See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257335151

Implications of failure criteria choices on the rapid concept design of

composite grillage structures using multiobjective optimisation

Article  in  Structural and Multidisciplinary Optimization · May 2013

DOI: 10.1007/s00158-012-0875-4

CITATIONS READS

2 117

3 authors:

Adam J. Sobey James I. R. Blake

University of Southampton University of Southampton
63 PUBLICATIONS   153 CITATIONS    102 PUBLICATIONS   315 CITATIONS   

SEE PROFILE SEE PROFILE

R. A. Shenoi
University of Southampton
298 PUBLICATIONS   2,855 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Feasibility study on manganese nodules recovery View project

Nuclear Maritime Propulsion View project

All content following this page was uploaded by R. A. Shenoi on 21 July 2015.

The user has requested enhancement of the downloaded file.

Struct Multidisc Optim (2013) 47:735–747
DOI 10.1007/s00158-012-0875-4

RESEARCH PAPER

Implications of failure criteria choices on the rapid concept

design of composite grillage structures using multiobjective
optimisation
A. J. Sobey · J. I. R. Blake · R. A. Shenoi

Received: 22 May 2012 / Revised: 8 October 2012 / Accepted: 21 October 2012 / Published online: 16 January 2013
© Springer-Verlag Berlin Heidelberg 2013

Abstract Grillage topologies are commonly used in many Asx,sy Axial rigidities of stiffeners
composite structural applications to produce low mass bs Crown thickness
designs that have a high stiffness. While composite fail- b,g Numbers of beams and girders
ure criteria are being compared in many different simple cs Web width
structures, for example plates and tubes, literature must Dsx,sy Stiffener rigidities
also compare more complicated applications, including gril- dna Cross sectional area to neutral axis
lages, as there are distinct differences in behaviour. This distance
paper therefore performs analysis of grillage structures with ds Web height
more up to date failure criteria, taken from the world wide E Young’s modulus
failure exercise, than previously investigated. The grillage Ef 1 Young’s modulus of fibre
theory selected is that of Navier theory with elastic equiva- G Shear modulus
lent properties due to its low computational expense for use I Second moment of area
with a genetic algorithm to optimise a composite structure. Icx Moment of inertia
The results take an example from leisure boatbuilding show- L,B Length and breadth of plate
ing the grillages produced from the different limit states, Ms Moments of stiffeners
comparing the cost and mass. The final results show that m,n Wave numbers
the method allows a rapid analysis of grillages and that the mσf Mean stress magnification factor
selection of the limit state has an important effect on the nb,g Number of beams or girders
optimised grillage topology. P Pressure
q(x,y) Pressure at a given point on plate
Keywords Failure criterion · Rapid structural analysis · S12 Shear strength in the ply
Grillage theory t Ply thickness
U, V , W, X, Ymn Coefficients for initial conditions
w Deflection
Nomenclature w̄ Non-dimensionalised deflection
XC ,XT Strength parallel to fibres
a,b Stiffener spacing YC ,YT Strength transverse to fibres
as Crown width , γ Stiffness
amn Coefficient for grillage analysis 1T Tensile failure strain
Ai,j Laminate stiffness terms 1C Compressive failure strain
ρ⊥ Slope of the longitudinal fracture
envelope
A. J. Sobey () · J. I. R. Blake · R. A. Shenoi
ρ⊥⊥ Slope of the transverse fracture
University of Southampton, University Rd,
Southampton, SO17 1BJ, UK envelope
e-mail: ajs502@soton.ac.uk σ Stress
736 A.J. Sobey et al.

σcri Critical stress tophat approach, an application of a grillage of stiffeners

σ1D Stress value for linear degradation is shown in the top left part of Fig. 1. The top right part
τ Shear stress of Fig. 1 shows the idealised representation of this grillage
υ Poisson’s ratio and at the bottom an idealisation of the stiffener geometery,
which provides necessary flexural and axial stiffness to the
structure.
1 Introduction Optimisation is often an approach that is taken to inves-
tigate how the different structures behave given a specific
Composite materials are used within a large number of condition. There are many available papers discussing opti-
products in many industries from aerospace through to auto- misation and more specifically genetic algorithms to solve
motive and boatbuilding among many others. They are more complex composite design problems. Genetic algo-
generally utilised for their ability to be created with proper- rithms have been used over gradient based methods as this
ties specific to the task required. They exhibit high strength allowed a large, complex search space to be investigated
to weight ratio, excellent corrosion resistance and a large without priming the result based on previous knowledge.
design freedom. To model the constraints of the material Furthermore genetic algorithms have a unique ability to deal
properties on the composite a number of different failure with alpha-numeric fields. A selection of literature for a
criteria are used. These failure criteria have been applied to range of applications follows. Satheesh et al. (2010) looked
different simple structural applications from tubes to pan- at the use of multiple failure criteria with genetic algo-
els. Many of these criteria are being investigated through the rithms for design optimisation of laminated plates. Kim and
work being performed by the World Wide Failure Exercise Kim (2010) looked at the optimal design of stiffened panels
(WWFE) and are collated in Hinton et al. (2004b) among from buckling. Kang and Kim (2005) looked at minimum
other examples. weight design of compressively loaded composite plates
Counter to the many benefits composite materials exhibit using nonlinear finite element analysis. Naik et al. (2008)
they also suffer from a relatively low modulus and therefore used a genetic algorithm to look at maximum stress and
there is a requirement for stiffeners to be utilised within the Tsai-Wu failure criteria for minimum weight design of com-
structure. Composite structures are often stiffened using a posite plates. Lopez et al. (2009) investigated the use of

