ ,tÞ ðmÞ
(d) To choose as the kinematic ‘‘time’’ t the non-dimensional
infinitesimal parameter t ¼a/L, thus making the critical time tc
the initial conditions at tc ¼0 are
equal to zero. The so-defined time plays the role of the e
lð0Þ ¼ lc ðnÞ vðs,0Þ ¼ 0 ðoÞ dðs,0Þ ¼ ðpÞ
first-order infinitesimal parameter in the asymptotic expres- 2
sions for l, v, r and d.
The non-linear system (a)–(p) is solved by the asymptotic
expansion method, according to which the unknowns in hand are
y expanded into power series involving the non-dimensional small
parameter t:
a
e/2 t2
lðtÞ ¼ l0 þ l1 t þ l2 þ
2
d
0 x
t2
vðs,tÞ ¼ v1 ðsÞt þv2 ðsÞ þ
2
-e/2
L t2
rðs,tÞ ¼ r0 ðsÞ þ r1 ðsÞt þr2 ðsÞ þ
2
y t2
dðs,tÞ ¼ d0 ðsÞ þ d1 ðsÞt þd2 ðsÞ þ ð39Þ
2
Let us explain in detail the calculations up to the second-order
d(0) terms. The remaining terms can be addressed in the same way.
z
b/2 0 -b/2 Order 0
Equations of order 0 are ðP0 6Pð0Þ, Q0 6Q ð0ÞÞ:
Cross-section at x = 0 s A 0,1½:
00
P0 r0 ¼ 0
00
Q0 r 0 ¼ 0 l1 r0 ¼ v1 v01 s
00
2
Fig. 3. Unloaded zone (shaded) during the plastic buckling of a cantilever beam. s A 1,2½: ET Ir0 þ L l0 r0 ¼ 0 l1 r0 ¼ v1
274
The boundary conditions are: r 00 ð0Þ ¼ 0, r0(2)¼ 0. x ¼ 2vð0,tÞ=e as follows:

The connecting conditions are: r0 ð1 Þ ¼ r0 ð1 þ Þ, r 00 ð1 Þ ¼ 0.
t5 t7 t9
l ¼ l0 þ l5 þ l7 þ l9
5! 7! 9!
Order 1
Equations of order 1 are ðP1 6P, d ð0Þ, Q1 6Q , d ð0ÞÞ: t5 t7 t9
x ¼ x5 þ x7 þ x9 ð40Þ
s A 0,1½:
00
P0 r1 þ P1 d1 r0 ¼ 0
00
5! 7! 9!
00 00
where xi ¼ 2vi ð0Þ=e for i¼5,7,9.
Q0 r 1 þ Q1 d 1 r 0 ¼ 0 By eliminating parameter t from relations (40) in a step-by-
step manner, one gets the following asymptotic expansion with
s
l2 r0 þ l1 r1 ¼ v2 v02 fractional exponents for the load l with respect to parameter x:
2
00 l ¼ l0 1 þc1 x þ c2 x7=5 þc3 x9=5 þ o x9=5 ð41Þ
s A 1,2½: ET Ir1 þ L2 ðl1 r0 þ v1 þ l0 r1 2l0 r0 Þ ¼ 0
where
l2 r0 þ l1 r1 ¼ v2 v01 ð2sÞ
0 12=5
The boundary conditions are: d1(1)¼0, r 01 ð0Þ ¼ 0, r1(2)¼0, 15 @ 15pEET
c1 ¼ 3, c2 ¼ pffiffiffi A
v01 ð0Þ ¼ 0, v1(2)¼0. 7 4 2 1ET
E
The connecting conditions are: r1(1 ) ¼r1(1 + ), v1(1 )¼
v1(1 + ), r 01 ð1 Þ ¼ r 00 ð1 Þ þ r 00 ð1 þ Þ. ET 0 14=5
65 48 pEET A
c3 ¼ E @ p15
ffiffiffi ð42Þ
Order 2 ET 4 2 1EET
42
Equations of order 2 are ðP2 6P, dd ð0Þ, Q2 6Q , dd ð0ÞÞ: E
00 00 00
s A 0,1½: P0 r2 þ 2P1 d1 r1 þ ðP1 d2 þP2 d21 Þr0 þ 2L2 ¼ 0 Expressions for c1 and c2 agree well with those proposed
by Cheng et al. [14]. On the other hand, our expression
00 00 00
Q0 r2 þ 2Q1 d1 r1 þ ðQ1 d2 þ Q2 d21 Þr0 þ2L2 l0 r0 ¼ 0 for c3 slightlypffiffiffidiffers from that of Cheng et al., which is
17 4=5
42ð15pðET =EÞ=4 2ð1ET =EÞÞ . They would be the same if one
s made ET ¼E in the first ratio in (42).
l3 r0 þ 2l2 r1 þ l1 r2 ¼ v3 v03
3 It should be noted that the fractional expansion (41) is only
00 valid if the maximum deflection-to-thickness ratio x ¼ 2vð0,tÞ=e is
sA 1,2½: ET Ir2 þ L2 ðl2 r0 þv2 þ 2l1 r1 8l1 r0 þ l0 r2 4l0 r1 þ 2l0 r0 Þ ¼ 0
small, in the same way as the time parameter t ¼a/L related to it
l3 r0 þ 2l2 r1 þ l1 r2 ¼ v3 ðv02 þ 2v01 Þð2sÞ
The boundary conditions are: d2(1)¼0, r 02 ð0Þ ¼ 0, r2(2)¼0,
Table 3
v02 ð0Þ ¼ 0, v2(2)¼0. Coefficients for deflection v in expansion (39).
The connecting conditions are: r2(1 )¼r2(1 + ), v2(1 )¼ v2(1 + ),
r 2 ð1 Þ ¼ 2ðr 01 ð1 Þ þr 01 ð1 þ ÞÞ.
0 v1 0
Equations (a)–(p) are solved sequentially, giving the li values v2 0
v3 0
together with the functions vi(s), ri(s) and di(s) involved in
v4 0
expansions (39). At order 0, one gets r0(s)¼C for sA ½0,1, where 8 4
>
> p eðEET Þ
C is unknown so far. At order 1, one gets r0(s) for sA ½1,2, and the >
< s A ½0,1
8ET
same expression for the critical stress l0 ¼ lT as in (34). At order v5 4
> p eðEET Þ
>
>
ps
: sin s A ½1,2
2, the value for constant C is determined together with v1(s) ¼0 8ET 2
8
for sA ½0,2 and l1 ¼ 0, and so on. Solving equations at order <0
> s A ½0,1
3 provides d1(s)¼0 for s A ½0,1, r1(s) for s A ½0,2, v2(s) ¼0 for v6 3p5 eðEET Þ ps
>
: ð2sÞcos s A ½1,2
8ET 2
s A ½0,2 and l2 ¼ 0. The calculations are carried out in a similar 8 6
>
> p e
way up to the 10th order. The higher the considered order, the >
>
< 32E2
ðEET Þð36E þ 40ET 21s2 ET Þ s A ½0,1
T
more intricate and lengthy the equations. Without going into v7 > 6
>
>
p e 2 ps
details, we give the resulting load factor l and deflection v, as : 32E2 ðEET Þð36E44ET þ 84sET 21s ET Þsin 2 s A ½1,2
>
T
summarized in Tables 2 and 3. 8
Lastly, expressions for the plastic boundary di(s¼0) (i¼ 0–8) in <0
> s A ½0,1
7
v8 > p e ðEE Þ½72E þ 24E þ sð36E þ 44E Þ42s2 E þ 7s3 E cosps s A ½1,2
the expansions (39) can be obtained simultaneously with coeffi- : 8E2
T
T T T T T
2
cients li and functions vi(s), as shown in Table 4. 8 8
>
> 3p e
>
> ðEET Þ½480E2 972EET þ 521E2T þ 24s2 ET ð9E10ET Þ þ 21s4 E2T s A ½0,1
>
> 64E3T
>
>
<
4.2. Asymptotic expansion of the bifurcated branch 3p8 e
v9 > 3
ðEET Þ½480E2 108EET 103E2T 288sET ð3EET Þ
>
> 64ET
>
>
>
> ps
From Tables 2 and 3, one can rewrite the truncated expansions >
: þ 24s2 ET ð9E þ 11ET Þ168s3 E2T þ 21s4 E2T sin s A ½1,2
2
for the load l and the maximum deflection-to-thickness ratio
Table 2
Coefficients for load factor l in expansion (39).
l0 l1 l2 l3 l4 l5 l6 l7 l8 l9
p2 ET I 0 0 0 0 3p6 ðEET ÞI 0 3p8 ðEET Þð36E25ET ÞI 0 9p10 ðEET Þð240E2 402EET þ 173E2T ÞI

4L2 16L2 64ET L2 64E2T L2
275
Table 4
Coefficients for the plastic boundary d(0,t) in expansion (39).
d0(0) d1(0) d2(0) d3(0) d4(0) d5(0) d6(0) d7(0) d8(0)
e 0 p2 e 0 p4 e 0 p6 e 63p6 e p8 e
2 ð6E5ET Þ ð360E2 579EET þ 221E2T Þ ðEET Þ ð12 600E3 29 853E2 ET þ 22 851EE2T 5597E3T Þ
8 32ET 256E2T 16ET 512E3T
0.012
0.011 1.03
0.01 ic
1.025 etr
0.009 ram
Pa
0.008 1.02
0.007 9/5
λ/λ0
0.006 1.015
ξ
0.005
0.004 1.01
0.003
1.005
0.002
0.001 1
0
0 0.1 0.2 0.3 0.4 0.5 0 0.002 0.004 0.006 0.008 0.01
t = a/L ξ
Fig. 4. Validity of expansion (41) in terms of the maximum deflection-to-thickness ratio x ¼ 2vð0,tÞ=e.
by the second relation in (40). Let us consider, as an example, the Table 5

validity domain in x for a beam with E/ET ¼2.4 (the same material Data for the finite element computation of the bifurcated branch.
as in Subsection 3.3).
L e b E ET s0
Fig. 4 displays x versus t and l=lT versus x, making use of
relations (40)–(42). As can be seen in Fig. 4(a), at the beginning 0.5 m 1 mm 5 cm 2.1 10 11
Pa 10
8.75 10 Pa 5 104 Pa
of the bifurcation process, the elastic unloading region with
length a spreads out very quickly up to about a third of the beam
length, while x remains very small, of the order of 0.001. methods [22–25] in order to bifurcate from a specific point of the
Moreover, Fig. 4(b) shows that the fractional expansion up to continuous bifurcation range and pass through load or displace-
the 9/5-order (41) can be identified to the parametric expansion ment limit points on the bifurcated paths.
up to the ninth order (40) with parameter t if x is small enough. The material is the same as in Subsections 3.3 and 4.2, except
Thus, in the sequel, we will limit ourselves to x r 0:01 (corre- for the yield stress: as the beam is very thin in the y-direction, s0
sponding to t r 0:51), so that the difference between the frac- has been given a very small value for the plastification to occur
tional and parameter expansions do not exceed 0.7%. before the beam buckles at the tangent modulus critical load lT .
The ratio E/ET ¼2.4 is small enough to obtain separate continua
of bifurcation points and simplify the analysis. In the case of a
4.3. Finite element validation rectangular cross-section, the so-called reduced modulus can be
expressed as follows:
Several finite element computations are presented hereafter in 4EET
order to check the previous analytical results obtained for the ER ¼ pffiffiffi pffiffiffiffiffi 2 ¼ 1:29 1011 Pa ð43Þ
ð E þ ET Þ
plastic post-buckling of a beam. The numerical computations are
conducted on a simply-supported beam, which actually gives In each continuous range of bifurcation points, the critical
the same results as a cantilever beam because of symmetry. The loads corresponding to the three moduli—the tangent modulus,
geometry and material data are given in Table 5 with the nota- the reduced modulus and the elastic modulus—are in ascending
tions defined in Fig. 3. One tip is fixed in the x-direction and the order:
other is loaded by a uniform compressive force l. Only the middle
n2 p2 ET I n2 p2 ER I n2 p2 EI
node at each tip is fixed in the z-direction, which enables a lTn ¼ , lRn ¼ , lEn ¼ ð44Þ
4L2 4L2 4L2
uniform stress state in the beam during the pre-critical deforma-
tions. Since the thickness is very thin, the numerical solution of whence the numerical values for the first mode:
the bending problem can be performed using the shell finite lT1 ¼ 3:6N, lR1 ¼ 5:3N, lE1 ¼ 8:64N ð45Þ
element described in Le Grognec and Le van [21]. One single finite
element in the z-direction is used, whereas 50 elements along the 4.3.1. Elastoplastic bifurcation diagrams
length are necessary to correctly represent finite rotations. Use In plasticity, each point of the continuous bifurcation range is a
is made of two Gaussian points in the x- and z-directions and potential bifurcation point, and the corresponding non-trivial rate
four Gaussian points through the thickness y-direction. Use is solution involves an elastic unloading zone whose shape depends
made of the same computer program as in Subsection 3.3. on the compressive force value. At the tangent modulus critical
Here, one also needs additional branch-switching and arc-length load, the concept of ‘‘comparison elastic solid’’ allows one to use
276
the same approach as in elasticity: the critical load lT is detected height e¼ 10 cm along the y-axis, the width b¼1 cm along the z-axis,
when the stiffness matrix becomes singular and the bifurcated and enforce the bending to take place in the xy-plane by prescribing
branch (which starts with a single unloading point) can be all the out-of-plane degrees of freedom to zero, see Fig. 7.
predicted by the corresponding eigenmode. If one wishes to The material is the same as before, except for s0 which is again
bifurcate at another point in the continuous range of bifurcation 5 107 Pa. The mesh for the new geometry involves 30 elements
points, the prediction is based on the eigenmode corresponding to along the x-direction and four elements along the y-direction. In each
the smallest eigenvalue of the tangent stiffness matrix. element, use is made of seven Gaussian points in both x- and
The obtained numerical results show that this procedure y-directions. Only two Gaussian points along the z-direction are used
enables one to correctly bifurcate in the whole continuous range for the numerical integration through the thickness, as the elasto-
of bifurcation points. In agreement with the well-known theore- plastic state is uniform along that direction. In theory, the plastic
tical results, the obtained curves are bounded by those corre- points are those with positive plastic strain rates. Here, in the
sponding to the tangent modulus critical load lT and the elastic numerical computations, they are detected in a discrete way: since
critical load lE . Fig. 5 plots the maximum displacement at the the unloading zone keeps growing (at least in the vicinity of the
middle of the beam versus the compressive force. It can be seen bifurcation point), the plastic points at increment n are numerically
that the bifurcation takes place with an increasing load when detected by comparing the equivalent plastic strains between steps n
lc A ½lT , lR ½, with a decreasing load when lc A lR , lE , and with a and n+1.
constant load at lR , as predicted by the theory. At the tangent modulus critical load, the unloading zone
In the initial post-critical zone (small deflections), say spreads out from a single point, as one advances on the bifurcated
V o 0:0005 m in Fig. 5(a), all the bifurcated branches converge branch. Applying the procedure above, the unloading zones
towards the asymptotic curve l ¼ lR , as predicted by the analy- obtained by numerical computations (symmetric with respect to
tical solution of El Koulani [7]. However, Fig. 5(b) shows that, in
the advanced post-critical zone where the deflections are larger
and the geometric non-linearities prevail, the bifurcated branches 3.7
tend towards another asymptote.
Let us now focus on the bifurcated branch arising from the first 3.69
tangent modulus critical load lT1 . Fig. 6 displays four different
3.68 7/5
ar
post-critical curves l ¼ lðxÞ:
ne
Li
3.67
the curve obtained from the finite element results;
the analytical fractional curve (41) including two non-linear 3.66
9/5
terms with exponents 7/5 and 9/5;
for comparison purposes, the fractional curves with only the 3.65
λ (N)
linear term and with the 7/5-order term, as obtained by nt

