Sei sulla pagina 1di 6

COMPUTATIONAL MECHANICS

WCCM VI in conjunction with APCOM`04, Sept. 5-10, 2004, Beifing, China


2004 Tsinghua University Press & Springer-Verlag

Topology Optimization for Shell Structures with Linear Buckling


Responses
M. Zhou
Altair Engineering, Inc., 2445 MacCabe Wav, Suite 100, Irvine, CA 92614, USA
e-mail: zhoualtair.com

Abstract Topology optimization has gained broad recognition as a powerIul tool Ior structural concept
design. At this design stage, it is desirable that structural stability concern is addressed. It has been shown
in the literature that Iiltering oI buckling modes in low density zone is challenging when linear buckling
responses are considered Ior topology optimization. In this paper, Iurther numerical diIIiculties are
discussed. This study leads to the conclusion that at the current state topology optimization with linear
buckling responses is only meaningIul Ior a subset oI quasi-topology problems where creation oI new
cavities is not permissible.

Key words: Shell Structure, Topology Optimization, Linear Buckling, Eigenvalue

INTRODUCTION

Topology optimization has become increasingly popular in industrial applications. In many cases,
tremendous cost savings can be achieved owing to its impact in the early stage oI the design process.
Research in this area has been one oI the Iocuses oI structural optimization in the last decade and
signiIicant progress has been made both in addressing the Iundamental approaches and in various
applications. An overview oI the development can be Iound in some survey papers and recent books |1|
|5|. It has been shown Irom both research literatures and industrial applications that it is important to
address structural stability concern during the concept design stage. Consideration oI linear buckling in
topology optimization appears to be the most direct approach in addressing structural stability
requirements. In general, two design Iormulations can be used: (a) buckling eigenvalues are constrained
with an lower bound so that buckling does not occur under given loads; (b) the lowest buckling eigenvalue
is maximized to increase the buckling load limit. For the later, since the optimization process tends to
make several buckling modes critical, a bound Iormulation should be used to solve the problem oI
maximizing the lowest eigenvalue oI a number oI buckling modes that are potentially critical.
A challenge Ior topology optimization with linear buckling responses lies in the Iact that numerically
critical buckling modes may appear in regions with low material density |6|. It has been noticed during this
study that Iurther diIIiculties exist: (a) the singular topology` phenomenon associated with the conditional
nature oI buckling constraints; (b) the signiIicant impact oI low density material to the accuracy oI
eigenvalues. The Iirst issue represents a major barrier that would be expected to mislead, in most case, the
optimization process as it prevents the removal oI material Irom areas with buckling weaknesses. As a
positive outcome, this study shows that topology optimization could produce meaningIul results Ior shell
structures with a given base thickness.

DESIGN PROBLEM
Linear buckling analysis is governed by the Iollowing eigenvalue problem:
G(u) - K] 0, (1)

where K represents the stiIIness matrix, and G the geometric stiIIness matrix that is dependent oI the
displacement vector u oI a given static load case. is the buckling eigenvector and the buckling
eigenvalue that represents the load Iactor when buckling occurs Ior a given load case. Buckling
eigenvalues can be considered as constraints in a single objective optimization Iormulation:

N i
i
M f
U
f
g
f
g
f
,..., 1 , 1
min

,..., 1 , 0 ) ( Subject to
) ( Minimize
=
=

(2)

where
i
represents the density variable oI the i-th element. In order to achieve discrete topology, the
SIMP power-law penalization oI stiIIness is used. The lower bound oI a buckling eigenvalue, representing
the buckling load Iactor oI the corresponding static load case, should be typically 1.0 Ior real load cases.
An alternative is to use a bound Iormulation to maximize the lowest value oI several critical eigenvalues.
Adjoint sensitivity analysis can be used to calculated the sensitivities oI buckling eigenvalues eIIiciently.



IMPACT OF LOW DENSITY MATERIAL
Example 1: To study the impact oI low density material, the 2D portal Irame shown in Fig.1 is used to
demonstrate the impact oI low-density material to the buckling eigenvalues. The Irame has a dimension oI
300x900 mm, and is modeled with 675 quad elements with a dimension oI 20 mm x 20 mm and unity
thickness. The Young`s modulus E is 200000 N/mm
2
and Poison`s ratio is 0.3. Two downward Iorces oI
1000 N are applied on the top two corners. The outer Irame has Iully dense material and the inner region
has a density
min
that can also be interpreted as the thickness Ior membrane structures. The Iirst buckling
eigenvalue corresponding to various
min
values is shown in Table 1:

Table 1. Impact of low densitv material

min
0.1 0.01 10
-3
10
-4
10
-5
10
-6
10
-7
10
-8

1
st
Eigenvalue 221.1 80.53 13.13 3.808 2.361 2.259 2.248 2.248

It can be seen that the impact oI the inner region started to disappear at a density as low as 10
-5
. ThereIore,
with a value
min
0.01, a penalty oI p3 is needed to eliminate the impact oI
min
according to this case
study. However, the challenge lies in the signiIicant impact oI low-density elements that can qualitatively
mislead the buckling load prediction Ior topology optimization where low density regions will be
interpreted as voids in the Iinal design. As it is shown in this example, a density oI 10
-3
oI the inner region
that has only 0.56 oI the volume oI the outer Irame increased the buckling load by 5.8 times. ThereIore,
when constraints are imposed on buckling eigenvalues, the impact oI low density structural regions could
mislead the optimization process to keep element densities Irom reaching
min
so that buckling load
capacity is artiIicially inIlated by relying on low density material.

Fig.1 From left to right. (a) Portal frame, (b) First buckling mode

BUCKLING MODES IN LOW DENSITY REGION
The problem that numerically critical buckling modes may appear in regions with low material density
was Iirst discussed by Neves et al. |6|. In order to suppress such buckling modes that have no structural
signiIicance, they suggested to ignore the geometric stiIIness oI elements with a density lower than a
cut-oII point
cut
. To avoid discontinuity, Bendsoe and Sigmund |5| suggested to use slightly diIIerent
penalization Ior stiIIness matrices K and G:
For stiIIness matrix K: E
K
|
low
(1 -
low
)
p
| E
0
; (3)
For geometric stiIIness matrix G: E
G

p
E
0
.
In the above Iormulation, both stiIIness matrices are penalized equally at Iull density, whereas the
geometric stiIIness matrix is penalized more than the stiIIness matrix when the density approaches
low
. It
could be noticed that the value
low
plays a critical role in this Iormulation and has similar ad-hoc
characteristics as the Iormulation using a cut-oII density Ior geometric stiIIness matrix. For a too low
low

value, the Iormulation may be insuIIicient in suppressing undesired buckling modes during the iterative
process. On the other hand, a too large
low
may produce physically inconsistent results Ior buckling
analysis at early iterative stage when intermediate densities are dominant.
Example 2: Considering the same portal Irame discussed in the previous section, the maximization oI the
lowest eigenvalue is considered. A max(min) bound Iormulation retaining the Iirst 10 eigenvalues is used
and the volume Iraction is constrained at 0.5. The lower bound oI density in eq.(2) is
min
0.01. Using the
penalization scheme in Eq.(3) with a penalty p3.0 and
low
0.01 leads to the Iinal design shown in
Fig.2(a). The Iinal lowest buckling eigenvalue is 4.16x10
2
. For the same structure without outer Irame as
none design domain,
low
0.01 resulted in the design shown in Fig.2(b), with critical buckling eigenvalue
at 3.58x10
2
. Increasing penalty to p4.0 while keeping
low
0.01 led to the design shown in Fig.2(c), with
critical buckling eigenvalue at 3.26x10
2
. A more reasonable design shown in Fig.2(d) with eigenvalue at
3.17 x10
2
was obtained with
low
0.001 and p3.0. It should be noted that owing to the presence oI
semi-dense elements and the low-density eIIect discussed earlier, the eigenvalues might be Iar Irom
accurate quantitatively. In particular, the design shown in Fig.(2c) demonstrates the potential weakness oI
the Iiltering scheme shown in eq.(3). Due to increased penalty oI geometric stiIIness, the buckling modes
oI the low density region below the disconnected top portion is artiIicially suppressed and the buckling
eigenvalue oI 3.26 x10
2
is an artiIicially inIlated value. Experiments with the method using a cut-oII value

cut
Ior geometric stiIIness matrix have shown that such a value could be very case dependent. It is to be
noted that Iactors such as move limits and other implementation details could also aIIect the Iinal design
outcome. However, it is clear that both Iiltering Iormulations do not correctly reIlect the physics Ior
density close to
low
or below
cut
. II we were to start with a design at the density
low
or below
cut
Ior all
elements, critical buckling modes would be artiIicially suppressed or ignored Ior both Iiltering approaches
discussed in this section.