Fig. 1 Example of a Grillage

stiffened plate in a
composite boat
Implications of failure criteria choices on the rapid concept design of composite grillage structures 737

optimisation for composites considering maximum stress, For the structural modelling of the stiffened plate Navier
Tsai-Wu and Puck failure criteria on a laminated plate. method grillage analysis has been adopted. This work is
While this research has concentrated on composite struc- originally covered in Vedeler (1945) and the has been shown
tural optimisation and the use of more stringent failure in Maneepan et al. (2007) to closely approximate the more
criteria none have looked at the structural response of accurate methods while being computationally more
grillages. efficient. Navier theory is based upon the deflection of
To find the mechanical response of a grillage a num- intersecting points found between longitudinal girders and
ber of theories can be used based on beam theories defined transverse beamswith a pressure applied to the panels from
by Clarkson (1965) which include orthotropic plate method the opposite side of the plate to the stiffeners, as shown in
and folded plate method. While these theories may lack the top right part of Fig. 1. From these deflections it is pos-
some of the complexities of modern techniques they are sible to determine the stresses within the stiffeners. This
rapid to solve and useful within highly computationally method has been used for many years and is combined with
expensive algorithms, for example genetic algorithms. Pre- elastic stress analysis as covered in Datoo (1991).
vious research has been performed that has investigated The grillage analysis uses the Navier summations of
using these different grillage theories on composite struc- points within the grillage to develop the deflection of the
tures including Nagendra et al. (1995), Eksik et al. (2007) stiffeners. This methodology has been performed based on
and Maneepan et al. (2007), but these have concentrated a panel under simply supported boundary conditions as this
on more basic failure criteria. Whilst grillage analysis is will allow a conservative estimate of the stresses and deflec-
an important method used for analysis of top-hat stiffened tions. The equation giving deflection of the stiffened plate
composite structures, the analysis of up to date failure cri- is assumed to be
teria for these structures has not before been introduced
∞ 
 ∞
and the effects of these criteria on the optimised structure mπ x nπy
must be investigated to determine the effectiveness of the w(x, y) = amn sin sin (1)
L B
m=1 n=1
method.
Research has been performed on the effects of complex
where the value of amn is a coefficient found from (2). The
failure criteria on simple structures but it is important to
coefficient amn is found based on the assumption that the
understand how these criteria affect more complex topolo-
change in potential energy from a small deflection will be a
gies. For the first time this paper investigates the effects
minimum. The coefficient amn is dependent on the flexural
of these failure criteria on more complex structures using
rigidities of the stiffeners (Dg,b ).
tophat stiffened plates as an example. This paper deter-
mines a method for rapid assessment of composite tophat
16P LB
stiffened grillage panels, commonly used for increased tor- amn =   (2)
sional stability, and performs an optimisation on an example Dg Db
π mn m (g + 1) 3 + n (b + 1) 3
6 4 4
taken from boatbuilding. The structural model that has been L B
developed uses Navier grillage with Third Order Shear
Deformation theory to ensure that the panels in between The moments can be found in the beams or girders (Mg,b )
the stiffeners do not fail. The optimisations have been per- from (3):
formed using constraints taken from those developed for
the WWFE, a buckling criterion and a deflection criterion ∂ 2w
Mg,b = −Dg,b (3)
where each has been used as a constraint independently ∂x 2
allowing an analysis of the effects of using these as the over-
all limit state. Results from the analysis show that the Puck Finally using the maximum moments in the grillage the
failure criterion appears to be the most conservative of the maximum stress σmax can be determined, where Es(i) is the
World Wide Failure Exercise but that this state is easier to Young’s modulus of the element of a stiffener, either girder
satisfy than the deflection limit state. or beam, Ms is the moment created in the stiffener, Zs is the
vertical distance of the centroid of an element to the neutral
axis and Ds is the structural rigidity of a stiffener:
2 Structural modelling
Es(i) Ms Zs
σmax = (4)
2.1 Grillage method Ds

The structural analysis is performed on a grillage under the The tophat stiffeners are idealised as shown in the bot-
loading condition shown in the top right part of Fig. 1. tom part of Fig. 1 with each stiffener being made up of 4
738 A.J. Sobey et al.