3.64 leme
Hutchinson [5]. it ee
Fin
All the fractional curves are drawn for x r 0:01, in accordance 3.63
with the validity domain in x, as discussed in Subsection
4.2. Fig. 6 shows that the finite element results are in very 3.62
good agreement with the fractional expansion (41). Over the 3.61
considered range of x, the difference is less than 0.1% in
absolute value. 3.6
3.59
4.3.2. Plastic boundary and unloading zone 0 0.002 0.004 0.006 0.008 0.01
Let us focus now on the evolution of the shape of the ξ
unloading zone in the beam along the bifurcated branch arising
Fig. 6. Tangent modulus bifurcation: comparison between different fractional
from the tangent modulus critical load. expansions and the finite element results. Linear: curve with only the linear term
Again, consider a simply-supported beam of length l¼ 2L¼1 m. in (41); 7/5: curve with terms up to exponent 7/5; 9/5: curve with terms up to
Here, in order to make the unloading zone visible, let us now take the exponent 9/5.
0.002 0.015
El Koulani’s analytical El Koulani’s analytical
asymptote (1996) 0.01 asymptote (1996)
0.001
0.005
V (m)
V (m)
0 λT
0
λE λT λR λE
λR -0.005
-0.001
-0.01
-0.002 -0.015
0 2 4 6 8 10 0 2 4 6 8 10
λ (Ν) λ (Ν)
Fig. 5. Elastoplastic bifurcation diagrams: (a) initial post-buckling and (b) advanced post-buckling.
277
λ For all practical purposes, there would be no need to go further

Y
than the 9/5-order term in the fractional expansion.
Z X
References
[1] M.J. Sewell, A survey of plastic buckling, in: Stability, University of Waterloo
Press, 1972, pp. 85–197.
[2] F.R. Shanley, Inelastic column theory, International Journal of Aeronautical
Sciences 14 (1947) 261–267.
[3] R. Hill, A general theory of uniqueness and stability in elastic-plastic solids,
Journal of the Mechanics and Physics of Solids 6 (1958) 236–249.
[4] D. Durban, Plastic buckling of plates and shells, in: Stability Analysis of Plates
λ and Shells, NASA CP-1998-206280, 1998, pp. 293–311.
[5] J.W. Hutchinson, Plastic buckling, Advances in Applied Mechanics 14 (1974)
Fig. 7. Geometry, mesh and boundary conditions for the analysis of the 67–144.
unloading zone. [6] A. Cimetie re, A. El Koulani, A. Léger, Buckling and post-buckling of an
elastoplastic simple model (Flambage naissant et post-flambage d’un mode le
simple en élastoplasticité), Comptes Rendus de l’Académie des Sciences de
d(0) 2a Paris 319 (1994).
[7] A. El Koulani, Continuation in moving boundary problems of the plastic
bifurcation type (Continuation dans les proble mes a frontie res libres de type
bifurcations plastiques), Ph.D. Thesis, University of Metz, France, 1996.
[8] A. El Koulani, A. Léger, Bifurcated solutions of the initial velocity problem for
Numerical result an elastoplastic beam (Solutions bifurquées du proble me en vitesses initiales
d(0) pour une poutre élastoplastique), Comptes Rendus de l’Académie des
2a Sciences de Paris 322 (1996).
[9] A. El Koulani, A. Léger, Post-critical behavior of elastoplastic beams: existence
and regularity of the bifurcated branches (Comportement post-critique des
poutres élastoplastiques: existence et régularité des branches bifurquées),
Analytical result Comptes Rendus de l’Académie des Sciences de Paris 324 (1997).
[10] Q.S. Nguyen, C. Stolz, On the asymptotic development method in plastic
Fig. 8. Unloading zone for the bifurcated branch arising from the tangent modulus buckling (Sur la méthode de développement asymptotique en flambage
critical load (first mode). plastique), Comptes Rendus de l’Académie des Sciences de Paris 300 (1985).
[11] S. Akel, On the buckling of elastoplastic structures (Sur le flambage des
structures élastoplastiques), Ph.D. Thesis, Ecole Nationale des Ponts et
the median plan of the beam) are in very good agreement with Chaussées, France, 1987.
the theoretical ones, as shown in Fig. 8 for the same load level. [12] A. Cimetie re, On the modelling and buckling of elastoplastic plates (Sur la
modélisation et le flambage des plaques élastoplastiques), Ph.D. Thesis,
University of Poitiers, France, 1987.
5. Conclusions [13] V. Tvergaard, A. Needleman, On the buckling of elastic-plastic columns with
asymmetric cross-sections, International Journal of Mechanical Sciences 17
Two new issues have been investigated in the field of the (1975) 419–424.
[14] Y.S. Cheng, W.D. Lu, H. Fang, Plastic post-buckling of a simply-supported
elastoplastic buckling and post-buckling of beams under axial column with a solid rectangular cross-section, Applied Mathematics and
compression. Mechanics 16 (8) (1995) 713–722.
(i) The first issue is the plastic bifurcation of a Timoshenko [15] Y.S. Cheng, H. Fang, W.D. Lu, Imperfection sensitivity analysis of a rectangular
column compressed into the plastic range, Applied Mathematics and
beam, where the transverse shear effects are taken into account. Mechanics 19 (1) (1998) 9–14.
The analysis is based on the 3D plastic bifurcation theory and [16] Q.S. Nguyen, Bifurcation and stability of irreversible systems obeying to the
proves to be an efficient way to deal with structures subjected to maximum dissipation principle (Bifurcation et stabilité des syste mes irré-
a uniform uniaxial pre-critical stress state, as shown in an earlier versibles obéissant au principe de dissipation maximale), Journal de Méca-
nique Théorique et Appliquée 3 (1) (1984) 41–61.
study on compressed plates and cylinders [17]. Here, the same 3D [17] P. Le Grognec, A. Le van, Some new analytical results for plastic buckling and
approach has been applied to a Timoshenko beam and enabled initial post-buckling of plates and cylinders under uniform compression,
one to obtain the tangent modulus critical load, the buckling Thin-Walled Structures 47 (8–9) (2009) 879–889.
[18] J.W. Hutchinson, On the post-buckling behavior of imperfection-sensitive
mode and the initial slope of the bifurcated branch. The trans- structures in the plastic range, Journal of Applied Mechanics 39 (1972)
verse shear effects may be significant depending on the cross- 155–162.
section geometry, as clearly shown in the finite element compu- [19] Q.S. Nguyen, Stability and Non-Linear Solid Mechanics, Wiley, 2000.
[20] P. Le Grognec, A. Le van, A unified treatment for the elastoplastic bifurcation
tation with a hollow rectangular section, where the Timoshenko
of structural elements, in: 21st International Congress of Theoretical and
theory gives better results than the Euler–Bernoulli one. Applied Mechanics Warsaw, Poland, August 15–21, 2004.
(ii) The second issue of the paper is a further analysis of the [21] P. Le Grognec, A. Le van, Elastoplastic bifurcation and collapse of axially
post-bifurcation of a Euler–Bernoulli beam, stemming from the loaded cylindrical shells, International Journal of Solids and Structures 45 (1)
(2008) 64–86.
tangent modulus bifurcation. Use has been made of a specific [22] E. Riks, An incremental approach to the solution of snapping and buckling
method in order to provide the fractional expansion of the post- problems, International Journal of Solids and Structures 15 (7) (1979)
critical branch, exhibiting one new non-linear term. Also, the 529–551.
[23] E. Riks, On formulations on path-following techniques for structural stability
shape of the elastic unloading region in the beam during the
analysis, in: New Advances in Computational Structural Mechanics, Elsevier,
bifurcation process has been obtained in the form of an asympto- 1991, pp. 65–80.
tic expansion. The finite element computations are in very good [24] W.F. Lam, C.T. Morley, Arc-length method for passing limit points in
agreement with the presented analytical results. structural calculation, Journal of Structural Engineering 118 (1) (1992)
169–185.
Two conclusions may be drawn from the fractional expansion [25] R. Seydel, Practical Bifurcation and Stability Analysis. From Equilibrium to
study of the post-bifurcated branch: Chaos, Springer-Verlag, 1994.
Compared to the 7/5-order term, the 9/5-order one brings

about a slight change and a better description of the post-
critical behavior.
278
Composites: Part A 42 (2011) 1390–1397
Composites: Part A
journal homepage: www.elsevier.com/locate/compositesa
Modeling of high speed RTM injection with highly reactive resin with on-line mixing
M. Deléglise a, P. Le Grognec a, C. Binetruy a,⇑, P. Krawczak a, B. Claude b
a
Ecole des Mines de Douai, Technology of Polymers and Composites & Mechanical Engineering Department, 941 rue Charles Bourseul, 59508 Douai Cedex, France
b
Renault, Polymer Engineering Department, 1 avenue du golf, 78288 Guyancourt Cedex, France
Article history: Structural composite manufacturing in automotive industry tends towards short cycle times for being
Received 9 February 2011 competitive with other material solutions. Development towards resin transfer molding process can be
Received in revised form 17 May 2011 considered when highly reactive resin with short curing cycle is used. Mixing of the resin system com-
Accepted 5 June 2011
ponents is then held at the mold inlet to reduce catalyzed resin life time. Cure kinetics and viscosity
Available online 14 June 2011
changes induced during filling are thus to be taken into account when injection process simulation is con-
sidered. In the case of on-line mixing of resin components, viscosity and degree of cure of the first
Keywords:
injected resin droplets are not the same than for the following ones. This topic will be addressed here
E. Resin transfer molding
C. Analytical modeling
for mixed constant flow rate and pressure injection schemes. A solution that could be implemented in
C. Computational modeling a RTM simulation package was developed and validated with an analytical solution and on a complex
B. Cure behavior shape automotive part demonstrator.
1. Introduction possible cycle times, but with an unavoidable incidence on part

