Sei sulla pagina 1di 20

Engineering Failure Analysis 14 (2007) 118–137

www.elsevier.com/locate/engfailanal

Damage tolerance based shape design of a stringer cutout


using evolutionary structural optimisation
a,b,* a,b c
R. Das , R. Jones , S. Chandra
a
DSTO Centre of Expertise in Structural Mechanics, Department of Mechanical Engineering, Monash University,
Wellington Road, Vic. 3168, Australia
b
Rail CRC, Department of Mechanical Engineering, Monash University, Wellington Road, Vic. 3168, Australia
c
Advanced Research Group, Structures Division and Centre for Civil Aircraft Design and Development,
National Aerospace Laboratories, Bangalore, India

Received 12 September 2005; accepted 21 November 2005


Available online 14 February 2006

Abstract

This paper uses a modified evolutionary structural optimisation (ESO) algorithm for optimal design of a stringer cutout
used in many pressured fuselages of transport aircrafts. Both stress and fracture based ESO algorithms are studied. Ini-
tially, a stress optimised shape of the stringer cutout is determined. It is found that the ESO method significantly reduces
the peak stress on the boundary. A residual strength based optimisation is then performed. The fracture based evolution-
ary structural optimisation algorithm uses residual strength as the design objective. The formulation used here allows for
numerous cracks to be located along the entire structural boundary. The fracture ESO algorithm considerably improves
the residual strength by decreasing the maximum stress intensity factor. It also reduces the variability in the stress intensity
factors for the optimum shape and produces a near uniform level of fracture criticality around the boundary. It is also
shown that the shapes optimised for stress and fracture strength may differ, and a fracture strength based optimisation
may produce a lighter design. This highlights the need to explicitly include fracture parameters in the design objective
function.
Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Structural optimisation; Evolutionary algorithm; Damage tolerance; Finite element analysis; Residual strength

1. Introduction

The requirement that flaws are detected before reaching their acceptable size is now a well established prac-
tice as part of design and maintenance guidelines in aerospace industry. For transport aircrafts one of the crit-
ical components is the pressurised fuselage that needs a ‘two-bay-crack criterion’ to be satisfied as part of the
certification requirement. This criterion specifies that a longitudinal crack present in the fuselage skin with a

*
Corresponding author. Tel.: +61 3 9905 3571; fax: +61 3 9905 1825.
E-mail address: rajarshi.das@eng.monash.edu.au (R. Das).

1350-6307/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfailanal.2005.11.008
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 119

length of two frame bays above a broken centre frame should not cause failure of the structure [10,1]. This
requirement is to ensure that neither the skin nor the frame experiences a fracture failure. For pressurised fuse-
lages in medium and long range aircrafts the two-bay-crack criterion is used to design large portions of the
upper and side shells of the fuselage.
Stringer cutouts are commonly found as part of bulkheads in pressurised fuselage. Fig. 1 shows a typical
stringer cutout used in a P3 Orion maritime aircraft. The stringer cutout passes through a shear clip. This
shear clip is attached to the bulkhead or (in some cases) integrally machined with it. The bulkheads are usually
constructed using sheet metals. In some aircrafts, e.g., KC-135, there is no shear clip as the fuselage was
designed without considering the damage tolerance criteria [3].
The stringer cutout is placed at a very critical location. It often provides crack initiation sites, and can lead
to failure or detachment of the bulkhead. As such, these stringer cutouts need a damage tolerance assessment
both in the design and during the development of structural rework geometries.
Enhancing the residual strength of the stringer cutout can increase its durability and prolong the inspection
intervals. One way of improving the residual strength is to adopt a ‘fracture based design’. The first step will be
to design a stress optimal shape to enhance the residual strength by minimising the peak stress. It is now
known that a stress optimal shape may have an improved residual strength, but it does not necessarily render
the best (optimal) residual static strength [6]. One approach to further enhance the residual strength will be to
perform an ‘explicit’ residual strength optimisation. This shape can not only be incorporated in the design
stage, but also be adopted, as closely as possible, during the ‘rework’ or ‘repair’ of an existing stringer cutout.
This will ensure an improved damage tolerant design, thereby reducing the likelihood of failure due to flaws
originating from any portion of the cutout. An additional aim of optimum shape design of an aircraft com-
ponent may be to monitor the change in weight of the resulting structure, although for general shape optimi-
sation weight is usually not the prime concern.
One approach of shape optimisation involves the use of mathematical programming methods, which
require computation of the gradients of the objective function. The process of gradient calculation is compu-
tationally expensive and often difficult in cases where the design objective function has discontinuities, or an
analytical expression of the gradient function is not easily obtainable. To overcome this, heuristic or non-tra-
ditional methods are becoming popular.
In this paper we use the evolutionary structural optimisation algorithm for shape optimisation of a typical
‘stringer cutout’. The fracture based ESO algorithm presented in [7] can predict near optimal shapes of struc-
tural components containing defects or cracks. An in-house optimisation software NASESO will be employed
to investigate shape optimisation using ESO. This software integrates the ESO algorithm with the FE solver
NE-NASTRAN to form an automated optimisation system. A stress based optimisation will be performed
first. We will subsequently consider the residual strength optimisation of the structure in the presence of flaws

Fig. 1. Stringer cutout from a P3 Orion maritime aircraft [2].


120 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

along its boundary. In this study we attempt to design the optimised boundary profile of the cutout to max-
imise its residual strength.