elements labelled 1 to 4 where elements 1 to 3 are attached It is then possible to find the coefficients of the boundary
to a large base plate forming element 4. Each of these conditions using the stiffness matrix [C] by substituting into
elements is made up of a number of different plies. the equations of motion the (5) to (6).
A grillage panel is then constructed of a number of these    
 0   Umn 
stiffeners on top of a flat base plate. The flexural rigidity can    
 0   Vmn 
be found using these elastic equivalent properties method    
which can be found in Datoo (1991). The flexural rigidity [C][] =  Qmn  [] =  Wmn  (7)
 0   Xmn 
can then be used to determine the stresses in the stiffeners    
 0   Ymn 
using the Navier grillage method.
The stresses and strains then allow the use of failure
2.2 Third order shear deformation theory
mechanisms to determine whether a given thickness of plate
will fail. The maximum stress and deflection from the plate
The grillage method that has been used finds the maximum
can therefore be used with the failure criteria to determine
stresses in the stiffeners by assuming that the entire load is
whether the plate is safe.
passed through to the stiffening members. It is also impor-
tant to make sure that the plate of the hull is thick enough
to withstand the expected loads between each of the stiffen- 2.3 Failure criteria
ers. This can be done computationally easily using classical
laminate plate theory and first order shear deformation the- Having determined the stress and deflection within the panel
ory for thin structures with uncomplicated layups. As more it must be determined whether these values are applicable
complex layups are required it is necessary to use higher within a real structure both in terms of the strength of the
order shear deformation theories but these are computa- structure and the serviceability, in this case deflection. Fail-
tionally more expensive. Plate analysis has been calculated ure criteria have therefore been selected to determine a limit
using third order shear deformation theory (Reddy 2004) to for the dimensions that will survive the given condition.
determine the properties required for the failure criteria as Further to previous work reported by Sobey et al. (2009)
this will allow the full benefits of using different layups in extra failure criteria have been added to the model to more
the material to be used. accurately model the behaviour of the composite materials.
The forces at each point on the plate, q(x,y), are deter- The failure criteria used came from the ‘World Wide Fail-
mined from (5): ure Exercise’ (WWFE) (Kaddour et al. 2004; Hinton et al.
2002; 2004a). The choice made for each failure type can be
∞ 
 ∞
seen from Table 1 and was based upon the findings of the
q(x, y) = Qmn sin αx sin βy (5) World Wide Failure Exercise.
n=1 m=1
The failure envelopes generated for maximum stress for
a given material property are also shown in Fig. 2a and c.
where Qmn is the lateral loading on the plate and is given by:
The exercise concluded that in the case of buckling cri-
 teria that they ‘did not address the prediction of buckling
4 L B mπ x nπy modes of failure’ (Soden et al. 2004). Buckling is a key part
Qmn (z) = q(x, y) sin sin dxdy
LB 0 0 L B of failure in hull stiffeners and therefore an Euler based rule,
(6) seen in (8) and (9), where the crown and web are assumed

Table 1 Failure criteria

Failure type Criteria

Predicting the Puck and Schurmann (1998, 2002), Liu and Tsai (1998) and
response of lamina Kuraishi et al. (2002)
Predicting final strength Puck and Schurmann (1998, 2002)
of multidirectional laminates
Predicting the deformation Zinoviev et al. (1998, 2002) and Puck and Schurmann (1998, 2002)
of laminates
Predicting final strength Puck and Schurmann (1998), Zinoviev et al. (1998), Liu and Tsai (1998)
of multidirectional laminates and Kuraishi et al. (2002
deformation of laminates
Implications of failure criteria choices on the rapid concept design of composite grillage structures 739

Fig. 2 Failure criteria stress

envelopes

(a) Puck criterion

(b) Tsai criterion (c) Zinoviev criterion

to be taken as clamped at both ends has been used to con- and the panels in between the stiffeners by the TSDT theory
strain the model for both the crown and the webs and is reported in Sections 2.1 and 2.2.
taken from Gambhir (2004). This criterion shows that the
maximum stress in the web or crown must not increase
beyond that defined using the material properties and the 2.3.1 Puck failure criteria
geometry.
The Puck failure criterion is based upon 3-D phenomeno-
6.97π 2 Es logical models where the development of the method is done
σcri,web =   (8)
12 1 − υ12
2 (d /c )2
s s through matching current theory to experimental results
(Table 2). The method is a composite laminate theory
6.97π 2 Es method which is nonlinear to solve. The Puck method is rec-
σcri,crown =   (9) ommended by the World Wide Failure Exercise to be used
12 1 − υ12
2 (a /b )2
s s
for predicting strength of unidirectional laminae and this
Furthermore an arbitrary deflection criterion of 10 % of method has been used as it gives a more conservative view
the length has been included to ensure that materials with for the failure of the laminates. Puck’s formulation is also
a low stiffness and cost cannot be selected without creat- used for predicting the initial strength of multidirectional
ing a thicker topology. These failure criteria had require laminates as other methods did not predict the failure well.
the inputs of material properties and maximum stresses and Puck is further recommended to be used to predict final
deflections provided for the stiffeners by the grillage theory strength of multidirectional laminates. While this method is

Table 2 Puck failure criteria

υ
Fibre failure in tension 1
1T 1 + Eff121 mσf σ2 = 1


 υ 
Fibre failure in compression 1
1C  1 + Eff121 mσf σ2  = 1 − (10γ21 )2

2

2
τ12 (+) YT 2 (+) σ2
Inter-fibre failure mode A S12 + ρ⊥ S21 + YσT2 + ρ⊥ S12 = 1 −
σ1
σ1D

(for transverse tension)  

Inter-fibre failure mode B 1 2 + (ρ (−) σ )2 + ρ (−) σ = 1 −
τ21 σ1
S21 ⊥ 2 ⊥ 2 σ1D

(for moderate transverse compression)  

2
2
τ21 σ2 YC σ1
Inter-fibre failure mode C (−) + YC (−σ2 ) =1− σ1D
2(1+ρ⊥⊥ )S21

(for large transverse tension)