quality if this reduction does not result from a dedicated analysis.
Due to their special features, such as high specific strength and The challenge consists of making robust the production process for
rigidity, composite materials are increasingly being used to replace economic manufacture at high volume by an improved under-
metallic components. They offer the potential to save weight, in- standing of the material/process/property relationships.
crease mechanical properties and shaping freedom, reduce the Resin transfer molding is generally known as a long cycle pro-
number of components in a part, and allow properties to be tai- cess and reserved to low to medium volume manufacturing batch.
lored to applications. For all of these reasons, composites are In this process, a dry fibrous preform is placed in a closed mold,
increasingly being used while also achieving a reduction in part and then resin is forced through the preform during the injection
cost. The transportation industry represents a large application step. Resin cures under the conjugated effect of heat and catalyst
area for fiber reinforced composites and is driven by numerous to form the final part that can be de-molded. The curing time usu-
interacting requirements that depend on the specific sector consid- ally represents 80% of the total cycle time, which places RTM
ered. For example, aerospace and aeronautics have traditionally among the medium series production processes. The automotive
been driven by high performance, where longer cycle times and in- industry is however interested in that technique to develop struc-
creased scrap levels are permitted. In the automotive industry tural composite parts. In order to render RTM competitive and
where high volume applications are looked for, fast and highly highly productive, cycle times need to be reduced by increasing
automated techniques are required, with a strong emphasis on the injection speed and reducing cure cycle, which infers mold
decreasing production and product costs. Thermoplastic-based heating to reduce resin viscosity during filling and to catalyze poly-
composites are generally suited to lower cycle times and hence merization kinetics. A beneficial effect for the injection is thus to
higher volume production due to the nature of the matrix, com- have a fast resin cure right after mold filling.
pared to thermoset matrices which require during processing time Considering equipment, RTM injection processes are generally
consuming cross-linking reaction. For polymer composites to com- conducted at a constant injection speed with injection pumps in
pete with metallic parts, cost reductions in manufacturing cycles order to insure filling times and production rates, explaining why
are necessary to balance the higher raw material cost. From a man- only few studies focused on constant pressure injection schemes.
ufacturing point of view, it is desirable to process the part at the However, a pressure limit is fixed to avoid mold opening or fiber
lowest possible pressures and temperatures and with the shortest preform washing. Limited pressure level is also required when con-
sidering injection on a flexible material or on an inflatable bladder
⇑ Corresponding author. Tel.: +33 (0)3 27 71 21 75; fax: +33 (0)3 27 71 29 81. in order to avoid deformation of a foam core when producing sand-
E-mail address: christophe.binetruy@mines-douai.fr (C. Binetruy). wich structures or parts with hollow cores [1,2]. In those particular
1359-835X/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compositesa.2011.06.002
279
M. Deléglise et al. / Composites: Part A 42 (2011) 1390–1397 1391
but not seldom cases, constant flow rate injection can transform to first injected resin droplets are not the same than for the following
a constant pressure injection. Filling time and start of the curing ones. This situation differs from the case where the full volume of
kinetics then become an issue. resin components are mixed together prior to injection where all
Generally, resin systems and molding conditions are chosen so the resin droplets share the same initial curing condition.
as to allow the complete filling of the mold before the reaction Finally, the simulation code is used for filling time prediction of
kinetics is too much advanced. In order for RTM process to be com- an automotive composite part produced in industrial conditions.
petitive in the automotive industry, cycle times have to be short, so Processes in the composite injection industry usually imply mixed
that the part can be removed from the mold cavity as early as pos- injection scheme, generally a constant flow rate injection is con-
sible. Resins used for this type of application are thus highly reac- ducted until inlet pressure reaches the maximum set level to avoid
tive, and cure kinetics may be started before the end of the filling fiber washing or mold deflection, and then injection is switched to
stage. The objective is to have a part cured when the last resin the constant pressure mode to complete the injection. Both
droplet is injected. Solutions have been proposed where the ratio schemes are addressed in the presented work.
of catalyst evolves with time as the mold is being filled [3–6],
the last resin droplets being more catalyzed than the first ones. 2. Analytical approach
The purpose of this solution was to have a homogeneous curing
throughout the part. This solution requires knowing the effect of Fast injection for composite part manufacturing with a highly
the amount of catalyst on resin kinetics and on the final degree reactive resin in a heated mold is considered. The resin system is
of cure reached so as to insure good part quality. In the matter dis- composed of two pre-heated components and mixing is conducted
cussed here, a constant resin to catalyst ratio system is injected in a at the mold inlet, so that no reaction kinetics is involved prior to
heated mold, the two components being mixed together right at injection. The mold is heated and maintained at a constant temper-
the mold inlet. Resin kinetics thus starts as soon as the resin enters ature during the injection. Because resin system is pre-heated at
the mold, affecting the viscosity and thus the molding parameters, the mold temperature, heat transfer effects at the mold inlet can
i.e. time and pressure. be neglected. The simulation is thus considered isothermal. In that
Optimization of the process cycle then implies accurate predic- case, resin kinetics can be described through the evolution of resin
tion of the injection time to insure complete mold filling before viscosity with time at a desired temperature.
gelling of the resin. During fast injection techniques such as in The parameters employed in this section are arbitrary and will
SRIM processes, viscosity changes are either considered non rele- only be used in the purpose of deriving a realistic analytical solu-
vant during injection stage or mixing of the resin components is tion, although no resin systems may fulfill the properties employed.
previously performed in a barrel [7,8]. However, before gelling of Derivation of an analytical solution is performed according to
the resin, polymerization has been initiated, inducing a viscosity the following assumptions:
increase during mold filling which generally can be expressed in
a generic constitutive equation with an exponential law [9–11]. – Resin viscosity change according to a given temperature, cure
Cure kinetics during composite manufacturing can be well de- and time is known.
scribed, taking into account the heat generated during the curing – Injection is conducted in isothermal conditions. Only viscosity
reaction, the part quality in terms of final degree of cure or residual change due to resin kinetics is considered.
stresses [12–20]. However, few developments consider the evolu- – The mold is rectangular and is injected via a line gate, which
tion of degree of cure or viscosity during mold filling in the case allows a 1D representation of the macroscopic fluid flow
of on-line mixing [21]. In liquid molding processes, gelling of the velocity.
resin starts after mold filling, whether the injection conditions
are isothermal or non-isothermal. So resin viscosity is considered Filling of a mold containing a porous medium is modeled with
constant (isothermal case) or varying with time and temperature Darcy’s law that links the pressure gradient rP in the mold to
(non-isothermal case) during filling. The proposed article considers the volume averaged velocity v, also known as Darcy’s velocity:
the isothermal filling with a highly reactive resin inducing a time
and space variation of viscosity during filling of the mold as resin Q K
v¼ ¼ rP ð1Þ
components are mixed directly at the injection head. A l
The process dealt with in this paper is placed between RIM pro-
where Q is the flow rate across the cross section A, l is the resin vis-
cess where reactive resin is injected at high speed in a mold but re-
cosity, and K the fibrous preform permeability tensor. Neglecting
sin properties are considered constant, and non-isothermal mold
the delay of saturation of fiber tows and density change, assuming
filling where constant inlet conditions are kept but heat exchange
fabrics are not deformed by the fluid flow, the mass conservation
due to a temperature difference between the mold and the resin
principle is expressed as:
and resin kinetics occur, influencing the resin viscosity level during
infusion. rv ¼0 ð2Þ
Non-isothermal injection modeling generally includes heat ex-
the governing equation for injection through fabrics viewed as a
change between mold and the resin/fibrous medium structure
porous medium becomes:
and the heat of resin reaction. Resolution of heat and degree of cure

advancement is linked to the constitutive equation through the K
viscosity term expressed as a function of heat only when cure r rP ¼0 ð3Þ
l
kinetics are considered negligible in the injection time, as a func-
tion of the degree of conversion when heat increased is only due In this analytical study, l is not constant, which leads to a more
to cure kinetics [13,15,17,20,21]. intricate expression of the filling time than usual. Let us assume
Prior to describe the algorithm to be implemented in the simu- that l depends on the polymerization time t p , following the expo-
lation code for flow pattern prediction, an analytical solution is de- nential form:
rived to predict for a 1D flow the transport of the degree of curing l ¼ l0 ektp ð4Þ
through the evolution of viscosity and meant to be compared with
the numerical solution. The specific case of on-line mixing of resin where l0 is the initial resin viscosity and k is a constant. This
components is addresses where viscosity and degree of cure of the expression gives a general representation of viscosity time history
280
1392 M. Deléglise et al. / Composites: Part A 42 (2011) 1390–1397
2
under isothermal curing [9–11]. This assumption has the advantage /l x
to allow derivation of an analytical solution for the constant pres- PðxÞ ¼ l expðktÞ exp k ð10Þ
t 2 kK 0 vr
sure injection scheme presented in Section 2.2. So the only purpose
of this viscosity equation form is to allow the derivation of an ana- or
lytical solution to validate the implementation of the simulation of
/v 2r x
an injection with space and time varying viscosity. The simulation PðxÞ ¼ l0 exp k ð11Þ
kK vr
code could then be used for any kind of viscosity time history
profile. where v r ¼ tinj
L
is the constant resin velocity and tinj the total filling
In the case of a 1D plate filled linearly along the x direction time of the total mold length L.
(mold length), as presented in Fig. 1, the filling of the porous med-
ium can be simply deduced from Eq. (1) as follows: 2.2. Constant pressure injection
@P /l
¼ vr ð5Þ
Next consideration is for processes involving constant pressure
@x K
injection schemes. When the viscosity l remains constant during
where / is the preform porosity, K is the permeability of the pre-
the process, even in the case of a prescribed constant pressure
form in the flow direction (x) and vr is the resin interstitial velocity
Pinj , the filling time is obtained by a straightforward integration
along the mold length defined by:
over the total length of the mold L and can be expressed as:
v
vr ¼ ð6Þ 1 /l 2
/ tinj ¼ L ð12Þ
2 KP inj
Then the mass conservation equation, taking in account the
above-cited assumptions, simply writes: Let us now consider the case where viscosity change must be ta-
ken into account. First, the current position x of a resin drop-
@v r let along the 1D flow direction, at time t, can be related to the
¼0 ð7Þ
@x corresponding polymerization time:
which allows to express the uniform velocity easier as follows: Z t
dl
x¼ vr dt ¼ lðtÞ lðt t p Þ ð13Þ
vr ¼ ð8Þ tt p
dt
In other words, it means that the particle considered at time t
thus only involving the position l of the flow front. has polymerized from the mold inlet to its current position x, what
From this process description, two boundary condition cases corresponds to a duration of tp. As the resin velocity is uniform (it
will be addressed separately, the constant flow rate injection and only depends on time), the traveled distance corresponds to the
the constant pressure injection strategies. flow front advance between t tp and the current time t.
By differentiating Eq. (13) with respect to tp, it gives:
2.1. Constant flow rate injection
dx ¼ dlðt t p Þ ¼ v r ðt t p Þdtp ð14Þ
For a constant flow rate injection, resin velocity is constant and By integrating Eq. (5), it leads to the nominal prescribed
time is directly linked to the flow extent via the resin velocity. pressure:
Injection time will not be affected by the resin viscosity increase Z Z
l l
during filling, however pressure field along the mold will be greatly @P /
Pnom ¼ dx ¼ v r l dx ð15Þ
modified. This pressure profile is derived from Eq. (5) that can be 0 @x K 0
written as:
Using Eq. (14), we get the following integro-differential equa-
@P /v r x tion, only involving the time-dependent velocity:
¼ l0 exp k ð9Þ
@x K vr Z t
KPinj
lðtp Þv r ðt tp Þdtp ¼ ð16Þ
Due to the constant resin velocity, the polymerization time here 0 /v r ðtÞ
is simply related to the position of the given resin fluid particle. At
Let us note wðtÞ the left-hand side of Eq. (16). Thanks to the vis-
each location, resin viscosity will remain constant until the mold is
cosity definition (Eq. (4)):
completely filled and resin front is stopped. The pressure field
along the mold length is obtained by integrating Eq. (9) from the
_
wðtÞ ¼ l0 v r ðtÞ þ kwðtÞ ð17Þ
mold inlet to the resin flow front located at a distance x = l where
an atmospheric pressure condition is assumed, thus obtaining:
where the dot superscript denotes a time derivative.
Finally, vr is solution of the ordinary differential equation:
l0 / 3
v_ r þ kv r þ v ¼0 ð18Þ
KPinj r
l Flow front location
and takes the following general expression:
1
v r ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l /
ffi ð19Þ
a e2kt kKP0 inj
where a is a constant of integration such that the velocity tends to

Fibrous infinity at initial instant t = 0:
preform l0 /
a¼ ð20Þ
Fig. 1. 1D flow representation. kKP inj
281
Then, the flow front position can be deduced, by integrating

Initial parameters (K, µ) and boundary conditions (gate
Eq. (19):
and vent location, injection flow rate)
sffiffiffiffiffiffiffiffiffiffiffi
KP inj pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l¼ arctanð e2kt 1Þ ð21Þ
kl0 / Flow front advance
with a null constant of integration in order to ensure l(0) = 0.

The analytical prediction of the filling time of the total length L
of the mold is then straightforward: New nodes no
Injection time filled ?
" sffiffiffiffiffiffiffiffiffiffiffi!#
1 kl0 /
t inj ¼ ln 1 þ tan2 L ð22Þ yes
2k KPinj
New viscosity
The exponential expression presented in Eq. (4) to represent the
at nodes
evolution of viscosity with time has the advantage of offering an
easily derived analytical solution. However, caution should be ta-
End of the
ken with the expression obtained. According to the analytical solu- K/µ update simulation
tion, there is a mathematical singularity without physical meaning
for which no solution can be derived and it occurs when theqtan-ffiffiffiffiffiffiffiffi Fig. 2. Constant flow rate injection with on-line resin components mixing
gent term goes to infinity, namely when the expression L kKPl0 / algorithm.
inj
equals an odd multiple of p2 . This first occurs when:
flow front advances in the mold. A time step is calculated and
p2 KPinj the filling status of the nodes is updated. The algorithm will run
k¼ 2 ð23Þ
4L l0 / a test on the nodes to see if their status has changed during the cal-
culation step. If they are newly filled, the elapsed time since the
Coefficient k is thus chosen with care for implementation of the
beginning of the injection is taken to calculate the new viscosity
analytical solution for validation of the numerical study.
such as expressed in Eq. (4) and is updated on the mesh properties.
3. Numerical approach
3.2. Constant pressure case
The main issue is the implementation of space and time varying
viscosity variation during injection. Actually, the first resin droplet Imposing a constant pressure at the injection gate implies that
injected will remain in the mold longer than the last resin droplet. resin velocity is not constant during mold filling, or more exactly
The part will then cure first close to the mold outlet and then at the that the travel time for a resin droplet from the injection gate to
mold inlet, which also implies that resin viscosity at the mold inlet a considered location increases. As for the constant flow rate injec-
and that at the mold outlet are not the same. The aim of the algo- tion case, spatial viscosity variations are considered, but for a con-
rithm is thus to consider the elapsed time spent in the mold for stant pressure injection the time increase to travel a given length
each resin droplet to be able to evaluate the resin viscosity at needs to be taken into account.
any time and location in the mold. The algorithm described here The algorithm for viscosity calculation for a constant pressure
can be implemented in any simulation code without changing injection is thus based on the calculation of the travel time for each
the fundamental equations. An open source code was used for resin droplet. Due to on-line mixing of the resin components, each
development of the method [22]. resin droplet has a different initial curing condition. In order to
Two cases are presented separately, the constant flow rate avoid particle tracking, the algorithm will consider the evolution
injection scheme and the constant pressure injection scheme. of the travel time for a resin droplet to reach a specific position
in the mold. A first injection simulation is performed in order to re-
cord the volume injected before each node is filled. During a sec-
3.1. Constant flow rate case
ond injection simulation, for each calculation step and for each
node that is filled, comparison of the injected volume with the vol-
Viscosity changes are first applied to mesh properties. In the
ume recorded during the first injection allows to define the viscos-
case of LIMS software, resin viscosity is a global input for the entire
ity change by considering the time taken to inject the considered
mesh. Viscosity variations are thus introduced through the perme-
volume from the viscosity equation expressed as a function of
ability defined for each mesh element. The ratio Kl appearing in
time. Fig. 3 highlights the main steps of that algorithm. The impor-
Darcy’s law (Eq. (1)) is thus considered as a whole. Change in vis-
tant parameter retrieved from the first injection is the volume in-
cosity is transferred to a change in permeability values without
jected to reach a particular node. During the effective simulation,
affecting the computed solution. In this section, only simple parts,
at each flow front advance, the time step and the volume injected
defined as parts that will induce a constant local velocity at all
during that time step are recorded. For each filled node, a sum on
locations in the mold, are considered. When a change of resin
the latest injected volumes and a sum on the corresponding time
velocity is possible, the numerical approach should be conducted
steps are conducted until the injected volume defined from the
as for a constant pressure injection scheme, described in the next
first injection is reached. Time is then retrieved and is used to cal-
paragraph.
culate the new local viscosity at the considered node.
In the case of a constant flow rate injection, viscosity only de-
pends on the spatial location. At each location, the time taken for
a resin droplet to flow from the inlet to the considered location will 3.3. Validation
be the same for the first droplet or any following droplet. The time-
dependent viscosity can then be transferred to a spatially depen- The numerical algorithm developed is tested on a plate, the
dent viscosity. The only output value required is thus the wetting objective being to compare numerical and analytical results in
time at every node location in the mesh. The general algorithm terms of flow front advancement during filling so as to validate
developed is schematized in Fig. 2. At each calculation step, the the implementation scheme of the spatial and time dependence
282
Preliminary simulation