2. Evolutionary structural optimisation

Evolutionary structural optimisation (ESO) is a heuristic optimisation method. The ESO algorithm mimics
Darwinian principles of evolution in naturally occurring structures. It has been observed that naturally occur-
ring species tend to achieve shapes that are close to ‘fully stressed’ configurations, as this leads to an optimal
material utilisation. The basic principle of ESO can be expressed as: if a portion of material in a structure does
not contribute effectively to the functioning of the structure with respect to its design objective(s), then it can
be removed from that region. This gradual removal process leads to a structure that meets the design objec-
tive(s) subject to constraints [11]. Since ESO removes material completely without distributing the material
locally in a specified region or altering its density, it is also called a hard kill or a (0,1) method.
Evolutionary structural optimisation removes inefficient material from a structure based on a set of prede-
fined criteria. Here the term ‘inefficient’ means that the material is not taking part, or contributing effectively,
to the overall performance of the structure. Finite element method is generally used to compute the structural
response. Material elimination is accomplished by progressively removing elements from the FE model. In
general, the objective function may be the stress, the stress intensity factor, the weight, or any other relevant
performance criteria. A contribution factor or fitness index is defined for each element, which measures the
contribution of that element towards the overall structural behaviour. Given an initial design domain, loads,
and constraints, ESO will remove those elements whose contribution factors are less than a reference value.
The updated structure is then re-analysed. Depending on the response of this new structure, the algorithm
again identifies elements with a low fitness number and eliminates them from the structure. This process is
continued until the resulting structure satisfies a convergence criterion.

3. Stress based formulation

One main concern of the designer is to avoid stress concentration in structural boundaries, especially in the
areas of high and/or rapidly changing curvatures. Peak stress locations can act as potential sites for crack ini-
tiation and can lead to fracture failure of the structure. Therefore, in stress based shape optimisation, a certain
portion of the structure is considered as the design boundary, and the shape of this is varied to obtain an opti-
mised shape with a reduced peak stress and/or a near uniform stress distribution. Stress optimised shapes can
be used as starting shapes for fracture strength optimisation, and hence, stress based shape optimisation has
enormous significance in any basic design optimisation process.
The general evolutionary structural optimisation algorithm, that removes material from everywhere in a
structure, has been modified to apply it to shape optimisation problems. The modified algorithm, called ‘Nib-
bling ESO’, only removes material from a specified design boundary/surface, whose shape is to be determined.
The elements lying on the structural boundary being optimised are termed as the design elements. In this case
the maximum principal stress at the centroid of each boundary (design) element is chosen as the ESO criteria.
At each iteration, the elements with a lower maximum principal stress are removed from the design boundary.
This process leads to a gradual reduction in the peak boundary stress and may in some cases (for simple
shapes) produce a nearly uniformly stressed boundary. In this approach, the ith element is removed from
the current structure if
rp;i 6 rp;ref ð1Þ
where rp,i is the maximum principal stress at the centroid of the ith element, rp,ref is the reference maximum
principal stress, rp,max is the peak maximum principal stress for all the design elements, EF is the elimination
factor.
In the present case the reference stress is calculated using the following equation:
rp;ref ¼ EF  rp;max ð2Þ
where EF is calculated as
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 121

EF ¼ SR þ FR  ð1  SRÞ ð3Þ
where SR is the ratio of the minimum to maximum value of the ESO criteria. In the stress based case,
SR = rmin/rmax, and FR is a control factor. Using this relation the reference maximum principal stress
becomes
rref ¼ rmin þ FR  ðrmax  rmin Þ ð4Þ
A convergence function measuring the uniformity of stress distribution around the boundary is used in this
work. It is given by
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3,
Pnb 2 Pnb 2
1 ðr  r Þ 1 i¼1 ðri  rmax Þ 5
F ¼ 4C 1  i¼1 i av
þ C2  ðC 1 þ C 2 Þ ð5Þ
rav nb  1 rmax nb  1

where ri is the maximum principal stress at the centroid of the ith element, rav and rmax are the average and
the maximum stress for all of the elements, and nb is the total number of elements currently existing on the
boundary being optimised at a given iteration.

4. Durability based formulation

It is now known that stress optimised shapes are not necessarily optimised for fracture strength [8,6]. Thus,
a modified formulation of the evolutionary structural optimisation algorithm for shape optimisation is devel-
oped for optimisation with residual (fracture) strength/durability as the design goal. This fracture based nib-
bling ESO, which we will term as the fracture ESO algorithm, removes elements from the boundary that
contains flaws. In this case cracks are considered to be present at all points around the boundary (Fig. 2).
The cracks can be placed either in every element or in certain specified elements on the design boundary. This
through-crack configuration in each element is schematically shown in Fig. 3. In a simple formulation the
cracks are assumed to run from the element boundaries to their centres. Hence, the crack length is the normal
distance from an element centre to its boundary.
The present formulation uses the stress field of the element centroid together with Kujawski approximation
[9] to compute the stress intensity factor associated with each crack. According to Kujawski’s formulation, the
stress intensity factor at the deepest point of an edge crack for un-notched bodies can be estimated by
pffiffiffiffiffi
K I ¼ Qrt pl ð6Þ

Fig. 2. Schematic of numerous cracks around the structural boundary.


