4TH INTERNATIONAL CONFERENCE

CIVIL ENGINEERING - SCIENCE AND PRACTICE

ŽABLJAK, 20-24 FEBRUARY 2012

Rasim Šehagi 1, Senad Medi2

FAILURE ANALYSIS OF A STEEL FRAME

Summary
In this work we will present failure analysis of a steel frame considering stress – strain
diagram with hardening and softening branches, as well as deformability of beam-to-
column connections. Steel frame is numerically modeled using beam finite elements.
Joints which are seen as potential plastic hinges are defined using M –  curve
determined assuming M-N interaction. All results were presented on diagrams that
show dependence of the observed joint displacement and the load level. Also, we
briefly mention the method of inequalities for determination of the ultimate load.
Key words
Failure analysis, plastic hinge, beam-to-column connection, M –  curve.

GRANINA ANALIZA ELINOG OKVIRA

Rezime
U ovom radu predstavit emo graninu analizu elinog okvira uzimajui u obzir radni
dijagram elika koji ukljuuje uticaj ovršavanja i omekšavanja elika u procesu
deformisanja, kao i popustljivost veze grede i stuba. Okvir je numeriki modeliran
koristei štapne konane elemente. Svi vorovi, koji su posmatrani kao potencijalni
plastini zglobovi, su definisani M –  krivom odreenom uz pretpostavku interakcije
sa normalnom silom. Svi rezultati su prikazani dijagramom zavisnosti posmatranog
pomjeranja i stepena optereenja. Takoer, daje se i kratki prikaz metode nejednaina u
odreivanju graninog optereenja.
Kljune rijei
Granina analiza, plastini zglob, veza greda – stub, M –  kriva.

1
Dipl.ing.gra., projektant, Calypso d.o.o., H. Kreševljakovia 7A, Sarajevo, BH, rasimsehagic@gmail.com
2
Dipl.ing.gra., asistent, Graevinski fakultet u Sarajevu, Patriotske lige 30, BH, senad_medic@yahoo.com

353
Civil Engineering - Science and Practice

1. INTRODUCTION

Plastic design of steel structures has several advantages over the elastic design, of
which the most important are simplified procedures, savings in the cost, and more realistic
representation of actual behavior of steel structures [1]. These advantages stem from the
fact that the plastic design fully exploits the important property of steel – ductility. The
elastic design method, also termed as allowable stress method is a conventional method of
design based on the elastic properties of steel. This design method limits the structural
usefulness of material up to a certain allowable stress, which is well below the elastic limit.
In the method of plastic design of a structure, the ultimate load rather than the yield
stress is regarded as the design criterion. The term plastic has occurred due to the fact that
the ultimate load is found from the strength of steel in the plastic range. Two processes
occurring simultaneously until the frame transforms into failure mechanism are
redistribution of forces (which is a benefit of ductility) and plastification. Plastification
means that some fibers enter the plastic regime in the constitutive diagram. Integrating the
stresses over the cross-section we get normal force and bending moment, which is usually
termed plastic moment. The cross-section at which the plastic moment acts is called plastic
hinge. After formation of a sufficient number of plastic hinges, the structure transforms into
mechanism (this type of analysis is called plastic hinge method).
Gabor von Kazinczy drew attention to a contradiction in the plastic hinge method as
early as 1931. The load P of the continuous beam is increased from zero to elastic limit
state so that the extreme fibers begin to become plastic at the point of the maximum
bending; as the load increases further, so plastic hinge form. The limit state of load-
carrying capacity of the continuous beam is then reached when plastic hinge form at points
1, 2 and 3 (see Fig.1). These three plastic hinges change the system with two degrees of
static indeterminacy into a simple cinematically determinate system. The load associated
with this series of plastic hinges is known as the ultimate load T or the plastic limit load T
of the continuous beam.

Fig. 1. Paradox of the plastic hinge method [2].

By contrast, a plastic hinge can only form at point 2 in the simply supported beam
(Fig. 1. – left). If we compare the ultimate loads of the two systems shown in Fig.1., then
the ultimate load of the continuous beam is twice that of the simply supported beam. If
l10 the continuous beam shown in Fig. 1. – right changes to a fixed-end beam with span l
and ultimate load T=8Mpl/l; this ultimate load matches that of the continuous beam. But if
we take this to the limit l1, the continuous beam is transformed into a simply supported

354
 GNP 2012

beam with span l and the ultimate load T=4Mpl/l; this ultimate load is only half that of the
continuous beam. And that is paradox of plastic hinge theory.
It was not until 1952 that Symonds and Neal cleared up the paradox of ultimate load
theory. They investigated the deflections, the relative rotations and strains at the plastic
hinge and discovered that, in particular, the rotation of the plastic hinge became
unacceptably large above a certain length of end span l1; at the limit l1, the deflection at
point 2 also tends to infinity, which means that the calculations lose their meaning and
cannot be regarded as a limit state for the simply supported beam [2].
We deliberately stress the importance of determining the deformations in the plastic
hinge analysis since assuming infinite deformability can lead to erroneous results. In a
typical hinge-by-hinge method sufficient rotational capacity is tacitly assumed so that the
moment in the plastic hinge never drops. However, this is not realistic.