740 A.J. Sobey et al.

Table 3 Zinoviev failure criteria 3.1 Genetic algorithms

Longitudinal tension failure σ1 = XT
Genetic algorithms are a multiobjective optimisation
Longitudinal compressive failure σ1 = XC
method that will allow fast resolution of results while
Transverse tensile failure σ2 = YT
searching a large search space. Genetic algorithms work by
Transverse compressive failure σ2 = XC
copying the process of DNA transfer in living organisms.
In-plane shear failure τ12 = S12
They then use the process of evolution to find the optimum
solution for a given search space.
recommended for most cases there is a requirement to use
other criteria at the same time to ensure that a conservative 3.1.1 Introduction
analysis is performed as this criteria is not conservative for
all cases. Embedded genetic algorithms have been used in the opti-
misation process as shown in Fig. 3 and developed from
2.3.2 Zinoviev failure criteria Coley (2001). The main genetic algorithm is used to opti-
mise the stiffener spacing in the longitudinal and transverse
The Zinoviev failure criterion is based on the development directions, the material type and layup angles. The embed-
of maximum stress theory (Table 3). This method is based ded algorithm is used to determine the optimal geometry of
on composite laminate theory and has a linear solution. the stiffeners through creation of the crown width and thick-
Zinoviev is recommended by the World Wide Failure Exer- ness as well as the web height and thickness. This means
cise to predict the deformation of laminates along with a that the embedded algorithm will develop an optimal stiff-
non-linear method such as Puck. ener geometry, with respect to mass and cost, for each given
stiffener spacing and material property generated in the
2.3.3 Tsai failure criteria main algorithm. The main algorithm will then determine the

The Tsai failure criterion is developed through an interactive

progressive quadratic failure criterion. This method is also
based on composite laminate theory and is linear in its solu-
tion. The Tsai failure criterion is used in conjunction with
Puck to determine the response of lamina. The Tsai failure
criterion is the best fit to the test data reported in Soden et al.
(2004) for the behaviour of the laminates. This criterion
underestimates the failure stress at given points and so the
Puck failure criterion can be used to check that failure does
not occur.
 2  2  
σ1 σ2 1 1
+ + − σ1
XT XC YT YC XT XC
   
1 1 2F1 2σ1 σ2
+ − σ2 + √
YT YC XT XC YT YC
 2
τ12
+ =1 (10)
S12

3 Optimisation

Due to the nature of modeling structures there is a compro-

mise between many different input variables all of which
must be manipulated correctly so that the optimum structure
for the design criteria can be created. Design is also a pro-
cess that will require rapid generation of results and for this
reason genetic algorithms have been used due to the large
search space available. Fig. 3 Genetic algorithm flow diagram
Implications of failure criteria choices on the rapid concept design of composite grillage structures 741

Table 4 Genetic algorithm

constraints Property Bounds Property Bounds

Long. stiffener spacing 0–10230 mm Ply Angles 0,90

Trans. stiffener spacing 0–2046 mm Ply Materials E-glass, Aramid,
Carbon, HM Carbon
Number of plies 0–32 Long. Crown width 0–102.3 mm
Long. crown height 0–20.46 mm Long. web width 0–20.46 mm
Long. web height 0–102.3 mm Trans. crown width 0–102.3 mm
Trans. crown height 0–20.46 mm Trans. web width 0–20.46 mm
Trans. web height 0–102.3 mm Plate thickness 0–102.3 mm
Stiffener base width 0-102.3 mm Stiffener base width 0–102.3 mm

reaction for the total grillage and compare these grillages with the best fitness. This is done by picking a tournament
over many generations until the optimal topology is found. size and from this selection choosing the fittest selections to
The constraints for these different properties are shown go through to the next round.
in Table 4.
This leads to a description of the objective function and
constraints in terms of the design variables used
4 Production modelling
minψ(cost, mass)
= 1/cost (ab , bb , cb , db , eb , ag , bg , cg , dg , eg , To determine the cost accurately it is important to model the
Nplies , a, b, CMaterial ) production route. For the example given later hand layup is
used for the production technique as it is the method most
+ 1/mass(ab , bb , cb , db , eb , ag , bg , cg , dg , eg ,
prevalence used within boatbuilding community. Production
Nplies , a, b, ρMaterial ) (11) modelling was originally performed using a parametric cost
subject to model taken from the SSA report by Shenoi et al. (2003) as
shown in Table 5.
σ < σlimit , σ < σcri andδ < δlimit (12)
This model has no cost for stiffeners and is for a sand-
wich plate. This has therefore meant that a stiffener cost
3.1.2 Initial population model has been attached to the main model replacing the
cutting and laying core section of the SSA production model
The first step in generating a solution for the genetic algo- for each longitudinal and transverse section and shown in
rithm is to develop the initial population of strings. The Table 6.
strings are made up of binary numbers, each section of
which represents part of the geometry of the stiffener. This
is done through a random number generator found in Press
(1986). The first algorithm is made up of 100 strings of 60
Table 5 SSA sandwich panel production model (Shenoi et al. 2003)
numbers the second algorithm is created from 100 strings of
80 numbers. Each algorithm runs for 300 generations as this Action Cost(mins)
was found to be a value for which an adequate convergence
Fairing compound 10 min/m2
occurred.
Smoothing fairing compound 60 min/m2
Apply release compound 10 min/m2 /ply
3.1.3 Exploiting operator Cutting cloth 10 min/m2 /ply
Laying cloth 5 min/m2 /cloth
The exploiting operator is the selection process which Cutting and laying core 60 min/m2 /core
chooses the strings to be used and those which will no Apply resin with brush or rollers 10 min/m2
longer be used. This is done using the criteria of a fitness Remove the components 30 min/m2
function which will normally be based on a function of the from the mould
inputs f (x) for maximisation or 1/f (x) for minimisation Quality inspection 3 min/m2
problems. The main exploiting operators are outlined below.
Trim 15 min/m/edge
Tournament selection is carried out by using only the values
742 A.J. Sobey et al.