122 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

σt

Structural boundary
l

de

Boundary element
Crack

Fig. 3. Crack configuration in each element.

where KI is the mode I stress intensity factor at the deepest point, Q is a shape factor and depends on the crack
type and problem geometry, Q = 1.12 for this problem with through-thickness cracks, l is the length of the
crack, and rt is the stress at the location of the crack tip normal to the direction of the crack. For small cracks
this stress (rt) can be approximated as the maximum principal stress at element centroids, since the centroids
of the boundary elements lie very close to the free edge with no normal stress. If we ensure that the element size
is such that the centroid of the element is at a distance of l from the edge of the boundary being optimised,
then the stress intensity factor for the crack in that element can be calculated using the maximum principal
stress at the element centroid (see Fig. 3). It is reported that the accuracy of Eq. (6) in computing the stress
intensity factors is within 5% [9].
However, the mesh pattern and density often may not allow the crack tip to lie close to the element centroid
so that the stress field at the element centroid can be used to compute the stress intensity factors. The present
work has proposed and implemented several alternative approaches depending on the specific nature of the
problem. In the case of cracks smaller than the half of the element size, the stress field at the crack tip was com-
puted either using the displacement field obtained from finite element analysis, or by interpolating from the
nodal stress values. For longer cracks the element containing the crack tip was identified and the stress field
at the crack tip location was evaluated for that internal element using the nodal displacement field. The ultimate
aim of all the measures was to compute the stress field at each crack tip location to a sufficient accuracy.
In this extended ESO approach, the stress intensity factor associated with the crack in each element is con-
sidered as the ESO criteria. An element is then removed if the stress intensity factor of the crack associated
with it satisfies the following condition:
K i 6 K ref ð7Þ
where Ki is the stress intensity factor associated with the ith crack and Kref is the reference stress intensity fac-
tor that is calculated using an equation similar to Eq. (2) and given by
K ref ¼ EF  K max ð8Þ
The elimination factor EF is calculated as
EF ¼ K min =K max þ FR  ð1  K min =K max Þ ð9Þ
where Kmin and Kmax are the minimum and the maximum stress intensity factors around the boundary being
optimised and FR is a control factor, decided adaptively by the optimiser.
In this study we will use the following equation, which is derived from Eq. (5), as a measure of convergence,
viz:
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3,
Pnb 2 Pnb 2
1 i¼1 ðK i  K av Þ 1 i¼1 ðK i  K max Þ 5
F ¼ 4C 1  þ C2  ðC 1 þ C 2 Þ ð10Þ
K av nb  1 K max nb  1
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 123

where K1, K2, . . ., Knb are the stress intensity factors associated with the cracks present on the boundary, Ki is
the stress intensity factor for the ith crack, Kav and Kmax are the average and the maximum stress intensity
factors around the boundary being optimised, nb is the number of cracks/elements in the boundary, and
C1 and C2 are constants and as previously are taken unity for this study.

5. Analysis and optimisation methodology

The steps used in this study for optimisation using the ESO algorithm with stress/fracture strength as the
design objective are as follows:

(i) A geometric model of the structure was constructed using any pre-processor. FEMAP was used for
developing the initial solid models in the present research.
(ii) A structural boundary or surface, whose shape is to be designed, was specified to the algorithm. Material
removal was permitted from this boundary only. A region around this boundary was also defined within
which the final shape/boundary was to be contained. This zone is termed as the design/optimisation
domain in this paper. Once any portion of the shape reaches the boundary of the optimisation domain,
it cannot be moved further. The optimisation category, design constraints, design criteria, and ESO
parameters were specified as the inputs to the optimiser through an initialisation/parameter data file.
(iii) Finite element analysis was performed to evaluate the stress distribution for this geometry. In this work
NE-NASTRAN was used as the finite element solver.
(iv) For residual strength optimisation, the stress intensity factors for all the cracks present on the design
boundary were computed. For stress based optimisation, the maximum principal stress evaluated at
the centroid of the elements was considered as the design criteria.
(v) The optimisation was performed by invoking the shape optimisation option in the in-house optimisation
code NASESO. The stress intensity factor for each crack on the optimisation boundary was compared
with the reference stress intensity factor calculated. An element was then removed from the structure if
the stress intensity factor of the crack contained within it was below the reference value. In the present
research this was accomplished by reducing the modulus of elasticity (E) of the element to a very low
value so that it (practically) could not take any load. Here the properties of the removed elements were
changed to: E = 0.1 MPa and m = 0.3. For stress optimisation, the maximum principal stress at the cen-
troid of the elements was chosen as the design criteria and used (instead of the stress intensity factor) in
deciding which elements were to be removed.
(vi) The objective function and the convergence function were monitored to check for convergence.
(vii) The updated structure was then re-analysed and steps (ii)–(vii) were continued iteratively until the solu-
tion converged.

An in-house software (NASESO) has been developed to perform optimisation using ESO. This program
implements the ESO algorithm for both topology and shape optimisation. It has been designed to work with
the FEMAP pre/post-processor. Modelling of the initial geometry, or design domain, can be performed in
FEMAP. Appropriate geometric design constraints are also specified interactively within FEMAP. The opti-
misation parameters, ESO criterion, design objective, response constraints, and optimisation category are
specified via an input data file. The program is currently interfaced with the finite element code NE-NAS-
TRAN to form an integrated structural optimisation system. The initial FE analysis model exported from
FEMAP acts as the input to this software system. The optimisation using the ESO method is then performed
iteratively until the convergence criteria are met, or the process is terminated by the user.