2. INPUT PARAMETERS FOR FAILURE ANALYSIS

2.1. GEOMETRY, LOADS AND MATERIAL PROPERTIES

The analyzed model is a steel frame 9000 mm wide and 4000 mm tall. All cross-
sections are IPE 550. Concentrated loads of 10 kN are distributed at every 1250 mm.
Numbers in circles represent characteristic joints in the frame and we analyze
displacements in joints 2 (horizontal) and 8 (vertical). Every load is multiplied by load
factor .

Fig. 2. Analyzed models.

Fig. 3. Elastic – plastic stress and strain distributions with strain – hardening.
At the cross-section level we need to define moment – curvature relationship. In
Fig.3. one can see typical stress-strain distributions obtained for pure bending. The uniaxial
constitutive relationship for steel employed in the analysis has following properties:

355
Civil Engineering - Science and Practice

Young's modulus is E=210000 N/mm2, yield stress is y=240 N/mm2, ultimate stress is
u=360 N/mm2, yield strain is
y= y/Ey, strain at ultimate stress is
u=0.1 and strain at
failure is
f=0.13 [3] (see Fig.4 - left).

Fig. 4. Stress – strain diagram from [3](left) and M –  curve with zero axial force (right).

The iterative procedure for determination of M –  curves is outlined in the next

algorithm (note that it includes the influence of axial force):
1. Impose the desired curvature .
2. Assume strain at the top fiber
u. Calculate
l from  = (
u +
l)/h. Now we
have deformations profile from which we easily determine the corresponding
stresses.
3. Find the section forces by performing the following integrations:

N ³ V dA
A
(1)

M ³ V ydA
A
(2)

4. Now we check if the equation (1) is satisfied (constraint equation). Namely,

internal axial force is either zero or equal to external axial force P for the case
of M – N interaction. If the residual is lower than the chosen tolerance
parameter, we adopt the solution and proceed to the next value of curvature.
Else, another
u is chosen and we loop again from the step 2 to the step 4.

Employing the previous method we have derived moment – curvature diagrams for
different levels of axial load. For this purpose, a code in Matlab was written ([4]). In order
to compare the obtained results with some established program, we have modeled the same
problem with the OpenSees ([5]) using its Steel01 material properties. Some characteristic
values are given in Table 1. and the complete diagrams are plotted in Fig.5. noting that the
Matlab diagram is shown simplified in order to clearly see the onset of yielding.
Table 1. Characteristic values for M –  curves.
N = 0 kN N = - 0.1 Ny N = - 0.2 Ny N = - 0.3 Ny
 M  M  M  M
Value
[rad/m] [kNm] [rad/m] [kNm] [rad/m] [kNm] [rad/m] [kNm]
Yield 4.15e-3 585.6 3.74e-3 527.3 3.32e-3 469.1 2.91e-3 410.8
Ultimate 0.363 917.5 0.327 886.5 0.294 853.1 0.265 812.2

356
 GNP 2012

Fig. 5. M –  curves obtained by OpenSees (top) and Matlab (bottom).

The calculated responses match very well. However, one can notice that the bending
moment in OpenSees can infinitely grow, which is a consequence of the defined
constitutive law. Namely, it is assumed that the stress never drops to zero which is not
realistic. Moreover, it can be dangerous making us believe that the load bearing capacity is
not exhausted, while the real cross section has lost its capacity long ago.
The effect of the axial force on the ductility and the yielding/ultimate moment is
very well known and the predicted response matches the expectations. However, for the
loads and the geometry that we analyze it can be neglected because the axial loads are well
below the axial force at yielding and the failure mechanism is predominantly dependant on
the bending moments.

357
Civil Engineering - Science and Practice

2.2. PROPERTIES OF CONNECTIONS – COMPONENT METHOD

Eurocode 3 ([6]) suggests a possible way to encompass the connection flexibility
into the global structural analysis by using the so called component method. The
application of the component method requires three basic steps: listing of the joint
components, evaluation of the force-deflection diagram of each individual component in
terms of the initial stiffness, strength and deformation capacity, and assembly of the
components in order to evaluate the behavior of the joint as a whole. The components are
loaded in tension, compression and shear [6].

Fig. 6. Components of beam-column joint [6](left, center), moment – rotation curve (right).

