Nguyen Enhancing Single-loop SRBTO AIAA
Transcript of Nguyen Enhancing Single-loop SRBTO AIAA
American Institute of Aeronautics and Astronautics
1
Enhancing Single-loop Approach for Component and
System Reliability-Based Topology Optimization
Tam H. Nguyen1, Junho Song
2, and Glaucio H. Paulino
3
University of Illinois at Urbana-Champaign, Urbana, IL, 61801
This paper introduces new single-loop algorithms developed for component and system
reliability-based design and topology optimization. A single-loop reliability-based
design/topology optimization (RBDO/RBTO) algorithm replaces the inner-loop iterations
that evaluate probabilistic constraints by a non-iterative approximation. The proposed
single-loop algorithms account for statistical dependence between the limit-states by using a
matrix-based system reliability (MSR) method to compute the system failure probability and
its parameter sensitivities. The SRBTO/MSR approach is applicable to general system
events including series, parallel, cut-set and link-set systems and provides the gradients of
the system failure probability to facilitate gradient-based optimization. In most RBTO
applications, probabilistic constraints are evaluated by use of the first-order reliability
method (FORM) for efficiency. In order to improve the accuracy of the reliability
calculations for RBDO or RBTO problems with high nonlinearity, we introduce a new
single-loop RBDO scheme utilizing the second-order reliability method (SORM). Moreover,
in order to overcome challenges in applying the proposed algorithm to computationally
demanding topology optimization problems, we utilize the multiresolution topology
optimization (MTOP) method, which achieves computational efficiency in topology
optimization by assigning different levels of resolutions to three meshes representing finite
elements, design variables, and material density distribution, respectively. The paper
provides a numerical example of three-dimensional topology optimization to demonstrate
the proposed CRBTO and SRBTO algorithms and applications. Monte Carlo simulations
are also performed to verify the accuracy of the failure probabilities computed by the
proposed approach.
Nomenclature
d = vector of design variables
C = compliance
volfrac = volume fraction
i = density element i
dn = design variable n
rmin = minimum length scale
c = “event” vector
p = “probability” vector ˆ
iα = negative normalized gradient vector
i = reliability index
βt
i = target reliability index
Q4/n25 = MTOP Q4 element with 25 density elements and 25 design variables
B8/n125 = MTOP B8 element with 125 density elements and 125 design variables
1 Doctoral Student, Department of Civil and Environmental Engineering, University of Illinois, Urbana, IL, 61801,
(AIAA Student Member) 2 Assistant Professor, Department of Civil and Environmental Engineering, University of Illinois, Urbana, IL, 61801
3 Donald Biggar Willett Professor in Engineering, Department of Civil and Environmental Engineering, University
of Illinois, Urbana, IL, 61801, (AIAA Member)
American Institute of Aeronautics and Astronautics
2
I. Introduction
tructural design/topology optimization aims to obtain an optimal set of structural design variables or topology
that fulfills the objective(s) of a structural design while satisfying given design constraints. In spite of active
research efforts and significant technological advances in the past decades1,2
, there still exist impediments to
adopting such optimization techniques in design practice. In particular, most of the efforts have been conducted in a
deterministic manner although uncertainties in loads or material properties may result in significant likelihood of
violating design constraints. In this study, this approach is referred to as deterministic topology optimization (DTO).
Recently, active research has been performed to achieve optimal topologies with acceptable probability of satisfying
given constraints. This approach is often termed as reliability-based topology optimization (RBTO). It is noted that
most research efforts on RBTO have been focused on satisfying the probabilistic constraint given for each failure
mode. In the current study, this approach is referred to as component reliability-based topology optimization
(CRBTO). A generic formulation for CRBTO problems is given as follows.
min (ρ( ; ))
. . [ (ρ( ; ), ) 0] , 1,...,
(ρ( ; ))
ti i
d
L U
f
s t P g P i n
dψ d
ψ d X
K ψ d u f
d d d
(1)
where dk is the vector of deterministic design variables; ( ;d) is the material density at the position
2 or
3 that is generally determined by a projection function fp(.) and the design variables, i.e. ( ;d)=fp(d); f(.) is the
objective function that often describes the volume, compliance or displacement of the structure; Xm is the vector
of random variables representing the uncertainties in the problem; gi(.), i=1,..,n is the i-th “limit-state function” that
indicates violating a design constraint given in terms of volume, displacement, or compliance by its negative sign,
i.e. gi(.) 0; t
iP is the constraint on the probability of the i-th limit state; K, ud and f respectively denote the stiffness
matrix, displacement vector, and load vector in the equilibrium condition; and dL and d
U are the lower and upper
bounds on d, respectively. For simplicity, the equilibrium condition and the bounds on the design variables will be
omitted in the following RBTO formulations of the paper. The probability constraint in Eq. (1) is described in terms
of either the reliability index (RIA)3 or the performance function, i.e. the
t
iP - quantile of the limit-state function
(PMA)4, which is obtained by use of a structural reliability analysis method such as the first-order reliability method
(FORM).