6. Structural configuration and optimisation problem

A typical bulkhead frame construction consists of top and bottom flanges, an optional auxiliary flange, and
the presence of necessary cutouts such as stringer cutouts. The bulkhead frame analysed here is circular in
shape, with a bottom flange of radius 975 mm, and a web with a radial width 75 mm and thickness
2.5 mm. A geometric model of a 106.2 mm long portion of the bulkhead frame is shown in Fig. 4. A schematic
124 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

Fig. 4. A typical bulkhead frame with a stringer cutout.

shape of a stringer cutout located near the top flange can also be seen in Fig. 4. Depending on the number of
stringers passing through a bulkhead, it can have a number of these cutouts located circumferentially. Hence,
minimising the inspection time can lead to considerable (maintenance) cost savings.
The design guidelines [10,1] require that the bulkhead frame be analysed considering loads generated by the
internal pressure of the fuselage. In the present study, the loading condition was obtained from [2]. These loads
were obtained from an analysis of a complete aircraft. The internal pressure creates a hoop stress of 50 MPa in
the direction shown in Fig. 5. Thus, the section of the bulkhead was modelled by applying a uniform stress of
50 MPa on one side, while the other end was constrained. The current loading generates a high stress concen-
tration along some parts of the stringer cutout that may act as potential crack initiation sites, or aid in prop-
agation of pre-existing cracks. It is noteworthy that the present bulkhead frame design incorporates a crack
arrest mechanism. The auxiliary flange in Fig. 4 serves as an intermediate structural element that is supposed
to restrict the propagation of a crack, emanated from the cutout, from the upper part to the lower part of the
web.

Fig. 5. Loads and constraints on the bulkhead frame.


R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 125

Let us now focus on the stringer cutout and various other components in relation to their influence on the
structural integrity of the frame. The top and bottom flanges provide additional stiffness to the frame. The
auxiliary flange, apart from resisting crack propagation, also strengthens the middle portion of the web
and is particularly effective in reducing the stress levels in the lower part of the cutout. However, the top flange
has the most prominent role in controlling the deformation and stress field around the cutout contour. It sig-
nificantly stiffens the top portion of the cutout. The deflections around the regions near points A and B (in
Fig. 5) are considerably reduced as the top flange resists the opening of section AB. Consequently, this alters
the stress distribution around that region from that of a cutout without the top flange.
In the present optimisation study a conservative design approach was adopted. The bulkhead frame was
analysed without the flanges. The rationale behind this was as follows: the basic purpose of the auxiliary flange
is to arrest crack propagation into the lower web. However, it increases the weight of the component. Hence, it
was thought that if the residual strength of the optimum shape without this flange was within an acceptable
limit, then a considerable weight saving could be achieved by removing the auxiliary flange. The reasons for
analysing the stringer cutout without the top and bottom flanges were twofold. Firstly, it was once again
instructive to investigate whether the residual strength of the cutout alone would be sufficient without the
additional supports from the top and bottom flanges so that the design of the flanges could be modified to
lighten them. Secondly, from past experience with these structures it was found that in many occasions the
skin/top flange itself failed. In the event of such a failure, the top flange would not be able to provide any stiff-
ening action. The effect of this on the cutout would be equivalent to having no flange at all. In this situation it
must be ensured that a flaw on the stringer cutout boundary should not lead to failure of the web. As such, it
was instructive to investigate this case to ensure that in the worst case scenario for this design, i.e., with no
auxiliary flange and failure of the skin, the flaws on the boundary of the cutout should not lead to failure
of the bulkhead frame. Furthermore, it is evident that a stringer cutout optimised without the flanges will have
a considerably improved residual strength when incorporated into a bulkhead frame with flanges.
The bulkhead frame without the flanges is shown in Fig. 6. Fig. 6 also shows the currently implemented
shape of the cutout. This shape was attained based on experience, and the spatial clearance required to allow
the passage of the stringer and to accommodate other features. The material of the bulkhead frame was
assumed to be an aluminium alloy with a Young’s modulus of 72,000 MPa and a Poisson’s ratio of 0.3.
We first analyse this currently existing shape, which will be termed as the ‘current shape’ throughout this

Fig. 6. Current shape of the stringer cutout.


126 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

study. It may be noted that the profile of the cutout in Fig. 4 is a schematic and is different from the one actu-
ally implemented (Fig. 6).
For stress based shape optimisation study the maximum principal stress on the cutout boundary was taken
as the design criterion. The maximum principal stress distribution for the currently implemented shape is
shown in Fig. 7. The deepest point of the boundary, i.e., point C in Fig. 7, had the highest stress concentration,
while the regions around points A and B were lowly stressed. The maximum value of the peak stress (i.e., the
peak maximum principal stress) was 291.74 MPa for this currentp design (see Fig. 7). The maximum stress
intensity factor for the current design was found to be 15.78 MPa m and occurred near point C (see Fig. 7).
The current shape has been derived based on experience, and it may seem that this shape is somewhat arbi-
trary. However, the prime reason for this shape is to permit the passage of the stringer through the cutout. The
aim of the present study was to produce a new design/shape rather than reworking the existing shape. This pro-
vided the opportunity to select another alternative shape as the ‘starting shape’ for the present optimisation
study. When using ESO for shape optimisation, it is instructive to start from a shape that just satisfies the geo-
metric design constraints, as this allows the optimisation algorithm to explore the design space more extensively.
As a result, more alternative designs can be evaluated. Consequently, we started from a simple ‘L-shaped cutout’
(see Fig. 8). The choice of this specific starting shape was guided by one of the geometric constraints of the prob-
lem. The geometry of this shape (Fig. 8(b)) was just sufficient to provide the adequate clearance space necessary
for the stringer and other equipments to pass through the cutout. Another geometric constraint was the restric-
tion on the maximum dimensions of the cutout. It was proposed that the final profile of the cutout must lie
within the boundary C1C2C3C4 shown in Fig. 8(a). To distinguish this starting shape from the ‘current shape’
(Fig. 6), we will term this shape as the ‘initial shape’.

Fig. 7. Maximum principal stress distribution for the current shape.