and vent location, injection pressure)
Flow front advance
New nodes no
filled ?
yes
Matrix update : list of filled nodes and

corresponding injected volume
End of the
simulation
Effective simulation

and vent location, injection pressure)
Flow front advance
New nodes
filled ? Back calculate time to
yes
inject the volume defined
from the first injection to
End of the reach each filled node
no
simulation
Update K/µ map
Fig. 3. Constant pressure injection scheme with on-line resin components mixing algorithm.
of viscosity through the evaluation of the length of the path fol- the trial and according to the resin viscosity evolution with time
lowed by resin. characterization.
A line injection is considered to derive an analytical solution.
The part is 0.5 m long, 0.2 m wide, and 2 mm thick. Preform per-
meability and fiber volume fraction are set to 1 108 m2 and 4. Study on an automotive industrial part
45% respectively. A constant 0.5 MPa injection pressure is applied.
The viscosity change with time considered is represented by Eq. (4) Implementation of the algorithm for constant pressure injection
with l0 = 0.1 Pa s and for two cases, where k is equal to 5 and 8. was used on a demonstrator representing a low B-pillar part using
Filling profiles for both cases are presented in Fig. 4, through the an industrial high-speed RTM process. Injection was performed in
representation of the flow front location along the mold length an aluminum mold heated at 80 °C. The final part presented in
with time derived from the analytical solution (Eq. (22)) and from Fig. 5 is manufactured with two preforms with an overlap section
the numerical study. A perfect match between analytical solution with a 3 mm thick cavity in the overlap zone and 1.5 mm thick cav-
and the implemented code is obtained. The constant viscosity case ity elsewhere. Permeabilities used were characterized in [2]. The
is also considered using a 0.1 Pa s viscosity, equivalent to l0. For permeabilities used for given fiber volume fractions Vf are reported
k ¼ 5; the highest viscosity in the mold is of 0.27 Pa s inducing in Table 1. Materials were shown to be in-plane isotropic. No sig-
an increase in the filling time of 25% compared to the constant vis- nificant shear was induced as no draping was done directly on
cosity injection case. For k ¼ 8; the highest viscosity in the mold is the mold. Moreover, the part is composed of two preforms with
1.70 Pa s, the final injection time in that case is 144% longer than simple deformation (simple curvature and short drawing effect),
for the constant viscosity case. In both cases, perfect match is ob- limiting in-plane shear of the fabric. On the contrary, local thick-
tained between the analytical and the numerical injection profile, ness variations and race tracking channels had to be taken into
as shown in Fig. 4. account. Changes in thicknesses were considered with a propor-
The numerical code is thus validated and can be applied on an tional law on fiber volume fraction. The equivalent permeability
industrial part, with the effective injection parameters used during Kc of the race tracking channel of height h is evaluated by:
283
Fig. 4. Numerical validation of the code: (a) with k = 5; (b) with k = 8.
Vent Table 1
Preform characteristics.
Vent / Gate location Preform 1.5 mm 3 mm
1.5 mm Vf Permeability Vf Permeability
(%) (m2) (%) (m2)
8 mm
Preform #1 glass biaxial stitched 38 1.6 109 41 2.5 1010
3 mm fabric
Preform #2 hybrid preform (glass 50 4.6 1011 53 3.8 1011
Gate 3 and carbon biaxial stitched
Gate 2 fabric)
time of the resin is then set at 67 s. The resin viscosity was tested
Injection Line with Brookfield viscosity measurement apparatus at constant tem-
perature fixed to be 80 °C, temperature of the mold during injec-
tion. Viscosity measurement was held until resin gel time. Epoxy
Gate 1 resin was considered Newtonian, as it is generally observed for
epoxy resin [25].
Fig. 5. Meshed part and zones of different thickness.
Best fit of the viscosity time history curve was obtained with
2 two polynomial curves:
h
Kc ¼ ð24Þ
12 For t < 50 s l ¼ 3 104 t2 0:0024t þ 0:93 ð25Þ
For this demonstrator, preforms were not fully adapted to the
mold geometry and a small gap created between the preform For t > 50 s l ¼ 247:5t2 2176t þ 4:3 105 ð26Þ
and the mold wall edges had to be taken into account for flow sim-
ulation. Resin will flow faster in those race tracking channels, dis- The injection unit is controlled for a constant flow rate injection but
turbing the flow front shape. This consideration is fully handled by switches to a constant pressure injection scheme when it reaches a
the code by evaluating the size of the race tracking channel and specified pressure value. This value is set to 1 MPa. During injection,
evaluating an equivalent permeability [23]. pressure increase at the inlet is recorded over time and is
A two components epoxy resin, pre-heated to 60 °C, is consid-
1000
ered. Mixing of the two components is performed at the mold inlet,
thus no polymerization reaction takes place before injection. Mold
is kept at the constant temperature of 80 °C. As the part thickness 100
Viscosity (Pa.s)
can be considered small compared to the in-plane characteristic

length of the part, resin temperature is considered constant at
the mold temperature after entering the mold. Afterwards, when 10
temperature increase generated by the heat reaction occurs, the
heat is assumed to be dissipated through the metallic mold walls
and it will not affect resin viscosity. Thus, only resin viscosity 1
0 20 40 60
change with time at a given temperature can be considered [24].
Resin viscosity evolution versus time during polymerization at
0.1
80 °C was characterized, and reported in Fig. 6. Resin viscosity is Time (s)
stable and close to 1 Pa s during 20 s and then starts rising slowly
up to 1.25 Pa s at 36 s before resin kinetics is clearly initiated. Gel Fig. 6. Viscosity versus time at 80 °C.
284
12 implemented as a boundary condition in the numerical code. This

actual inlet pressure profile is represented in Fig. 7. Simulations
10
are then run using the constant pressure implementation scheme
Pressure (bar)
8 and compared with the injection time and/or profiles obtained on

the injected parts.
6
Multiport injection offers the possibility to reduce mold pres-
4 sure, run faster injections or inject larger parts. Several scenarios
were tested among isochronal and sequential strategies. The injec-
2
tion gates are the one located on a line-injection channel at the
0 bottom of the part (Gate 1) and the two pin gates while a point
0 20 40 60 80 vent is located at the top. Two families of injection were con-
Time (s) ducted, one until complete filling of the part whenever process
conditions make it possible and the second one is stopped after
Fig. 7. Recorded inlet pressure profile versus time and used as boundary condition
30s (short shot) in order to compare the observed flow front pat-
in numerical simulations.
tern at that time with the one obtained from the simulation. The
part obtained for preform #1 is shown in Table 2 along with the
Table 2
Comparison between simulated and injected flow patterns.
Produced parts Simulated flow patterns

Preform 1 – time t = 30 s
3 Injection points: Gate 1 opened at time t = 0; Gate 2 at t = 6 s and Gate 3 at t = 12 s
Preform 2/case 1 – time t = 67 s

3 Injection points: Gate 1, 2 and 3 opened at time t = 0
Preform 2/case 2 – time t = 67 s

3 Injection points: Gate 1, 2, 3 and 4 opened at time t = 0
285
simulated flow profile after 30 s. The flow front shape shows good take into account changes in the flow front path that can occur
agreement with the simulation. Complete filling of the glass fiber for complex part shape manufacturing when the flow is split due
part made with preform #1 was obtained after 57 s during trial, to presence of inserts or non homogeneous filling for a radial injec-
while simulation gives a total filling time of 58 s. The same injection when a vent is closed before others.
tion conducted with a constant viscosity of 1 Pa s, corresponding
to the initial resin viscosity, gives an injection time of 50 s, 13% less Acknowledgments
time than with the varying viscosity model.
Injection of the hybrid glass/carbon fiber preform #2 was con- The authors would like to thank Renault and Alcan Airex for
ducted using the line injection and the two secondary injection providing the automotive part mold and the trials required for
gates (case 1 in Table 2). Because of the low permeability of this the use of the code and the University of Perrugia for resin viscosity
preform, injection velocity is slower in this case than in the glass data. This study was carried within the frame of the SuperLIGHT-
fiber preform case. For preform #2 in Table 2, flow stopped after CAR collaborative Research & Development project funded by the
67 s due to the resin polymerization and the viscosity increase European Commission under the 6th Framework Program.
(Fig. 6). The injection pressure at that level is not sufficient to in-
duce a flow front advance. The part obtained is shown in Table 2 References
along with the simulated flow profile after 67 s. Again, the flow
front shape predicted by the simulation is in reasonable agreement [1] Binetruy C, Advani SG. Foam core deformation during liquid molding of
sandwich structures: modeling and experimental analysis. J Sandwich Struct
with the one obtained during the real process. Finally a fourth inlet Mater 2003;5(4):351–76.
gate is added in case 2 to impregnate preform #2, however it has [2] Deleglise M, Binetruy C, Krawczak P. Simulation of LCM processes involving
been shown that the hybrid preform #2 with lower permeabilities induced or forced deformations. Composites: Part A 2006;37(6):874–80.
[3] Comas-Cardona S, Ziaee S, Advani SG. Spatially homogeneous gelation in liquid
could not be fully impregnated with the current gating system. The composite molding. Polym Eng Sci 2002;42:1667–73.
filling patterns reported in Table 2 show a good agreement be- [4] Devillard M, Laut A, Advani SG. On-line mixing during injection and
tween the predicted and the real injection flow front patterns. simultaneous curing in liquid composite molding processes. Polym Compos
2005;26(1):74–83.
The numerical flow advancements correlate well with experimen- [5] Shojaei A, Farrahinia H, Pishvaie SMR. Effect of system of initiators on the
tal profiles. The difference between experimental and numerical process cycle of nonisothermal resin transfer molding–numerical
flow profiles are due to the non regularity of the preform cut and investigation. Composites Part A: Appl Sci Manuf 2010;41(1):138–45.
[6] Kim SK, Kim DH, Daniel IM. Optimal control of accelerator concentration for
the positioning of the preform in the mold that is made by hand
resin transfer molding process. Int J Heat Mass Transfer 2003;46:3747–54.
and is not accurate. This results in pinching the preform in some [7] Polushkin EY, Polushkina OM, Malkin AYa, Kulichikhin VG, Michaeli W, Kleba I,
places, inducing a local increase in fiber volume fraction and by et al. Modeling of structural reaction injection molding. Part II: Comparison
with experimental data. Polym Eng Sci 2002;42(4):846–58.
consequence a decrease in local permeability. The experimental
[8] Kuznetsov V, Malkin AYa, Michaeli W, Kleba I, Blaurock J. Modeling of
flow front is thus disturbed whereas a regular flow is numerically structural reaction injection molding. Part I: Mathematical model. Polym Eng
obtained. Sci 2001;41:850.
[9] Stolin AM, Merzhanov AG, Ya A. Non-isothermal phenomena in polymer
engineering and science: a review. Part II: Non-isothermal phenomena in
polymer deformation. Polym Eng Sci 1979;19(5):1074–80.
5. Conclusion [10] Polushkin EY, Polushkina OM, Malkin AYa, Michaeli W, Kleba I, Blaurock J. The
rheokinetics of a fast-curing polyurethane. Int J Appl Mech Eng 1999;4:85–90.
[11] Chick JP, Rudd CD, Van Leeuwen PA, Frenay TI. Material characterization for
Simulation of the injection of a highly reactive resin leads to
flow modeling in structural reaction injection molding. Polym Compos
additional complexity when considering an on-line mixing injec- 1996;17(1):124–35.
tion head and constant pressure driven mode. This unusual injec- [12] Ngo ND, Tamma KK. Non-isothermal 2D flow/3D thermal developments
tion scheme is motivated by the specific requirements of high encompassing process modelling of composites: flow/thermal/cure
formulations and validations. Int J Numer Methods Eng 2001;50:1559–85.
speed and robust manufacturing of structural composite parts in [13] Fei Shi, Xianghai Dong. 3D numerical simulation of filling and curing processes
automotive industry. Viscosity changes are not spatially homoge- in non-isothermal RTM process cycle. Finite Elem Anal Des 2011;47:764–70.
neous any longer and should be considered when resin kinetics [14] Shojaei A, Ghaffarian SR. Modeling and simulation approaches in the resin
transfer molding process: a review. Polym Compos 2003;4(4):525–44.
is involved during the injection phase. A methodology for predic- [15] Lin R, Lee LJ, Liou M. Non-isothermal mold filling and curing simulation in thin
tion of flow pattern and pressure field is proposed in this paper. cavities with preplaced fiber mats. Int Polym Process 1991;VI:356–69.
Digital simulation software is used in combination with the pro- [16] Mal, Couniot, Dupret. Non-isothermal simulation of the resin transfer
moulding process. Composites Part A: Appl Sci Manuf 1998;29:189–98.
posed algorithm by considering viscosity as a function of time for [17] Farina, Preziosi, Summerscales. Non-isothermal injection molding with resin
each location in the mould. Although development was performed cure and preform deformability. Composites Part A: Appl Sci Manuf
for two injection cases, constant flow rate and constant pressure 2000;31:1355–72.
[18] Ledru Y, Bernhart G, Piquet R, Schmidt F, Michel L. Coupled visco-mechanical
injections, mixed injection scheme can also be considered by com-
and diffusion void growth modelling during composite curing. Compos Sci
bining the two algorithms. The implementation algorithm thus Technol 2010;70(15):2139–45.
presented could be adapted to other commercial codes dedicated [19] Bruschke M, Advani SG. A finite element/control volume approach to mold
filling in anisotropic porous media. Polym Compos 1990;11:398–405.
to the numerical simulation of RTM process.
[20] Kendall KN, Rudd CD, Owen MJ, Middleton. Characterisation of the resin
This article presents resin kinetics as a function of time, when transfer moulding process. Compos Manuf 1992;3(4):235–42.
constant temperature can be considered. For thick parts, tempera- [21] Markus H, Breyer C, Niedermeier M, Ermanni P. A new kinetic and viscosity
ture variation due to the exothermic reaction of cure needs to be model for liquid composite molding simulations in an industrial environment.
Polym Compos 2004;25(3):255–69.
considered and the use of a non-isothermal model is needed along [22] Simacek P, Advani SG. A numerical model to predict fiber tow saturation
with resin characterization. In that case, special care is required for during liquid composite molding. Compos Sci Technol 2003;63:1725–36.
characterization of the resin and fabric. Non-isothermal modeling [23] Bickerton S, Advani SG. Characterization and modeling of race-tracking in
liquid composite molding processes. Compos Sci Technol 1999;59:2215–29.
of the on-line mixing high speed RTM process is also required for [24] Trochu F, Ruiz E, Achim V, Soukane S. Advanced numerical simulation of liquid
thick preforms where the heat generated during polymerization composite molding for process analysis and optimization. Composites: Part A
due to mass effect is not negligible. In addition, this algorithm is 2006;37:890–902.
[25] Advani SG, Sozer EM. Process modeling in composites manufacturing. New
based on the assumption that the resin travel path does not deviate York: Marcel Dekker; 2003.
during mold filling. Evolution of algorithm would be needed to
286
Composite Structures 93 (2011) 2795–2807
Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
Homogenization of the elastic behavior of a layer-to-layer angle-interlock composite