While most research efforts in the literature have been focused on CRBTO, in certain circumstances, the
probabilistic constraint should be given on a system failure event, i.e. a logical (or Boolean) function of multiple
failure modes. For example, the failure of a topology design can be defined as an event that at least one of the
potential failure modes occurs. This approach is termed as system reliability-based topology optimization (SRBTO).
SRBTO introduces additional complexity to reliability calculations especially when component events are
statistically dependent, or when the system event is not a series (i.e. union of events) or parallel system (i.e.
intersection of events). A generic formulation for SRBTO is as follows.
min (ρ( ; ))
. . ( ) (ρ( ; ), ) 0 , 1,...,k
t
sys i sys
k i C
f
s t P E P g P i n
dψ d
ψ d X (2)
where P(Esys) is the probability of the system failure event; Ck is the index set of the components (limit-states) in the
k-th cut-set; and t
sysP is the constraint on the system failure probability. Any type of system event may be considered
in SRBTO but, for illustration purpose, Eq. (2) shows a cut-set system formulation that can represent series, parallel,
and cut-set systems. A limited number of studies have been performed on SRBTO5 because calculation of system
probability and its parameter sensitivities introduces additional complexity to the topology optimization that already
requires high computational cost.
In the research efforts to apply the single-loop RBTO approaches to a wide range of problems, we identified the
needs to enhance the accuracy of the failure probabilities computed by the first-order reliability method, and the
applicability of the approach to large-size or three-dimensional topology optimization problems.
S
American Institute of Aeronautics and Astronautics
3
II. SORM-Based Component Reliability-Based Topology Optimization
Let us consider the following formulation of a single-loop approach6 for CRBTO problems in Eq. (1):
T,
min (ρ( ; ))
. . g g ρ( ; ), ( ) 0, =1,...,
(ρ( ; ), ( ))ˆ ˆwhere β ( ) and
(ρ( ; ), ( ))
ti
i
ti iP
t t t t ii i i i
i
f
s t i n
g
g
d
xx u
x u u
ψ d
ψ d x u
ψ d x uu α α J
ψ d x u
(3)
At each step of the design iterations, iu is
obtained by use of the KKT condition6 using
the target reliability index βt
i . Then, an
approximate most probable point (MPP) t
iu is
obtained by scaling the negative normalized
gradient vector at iu u , ˆ t
iα by βt
i as shown
in Fig. 1. The validity of this approximate
MPP is checked at the final step of the design
iterations. As this single-loop CRBTO uses
the first-order reliability method (FORM), the
reliability calculations can be inaccurate for
highly nonlinear limit-states7.
We thus modify this procedure to improve
the accuracy of the above single-loop
approach. Instead of finding the approximate
MPP on the surface of the sphere with the
fixed radius βt
i, the radius is updated at each
step of the design iterations by the ratio of βt
i
to the reliability index by a more accurate
reliability method such as the second-order
reliability method (SORM). The formulation of the proposed scheme is as follows.
( ) T,
( )
min (ρ( ; ))
. . (ρ( ; ), ( )) 0 1,...,
(ρ( ; ), ( ))ˆ ˆat the -th step: β ( ) and
(ρ( ; ), ( ))
1
ti
i
ti iP
t t k t t ii i i i
i
ti
t k ti i
f
s t g g i n
gk
g
k
d
xx u
x u u
ψ d
ψ d x u
ψ d x uu α α J
ψ d x u
( 1)
( 1)( )otherwiset k
it k SORMi
(4)
where ( )βt k
i is the radius used for the scaling at the k-th step of the iterations; and ( 1)( )βt k SORM
iis the reliability index
by SORM, which is obtained based on the principal curvatures around iu u at the (k─1)th step. The convergence
of the radius ( )βt k
i indicates that the reliability index by SORM approaches the target reliability index βt
i. Compared
to similar techniques in the literature8,9,10
, our study focuses on implementation of the idea of updating into the
single-loop RBTO based on FORM. In this study, the improved CRBTO by SORM in Eq. (4) is termed as “SORM-
based CRBTO.”