R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 127

Fig. 8. (a) Schematic of a part of the bulkhead web with the starting shape of the stringer cutout, (b) geometric details of the optimisation
domain (in mm).

In this study we used a parametric coordinate system along the boundary surface of the stringer cutout. In
this system one point on the surface was chosen as the origin, and the parametric position of any point on the
boundary was calculated as the distance of that point from the origin measured along the curved boundary. In
all the figures showing the parametric system (position), point A was assumed to be the origin unless specif-
ically mentioned otherwise. In the following sections we will use the evolutionary structural optimisation algo-
rithm to design the profile of the stringer cutout. Initially, a stress optimised shape will be derived. A damage
tolerance based optimisation will then be performed to maximise the residual strength of the resulting shape.

7. Stress optimisation

In this study the starting shape of the stringer cutout was taken as shown in Fig. 8. The initial geometric
model was meshed with 21 noded brick (hexahedral) elements. The FE model contained 24,414 elements and
124,277 nodes. The objective function was the peak value of the maximum principal stress around the surface
of the stringer cutout. The maximum principal stress distribution is presented in Fig. 9. The maximum stress
value for this ‘initial shape’ was 308.83 MPa, slightly higher than that of the ‘current shape’. The ESO algo-
rithm was applied to reduce the maximum stress along the cutout profile.
In the initial shape the low stress zones along boundaries AB, BC, CD, and EF (in Fig. 8) represent the
regions that are not effectively carrying loads. Consequently, the optimisation algorithm began removing
material from these regions. An element was removed from the cutout surface if its maximum principal stress
was lower than a reference value (Eq. (2)). As the material elimination continued, the load path altered result-
ing in a change in the stress field. The peak value of the maximum principal stress (rmax) decreased as
the shape changed, which can be seen in the convergence history presented in Fig. 10. The maximum stress
reduced significantly from an initial value of 308.83 MPa to a minimum value of 204.44 MPa at iteration
128 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

Fig. 9. Finite element model with maximum principal stress distribution for the initial shape.

320

300
σ p ,max (MPa)

280

260

240 Optimum point

220

200
0 20 40 60 80 100 120 140 160
Iteration number

Fig. 10. Variation of the objective function (maximum principal stress) for stress optimisation.

number 138, after which it showed a relatively rapid rise. Therefore, the design point corresponding to itera-
tion 138 represented the ‘near’ optimum point (see Fig. 10), and the corresponding shape was taken as a ‘near’
optimal shape.
The ‘near’ optimum shape produced by ESO had irregular boundaries and will be termed as the unsmooth
shape. This shape needed to be post-processed to produce a realistic, manufacturable design. To this end, a
curve was fitted through the boundary surface nodes of the ESO produced optimal finite element model to
generate a preliminary shape. This shape was further smoothed by fitting a spline using the preliminary shape
as the basis. This smoothed shape was taken as the stress optimised design. This final shape is shown in Fig. 11
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 129

along with the maximum principal stress distribution. The optimal profile turned out to be nearly symmetric
as should be the case with the given structural geometry and boundary conditions.
The variation in the maximum principal stress around the stringer cutout boundary is presented in Fig. 12
for all of the shapes considered in this study. This includes the currently implemented shape (current shape),
the starting shape for optimisation (initial shape), the ESO produced irregular shape (unsmooth shape), and the
final post-processed shape (smoothed shape). For the current shape (Fig. 7), the maximum principal stress var-
ied along the boundary in a relatively smooth fashion reaching its peak value at point P1 in Fig. 12, which
corresponds to point C (in Fig. 7) having a high stress concentration. In the initial shape, i.e., the L-shaped
cutout, the stress distribution was fluctuating due to the presence of sharp corners, i.e., points B, C, D, and

Fig. 11. Final stress optimised shape with maximum principal stress distribution.

350
Q1 P1
300 Q2 Current shape
R1
250
R2 Initial shape
(MPa)

200

150 Unsmooth stress


σp

optimised shape
100
Q3 S1
Smoothed stress
50 optimised shape

0
0 20 40 60 80 100
Parametric position (%)

Fig. 12. Maximum principal stress distribution around the stringer cutout boundary.
130 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

1.4

Convergence function
1.3
1.2
1.1
1
0.9 Optimum point

0.8
0.7
0.6
0 20 40 60 80 100 120 140 160
Iteration number

Fig. 13. History of the convergence function for stress optimisation.

E in Fig. 8. The inner corners B, D, and E acted as stress concentrators. From peaks Q1, Q2, and Q3 in Fig. 12
representing the stresses at corners D, E, and B (Fig. 8), respectively, it can be seen that the maximum prin-
cipal stresses at these regions were 52.35, 308.83 and 292.27 MPa, respectively. The ESO algorithm reduced
the stress peaks Q1 and Q2 substantially to R1 and R2, respectively. The maximum principal stresses corre-
sponding to points R1 and R2 were 204.44 and 199.08 MPa, respectively. As such, a reduction of about
34% in the objective function (peak stress) was achieved.
It was found that the ESO generated ‘near’ optimum shape had a rapid fluctuation in the maximum prin-
cipal stress, as reflected in Fig. 12. This is a common feature with this algorithm. The generation of a jagged
boundary leads to alternating high and low stress zones at inward and protruding element corners. However,
depending on the element orientation and removal pattern, some parts of the boundary may be (locally)
smoother. In this case the boundary surface was quite irregular between 25–42% and 60–75% of the paramet-
ric location. In other places it had a relatively smooth variation in stress (see Fig. 12). The stress distribution
for the post-processed final smoothed shape is also shown in Fig. 12. The elimination of the irregularities
resulted in a relatively smooth variation in the maximum principal stress around the boundary. The smoothing
further reduced the maximum principal stress to 193.64 MPa. This corresponds to a stress reduction of 33.6%,
from 291.75 MPa in the current shape to 193.64 MPa (point S1 in Fig. 12) in the final optimal shape.
The convergence of the optimisation process was monitored using the convergence function given in Eq. (5)
(see Fig. 13). The location of the optimum point in the ‘near’ constant portion of the curve indicates that the
solution converged. As the optimisation progressed, the convergence function decreased from an initial value of
1.28–0.69 at the optimal point, signifying a greater uniformity in stress distribution around the final optimal
surface. It is noteworthy that for a simple boundary the entire boundary surface may be ‘nearly’ uniformly
stressed. However, this phenomenon largely depends on the specific problem, boundary conditions, and design
constraints. In the present case, the stringer cutout surface was not permitted to move beyond the lines C1C2 in
the top and C3C4 in the bottom (see Fig. 8). This inevitably led to creation of low stress zones near the upper and
lower sections of the optimal cutout, introducing non-uniformity in the stress pattern around the boundary.