The accuracy of the method depends on accuracy of the description of the basic
components by independent springs and on the quality of assembling. It is assumed that the
component properties are independent and therefore they can be obtained easily. Behavior
of the components can be described by non-linear force – deformation curves including
various effects (strain hardening, contact between elements, membrane effect…), but it can
be simplified to bilinear or tri-linear model. The design moment resistance Mj,Rd is obtained
following the slope of initial stiffness Sj,ini (up to 2/3 Mj,Rd), then tracing the secant stiffness
Sj up to Xd where it reaches a plateau and yields until 0.015 rad. Thereafter the stiffness is
degraded by linear softening up to 0.017 rad
Bending moment resistance of joints is limited by the weakest component in
tension, compression and shear. When elastic distribution is assumed, shear force is
transferred by all bolts and the check for combined shear and tension is required. In this
case, the tensile bolt force should include the increase caused by prying of the bolts. For
elastic-plastic and plastic force distributions, the shear force is usually assigned to bolt
rows, which do not carry bending moment. As an alternative, combination of tension and
shear can be considered.
Curves that describe the behavior of the beam-to-column connection are calculated
using the method of components. One curve is formed under the assumption that the shear
force is carried by the lower bolt row and that component is not taken into account in the
calculation process. Other curve is formed assuming that the shear component affects
bending moment resistance. Deformation capacity of the analyzed joint is limited to 0,015
rad as recommended by Horvati [7] (see Fig 6 - right).

358
 GNP 2012

3. METHOD OF INEQUALITIES

Fig.7. Failure mechanisms of steel frame.

Before proceeding to the final chapter we briefly mention the method of inequalities
which can be used to directly determine the ultimate load bearing capacity. It is based on
the virtual work statement by which the sum of all virtual works of the generalized external
and internal forces equals to zero (see eq. (3), [8, 9]). The basis of the method is the fact
that for any frame structure one can form certain number of failure mechanisms with the
corresponding load factors. Mechanisms shown in Fig.7. are considered for this system and
the one with the minimum load factor is the one determining the load bearing capacity.
n _ elem

¦ FiG ui 
i
¦ ³ M GN dx  ¦ M GI
j k
k k
0 (3)
x

This problem is reduced to a system of inequalities which is actually a linear programming

problem. Solution to this problem will not be considered here.

4. RESULTS AND CONCLUSIONS

In the following figures we show the results obtained employing load control
analysis for 2 different load cases (Fig.8) and by using displacement control analysis only
for the vertical loading program (Fig.9). The analyses were performed with the help of
SAP2000 ([10]) assuming different properties of the beam-column connection, P –  effect
and geometrical imperfections as well as the previously defined moment – curvature
relationship. Plastic hinges can form only in the nodes of the FE model, which is
satisfactory for the loading pattern with concentrated forces.

Fig. 8. Results for model 1(left) and model 2(right).

359
Civil Engineering - Science and Practice

Fig. 9. Load – displacement curve for displacement control analysis.

After the occurrence of yielding at the critical cross section, the slope in the load –
displacement diagram decreases and the process continues so forth until the global stiffness
matrix becomes singular. In Fig.9. there are jumps in the diagram because the program
unloads the hinges whose capacity has been exhausted. Thereafter, the loads are
redistributed to the elements which can still sustain additional load. Typically, for vertical
load only, the frame transforms into a simple beam after yielding in the connections which
are weaker than the original cross-section. The deformable connection analyzed according
to EC-3 procedures, can have significant effects on the ultimate load bearing capacity. We
put special accent on the shear component of the analyzed joint. P –  effect as well as the
assumed geometrical imperfections ( = 1/200) reduce the ultimate load but only slightly.
The limit load analysis of a structural system plays an important role in
performance-based design procedure. Generally, frames fail by exhibiting localized failure
in a limited number of critical zones. By using beam elements, we are constrained to
choosing nonlinear spring models in order to catch the inelastic response. However, it is
very much possible that the local buckling precludes the possible rotations assumed in this
work. In that case, refined models using nonlinear shell elements should be employed.

LITERATURE

[1] W.F. Chen, I. Sohal: "Plastic Design and Second – Order Analysis of Steel Frames", Springer
- Verlag, Berlin,1995, 509p
[2] K.E. Kurrer: "Geschichte der Baustatik", Ernst&Sohn, Berlin, 2007, p. 539
[3] J. Dujc, B. Brank, A. Ibrahimbegovic: “Multi-scale computational model for failure analysis
of metal frames that includes softening and local buckling”, CMAME, Vol. 199, 2010,
p.1371-1385
[4] MATLAB: http://www.mathworks.com
[5] OpenSees: http://opensees.berkeley.edu
[6] Eurocode 3: Design of steel structures – Part 1.8: Design of joints, CEN, December 2003
[7] D. Horvati: "Spregnute konstrukcije elik - beton", MASMEDIA, Zagreb, 2003, p.232
[8] M. Koji: "Primenjena teorija plastinosti", Mašinski fakultet Univerziteta „Svetozar
Markovi“ i Graevinski fakultet u Sarajevu, 1979, 329p
[9] M. Jirasek, ZP. Bažant: “Inelastic Analysis of Structures”, John Wiley & Sons, New York,
2002, 734p
[10] SAP2000 v.14: CSI, Berkeley, 2009

360