Figure 1. Approximation in the single-loop algorithm.
g , ( ) 0d X U
KKT
MPP
1U
2U
iU
ˆ t
iα
ˆβt t
i iα
βt
i
t
iU
*
iU
American Institute of Aeronautics and Astronautics
4
III. System Reliability-Based Topology Optimization using MSR Method
When the limit-states in an SRBTO problem are assumed to be statistically independent5, the system probability can
be computed by algebraic calculations of the probabilities of the individual limit-states. For example, let us describe
the failure probability of a series system event by use of the inclusion-exclusion formula
1
1
1 2
1 1 11
( ) ( ) ( 1) ( )n n n n
n
i i i j n
i i j ii
P E P E P E E P E E E (5)
where Ei denotes the failure event of the i-th limit-state. When these events are statistically independent of each
other, each joint probability in Eq. (5) can be computed by the product of the component probabilities, e.g.
P(E1E2E3)=P(E1)P(E2)P(E3). If there exists significant statistical dependence between limit-states due to shared or
correlated random variables, one needs to use system reliability analysis methods that can account for statistical
dependence in computing the system failure probability. It is desirable to obtain parameter sensitivities of the system
failure probability to facilitate the use of gradient-based optimization algorithm. However, computation of the
parameter sensitivities of a system failure probability is challenging when component events are statistically
dependent or the system event is not a series or parallel system.
The authors have recently applied the matrix-based system reliability (MSR) method to general reliability-based
design optimization problems11
. The current study aims to use the MSR method for topology optimization. The
MSR method12
computes the probability of a general system including series, parallel, cut-set and link-set system
and its parameter sensitivities by systematic matrix calculations. Consider a system event whose i-th component,
i=1,…,n has two distinct states, e.g., failure or survival. Then, the sample space can be subdivided into N=2n
mutually exclusive and collectively exhaustive (MECE) events, denoted by ej, j=1,..,N. Then, any system event can
be represented by an “event” vector c whose j-th element is 1 if ej belongs to the system event and 0 otherwise. Let
pj=P(ej), j=1,…,N denote the probability of ej. Due to the ej’s mutual exclusiveness, the probability of the system
event Esys, i.e. P(Esys) is the sum of the probabilities of ej’s that belong to the system event. Therefore, the system
probability is computed by the inner product of the two vectors, that is
T
T
(independent components)
( )( ) ( ) (dependent components)sysP E
f dS
s
c p
c p s s s (6)
where p is the “probability” vector that contains pj’s, j=1,…,N; S denotes the random variables identified as the
sources of statistical dependence between components, termed as common source random variables (CSRVs)12
. For
a given outcome of CSRVs, the component events are conditionally independent of each other, which allows us to
use the efficient procedure to construct the probability vector that is applicable to independent components; p(s)
denotes the probability vector constructed by use of the conditional failure probabilities of the limit-states given S=s,
i.e. Pi(s)=P(Ei|S=s) instead of Pi=P(Ei); and fs(s) is the joint probability density function (PDF) of S. Song and
Kang12
developed matrix-based procedures to construct the vectors c and p efficiently; to compute conditional
probabilities and component importance measures; and to evaluate parameter sensitivities of the system failure
probability.
A single-loop SRBTO approach13
employing the MSR method is formulated as follows.
,
T
T
T
min (ρ( ; ))
. . (ρ( ; ), ( )) 0 1,...,
( ) ( ) dependent ( ; )
independent
ˆ ˆwhere β ( ) and
t
ti
ti iP
t tsyst
syst t
sys
t t ti i i i
f
s t g g i n
f d PP E
P
d P
Ss
ψ d
ψ d x u
c p s s sP
c p
u α α ,
(ρ( ; ), ( ))
(ρ( ; ), ( ))i
t i
i
g
g
xx u
x u u
ψ d x uJ
ψ d x u
(7)
American Institute of Aeronautics and Astronautics
5
IV. SORM-Based System Reliability-Based Topology Optimization
The single-loop SRBTO/MSR approach in Eq. (7) is improved by enhancing the accuracy of component reliability
analysis results that are used for the system reliability analyses. The formulation of the SORM-based SRBTO/MSR
is as follows.