8. Residual strength optimisation

In this section we will determine a shape that will maximise the residual strength of the structure allowing
for flaws around the boundary of the cutout. To this end, we considered the maximum stress intensity factor
associated with the flaws around the boundary as the design objective function. The aim of this optimisation
study was to minimise the objective function so as to maximise the residual (fracture) strength. The fracture
based ESO formulation, presented in Section 4, was applied to optimise the stringer cutout with residual
strength as the design objective. In this approach, flaws were assumed to be present in the elements lying
on the surface/boundary being optimised. In this study we considered 1 mm long through cracks placed
around the boundary (see Fig. 14). The stress intensity factors for cracks contained in all the boundary ele-
ments were evaluated. The elements with cracks that had lower stress intensity factors compared to the refer-
ence value, as calculated using Eq. (8), were progressively removed from the cutout boundary.
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 131

Fig. 14. Flaws around the stringer cutout boundary for residual strength optimisation.

In this study the starting shape was the same as that of stress optimisation, i.e., an ‘L-shaped cutout’ (see
Fig. 8). With fracture based ESO, material was removed from the portions of the boundary that had a lower
fracture criticality, i.e., the stress intensity factors associated with the flaws were relatively low. The regions
from which material removal commenced were evidently boundaries AB, BC, CD, and EF (in Fig. 8) as cracks
at these locations had low stress intensity factors due to the low stress field. The variation in the objective func-
tion, i.e., the maximum stress intensity factor (Kmax) associated with all of the cracks lying along the design
boundary is shown in Fig. 15.
As the p shape changed, the maximum stress intensity factor reduced. It attained a minimum value of
11.78 MPa m at iteration 164, after which it showed a steep increase. The corresponding shape was taken p
to be the preliminary (unsmooth) optimal profile. The starting L-shaped cutout had a Kmax of 14.84 MPa m.
Hence, the optimisation process resulted in a 20.6% reduction in Kmax in the (unsmooth) optimal shape
(Fig. 16). This non-smooth profile, which had irregular boundaries, was smoothed to produce a practical
shape by fitting a curve through the boundary p points (nodes). The smoothed profile is shown in Fig. 17. This
final optimal shape has a Kmax of 10.46 MPa m, and was again approximately symmetric, which is physically
intuitive in accordance with the given loading and boundary constraints.
Let us now study the fracture criticality of the various cutout profiles analysed here. The distribution of the
stress intensity factor, associated with the cracks present around the boundary, points out regions that are
more fracture prone. Figs. 18 and 19 present the stress intensity factor distribution around the boundary
for the various stringer cutout shapes, viz: the current shape, initial shape, unsmooth optimal shape, and
smoothed optimal shape. In Fig. 18 the stress intensity factor distributions at various stages of the fracture

16

15
)
1/2

14
K max (MPa m

13

12

11
Optimum point
10
0 20 40 60 80 100 120 140 160 180 200 220
Iteration number

Fig. 15. Variation in the objective function (maximum stress intensity factor) for residual strength optimisation.
132 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

Fig. 16. ESO produced residual strength optimised shape.

Fig. 17. Smoothed residual strength optimised shape with maximum principal stress distribution.

strength optimisation are compared. The initial shape, which contains sharp geometry changes, had a wide
range of variation of the stress intensity factors. Boundary DE in Fig. 8 was the highly stressed region (see
Fig. 9), and consequently the cracks along DE had higher stress intensity factors. This can be observed
between 45% and 70% parametric locations
p of the boundary (Fig. 18). The stress intensity factors at points
D and E were 14.84 and 14.26 MPa m, respectively.
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 133

18
16 P1 P2
14 Q3 Initial shape
Q2
Keff (MPa m )
Q1
1/2

12
10 Unsmooth SIF
8 optimised shape
6
Smoothed SIF
4 optimised shape
2
0
0 20 40 60 80 100
Parametric position (%)

Fig. 18. Stress intensity factor distribution around the stringer cutout boundary at various stages of optimisation.

18
R1
16 R2
14 Current shape
)
1/2

12 S1
Keff (MPa m

10 Stress
8 optimised shape

6
SIF optimised
4 shape
2
0
0 20 40 60 80 100
Parametric position (%)

Fig. 19. Comparison of the stress intensity factor distributions for the current, stress optimised, and residual strength optimised shapes.