Patrick Lapeyronnie a, Philippe Le Grognec a,⇑, Christophe Binétruy a, François Boussu b
a
Ecole des Mines de Douai, Polymers and Composites Technology & Mechanical Engineering Department, 941 rue Charles Bourseul – BP 10838, 59508 Douai Cedex, France
b
ENSAIT, GEMTEX (Laboratory of Engineering on Textile Materials), 2 Allée Louise et Victor Champier, 59056 Roubaix, France
Article history: Composite materials reinforced with 3D layer-to-layer angle-interlock fabrics are increasingly employed
Available online 12 June 2011 due to their significant resistance to delamination and impact damage, which is not observed in classical
2D laminated composites. However, the prediction of the mechanical behavior of such composites is
Keywords: challenging due to the intricate fibrous architecture. The structure is intimately linked to its history of
Angle-interlock composite manufacturing which induces changes in the reinforcement geometry. The purpose of this work is to
Periodic homogenization assess the equivalent membrane and bending elastic moduli of the shell-type structure by an asymptotic
Yarn damage
homogenization procedure on a periodic unit cell, in the framework of the Love–Kirchhoff plate theory. A
Love–Kirchhoff plate
Finite element modeling
specific Python program using Abaqus software package is developed, allowing for parameterized geo-
metrical modeling and mechanical analysis in a systematic and efficient way. This modeling and simula-
tion tool enables to consider the real composite architecture after infusion and the yarn damage during
weaving. The effective properties are finally validated using numerical computations on 3D heteroge-
neous plates and by comparison with experimental tests.
1. Introduction (40–50%), due to the presence of unavoidable spaces between

yarns is inherent to 3D structures, with a rare threshold of 60% [6].
3D reinforcements have been studied for many years in replace- Among 3D interlock structures, the layer-to-layer angle-
ment of laminate structures in composite materials. For instance, interlock configuration provides the best answer to these mechan-
3D interlock geometrical arrangements seem to be a more suitable ical requirements together with formability [3]. But very few
solution to the delamination and impact damage, among other op- industrial applications involving these types of reinforcement were
tions, than a classical stacking of 2D plies. Accordingly, 3D woven attempted until now. The main reason is the present lack of knowl-
fabrics may substitute 2D weaves in many applications where they edge about their mechanical behavior together with the difficulties
are not appropriate anymore. Indeed, numerous advantages of 3D encountered when dealing with their geometrical modeling. Their
reinforced composites have already been pointed out. Complex use just recently began to spread in aeronautics and aerospace
near-net-shape fabrics can be manufactured with better mechani- structures (parts prone to impacts, such as rotor blades and landing
cal properties and a decrease of the cutting needs as well as a gears). The analysis of ballistic protection for bullet-proof vests and
reduction of the manufacturing and material costs [1]. They also vehicles is under consideration. Future applications could be front
display a better ballistic and low speed impact resistance with im- ends and leading edges of ships and boat hulls, sensitive to slam-
proved post-impact properties and a good resistance to crack prop- ming. All these applications are revealing the potential of 3D inter-
agation [2]. Cox and Dadkhah put forward a better delamination lock composites and show the growing interest of industrials for
resistance, higher ultimate tensile and compressive strains and a this type of materials.
good formability, compared to 2D-ply based laminates [3]. Kuo The use of 3D composites requires a realistic geometrical and
et al. pointed out the almost elimination of weak damaged planes material representation in order to perform an accurate mechani-
by dissipating loads in all the directions [4]. However, it is worth cal analysis. To do so, the study of the weaving and of the manufac-
mentioning a few drawbacks. In 3D reinforcements, the crimp turing process leads to a proper understanding of the final
due to the network of yarns is much more important than in 2D composite structure. During the weaving of a 3D fabric, a global
weaves and so is the damage due to the weaving and handling of distortion may occur because of its thickness, the residual stresses
yarns [5]. Furthermore, an intermediate fiber volume fraction in the yarns and the interlacing [7]. Contrary to the single ply of a
2D woven fabric, the layers of a 3D fabric may translate horizon-
tally due to the take-up motion (Fig. 1). Another modification is
⇑ Corresponding author. Tel.: +33 3 27 71 23 21; fax: +33 3 27 71 29 81. yarn crimp which is a local deformation of the yarn path, due to
E-mail address: philippe.le.grognec@mines-douai.fr (P. Le Grognec). yarn interlacing and tension. Yarn crimp may be emphasized
doi:10.1016/j.compstruct.2011.05.025
287
2796 P. Lapeyronnie et al. / Composite Structures 93 (2011) 2795–2807
Fig. 1. Idealized architecture (a) and modified architecture after taking-up from the loom (b) [7].
during manufacturing processes (RTM, for example) due to com- and weft layers with different orientations. That decomposition is
paction and nesting [8,9]. This phenomenon is already observed clearly inappropriate here since it does not take into account the
in laminate manufacturing [10] and may be more important in dis- waviness of the woven yarns and the ply interconnectivity. Thus,
torted yarns of 3D woven fabrics (Fig. 2). the classical laminate theory cannot be used anymore in the case
All these modifications of the architecture (crimp, compaction, of interlocks or 3D fabrics in general.
nesting), inherent to the manufacturing of 3D composites, lead to Specific semi-analytical and numerical approaches have been
a decrease of the mechanical properties. Their consideration in a proposed to study the mechanical behavior of 3D reinforced com-
model is necessary to improve the prediction of the mechanical posites, as a replacement for the well-known laminate theory. The
behavior. Orientation Averaging Model is an analytical approach developed
As far as data are concerned, not only the geometry but also the by Cox et al. and based on a volume averaging of the straight yarn
material properties play a major role in the mechanical analysis of matrices [3]. The waviness is taken into account through rigidity
a textile reinforced composite. The influence of yarn damage due to loss factors for each type of yarn, known as knock-down factors.
the weaving is an important feature which should be taken into ac- Some approximate numerical averaging approaches have also
count to get closer to the true homogenized behavior of the com- been developed to assess the elastic properties of such composites
posite material. That damage may be much more important in a using discretization methods such as the XYZ model [13], or using
3D weaving than for a classical 2D one. Lee et al. studied the tow direct methods of 3D calculation on a modified structure with sim-
damage during the weaving of 3D carbon-fiber orthogonal fabrics pler material properties, such as the Binary Model [14]. In both
[11]. It is due to abrasion and breaking of fibers when sliding in models, the in-plane and through-the-thickness moduli are still
the loom. The authors compared single carbon composite yarns be- overestimated, due to a coarse geometrical modeling (assembled
fore and after weaving and pointed out a decrease of 12% of the micro-blocks for the XYZ model or yarns with 2-node elements
mechanical strength. Archer et al. also studied the damage occur- for the Binary Model).
rence during the weaving of carbon-fiber angle-interlock fabrics More recently, specific textile modeling softwares, such as
[12]. The tensile strength of dry yarns from the fabric was found Wisetex [15], TexGen [2] or TexEng, have been developed. They
to be 9.4% lower than the nominal value before weaving. The generally include both geometrical modeling (up to the filament
importance of damage in this study will be discussed in more scale) and mechanical simulation. For example, numerical results
details in Sections 3.2 and 4.1. in Wisetex generally show a good consistency with experimental
The final architecture in the composite material is thus data but may often lead to discrepancies due to the limitations of
dependent of the parameters of the weaving and of the composite the homogenization scheme employed.
manufacturing process. It is essential to note that a complete In this work, the objective is to ease the mechanical study of a 3D
mechanical study must be based on the whole manufacturing his- interlock composite, using a high-performance tool. The final state
tory. Most of the authors only consider one of the aspects of the of the interlock fabric in the composite depends on the manufactur-
manufacturing previously cited. As for example, some mechanical ing parameters (weaving and process), which may produce an
methods are listed below. architecture different from the original definition before weaving.
The classical laminate theory is the basis of 2D composite Thus, on account of numerous existing geometries, a parameterized
mechanics and may be used as a starting point in the assessment architecture is developed by means of a Python program, using
of the effective mechanical behavior of 3D interlock composites. Abaqus software in the context of the finite element method. In-
Indeed, in a first approach, the interlock fabric could be considered deed, due to the strong heterogeneity and the complex architecture
as a laminate of 2D fabrics constituted of identical successive warp of such materials, a rigorous periodic homogenization scheme is
Fig. 2. Waviness change (a) and cross-section variation (b) due to compaction [10].
288
P. Lapeyronnie et al. / Composite Structures 93 (2011) 2795–2807 2797
preferred here to assess the membrane and bending elastic behav-

iors of the equivalent shell-type structure, in the framework of the
Love–Kirchhoff plate theory. Moreover, special attention is paid to
the actual textile geometry and material parameters thanks to opti-
cal observations and measurements of mechanical properties of
yarns extracted from woven fabrics. Hence, this paper proposes a
general numerical tool facilitating the meso-scale geometrical
modeling and the determination of the effective elastic properties
of a 3D interlock composite. The numerical solutions are then com-
pared to experimental tensile and bending results for validation.
2. Composite materials reinforced with 3D interlock woven