Fig.2 Form left to right. (a) With outer frame (
low
0.01, p3.0), (b) Without outer frame (
low
0.01,
p3.0), (b) Without outer frame (
low
0.01, p4.0), (c) Without outer frame (
low
0.001, p3.0).
SINGULAR TOPOLOGY
The conditional nature oI buckling constraints causes the so-called singular topology phenomenon. This
diIIiculty has been discussed in the context oI trusses by Zhou |7| and Rozvany |8| Ior local and system
buckling constraints, respectively. The phenomenon oI singular topology was Iirst discovered by Sved and
Ginos in 1968 in the context oI stress constraints Ior trusses |9|. Relaxation Iormulations oI conditional
constraints have been suggested in the literature to improve the connectivity oI the Ieasible design space
(see, e.g., Rozvany |8|, Cheng and Guo |10|). However, the Iundamental problem remains unsolved at the
current state.
Example 3: To illustrate this phenomenon, let`s consider the two bar truss example shown in Fig.3. The
load P, lengths oI the bars, Young`s modulus are: P1, L1, E1. The Euler buckling stress can be
expressed as
b
k
i
x
i
/L
2
, where k
i
is a constant depending on the geometry oI the cross-section, x
i
the cross
sectional area. Given k
1
1 and k
2
1/4, the buckling stresses oI the two bars are:
b1
x
1
,
b2
x
2
/4. The
design problem is to minimize the material volume subject to buckling stress constraints:

. 0 4 /
2
)
2 1
/( 1
2

, 0
1
)
2 1
/( 1
1
Subject to
2 1
Minimize
+ =
+ =
+ =
x x x g
x x x g
x x J
(4)
The optimal solution A that can be Iound in the Ieasible design domain is: x
1
1/50.447, x
2
4/51.789
and V2.236. Obviously, the optimal solution is solution B where bar 2 is eliminated: x
1
1, x
2
0 and
V1.0. The solution C with bar 1 eliminated is: x
1
0, x
2
2 and V2.0. On hand oI this simple example, it
becomes very easy to understand the diIIicult associated with singular topology the gradient-based
optimization process will not be able to reach 'singular designs B and C that are disconected Irom the
continuous Ieasible domain. While Ior stress constraints it have been shown that singular designs are
connected to the Ieasible domain along the axes, the connection oI singular disigns Ior the buckling
constraints in eq.(4) has to strech to the limiting value oI inIinity (i.e., the line Irom point B to x
1
and
x
2
0 is Ieasible and will meat the continuous Ieasible domain at inIinity Ior x
1
).

Fig.3 Two-bar truss and its design space

Example 4: A wing droop nose rib oI Airbus 380 shown in Fig.4(a) is studied (Krog et al. |11|). The
structure has a dimension oI approximately 1.5x1.1 m. It is modeled with 9397 shell elements. The
thickness oI both the web and the Ilange is 6.0 mm. The topology oI the web should be optimized to
maximize the lowest buckling eigenvalue subject to a volume Iraction constraint oI 0.3. 6 static load cases
are considered. A max(min) Iormulation retaining Iirst 10 buckling modes Ior every static load case is
used. A penalty p4.0 and a density cut-oII value
low
1.5 Ior geometric stiIIness are used. At the initial
design with uniIorm density at 0.3, the lowest eigenvalue is 0.00998. At the Iinal design, the lowest
eigenvalue is 0.197 and there are 20 eigenvalues in the range 0.197 and 0.213. Owing to extensive
intermediate density, the eigenvalues may likely be inaccurate, both Ior the initial design and Ior the Iinal
design. Fig.4(c) shows the Iinal design. For comparison, the design oI using compliance Iormulation is
shown in Fig.4(b) |11|. It can be seen that the Iinal buckling design is very noisy. In particular, it can be
noticed that material concentrations around holes are developed. This is likely due to the phenomenon oI
singular topologies as these areas are buckling critical Irom the very beginning. Such initially critical
buckling modes could dominate the optimization process, and hence prevent removal oI material Irom
areas oI buckling weakness. Similar to the 2-bar truss example, this tendency could prevent reshaping
existing holes so that buckling perIormance could be improved. Based on the discussion about diIIiculties
and experiments oI practical examples, it could be concluded that general topology optimization involving
buckling responses remains a challenging research topic.
In general, it is more meaningIul to constrain buckling eigenvalues so that structural instability does not
occur under given load conditions. However, such Iormulation may be inIeasible at an early stage while
seeking Ior material distribution over a large area. This example showed that the Iinal eigenvalues are Iar
below the critical bound oI 1.0. As it will be shown in Example 5 in the next section, buckling constraints
are usually met by allowing development oI ribs oI a much greater depth (60 mm depth Ior the ribs instead
oI the 6 mm depth Ior the web). ThereIore, Ior shell structures, it appears to be diIIicult to address overall
material layout and stability concern simultaneously.