The shape obtained from ESO consisted of an irregular geometry that influenced the local stress field
around the boundary. As a result, the stress intensity factors, which are sensitive to the (local) stress field,
showed considerable oscillations, see parametric positions 20–80% in Fig. 18. Peaks Q1, Q2, and Q3 in
Fig. 18 correspond to cracks located at parametric positions 29%, 42% and 61% on the unsmooth optimal p
shape shown in Fig. 16. The stress intensity factors at these locations were 11.20, 11.78 and 11.72 MPa m,
respectively. Hence, the localised peaks (points P1 and P2) for the initial shape were distributed into a number
of alternate peaks and valleys in the ‘unsmooth’ optimal shape.pThis jagged shape was smoothed, which fur-
ther reduced the maximum stress intensity factor to 10.46 MPa m (Fig. 18). It can also be observed that the
final optimal shape had a relatively smooth variation in the stress intensity factor.
The variation in the stress intensity factor along the cutout for the current shape, the stress optimised shape,
and the residual strength optimised shape is shown in Fig. 19. For the current cutout (Fig. 6) the stress field in
Fig. 7 shows a high stress concentration around point C. Consequently, the region around point C acted as the
most fracture critical location. This is also reflected in the stress intensity factor distribution shown inpFig. 19.
The values at the two peaks in the stress intensity factor distribution were 15.78 and 14.52 MPa m, and
occurred at points R1 and R2 in Fig. 19. It is noteworthy that the stress optimal shape had a considerably
improved p residual strength. The maximum stress intensity factor for the stress p optimised shape was
12.27 MPa m, whereas the value of Kmax for the current shape was 15.78 MPa m. However, optimisation
using the residual strength as the explicit design objective led to a further reduction (by 14.8%) in the maxi-
mum stress intensity factor, bringing down the peak value at S1 in Fig. 19.
134 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

1.4
1.3

Convergence function
1.2
1.1
1
Optimum point
0.9
0.8
0.7
0.6
0 20 40 60 80 100 120 140 160 180 200 220
Iteration number

Fig. 20. History of the convergence function for residual strength optimisation.

The variation of the convergence function with optimisation cycles (iterations) is shown in Fig. 20. The fact
that the optimum point lies on the ‘near’ constant part of the curve indicates a stable convergence. The values
of the convergence function for the initial and the optimal designs were 1.17 and 0.73, respectively. A consid-
erable reduction in the convergence function shows that the optimal shape had a more uniform stress intensity
factor distribution. This is reflected in Fig. 19, which compares the stress intensity factor distributions for the
initial and the final optimised shapes. It is also noteworthy that in the vertical portion of the optimal cutout
boundary (parametric locations 25–75%), the residual strength optimised profile had a SIF distribution pat-
tern that was more uniform than that of the stress optimised profile.
The different stringer cutout profiles are compared in Fig. 21. It can be observed that the stress and the
residual strength optimised shapes differ somewhat, justifying the need to incorporate an ‘explicit’ residual
strength based optimisation into the design. A summary of the key parameters (objective function values)
is presented in Tables 1 and 2. The maximum stress intensity factor for the ‘explicit’ residual strength opti-

50

45

40

Current shape
35

30 Initial shape
Y (mm)

25
Stress optimised
shape
20
SIF optimised
15 shape

10

0
0 5 10 15 20 25
X (mm)

Fig. 21. Comparison of the current, initial, stress optimised, and residual strength optimised shapes.
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 135

Table 1
Summary of the structural responses for various shapes of the stringer cutout: stress optimisation
Stringer cutout shapes rmax (MPa) Volume of the web (mm3)
Current shape 291.75 18,986.88
Initial shape 308.83 19,226.43
Stress optimised shape – unsmooth 204.44 17,933.66
Stress optimised shape – smoothed 193.64 18,135.80
Residual strength optimised shape 216.79 17,899.98

Table 2
Summary of the structural responses for various shapes of the stringer cutout: residual strength optimisation
p
Stringer cutout shapes Kmax (MPa m) Volume of the web (mm3)
Current shape 15.78 18,986.88
Initial shape 14.84 19,226.43
Stress optimised shape 12.27 18,135.80
Residual strength optimised shape – unsmooth 11.78 17,926.75
Residual strength optimised shape – smoothed 10.46 17,899.98

mised shape was 15% lower than that of the stress optimised shape. The ESO algorithm implemented here
removed material from the starting shape, and hence the resulting shapes were lighter (see Tables 1 and 2).
However, the most notable feature is that the fracture strength optimised shape was lighter than the stress
optimised shape. This supports the previous findings of the literature [4,5].
It is of interest to note that in the current shape the maximum stress was 291.75 MPa, which was lower than
that of the shape used as the starting geometry for optimisation (initial shape). In contrast, the maximum stress
intensity factor for the worst case crack in the current shape was higher than that of the initial shape (see Table
2). This is because the stress intensity factor is largely related to the stress pattern surrounding the crack tip
region, rather than the stress at the boundary. This demonstrates that boundary stress fields cannot always be
used to compare the fracture critically between different shapes. Hence, this signifies that a structure optimised
for stress may not necessarily have an optimum residual strength.
Another important aspect is that post-processing of the ESO produced preliminary shape is an integral part
of shape optimisation. Otherwise, the ‘uneven’ boundary profile can drastically reduce the performance of the
structure. For example, the maximum stress intensity factor for the unsmooth residual strength optimised
shape was higher than that associated with the smoothed SIF optimised shape (Table 2). Post-processing
reduces the unevenness, and in this case it reduced Kmax to a lower value for the final smoothed residual
strength optimised shape.