fabrics
3D interlock structures are fabrics, which means that they are

directly woven in the three dimensions with Z-yarns along the
thickness linking different layers together. For that matter, Bogd-
anovich and Mohamed pointed out the difference between multi- Fig. 3. Layer-to-layer angle-interlock architecture.
layer textiles which generate 3D structures thanks to the stacking
of plies and fabrics like interlocks which are basically 3D [16].
Many types of architectures can be defined but there is no general 3. Influence of manufacturing parameters on the definition of
data base for now. Boussu et al. recently suggested a unique the 3D interlock composite
parameterized architecture thanks to a complete inventory of the
geometric parameters in order to improve the representation and An analysis of the weaving and infusion processes is required
classification of 3D interlock structures and better identify the ade- for a correct description of the geometry and material properties
quate architecture relative to given mechanical requirements [17]. of the reinforced composite. First, the weaving itself causes a mod-
Indeed, all these parameters, among which the number of layers, ification of the architecture compared to the theoretical one, and
the number of yarns by weave repeat and the binding depth, are yarn damage. Then, the infusion process is also supposed to affect
essential in any geometrical and/or mechanical modeling. Three the final geometry of the reinforcement in the composite material
families can be distinguished among the 3D woven structures: due to transverse compaction.
– the orthogonal interlock where the Z-yarns are perpendicular to 3.1. Effect of tensioning on the architecture during weaving
the weaving plane and go through the whole thickness only
between two columns of weft yarns; All the fabrics used in this study are produced with the labora-
– the through-thickness angle-interlock where the Z-yarns go tory-scale ARM-Patronic loom, dedicated to the weaving of fabric
through the whole thickness across more than two columns of samples (Fig. 4). The fibers are made of carbon HexTowÒ IM7 from
weft yarns; Hexcel (Table 2).
– the layer-to-layer angle-interlock where the Z-yarns link at The weaving on classical looms can be divided into the follow-
least two plies. ing steps: warp beam winding (or warp creel set-up), warp let-off,
warp tensioning, shedding, weft insertion, beating, and taking-up
The latter configuration seems to be more suitable to fulfill of the finished fabric [18]. In this case, a few differences are note-
many mechanical requirements and formability ones. Moreover, worthy. There is no warp beam winding and let-off. The pristine
as the orthogonal and through-thickness angle-interlock structures bobbins are set up on a creel, the yarns pass through an align-
have been studied by many authors up to now [4,6,13,14], the ment device and then through the eyes of the heedles. Thanks
choice of the little-known layer-to-layer architecture has been re- to the successive raising and lowering of the warp yarns, the weft
tained in this study as it allows to investigate the effect of the ori- yarns can be inserted and beaten. The advancement of the weav-
entation and waviness of binding yarns on the mechanical ing is possible thanks to the translation of the set heedles/reed.
behavior of the composite material. The produced reinforcement has binding warp yarns and straight
In the sequel, the architecture considered is an 8-layer rein- weft ones and exhibits some regularity problems. Thereby a
forcement with carbon fibers. The repeated pattern is defined as tensioning system is added at the end of the loom in order to
follows: the weft binders link two successive plies together. The get a more regular and controlled fabric. But the respective roles
paths of the yarns and the section shape are defined using geomet- of the yarns are reversed with this final tension: the warp yarns
rical parameters, such as the fiber volume fraction, waviness, crimp become straight whereas the weft ones turn out to be the linking
and nesting of the yarns due to the weaving and composite pro- yarns. This new architecture is then studied in the sequel. This
cess. A 3D realistic representation of the interlock architecture reversal is noticed through the cross-section observations (Figs. 6
has been defined on the basis of optical observations and measure- and 5) and it may change the material strength according to the
ments achieved on cross-sections of the composite (Fig. 3). direction. This geometrical change has been implemented in the
3D interlock fabrics are usually woven on classical 2D looms proposed numerical tool in order to keep a wide range of
[16]. The reinforced composite can then be manufactured by interlock representations.
several Liquid Composite Molding (LCM) processes: RTM, VARTM The effect of tensioning on the cross-section shape is also ana-
or Infusion. The latter has been selected because of its growing lyzed. The first interlock fabric displayed lenticular shape for both
popularity in the industry for the ease of preparation thanks to types of yarn. The tensioning creates two section shapes: lenticular
the lack of the upper part of the mould and the low-level deforma- for the warp yarns and trapezoidal for the weft ones. This change is
tion of the structure when submitted to moderate vacuum. This also due to the beating, dependent of the manual action of the
manufacturing process is discussed in more details in Section 3.3. weaver. The straight warp yarns make up parallel channels where
289
Fig. 4. Scheme of a weaving loom for 3D interlock fabric.
as the number of layers. Warp tows open a shed thanks to the

ertical movements of the heedles in order to stack the weft yarns
on the same vertical orientation.
As an illustration, in Fig. 7, the warp yarns are moved in order to
open a path for the manual insertion of the weft yarn 1, and then
replaced to lock in this yarn and open a path for the weft yarn 2,
and so on for the whole column of weft yarns. Those repeated yarn
motions through the eyes of the heedles give rise to damage by
friction (Fig. 8): in the eyes, against the heedles and against the
other yarns (the wide warp tows are in contact). The warp fila-
ments are weakened due to abrasion. The effect of the reed makes
Fig. 5. Fabric woven without tensioning system – wavy warp yarns.
the damage deeper due to the friction against the warp yarns un-
der tension. The weft tows undergo another effect of the loom.
They are submitted in the rapier to multiple torsion efforts during
the shedding.
As a consequence, the material properties of the yarns and
therefore those of the manufactured fabric and composite decrease
significantly.
The second influence is the waviness of the weft yarns. They are
inserted in the shed and blocked by the warp tows. Then they be-
come wavy as the warp ones are still straight. At the interlacing
points between the warp and weft yarns, a flexural loading occurs
in the latter. As the yarn is thin (6 k carbon yarn), the curvature ra-
dius is low, what may create stress concentration regions and so
promote damage. Furthermore, during the beating, the reed packs
two or three times the weft yarn, what might also increase the pre-
vious damage.
Fig. 6. Fabric woven with tensioning system – straight warp yarns. The recognition and the quantization of these phenomena are
important in order to make the numerical model closer to the real-
ity of the composite. Tensile tests are thus performed on single
the untightened weft yarns conform to each other with a trapezoi- yarns before and after weaving to evaluate the stiffness decrease
dal section shape. This high density of weft yarns allows better due to the weaving (results are presented in Section 4.1).
mechanical properties in this direction. It is essential to keep in
mind that the second architecture is a particular case, due to ten- 3.3. Composite manufacturing process
sion and beating, of a more general geometry with binding warp
yarns and lenticular cross-sections for all yarns. The tensioning Composite samples are manufactured prior to characterization.
parameter allows to change the architecture without modifying The choice of infusion (vacuum pressure inferior to 0.1 MPa) was
the definition of the fabric. In the sequel, a lenticular section shape made among the different LCM processes available, as it is a simple
is attributed to all the yarns in order to keep the general purpose of closed-mould technique and induced a minimum of deformations.
this study.
3.2. Effects of 3D weaving on carbon tows
Two aspects are studied in particular: the yarns passing into the
loom and the waviness and crimp of the weft tows. The handcraf-
ted weaving allows to be more careful with the monitoring and the
change of broken yarns. Warp yarns can be damaged during the
weaving, which in turn may influence the final strength of the
material. The weft yarns are inserted manually and the operator
is assisted for the warp yarns handling. Contrary to 2D fabrics,
yarns are handled many times in 3D interlock fabrics – as many Fig. 7. Order of insertion of weft yarns [19].
290
Table 1
Properties of Epolam 5015 epoxy resin.
Maximum stress at failure 105 MPa

Young’s modulus 3 GPa
Failure strain 6%
Viscosity at 25 °C 0.21 Pa.s
Fig. 8. Yarn damage in the raising/lowering device.
Table 2
Properties of HexTowÒ IM7 carbon fibers.
All the different plies are laid down on a flat mould, as shown in
Fig. 9, and the vacuum bag is sealed. The resin is an Epolam 5015 Tensile strength 5310 MPa
Young’s modulus 276 GPa
epoxy, mixed with the 5016 hardener (Axson Technology, Table 1).
Failure strain 1.8%
Its viscosity at 25 °C is low (0.21 Pa.s) to ensure a good impregna- Fiber density 1.78
tion of the fabric. Then, the composite is post-cured during 2 h at Number of filaments 6000
50 °C to complete the polymerization. Fiber diameter 5.2 lm
3.4. Microscopic observations

The other purpose of the microscopic observations is to mea-
The aim of this subsection is to study the quality of the compos- sure the dimensions of the yarns in the composite. The shape of
ite, estimated by the quantity of porosities and to obtain the yarn the cross-sections is found to be lenticular for the warp yarns
dimensions. Therefore, cross-sections of the 3D interlock compos- and trapezoidal for the weft ones, as seen in Fig. 10. As mentioned
ite are sampled in different locations and observed under an opti- in Section 3.1, a lenticular shape is set to all the cross-sections, in
cal microscope. By definition, a porosity is a micro-cavity due to order to keep the general nature of the geometry. It can be noticed
the trapping of air, gas or solvent during the impregnation stage. that the numerical lenticular weft section is equivalent to the real
As the porosity scale of the fabric is double (filament and yarn one in terms of surface area (Fig. 12). The spaces between yarns
scales), two types of porosity exist: the micro-porosity (between and the dimensions of the cross-sections (width and thickness)
the fibers inside a tow) and the macro-porosity (between the are measured accurately. The averaged dimensions are listed in
yarns). The generation of porosities is a complex phenomenon Table 3.
which depends on the velocity and orientation of the flow front, It should also be mentioned that the cross-sections of the weft
the capillary pressure and the wettability of the fibers, among yarns are vertically misaligned due to the action of the reed during
other things. The quantity of porosities defines the quality of the the weaving and the residual stresses in the yarns (Fig. 10). In this
manufacturing which influences the mechanical behavior of the study, the porosities and the nesting are not considered at first to
composite. simplify the geometrical modeling but they could be implemented
Cross-sections in the interlock composite plate are sampled in in a further simulation step.
different locations, polished and then observed under a microscope Calcination tests are performed to measure the fiber fractions in
equipped with a CCD camera. The porosity volume fraction is cal- the composite. It also allows to distinguish the fractions in both
culated from the measured porosity surfaces (Fig. 10). The size of directions (Table 4). They can be then compared to fractions in
every single porosity on each sample has been measured and its the numerical model, in order to assess the accuracy of the
nature (micro- or macro-porosity) is recorded. It should be noted modeling.
that Fig. 10 displays only a third of the actual sample size. The fiber volume fractions inside the yarns are calculated from
The global averaged porosity rate is low, around 0.6%, with microscopic observations of cross-sections of binarized black and
some samples showing maximal porosity levels of 2%. It proves white pictures (Fig. 13). These fractions are used for the calculation
the good quality of the material. This very low rate also justifies of the moduli of the yarns in the numerical modeling (Table 5). The
a numerical study of the elastic mechanical behavior of the com- difference of fiber fractions in both directions is the consequence of
posite, without modeling the porosities and their inelastic behav- a large number of weft yarns which creates a dense binding web
ior. The samples also show that the porosities are mostly located and slightly compacts the warp yarns.
close to the interlace of the warp and weft yarns, as it is depicted
in Fig. 10 and schematized in Fig. 11. Even though the porosity lev- 4. Experimental results
els are low, the porosities may be strongly concentrated. Their
presence between the warp and weft yarns can be the location of 4.1. Tensile tests on single yarn composite
stress concentration and thus of damage initiation, such as a deco-
hesion phenomenon. The yarn/yarn and yarn/matrix decohesion Some tests allow to assess the mechanical fiber properties, such
might be affected by these porosities. as tensile tests on dry yarns or on dry yarns with impregnated end
Fig. 9. Experimental device used for infusion.
291
Fig. 10. Microscopic observation of cross-sections of the interlock composite in the weft direction.
Fig. 11. Macro-porosities locations in the weft direction.
Table 4
Fiber volume fractions in the whole composite and split into the warp and weft
directions.
Composite Warp yarns Weft yarns

Vf 42% 7% 35%
Fig. 12. Equivalent lenticular and real trapezoidal weft cross-sections.

the addition of a term accounting for the matrix, issued from the
mixture law:

Table 3 DP qf 1 Vf
Geometric parameters of the 3D interlock composite. EfL ¼ Em ð1Þ
D MUL Vf
Warp lens width 1.313 mm
Warp lens thickness 0.139 mm where DP is the load variation, D the strain variation, qf the fiber
Weft lens width 1.152 mm density, MUL the mass by unit length, Vf the fiber fraction of the sin-
Weft lens thickness 0.206 mm gle composite yarn and Em the matrix modulus.
Horizontal space between warp yarns 3.554 mm Table 6 displays the Young’s moduli for all the types of yarn. The
Horizontal space between weft yarns 0.02 mm
moduli of fibers from the bobbin and of warp yarns are very close to
Total thickness 2.6 mm
each other (1% of difference). There is no damage of the warp yarns
during the weaving, despite of what was expected. It might be due
to the low number of warp yarns which are thus not in close contact
tabs. The first type of specimen shows the difficulty of the tensile to each other and against the heedles thanks to the alignment de-
simultaneity on all the fibers thanks to a capstan. The second spec- vice which allows the differentiation of the warp yarns. Conversely,
imen implies stress concentrations at the interface between dry a loss of about 8% of the longitudinal modulus is noticed for the weft
yarn and impregnated yarn, where rupture occurs without any sol- tows, due to beating and waviness as mentioned in Section 3.2. The
licitation of the dry fibers. Thus, the ASTM D4018-99 standard low dispersion of the results proves their reliability.
method is followed to obtain the mechanical properties of the car- Values in Table 6 will be used as input data in Section 5, instead
bon fibers from tests on single composite yarns. Three types of of the values simply deriving from the classical mixture law using
sample are manufactured: one with yarns from the bobbin and the fiber manufacturer moduli.
two with woven yarns from the fabric in both directions. The tows
are dipped into the resin and kept under tension during curing. 4.2. Tensile and flexural tests on 3D interlock composite samples
Cardboard end tabs are then added onto the specimens to prevent
from slipping in the jaws. The alignment of the yarns with the ten- Experimental tests on composite samples cut out from the in-
sile direction and the jaw clamping effect (to avoid any slippage) fused composite plate have been conducted in order to validate
were controlled with great attention. A series of five tests for each numerical results.
yarn type is performed on a 1474 Zwick machine with a 1 kN force The monotonous static tensile tests are performed on an 8-layer
cell and the deformations are measured using an extensometer. As interlock composite according to the ISO 527-4 standard. Rectan-
the fiber fraction is low in the yarns (15%), the formulae from the gular specimens without end tabs (250 25 mm2) are tested on
standard for the tensile modulus of the fibers is changed with a 1474 Zwick machine with a 100 kN force cell. Strains are mea-
292
Fig. 13. Microscopic transversal cross-section of a warp yarn.
Table 5 5.1. Periodic homogenization