Table 6 SSA production model Table 8 Validation of Navier method grillage analysis—Shear stress

Action Cost(mins) Property Datoo (1991) Maneepan et al. (2007) Current

(MPa) (MPa) (MPa)
Cutting cloth 10 min/m2 /ply
Laying cloth 5 min/m2 /cloth τ1 99 98.72 98.72
Cutting and laying core 60 min/m2 /core τ2 101 102.76 102.76
Apply resin with brush or roll 10 min/m2

17.7 kN/mm2 . A shear force of Q = 10.0 kN is found in

The time for each action has been transformed into a cost the stiffeners. The thickness of the flanges are 1.0 mm and
by using a wage of £ 20/hr. To determine the raw material the thickness of the web is 0.5 mm. τ1 is the shear stress at
costs for the stiffeners cost per kg for each material has been the corner of the crown element, τ2 is the shear stress at the
used developed from a database of materials form Lloyds neutral axis of the cross section (Table 8).
Register. These values had no deviation from the results found in
Maneepan and there is only a small deviation found com-
pared to the results found in Datoo. It is therefore considered
5 Applications that the grillage theory is capable of calculating the shear
stress.
5.1 Structural verification Finally the elastic equivalent properties were compared
to Datoo (1991) using lamina properties E1 = 140 kN/mm2 ,
Verification of both parts, Navier grillage analysis and E2 = 10 kN/mm2 , G12 = 5 kN/mm2 , υ12 = 0.3 and a ply
TSDT, of the first principles structural analysis method was thickness = 0.125 mm for each of the 8 plies all having a 0◦
performed. The results from the grillage method have been ply angle where the result was identical to Datoo’s value of
compared to those found in Clarkson (1965) for a panel 140 GPa.
with a length and width of 3810 mm. The panel consisted For the validation of third order shear deformation the-
of 4 transverse beams and longitudinal girders with dimen- ory a layup of [0/90/90/0] has been used. The length to
sions 254 mm deep 127 mm wide with 18.288 mm thick width ratio (L/B) of the plate is equal to 1.0 and the length
flanges and 9.144 mm thick webs and a pressure of 137.9 to thickness ratio (L/t) is 100. The material properties are
kPa was applied to each panel. The results are presented in E1 = 175 GPa, E2 = 7 GPa, G12 = G13 = 3.5 GPa,
Table 7. G23 = 1.4 GPa, and υ12 = υ13 = 0.25. The load acting on
These results were obtained with a wave number of 11, the plate is q0 = 50 kPa.
shown in (1) as m and n, as it is this minimum value at which From the validation of the third order shear deforma-
the deflection converges. tion theory it is possible to see that the results have at most
These values were found to be close to results found in a 1 % deviation from those given in Reddy (2004) at a
Maneepan as can be seen in Table 7. Furthermore these val- value for the wave numbers of eleven showing Third order
ues are similar to Clarkson, using the folded plate method, Shear Deformation Theory has been modelled accurately
which has been compared to experimental results but also (Table 9).
remain conservative. The grillage method was deemed valid
for the stiffener modelling. 5.2 Genetic algorithm verification
A validation of the shear stress has been made in com-
parison with a theoretical rectangular box beam found in Genetic algorithms can be tested to determine if the optimi-
Datoo (1991). The web height is 50 mm and the flange sation that has been carried out reaches the optimum value.
widths are 200 mm. The Young’s modulus of the flanges
are 54.1 kN/mm2 . The Young’s modulus of the web is
Table 9 Validation of Third Order Shear Stress Deformation Theory
(TSDT)
Table 7 Validation of Navier method grillage analysis—Stress
L/t Reddy(w̄ × 102 ) TSDT(w̄ × 102 )
Property Clarkson (1965) Maneepan et al. (2007) Current
10 1.0219 1.0102
Deflection 9.63 mm 9.93 mm 9.87 mm 20 0.7572 0.7546
Stress 165.52 MPa 171.19 MPa 170.13 MPa 100 0.6697 0.6696
Implications of failure criteria choices on the rapid concept design of composite grillage structures 743

Shear Deformation Theory for assessing the plates between

the stiffeners. Each failure model has been used as the
assessment for failure or success separately to determine
how these failure criteria affect the optimal structure.
For each different failure criterion a simple study has
been performed on a horizontal section of hull at the bot-
tom of the boat. This has been carried out using a grillage
panel length of 24 m and width 2 m. The structural anal-
ysis has been implemented using the pressure, 131.47 kPa,
from Lloyd’s Register for Special Service Craft as this gives
the most conservative estimate ensuring that the masses and
costs used within the comparison are likely to be for the
worst case scenario.