9. Discussion

This work has demonstrated that evolutionary structural optimisation can be an effective tool for shape
optimisation problems with stress and residual strength as the design criteria. An ESO based optimisation
results in a reduction in the peak stress or peak stress intensity factor, and a relatively uniform stress or stress
intensity factor distribution. The shape optimisation case study undertaken demonstrated the capability of the
fracture based evolutionary structural optimisation algorithm to provide optimum solutions with residual
strength as the design criteria.
Evolutionary structural optimisation has many advantages. It uses the finite element method to perform
structural analysis, and hence the response of a complex structure can be evaluated. ESO works by removing
elements from lowly stressed regions. It uses a fixed mesh, i.e., the initial finite element model throughout the
optimisation process, thus avoiding remeshing and the associated mesh distortion problems.
The present formulation can handle a complex geometry without major modification in the algorithm or
computer program. The fact that ESO produced shapes do not greatly depend on the initial/starting shape
makes it suitable for those problems, where it is difficult to guess a good starting shape, i.e., a solution close
to the optimum. The implementations of many optimisation methods require the geometric model to be built
appropriately. For example, the biological algorithm needs a shape to be represented by a set of design points
136 R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137

or nodes [6]. On the other hand, one benefit of this formulation is that an initial FE model created using any
pre-processor can be used as input to the algorithm. This approach is therefore applicable to structures where
solid models or geometric descriptors are not available.
In the present approach the optimiser can be externally interfaced with the FEM software, and hence any
commercial or in-house finite element program can be used for the structural analysis without requiring the
access to the source code of the finite element program. It is also a gradient-less method, and is applicable
to a wide range of objective functions, even in cases where the objective function is non-differentiable.
The fracture based ESO algorithm studied here enables an effective modelling of the stress intensity factor
variation along the boundary. As the shape and the length of the boundary gradually change due to element
removal, the number of cracks is automatically adjusted according to the boundary length. However, the pro-
posed method needs the shapes to be post-processed to generate smooth optimal contours for realistic
applications.
The convergence or computational efficiency of evolutionary structural optimisation depends on the start-
ing design, mesh density, and material removal parameter (control factor). A good guess of the starting shape/
solution can result in a rapid convergence. Mesh density also contributes to the computer time required. A
finer mesh may lead to less material removal per optimisation cycle, thus increasing the computational effort.

10. Conclusions

This paper has evaluated the use of a fracture based evolutionary structural optimisation algorithm as a
practical tool for optimum shape design problems. This has been demonstrated through the design of a strin-
ger cutout profile present in the bulkhead of the P3-Orion aircraft. The maximum stress and residual strength
were considered as the design objectives in separate shape optimisation studies. The stress optimisation
reduced the maximum boundary stress considerably, and the fracture strength optimisation led to a significant
improvement in the residual strength of the structure. The need for post-processing the design boundary
obtained directly from the ESO method is essential for real life applications.
The present study has shown that residual strength and stress optimised shapes may be different. This
implies that the use of the maximum stress as the optimisation criteria alone may not be sufficient for some
applications. This is particularly significant in cases where the crack size can be large, e.g., mining or marine
industry. In such applications it is necessary to explicitly consider residual strength as the optimisation objec-
tive. The residual strength optimised shape should then be analysed to ensure that it satisfies the maximum
design strength criteria. The ‘stringer cutout problem’ considered here also demonstrates that residual strength
based optimisation may lead to a structure that is lighter than a structure optimised for stress. This example
establishes the evolutionary structural optimisation method as a powerful and robust design tool for shape
optimisation of real life structures with stress and residual strength as the design criteria.

Acknowledgements

This work was performed under the publication grant awarded by Monash University. The authors
acknowledge the support of Noran Engineering, distributors of NE-NASTRAN in integrating NE-NAS-
TRAN into the present optimisation system.

References

[1] Ball DL. The role of non-destructive testing in aircraft damage tolerance. The American Society of non-destructive testing 2003.
[2] Chandra S. Integrating damage tolerance criteria in design process and product life cycle management (PLM) architectures.
Bangalore, India: National Aerospace Laboratories; 2004.
[3] Cope DA, Lacy TE, Luzar J, Horn WJ. Residual strength assessment of KC-135 fuselage panels. In: Proceedings of the 5th joint
NASA/FAA/DoD conference on aging aircraft, Kissimme, FL, USA, 2001.
[4] Das R, Jones R. Designing structures for optimum fracture strength. In: Proceedings of the international conference on failure
analysis and maintenance technologies, Brisbane, Australia, 2004.
[5] Das R, Jones R, Peng D. Optimisation of damage tolerant structures using 3D biological algorithm. In: First international conference
on engineering failure analysis, Lisbon, Portugal, 2004.
R. Das et al. / Engineering Failure Analysis 14 (2007) 118–137 137

[6] Das R, Jones R, Peng D. Optimisation of damage tolerant structures using a 3D biological algorithm. Eng Fail Anal
2005;13(3):362–79.
[7] Das R, Jones R, Xie YM. Design of structures for optimal static strength using ESO. Eng Fail Anal 2005;12(1):61–80.
[8] Jones R, Chaperon P, Heller M. Structural optimisation with fracture strength constraints. Eng Fract Mech 2002;69:1403–23.
[9] Kujawski D. Estimation of stress intensity factors for small cracks at notches. Fatigue Fract Eng Mater Struct 1991;14(10):953–65.
[10] Schmit HJ, Schmidt-Brandecker B, Tober G. Design of modern aircraft structure and the role of NDI. NDT.net 1999;4(6).
[11] Xie YM, Steven GP. Evolutionary structural optimization. Heidelberg: Springer; 1997.

Potrebbero piacerti anche