Fiber volume fractions in both types of yarn.
Warp yarn Weft yarn The approach described in this work has already been applied
V yarn 65% 60% by Buannic et al., among others, in the framework of sandwich
f
structures [20]. Dealing with asymptotic homogenization, two
scale parameters are introduced in the formulation of the 3D prob-
lem of a thin periodic structure: e, the thinness of the plate and l,
Table 6 the periodicity of the plate, which are supposed to be of the same
Mechanical properties of yarns before and after weaving.
order of magnitude. Then an asymptotic expansion method can be
Young’s modulus used with only one parameter. Two coordinate systems yi
Yarn from the bobbin 289 GPa (±6%) (i = 1, 2, 3) and xa (a = 1, 2) are defined (with subscript 3 for the
Warp yarn from the fabric 287 GPa (±2%) thickness direction), corresponding to the microscopic and macro-
Weft yarn from the fabric 264 GPa (±6%) scopic scales, respectively. Next, the displacement field solution of
the 3D elasticity problem is searched under the following form
(with ya = xa/e (a = 1, 2)):
Table 7
Elastic tensile and flexural moduli of the interlock composite. uðxÞ ¼ uð0Þ ðx1 ; x2 Þ þ euð1Þ ðx1 ; x2 ; y1 ; y2 ; y3 Þ
Tensile moduli Flexural moduli þ e2 uð2Þ ðx1 ; x2 ; y1 ; y2 ; y3 Þ þ ð2Þ
Warp (1) Em
1 ¼ 12:3 GPa ð5%Þ Ef1 ¼ 17:3 GPa ð10%Þ The 3D problem is thus decomposed in a sequence of 3D micro-
Binding weft (2) Em
2 ¼ 64:5 GPa ð7%Þ Ef2 ¼ 58:1 GPa ð4%Þ scopic and 2D macroscopic sub-problems, derived at each order of
the infinitesimal parameter in the asymptotic expansion. Solving
the 0th-order microscopic problem leads to the homogenized con-
stitutive relations in the framework of the Love–Kirchhoff plate
sured using an extensometer. The flexural tests are performed
theory, as explained in the sequel.
according to the ISO 14125 standard. The rectangular specimens
The local strain field in the unit cell is additively decomposed
(100 15 mm2) are loaded with a constant speed of 4 mm/min.
into the macroscopic strain E + y3K (where E represents the
The equivalent tensile (m) and 3-point flexural (f) moduli of the
membrane macrodeformation and K the curvature one) and a
8-layer interlock composite in the warp (1) and weft (2) directions
perturbation ⁄ which accounts for the effect of the heterogene-
are averaged each from a series of five tests (Table 7). These exper-
ities. This fluctuation of the strain tensor derives from a peri-
imental results will be compared to numerical ones in Section 6.
odic displacement vector uper and displays a null average value
in the cell volume. The following elasticity problem can then be
5. Numerical prediction of the equivalent mechanical behavior solved, with any enforced macroscopic strain field, and the
displacement uper, the microscopic strains and stresses r as
Full 3D finite element simulations of 3D interlock composites is unknowns:
a computationally intensive task due to the strong heterogeneity of
8
such structures. An alternative consists in replacing the 3D heter- > divðrÞ ¼ 0
>
>
ogeneous material by an equivalent homogeneous plate suitable >
> r¼C:
>
>
for structural applications. As the interlock geometry is periodic >
< ¼ E þ y K þ ðuper Þ
ij ij 3 ij ij with i; j ¼ 1; 2
in the fabric plane (in both directions), the calculation of the ð3Þ
effective properties can be performed using an homogenization > r n ¼ 0 on @ X3
>
>
>
>
>
procedure on a periodic unit cell. Numerical finite element compu- > uper y1 and y2 periodic
>
:
tations on an elementary volume involving different simple load- r n anti-periodic
ings will enable to derive the macroscopic elastic in-plane
moduli (or stiffnesses). The periodic homogenization technique is where C represents the local fourth-order elasticity tensor (which
briefly presented below and then applied to a periodic unit cell depends on the material at the local scale) and @ Xi stands for the
of the 3D interlock composite. boundary surfaces of the unit cell with normal ± yi.
293
The macroscopic constitutive relations of the plate equivalent

problem are classically written using the macroscopic in-plane
stress resultants {N} and moments {M}, which are simply derived
from the average stresses by integration over the thickness. The
homogenized constitutive equation can then be written as follows,
in matrix notation:
" #
fNg ½A ½B fEg
¼ ð4Þ
fMg ½BT ½D fKg
where [A], [B] and [D] represent the effective stiffness matrices of
the tensile, coupled and bending behaviors, respectively and {E}T =
{E11, E22, 2E12}, {K}T = {K11, K22, 2K12}, in the frame of the Love–
Kirchhoff plate theory.
5.2. Solution procedure
The constitutive matrix in Eq. (4) is naturally symmetric so that

only 21 components have to be determined. The 21 macroscopic
strain fields applied on the heterogeneous unit cell are defined
on an homogeneous cell, with the same mesh as the one used for
the heterogeneous cell. The homogeneous cell is isotropic and ru-
led by the Hooke’s law with the Lamé coefficients (k, l):
mE
r ¼ ktrðÞI þ 2l with k ¼ and
ð1 þ mÞð1 2mÞ
E
l¼ ð5Þ
2ð1 þ mÞ
where r and are the stress and strain tensors, E and m the Young’s
modulus and Poisson coefficient of the isotropic homogeneous
material, and I the identity tensor. The volume f and surface N load-
ings are deduced from the local equilibrium equations, defined with
boundary conditions on the lateral faces of the cell:
Fig. 14. Tensors, applied loads and strain fields in membrane.
divðrÞ þ f ¼ 0
ð6Þ
rn¼N The use of the described homogenization method makes neces-
where n is the outward normal unit vector from a given face. These sary the periodic modeling of the interlock composite. However,
boundary conditions are applied on the unit cell to create the differ- this periodicity cannot be respected because of some previously
ent strain fields. Figs. 14 and 15 show the six elementary strain noticed factors such as nesting, inhomogeneous yarns, yarn crimp
fields in membrane and curvature, respectively. and porosity. Thus, the geometry is idealized until then, assuming
As a consequence, the 21 successive macroscopic strains are ap- the absence of these factors. The modeling might not be represen-
plied as initial deformation states on the heterogeneous unit cell, tative of the real material due to these imperfections, which can be
each representing one of the membrane or curvature macrodefor- analyzed statistically and implemented. It is a subtle balance
mations or a linear combination of them, and the equilibrium state between the needs of a method and the representation of the
is reached under periodic boundary conditions. In each loading material.
case, the strain energy W is deduced and employed for the The unit cell is chosen as the smallest volume representing the
determination of the constitutive moduli, through the following whole structure by periodicity. The geometrical modeling of the
relation: unit cell is straightforward thanks to modifiable input parameters
Z such as the dimensions of the yarns, horizontal spaces between
1 yarns, and the mesh size.
W¼ r : dX
2 X As mentioned above, the longitudinal Young’s moduli of
S impregnated yarns are simply derived from the mixture law, using
¼ fEgT ½A fEg þ fKgT ½D fKg þ 2fEgT ½B fKg ð7Þ
2 the experimental values of fiber moduli found in Section 4.1 (Ta-
ble 6) and fiber volume fractions in each yarn type (Table 5). The
where S is the cell surface area in the reference plane. effective modulus of an undamaged tow (187 GPa) is assigned to
the warp tows whereas the binding weft tows are considered as
5.3. Finite element simulation damaged with a longitudinal modulus of 173 GPa. The other mod-
uli are calculated from the HTA carbon fiber properties used by Cox
5.3.1. Geometrical modeling and material properties et al. [3], which are supposed equivalent to these of IM7 carbon fi-
The finite element computations are performed on Abaqus soft- bers (Table 8 where L represents the fiber direction and T, T0 the
ware, using a home-made script Python program so as to automate transverse directions of a yarn).
the modeling procedure and the post-processing, namely the cal-
culation of the effective constitutive matrix. The reinforcement 5.3.2. Numerical results
considered features seven weft layers and eight warp layers The above-mentioned general procedure leads to the complete
(Fig. 3). The vertical space between yarns is null. stiffness matrix:
294
Fig. 15. Tensors, applied loads and strain fields in curvature.
2 3
51; 386 3113 3:9 11; 700 3362 3:4 It can be noticed that the components of sub-matrix [B] as well
6 3113 151; 925 56 8638 34; 406 14 7 as the terms A13, A23, D13 and D23 are very low in comparison with
6 7
6 7 the others, excepted for B11 and B22 (Eq. (4)). The assumption of the
6 3:9 56 6493 3:7 12 1118 7
½C ¼ 6
6 11; 700 8638
7
null Bij terms leads to the decoupling between the membrane and
6 3:7 35; 751 4858 18:4 7
7
6 7 flexural behaviors. The case of non null B11 and B22 might be due to
4 3362 34; 406 12 4858 87; 346 27:6 5
the volume unbalance introduced by the new architecture. In the
3:4 14 1118 18:4 27:6 4333
case of the former interlock (binding warp and straight weft), the
ð8Þ Bij terms were null, what might be due to its balanced architecture
295
(equivalent fiber volume fractions in both directions). Further-

more, the effective moduli calculated with null Bij terms are the
same values as those calculated with the above sub-matrix [B].
Then, from the remaining terms, one can deduce the following
in-plane elastic moduli, corresponding to the membrane (m) and
flexural (f) behaviors, respectively:
Em
1 ¼ 17; 303 MPa Ef1 ¼ 16; 318 MPa
Em
2 ¼ 51; 116 MPa Ef2 ¼ 39; 867 MPa
Gm
12 ¼ 2189 MPa Gf12 ¼ 1993 MPa
mm12 ¼ 0:0205 mf12 ¼ 0:0556
mm21 ¼ 0:0606 mf21 ¼ 0:135
where 1 stands for the warp direction and 2 for the weft one.
6. Validation and analysis
In this section, a comparison between the deduced 2D equiva- Fig. 16. Comparison between 3D numerical and 2D equivalent analytical responses
lent behavior and the 3D heterogeneous plate is performed in in tension of 1/2/4 cells in the weft direction.
membrane and flexural modes, in both directions. Then, a second
comparison is performed with experimental tensile and bending
tests.
The 3D heterogeneous plate is created by assembling unit cells
together, only in the loading direction. The numerical simulation of
the plate is restricted by the number of elements and so the num-
ber of merged unit cells. Simulations are achieved with one, two
and four unit cells and the corresponding deformed shapes are
compared with the analytical solutions based on the 2D equivalent
properties.
In the weft direction, for a same tensile load, Fig. 16 shows that
an improvement of the accordance between the 3D and 2D equiv-
alent plates is achieved with an increasing number of cells (the
maximal error at the plate end is 43%, 35% and 21%, respectively
for a 1-, 2- and 4-cell structure). In the warp direction, adding
one cell decreases significantly the error from 73% (Fig. 17) to
13% (Fig. 18) at the end of the structure. These results show the
importance to use structures with large dimensions, to keep the
scale separation effective (e.g. l 1, as defined in Section 5.1).
The local increase of 3D deformed shapes at the structure end is
due to the side effects in the matrix.
In pure bending, the plates are composed of two cells in both
directions and a moment is applied at the end of the structure. Fig. 17. Comparison between 3D numerical and 2D equivalent analytical responses
There is a good agreement in the weft and warp directions be- in tension of one cell in the warp direction.
tween the two models, with only 11% (Fig. 19) and 22% of error
(Fig. 20), respectively. In the warp direction, the error might be fur-
ther decreased with the add of cells in order to reach a longer The equivalent elastic properties are already in good agreement
structure as in the weft direction. with the 3D simulations in both tensile and pure bending cases, for
In the simple bending case with two cells, a difference of 22% is a limited number of cells. A higher number of assembled cells
obtained in the weft direction (Fig. 21) whereas 60% of error is ob- would be required to really conclude about the accuracy of the re-
tained in the warp direction (Fig. 22). The transverse shear effects sults. However, this work was not achieved due to the limitation in
are not considered in the Love–Kirchhoff theory, which may ex- computation resources.
plain the discrepancies, but this feature can be introduced using 2D equivalent results are then compared with experimental
higher-order terms in the asymptotic expansion to get better tensile and 3-point bending tests in the warp and weft directions,
predictions. in the form of stress–strain and force–displacement curves (Figs. 23
and 24).
In the tensile case, the weft and warp averaged experimental
Table 8 curves show a non-linear behavior likely due to the damage initi-
Mechanical properties of constituents and impregnated yarns. ation: yarn/matrix decohesion and friction. In the bending test,
IM7 carbon fiber HTA carbon fiber [3] the composite material has a pure elastic behavior until rupture.
The experimental and equivalent elastic moduli are compared in
EL ET, E0T GLT, G0LT G0TT mLT, m0LT m0TT
Table 9. Correlation is good in bending in the warp direction but
Fiber 287/264 GPa 17 GPa 55 GPa 6.7 GPa 0.25 0.27 could be improved in tension with a more precise modeling of
Resin 3 GPa – 1.1 GPa – 0.35 –
the yarns (compaction and cross-section variation). In the weft
Yarn 187/173 GPa 8.4 GPa 4.8 GPa 2.9 GPa 0.28 0.44
direction, in both types of loading, the numerical moduli underes-
296
Fig. 18. Comparison between 3D numerical and 2D equivalent analytical responses Fig. 20. Comparison between 3D numerical and 2D equivalent analytical responses
in tension of two cells in the warp direction. in pure bending of two cells in the warp direction.
Fig. 19. Comparison between 3D numerical and 2D equivalent analytical responses Fig. 21. Comparison between 3D numerical and 2D equivalent analytical responses
in pure bending of two cells in the weft direction. in simple bending of two cells in the weft direction.
timate the experimental ones. These differences are mainly due to (with damaged/undamaged yarns or totally undamaged yarns)
the numerical representation of the fiber volume fractions. Indeed, which will then overestimate the experimental values.
the numerical fiber volume fractions are lower of 4.4% in warp and
19% in weft direction than the measured ones in the composite 7. Conclusion
(Table 4). In order to increase these values in the model, without
changing the yarn dimensions, it is necessary to consider the nest- 3D interlock composites make them candidates for specific
ing and thus a more densely packed architecture. applications thanks to their delamination and impact resistance.
Furthermore, for comparison purposes, the predictions with all Their mechanical study needs new approaches different from the
yarns considered as undamaged (namely a longitudinal modulus of ones based on the laminate theory. In this work a numerical tool
187 GPa) are plotted in the tensile and bending cases of the weft has been developed for the mechanical study of a carbon/epoxy
direction (Weft num 2D (undamaged yarns) curves). The average layer-to-layer angle-interlock composite. The realistic modeling
moduli would then be 55 GPa (tensile) and 42.7 GPa (bending): of the material in terms of geometry and constituents properties
the difference of 7% with the first damaged equivalent moduli is essential for an accurate prediction of the mechanical behavior.
shows the importance of taking into account the real material An analysis of the whole manufacturing of the composite high-
properties of the yarns. The latter moduli with undamaged yarns lights how the weaving and infusion process parameters can influ-
seem to improve the agreement with the experimental results. ence the final architecture. Some of these parameters such as
However, as mentioned above, the improvement of geometrical dimensions, fiber fractions, yarn damage during weaving, are
modeling thanks to nesting will increase the equivalent moduli implemented into the automatic and parameterized modeling of
297
Table 9
Tensile and bending experimental and equivalent moduli.
Tension Bending
Experimental Numerical Experimental Numerical
Warp 12.3 GPa 17.3 GPa (+30%) 17.3 GPa 16.3 GPa (8%)
Weft 64.5 GPa 51 GPa (20%) 58 GPa 39.8 GPa (32%)
the architecture of a unit cell. The results are the effective elastic
in-plane constants predicted from a periodic homogenization
method. There is a good agreement of the in-plane behaviors be-
tween 3D and 2D equivalent models, except for the simple bending
behavior where the transverse loadings were not included. The
interlaminar shear behavior requires more advanced develop-
ments in the homogenization method, combined with short-beam
experimental tests. The determination of the through-the-thick-
ness properties is an important point for impact resistance. These
elastic constants can be then used in shell-type simulations. The
effective moduli are also compared to experimental results and
Fig. 22. Comparison between 3D numerical and 2D equivalent analytical responses show good agreement for the warp flexural modulus but could
in simple bending of two cells in the warp direction. be improved in tension mode thanks to a better geometry repre-
sentation. For the weft tensile and flexural moduli, the nesting will
be one of the parameters which will increase the fiber fraction and
improve the experimental comparison. Hypotheses have been
introduced to ease the numerical implementation at this first stage
of a more general development.
Acknowledgement
The authors thank the Nord-Pas-de-Calais Council for financial