Fig. 4 Validation of genetic algorithm using different starting points

6.1 Puck

The first optimisation was undertaken using the Puck fail-

This is investigated by starting the algorithm at different ure criterion of Section 2.3.1. The topology of the stiffened
points and determining if, at the finish, all the algorithms panel that was produced using the Puck failure criteria with
reach a similar fitness function. Genetic algorithms require the first principles method, shown in Section 2.1 can be
that the best fitness value, after each generation, will grad- seen in Table 10. The optimised thickness of the stiffened
ually increase. This leads to a distinctive “handgun” shaped plate and the spacing of the transverse and longitudinal stiff-
graph when fitness function is plotted against generation. If eners from the Puck criteria analysis are also reported in
the same algorithm is started from different points this will Table 10.
lead to the optimisation reaching similar fitness functions as The web thickness of 0.86 mm is small in comparison
shown in Fig. 4. to those that would be expected in a real application. Due
As can be seen from the examples in Fig. 4 the graph to the Puck failure criteria being stress-based the optimisa-
follows the distinctive genetic algorithm shape where each tion is attempting to reduce the maximum stress within the
of the individual strands reaches a similar final result from whole panel. This involves an increase in the moment of
whichever starting position. This shows that the algorithm inertia in the stiffeners and for a low stress in the stiffener a
is working correctly therefore validating the optimisation. high neutral axis is therefore required. The thickness of the
web therefore does not affect the moment of inertia as much
as the thickness of the crown and its distance from the plate.
6 Results The panel topology has a wide stiffener spacing and a small
plate thickness. Out of plane pressure on the panel devel-
For each failure criterion reported a comparison has been ops a stress that is not as complex as a real life situation,
made between the optimised structures that can be created lacking axial, torsional and shear forces, leading to thin-
for the constraints of those failure criteria. The grillage that ner panel dimensions and a large stiffener spacing, than is
is generated is developed for use within leisure boats, as an realistic.
example. This optimisation used a genetic algorithm reduc-
ing both the mass and the cost of the developed structures.
The cost was developed using a simple parametric model. 6.2 Tsai
The genetic algorithm used a weighting of 0.5 for cost and
mass. The choice of structural models comprises of Navier The second optimisation was undertaken using the Tsai
Grillage theory for assessing the stiffeners and Third order failure criterion of Section 2.3.3. The optimised topology

Table 10 Panel topology for

puck failure criteria Stiffener type Web Web Crown Crown Stiffener Plate Ply
height thickness width thickness spacing thickness angles

Longitudinal 100.7 mm 0.86 mm 5.6 mm 1.78 mm 2200 mm 1.2 mm 0/0/90

Transverse 36.1 mm 4.16 mm 5.6 mm 2.78 mm 570 mm 1.2 mm
744 A.J. Sobey et al.

Table 11 Panel topology for

Tsai failure criterion Stiffener type Web Web Crown Crown Stiffener Plate Ply
height thickness width thickness spacing thickness angles

Longitudinal 38.3 mm 0.02 mm 1.1 mm 6.28 mm 430 mm 0.5 mm 0/90

Transverse 71.3 mm 0.14 mm 12.5 mm 4.9 mm 40 mm 0.5 mm

produced using only the Tsai failure criterion can be seen in 6.3 Zinoviev
Table 11.
This failure criterion again produced a topology with a The third optimisation was run using the Zinoviev failure
thin web thickness due to the nature of the optimisation criterion of Section 2.3.2. The topology of the optimum
attempting to reduce maximum stress but web thickness stiffener plate using only the Zinoviev failure criterion can
having little effect on the neutral axis and hence on this be seen in Table 12.
value. The stiffeners themselves are small in comparison The Zinoviev criterion produced a similar panel to the
to those from the Puck criterion. Furthermore the stiffener other failure criteria selected from the World Wide Failure
spacings are narrow and the panel thickness is thinner than Exercise producing a high stiffener with a thin web and a
the other cases. The criteria of the World Wide Failure Exer- thick crown. The Zinoviev criterion did show a large dif-
cise are similar, being reliant on the maximum stress and ference between the transverse and longitudinal stiffeners.
having been produced to fit the same experimental data, it The longitudinal stiffener shape did not appear to make
would be expected that the panel produced would be simi- much of a difference to the plate strength with the trans-
lar to that of the Puck criterion. The mass produced using verse stiffeners providing most of the strength which is why
the Tsai failure criterion is small and the cost quite large the longitudinal crown is small. Furthermore the stiffener
compared to the other plates, as seen in Figs. 5 and 6, and spacings are a large distance apart with a thin panel. The
therefore it is likely that the different shape of the fail- stiffener topology was most similar to that produced using
ure envelope, shown in Fig. 2b, led the evolution of the the Puck failure criteria. This result is to be expected as the
genetic algorithm down a different route. This is shown envelope, shown in Fig. 2c, within in which the combined
from the small stiffener spacing. The extra stiffeners there- stresses would not cause a failure for the Zinoviev and Puck
fore allowed a reduction in the stiffener size but created criteria are similar.
extra cost. This shows that for a large number of stiffeners
a high cost is incurred but a low mass is possible. Due to 6.4 Deflection
the weightings being equal between the mass and the cost
it was possible to gain a similar fitness function through a The fourth optimisation was run using the arbitrary failure
plate that concentrated on low mass and high cost, high cost criterion of 10 %. This failure criterion meant that the max-
and low mass or a compromise. imum deflection of the plate could be no more than 10 % of

Fig. 5 Comparison of cost for failure criteria Fig. 6 Comparison of mass for failure criteria
Implications of failure criteria choices on the rapid concept design of composite grillage structures 745

Table 12 Panel topology for

Zinoviev failure criteria Stiffener type Web Web Crown Crown Stiffener Plate Ply
height thickness width thickness spacing thickness angles