Fig.4 From left to right. (a) Wing troop nose, (b) Compliance design, (c) Buckling design


SHELL STRUCTURE WITH A BASE THICKNESS
For shell structures with a given base thickness, the diIIiculties associated with (1) singular topology and
(2) buckling modes Ior low density zones do not exist since new cavities will not be introduced. Moreover,
the impact oI low density material to the accuracy oI buckling analysis is also limited as the overall
thickness oI the shell is usually much smaller than the dimension oI the structure.
Example 5: The same wing droop nose considered in example 4 is studied. The web layout, shown in
Fig.5, has been deIined based on the compliance design in Fig.4(b). The base plate thickness is T06mm,
and the Ilange is 60 mm high. The optimization problem is to optimize layout oI ribs oI 60 mm depth on the
web so that the weight oI the structure is minimized and buckling constraints are met. The shell elements
are modeled with a solid base thickness oI T06 mm and a designable region between T0 and T160 mm.
The initial design was assigned with a density oI 0.1 that yielded a weight oI 53.1 kg. With the lowest
eigenvalue at 0.113 Ior the initial design, the buckling constraints are severely violated. The Iinal design
shown in Fig.5(a) satisIied all buckling constraints, with 5 eigenvalues at critical value 1.0 and 7 more
below 1.3. The weight is reduced to 40.5 kg. In order to eliminate ambiguity oI analysis due to
intermediate density, the density was rounded up to 1.0 above the threshold oI 0.3 and reduced to zero Ior
the rest. The resulting discrete design is shown in Fig.5(b) and its weight is 42.9 kg. The lowest buckling
eigenvalue Ior this design is 0.97 and 7 eigenvalues were below 1.3. Note that the modeling oI deep ribs
using shell elements could be expected to yield reasonable analysis accuracy Ior the buckling modes oI the
web since it is similar to beam modeling. However, the lateral buckling oI ribs is not captured with this
model.



Fig.5 From left to right. (a) Buckling design with intermediate densitv, (b) Discrete design for reanalvsis

CONCLUSION
DiIIiculties associated with buckling responses in topology optimization are discussed in this paper. In
particular the so-called singular topology` phenomenon due to the conditional nature oI buckling
responses represents a Iundamental barrier that prevents a consistent and meaningIul solution. However, it
has been shown in this study that topology optimization can be used to optimize the layout oI stiIIening
ribs Ior shell structures with none zero base thickness. The most important characteristic oI this class oI
problem is that the buckling responses are not conditional. ThereIore, the phenomena oI (1) singular
topology` and (2) structurally irrelevant buckling modes in low density zones do not exist. This class oI
problems is quite meaningIul Ior aerospace and automotive applications as buckling weaknesses represent
a critical aspect oI shell structures that need to be addressed with well-designed stiIIening ribs. MeaningIul
results were achieved Ior real industrial examples. The study in carried out on the soItware platIorm Altair
OptiStruct |12|.

REFERENCES
|1| Bendsoe, M. 1995: Optimization oI structural topology, shape and material. Springer-Verlag, Berlin
|2| Rozvany, G.I.N.; Bendsoe, M.; Kirsch, U. 1995: Layout optimization oI structures. Appl. Mech. Rev.,
48, 41-119
|3| Sigmund, O.; Petersson, J. 1998: Numerical instabilities in topology optimization: A survey on
procedures dealing with checkerboards, mesh-dependencies and local minima. Struct. Optim., 16,
68-75
|4| Thomas, H.L., Zhou, M., Schramm, U. 2002: Issues oI commercial optimization soItware
development. Struct. Multidisc. Optim., 23, 97-110
|5| Bendsoe, M.P; Sigmund, O. (2003): Topology Optimization Theory, Methods and Applications.
Springer, Berlin.
|6| Neves, M.M., Rodrigues, H.C.; Guedes, J.M. 1995: Generalized topology design oI structures with a
buckling load criterion. Struct. Optim., 10, 71-81
|7| Zhou, M. 1996: DiIIiculties in truss topology optimization with stress and local buckling constraints.
Struct. Optim., 11, 134-136
|8| Rozvany, G.I.N. 1996: DiIIiculties in truss topology optimization with stress, local buckling and
system buckling constraints. Struct. Optim., 11, 213-217
|9| Sved, G., Ginos, Z. 1968: Structural optimization under multiple loading. Int. J. Mech. Sci., 10,
803-805.
|10|Cheng, G., Guo, X. 1998: -relaxed approach in structural topology optimization. Struct. Optim., 13,
258-266
|11|Krog, L., Tucker, A., Rollema, G. 2002: Application oI topology, sizing and shape optimization
methods to optimal design oI aircraIt components. Proc. Altair Hvperworks 3
rd
UK Conference,
November, 2002, 11/1-11/12
|12|Altair OptiStruct 2003: Users manual. Altair Engineering, Inc., Troy, MI

Potrebbero piacerti anche