support of this work.
References
[1] Zhang C. Characterization and modeling of 3D woven composites. PhD thesis.

USA: Graduate Faculty of North Carolina State University; 2003.
[2] Sherburn M. Geometric and mechanical modelling of textiles. PhD thesis. UK:
School of Mechanical, Materials and Manufacturing Engineering, University of
Manchester; 2007.
[3] Cox BN, Dadkhah MS. The macroscopic elasticity of 3D woven composites. J
Compos Mater 1995;29:785–819.
[4] Kuo WS, Fang J, Lin HW. Failure behavior of 3D woven composites under
transverse shear. Composites: Part A 2003;34:561–75.
[5] Lee L, Rudov-Clark S, Mouritz AP, Bannister MK, Herszberg I. Effect of weaving
Fig. 23. Comparison of experimental and 2D equivalent curves under tensile load in damage on the tensile properties of three-dimensional woven composites.
the warp and weft directions. Compos Struct 2002;57:405–13.
[6] Callus PJ, Mouritz AP, Bannister MK, Leong KH. Tensile properties and failure
mechanisms of 3D woven GRP composites. Composites: Part A 1999;30:
1277–87.
[7] Yi HL, Ding X. Conventional approach on manufacturing 3D woven preforms
used for composites. J Ind Textiles 2004;34:39–50.
[8] Cox BN, Dadkhah MS, Morris WL. On the tensile failure of 3D woven
composites. Composites: Part A 1996;27A:447–58.
[9] Mahadik Y, Robson Brown KA, Hallett SR. Characterisation of 3D woven
composite internal architecture and effect of compaction. Composites: Part A
2010;41:872–80.
[10] Potluri P, Sagar TV. Compaction modelling of textile preforms for composite
structures. Compos Struct 2008;86:177–85.
[11] Lee B, Leong KH, Herszberg I. The effect of weaving on the tensile properties of
carbon fibre tows and woven composites. J Reinf Plast Compos
2001;20:652–70.
[12] Archer E, Buchanan S, McIlhagger AT, Quinn JP, Morgan M. An investigation on
the effect of 3D weaving on carbon fiber tows, fabrics and composites. In:
Advani, Gillepsie, editor. TexComp 9 – Recent advances in textile composites;
2008, p. 371–9.
[13] Tan P, Tong L, Steven GP. Micromechanics models for mechanical and
thermomechanical properties of 3D through-the-thickness angle interlock
woven composites. Composites: Part A 1999;30:637–48.
[14] Cox BN, Carter WC, Fleck NA. A binary model of textile composites – I.
Formulation. Acta Metall Mater 1994;42:3463–79.
[15] Verpoest I, Lomov SV. Virtual textile composites software WiseTex: integration
with micro-mechanical, permeability and structural analysis. Compos Sci
Technol 2005;65:2563–74.
Fig. 24. Comparison of experimental and 2D equivalent curves under flexural load [16] Bogdanovich AE, Mohamed MH. Three-dimensional reinforcements for
in the warp and weft directions. composites. SAMPE J 2009;6(45):8–28.
298
[17] Boussu F, Legrand X, Serret A. General clustering of warp interlock structures. [19] Tsai KH, Chiu CH, Wu TH. Fatigue behavior of 3D multi-layer angle
In: 1st International conference of intelligent textiles and mass customization, interlock woven composite plates. Compos Sci Technol 2000;60:
Casablanca, Morocco, November 15–17, 2007, p. 83–90. 241–8.
[18] Rudov-Clark S, Mouritz AP, Lee L, Bannister MK. Fibre damage in the [20] Buannic N, Cartraud P, Quesnel T. Homogenization of corrugated core
manufacture of advanced three-dimensional woven composites. Composites: sandwich panels. Compos Struct 2003;59:299–312.
Part A 2003;34:963–70.
299
Titre : Contribution à la modélisation du comportement mécanique de structures minces
pour la prédiction de divers types d’instabilité – Applications multi-matériaux.
Résumé : Mes travaux de recherche s’articulent autour de la modélisation théorique et

numérique du comportement non-linéaire des matériaux et des structures, et s’intéressent
en particulier au flambement de structures élastoplastiques ou composites. Ma princi-
pale activité a consisté à développer un certain nombre de programmes de calcul ou de
modélisation par éléments finis. Le premier thème abordé est la plasticité. Une formulation
générale de l’intégration locale a été mise en œuvre pour un critère de plasticité et une
loi d’écrouissage arbitraires. Les applications visées sont la prédiction des phénomènes de
rochet et de striction. Les derniers développements concernent l’identification d’un modèle
3-phases pour des films polymères en grandes transformations. Le second thème traite
de l’homogénéisation périodique de matériaux composites à renforts 3D. L’objectif est de
remplacer la structure mince hétérogène par une plaque homogène équivalente et ses rai-
deurs effectives en élasticité. Une réflexion sur l’endommagement de ces matériaux par
décohésion fibres/matrice est en cours.
L’essentiel de mes travaux porte sur la modélisation du flambement et du post-
flambement, qui repose conjointement sur une démarche de résolution analytique et un
programme élément fini coque doté de méthodes numériques de longueur d’arc et de bi-
furcation. Les thèmes précédemment abordés permettent de s’intéresser au flambement
plastique ainsi qu’au flambement local/global de matériaux hétérogènes. La poursuite de
ces travaux tend à combiner plasticité, homogénéisation et flambement, afin de définir des
outils de dimensionnement efficaces pour des structures telles que les matériaux sandwich.
Mots-clés : Flambement et post-flambement, plasticité en grandes transformations, ho-

mogénéisation périodique, modélisation analytique et numérique.
Title : On the modeling of the mechanical behavior of thin structures for the prediction
of various instability phenomena – Applications to a wide range of materials.
Abstract : My field of investigation is around theoretical and numerical modeling of the

non-linear response of materials and structures, and particularly deals with the buckling
behavior of elastoplastic or composite structures. My leading activity consisted in develo-
ping a number of finite element calculation or modeling tools. Plasticity is the first topic
addressed. A general formulation for local integration was carried out, allowing for an ar-
bitrary hardening model to be used. The intended applications refer to the ratcheting and
necking phenomena. Later works are about the identification of a 3-phase model devoted
to polymer films in finite plasticity. The second topic deals with periodic homogenization
of 3D reinforced composite materials. The objective is to replace the thin heterogeneous
structure with an equivalent homogeneous plate characterized by its elastic effective stiff-
nesses. A damage analysis of such materials due to fiber/matrix debonding is in progress.
My main research field is about buckling and post-buckling. Two resolution tools are
jointly used : an analytical approach and a shell finite element program including arc-length
and branch-switching methods. Previous experience in plasticity and homogenization en-
ables to deal with plastic buckling and with the local/global buckling of heterogeneous
materials. In my future work, plasticity, homogenization and buckling will combine, in
order to define efficient design tools for materials such as sandwich structures.
Keywords : Buckling and post-buckling, finite plasticity, periodic homogenization, ana-

lytical and numerical modeling.

2011 219

Transféré par

Droits d'auteur :

Formats disponibles

2011 219

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

2011 219

Transféré par

Droits d'auteur :

Formats disponibles

N° d’ordre : 40673

UNIVERSITÉ DE LILLE I SCIENCES ET TECHNOLOGIES

HABILITATION À DIRIGER DES RECHERCHES

Maı̂tre-Assistant à l’École des Mines de Douai

CONTRIBUTION À LA MODÉLISATION DU COMPORTEMENT

Soutenue le 25 novembre 2011

M. KARAMA Moussa Professeur, ENIT Rapporteur

Je tiens tout d’abord à remercier Professeur Zitouni Azari de l’École Nationale

Je remercie Patricia Krawczak, responsable du département Technologie des Polymères

Table des figures x

Liste des tableaux xii

Partie I : Présentation générale 5

3 Synthèse des activités de recherche 14

5 Production scientifique et technique 18

6 Collaborations académiques et industrielles 24

Partie II : Présentation des travaux de recherche 27

1 Plasticité en grandes transformations 29

2 Homogénéisation périodique de matériaux composites 57

3 Instabilités géométriques : flambement et post-flambement

Conclusions et perspectives 125

Partie III : Recueil de publications 137

Ce mémoire synthétise l’ensemble de mes travaux de recherche menés à l’École Cen-

Depuis décembre 2003 : Enseignant-Chercheur au Département Technologie des

De septembre 2002 à août 2003 : Attaché Temporaire d’Enseignement et de Re-

D’octobre 2001 à août 2002 : Attaché Temporaire d’Enseignement et de Recherche

D’octobre 1998 à septembre 2001 : Allocataire de recherche MENRT et Moniteur

D’octobre 1997 à juillet 1998 : Scientifique du Contingent à l’École Nationale des

De juillet 1996 à octobre 1996 : Ingénieur Stagiaire à Citroën La Janais, Rennes.

Novembre 2001 : Thèse de Doctorat de l’École Centrale de Nantes et de l’Université

Septembre 1997 : Diplôme d’Études Approfondies en Génie Mécanique option

De 1992 à 1994 : Mathématiques Supérieures et Spéciales M’ au Lycée Chateaubriand,

Juin 1992 : Baccalauréat série C. Mention Bien.

Compétences Scientifiques et Pédagogiques

Mécanique non-linéaire des structures :

Références Industrielles et Académiques (collaborations)

– CETIM, IFREMER, DGA, AXS Analyse de Structures, bureau d’études PRODESS,

– 18 publications (dont 16 dans des revues à comité de lecture) + 1 soumise.

– Systèmes d’exploitation : MS Windows, Unix, Linux.

– Anglais : Lu, parlé, écrit.

2.1 Statuts et charges d’enseignement

ATER à l’École Centrale de Nantes (2001-2002) : J’ai poursuivi, l’année suivante,

ATER à l’Université de Nantes (2002-2003) : J’ai effectué ensuite un service d’en-

Enseignant-Chercheur à l’École des Mines de Douai (depuis décembre 2003) :

2.2 Nature des enseignements

Les disciplines enseignées sont présentées ci-dessous, en décrivant brièvement, pour

Travaux Dirigés de Mécanique des Solides : Le contenu de cet enseignement, qui

Travaux Dirigés de Mécanique des Vibrations : Le contenu de ces travaux dirigés

Cours et Travaux Dirigés de Mécanique des Milieux Continus/Élasticité : Ce

Travaux Pratiques de Résistance des Matériaux : L’objectif est de mettre en pra-

Travaux Pratiques de Méthodes Numériques en Mécanique Non-Linéaire :

– Formation professionnelle ”Modélisation numérique par éléments finis” pour un

Enseignement dans le cadre d’un accord Erasmus : Missions à l’Université ”Du-

Projets de Découverte de la Recherche : Initiation à la recherche des élèves-

Jurys : Participation à des jurys d’admission des élèves-ingénieurs, d’Études Bibliogra-

Mon premier axe de recherche s’intéresse à l’identification et à la modélisation

Mon second axe de recherche concerne la modélisation de la réponse de structures

4.2 Thèses de Doctorat

Th2 - F. Zeng, Contribution à la modélisation du comportement mécanique en