Longitudinal 91.9 mm 0.06 mm 0.2 mm 17.22 mm 1130 mm 1.5 mm 0/90

Transverse 95.3 mm 0.02 mm 22.4 mm 2.02 mm 2200 mm 1.5 mm

Table 13 Panel topology for

deflection Stiffener type Web Web Crown Crown Stiffener Plate Ply
height thickness width thickness spacing thickness angles

Longitudinal 45.8 mm 0.84 mm 6.6 mm 13.86 mm 430 mm 0.7 mm 0/90

Transverse 83.5 mm 0.52 mm 23.5 mm 1 mm 170 mm 0.7 mm

Table 14 Panel topology for

buckling Stiffener type Web Web Crown Crown Stiffener Plate Ply
height thickness width thickness spacing thickness angles

Longitudinal 37.2 mm 2.34 mm 33.9 mm 6.76 mm 2130 mm 0.1 mm 0/90/0/90

Transverse 45.9 mm 2.94 mm 82.7 mm 2.1 mm 2200 mm 0.1 mm

its length. The resulting optimised stiffener topology can be to the stiffeners only. The resulting optimised topology for
seen from Table 13. the buckling criteria is given in Table 14.
This optimised plate topology is similar to those found The buckling criteria developed a stiffener topology dif-
using the World Wide Failure Exercise criteria and this is ferent to those found using the other failure criteria. The
due to the stress in the panel being based upon the deflec- main difference with this criterion was it developed a stiff-
tion. This requirement means that to minimise one output, ener web thickness and crown height that was thicker than
stress, a similar topology will be required to minimise the the corresponding dimensions found using the other criteria
other, deflection. This therefore meant for a low deflection as seen in Section 2.1. This is due to buckling being depen-
it was also important to have a high neutral axis. Further to dant on the equivalent thickness of the stiffening elements
this criterion, the material property would have been more in comparison the length of those elements. For buckling
important as the material’s stiffness would have made a dif- not to occur there is still a requirement that the stress was
ference to the deflection. Since the material selected was low and therefore it can be seen that the stiffening elements
E-glass, due to its low cost, the stiffening elements were produced during the optimisation were tall to increase the
required to be much larger due to the poor stiffness char- neutral axis. The panel topology developed a wide stiffener
acteristics of the material or a larger number of stiffeners spacing and a thin panel thickness as these criteria did not
are required. This showed that in terms of the grillage plate, affect the buckling of the stiffener.
the failure criteria chosen and the predicted pressure that the
constraints for deflection were more important than those
of stress. The deflection criterion requires that the stiffener
spacing is small as more stiffeners created a less flexible 7 Discussion
panel. Furthermore the thickness of the panel is small as this
part of the topology did not affect the deflection of the plate. A method for first principles structural modelling has
been developed and verified. The model has been attached
to an optimisation algorithm and different failure models
6.5 Buckling have been tested to determine the manner in which they
affect the optimum structure. A review of these results for
Finally buckling failure criteria have been applied to the the mass and cost of this section are shown in Figs. 5
stiffeners on the grillage. The buckling criterion was applied and 6.
746 A.J. Sobey et al.

The models of the World Wide Failure Exercise, Puck, limits is not fully taken into account. From the analysis per-
Tsai and Zinoviev, and maximum deflection criteria devel- formed it is possible to see that the failure criteria affected
oped a similar topology with small web thickness and crown the optimised structures in a different manner. While the
height. These results show a similarity between them except method used is not as complex as others that can be chosen
for the Tsai failure criterion where the genetic algorithm the method was rapid allowing the use of the model with-
may have followed a different evolutionary route. This was out developing a metamodel. It will be important for future
due to the use of a 50/50 weighting between mass and cost. analysis to compare the final results to that done using a
It was therefore possible to gain a similar fitness function finite element analysis comparing and contrasting these dif-
result between those that used a small stiffener spacing, and ferent methods for use in rapid initial grillage design. While
therefore generated a lower mass and higher cost, and those the results of the grillage analysis are not as complex of
that used a larger stiffener spacing for a lower cost but a those used within FEA it can also be seen that changes to
higher mass. The buckling failure criterion ensured that the the failure criteria have a large effect on the design of the
stiffener web and crown thicknesses were larger than for the grillage.
other cases. This result could be improved through the intro-
duction of more generations within the optimisation. This
can be seen from the manner in which it varies from the 8 Conclusion
other topologies with a higher cost but a lower mass than
the other World Wide Failure Exercise criteria. A fault of A method for rapid analysis of composite structures has
the genetic algorithm is that it may find a close to optimum been proposed and validated. This method has been used
solution. A comparison of the fitness functions show that for the first time to assess the effects of a number of fail-
the Tsai failure criterion was slightly below that of the Puck ure criteria on complex tophat stiffened structures. While it
failure criterion supporting this argument, as does Fig. 4 can be seen that the deflection criteria produced the most
which shows that the final fitness function was slightly dif- constrained overall result it is possible to see that the Puck
ferent for optimisations that were run from different starting criteria was the most constraining of the world wide fail-
points. This behaviour can be reduced through an increase ure exercise criterion. The results show further indicate that
in generations. this method of modelling could be utilised in concept design
The implications for these results are that the criteria used work allowing investigation of complex structures at an
show a difference in the optimum geometry that was devel- early stage. This method will allow rapid analysis of struc-
oped showing the method developed works to a reasonable tures than that obtainable by combining FEA and genetic
level of complexity. As expected for the e-glass composite algorithms and allowing a less constrained analysis than
materials the serviceability limit state is most difficult to sat- using metamodels.
isfy with a low cost and mass than that of ultimate strength.
It is important to note, as none of the failure criteria devel-
oped match all failure modes, as can be seen in Table 1, that References
each of these failure criteria individually do not incorporate
the entire strength limit state. It can be seen however from Clarkson J (1965) The elastic analysis of flat grillages. Cambridge
the WWFE (Kaddour et al. 2004; Hinton et al. 2002, 2004a) University Press
that the Puck failure criteria in particular match many of Coley DA (2001) An introduction to genetic algorithms for scientists
and engineers. World Scientific
these failure criteria well and therefore represents a criteria Datoo MH (1991) Mechanics of fibrous composites. Elsevier Science
close to matching this limit state. This is due to its phenome- Publishers Ltd., Essex, England
logical nature allowing effects noticed in experiments to be Eksik O, Shenoi RA, Moy SSJ, Jeong HK (2007) Finite element anal-
replicated within the theory allowing a more accurate repre- ysis of top-hat stiffened panels of fibre reinforced plastic boat
structures. Mar Technol 44:16–26
sentation of the real life phenomena. While the Puck criteria Gambhir ML (2004) Stability analysis and design of structures. Springer
is the most conservative of the World Wide Failure Exercise Hinton MJ, Kaddour AS, Soden PD (2002) Evaluation of failure pre-
criteria its use within structural design can be difficult due diction in composite laminates: background to ‘part b’ of the
to the requirement for a large amount of material data. This exercise. Compos Sci Technol 62:1481–1488
Hinton MJ, Kaddour AS, Soden PD (2004a) Evaluation of failure
requires estimates of the properties required to implement prediction in composite laminates: background to ‘part c’ of the
new materials or the use of previously used materials only. exercise. Compos Sci Technol 64:321–327
Grillage theory has advantages in the speed with which Hinton MJ, Kaddour AS, Soden PD (2004b) Failure criteria in fibre
the results can be processed while giving a good indication reinforced polymer composites. Elsevier Ltd.
Kaddour AS, Hinton MJ, Soden PD (2004) A comparison of the
for the final structural topology. Though these results can predictive capabilities of current failure theories for composite
be useful for initial design the complexities for the man- laminates: additional contributions. Compos Sci Technol 64:449–
ner in which stiffener spacing and geometry affect the final 476
Implications of failure criteria choices on the rapid concept design of composite grillage structures 747

Kang J, Kim C (2005) Minimum-weight design of compressively Puck A, Schurmann H (2002) Failure analysis of frp laminates by
loaded composite plates and stiffened panels for postbuckling means of physically based phenomenological models. Compos Sci
strength by genetic algorithm. J Compos Mater 69:239–246 Technol 62:1633–1662
Kim S, Kim C (2010) Optimal design of composite stiffened panel Reddy JN (2004) Mechanics of laminated composite plates and shells,
with cohesive elements using micro-genetic algorithm. Compos 2nd edn. CRC Press
Struct 44:369–387 Satheesh R, Narayan Naik G, Ganguli R (2010) Conservative design
Kuraishi A, Tsai SW, Liu KKS (2002) A progressive quadratic failure optimization of laminated composite structures using genetic algo-
criterion, part b. Compos Sci Technol 62:1683–1695 rithms and multiple failure criteria. J Compos Mater 44:369–
Lopez RH, Luersen MA, Cursi ES (2009) Optimization of laminated 387
composites considering different failure criteria. Compos, Part B Sobey AJ, Blake JIR, Shenoi RA (2009) Optimisation approaches to
Eng 40:731–740 design synthesis of marine composite structures. Schiffstechnik
Liu K, Tsai SW (1998) A progressive quadratic failure criterion for a Bd.54 - STR 56(1):24–30
laminate. Compos Sci Technol 58:1023–1032 Shenoi RA, Dulieu-Barton JM, Jeong HK, Blake JIR (2003) Manual
Maneepan K., Shenoi JIR, Blake JIR (2007) Genetic algorithms of design and production best practice. Technical Report No. 55,
(gas) based optimisation of frp composite plated grillages in ship University of Southampton
structures. Int J Marit Eng 149:1–19 Soden PD, Kaddour AS, Hinton MJ (2004) Recommendations for
Nagendra S, Jestin D, Gurdal Z, Watson LT (1995) Improved genetic designers and researchers resulting from the world-wide failure
algorithm for the design of stiffened composite panels. Comput exercise. Compos Sci Technol 64:589–604
Struct 58:543–555 Vedeler G (1945) Grillage beams. Grondahl and Son
Naik GN, Gopalakrishnan S, Ganguli R (2008) Design optimization of Zinoviev PA, Grigorviev SV, Lebedevab OV, Tairova LP (1998) The
composites using genetic algorithms and failure mechanism based strength of multilayered composites under a plane-stress state.
failure criterion. Compos Struct 69:354–367 Compos Sci Technol 58:1209–1223
Press WH (1986) Numerical recipes. Cambridge University Press Zinoviev PA, Lebedeva OV, Tairova LP (2002) A coupled analysis of
Puck A, Schurmann H (1998) Failure analysis of frp laminates by experimental and theoretical results on the deformation and failure
means of physically based phenomenological models. Compos Sci of composite laminates under a state of plane stress. Compos Sci
Technol 58:1045–1067 Technol 62:1711–1723

View publication stats