,
T ( )
( )
T ( )
min (ρ( ; ))
. . (ρ( ; ), ( )) 0 1,...,
( ) ( ) dependent ( ; )
independent
where
t
ti
ti iP
t SORM tsyst SORM
syst SORM t
sys
i
f
s t g g i n
f d PP E
P
d P
Ss
ψ d
ψ d x u
c p s s sP
c p
uT
,
(ρ( ; ), ( ))ˆ ˆβ ( ) and
(ρ( ; ), ( ))i
t t t t ii i i
i
g
g
xx u
x u u
ψ d x uα α J
ψ d x u
(8)
where ( )t SORM
P is the vector of the component failure probabilities by SORM at the approximate MPPs,
, 1,...,t
i i nu ; ( ) ( )t SORMp s and ( )t SORM
p respectively denote the conditional and marginal probability vectors
constructed by use of the SORM reliability indexes ( )t SORM
iinstead of t
i; and
iu is obtained by use of the KKT
condition using βt
i. The only change from Eq. (7) is that the probability vector is constructed by use of the SORM
reliability indexes instead of the FORM reliability indexes at the approximate MPPs.
V. RBTO using Multiresolution Topology Optimization
A main challenge in adopting topology optimization technique in design practice is its large computational cost.
Topology optimization often employs the material distribution method, which rasterizes the domain via the density
of pixels/voxels. This approach often requires a large number of design variables, especially in three-dimensional
applications. Most of the research efforts to overcome this challenge focus on finite element analysis since it
constitutes the dominant computational cost in topology optimization. For example, researchers make use of
powerful computing resources such as parallel computing14
or fast iterative solvers15
. These studies employ the same
resolution for the finite elements mesh and the design mesh during optimization process. The authors recently
developed a multiresolution topology optimization (MTOP) approach16
for handling large-scale problems with
relatively low computational costs. This approach uses distinct meshes for finite element, density and design
variables. In particular, the analysis is performed on a coarser finite element mesh, optimization is performed on a
fine design variable mesh, and element densities are defined on a fine mesh. This multiresolution approach enables
high resolution optimal designs with significantly reduced computational costs. This paper introduces a framework
to integrate the RBTO approach using MTOP to handle large-scale topology problem.
Topology optimization of a continuum structure aims to find the optimal distribution of the material densities in
a specific domain, which are considered as design variables. The basic formulation of “minimum compliance” (i.e.
maximum stiffness) topology optimization problem is expressed in the following discrete form using finite element
method:
Tmin (ρ( ; )) (ρ( ; ), )
. . (ρ( ; )) ρ( ; )
d d
s
f C
s t V dV V
dψ d ψ d u f u
ψ d ψ d (9)
where T( , ) dC u f u is the compliance of the continuum; ( )V is the total volume; and Vs is the prescribed volume
constraint. A desirable solution of topology optimization specifies the density at every point in the domain as either
0 (void) or 1 (solid). However, since it is impractical to perform such an integer optimization, the problem is relaxed
such that the density can have any value between 0 and 1. For example, in the Solid Isotropic Material with
Penalization (SIMP) approach17,18
, the constitutive matrix is parameterized using solid material density as follows
American Institute of Aeronautics and Astronautics
6
0( ) ρ( ; )p
D ψ ψ d D (10)
where 0D is the constitutive matrix of the material in the solid phase, corresponding to the density ( ) 1x ; and p is
the penalization parameter. To prevent singularity of the stiffness matrix, a small positive lower bound, e.g.
min=103, is placed on the density. Using the penalization parameter p>1, the intermediate density approaches either
0 (void) or 1(solid). In the conventional element-based approach, the density of each element is represented by one
value e and the global stiffness matrix K in Eq. (1) is expressed as
T
1 1
(ρ ) (ρ )el el
e
N N
e e e
e e
dK K B D B (11)
where Ke( e) is the stiffness matrix of the element e; B is the strain-displacement matrix of shape function
derivatives; and D( e) is the constitutive matrix.
Different from conventional element-based approaches which use the same mesh for finite element analysis and
design (Fig. 2a), the MTOP approach utilizes three different meshes: a relatively coarse finite element mesh to
perform the analysis, a fine design variable mesh to perform the optimization, and a fine density mesh to represent
material distribution and compute the stiffness matrices. The density mesh is finer than the finite element mesh so
that each finite element consists of a number of density elements (sub-elements). Within each density element, the
material density is assumed to be uniform. For example, Fig. 2a shows a conventional element-based approach Q4/U
element while Fig. 2b shows an MTOP Q4/n25 element where “n25” indicates that the number of density elements
per a Q4 element is 25 (coincides with the number of design variables).
The MTOP approach introduces a scheme to obtain the element stiffness matrix from corresponding density
elements. The stiffness matrix is computed by sum of the stiffness integration in each density element. The
integration of the stiffness in each density element is performed as standard finite element method. As a result, the
formulation for the stiffness matrix integration is expressed as follows.
0
0
1 1
T T T 0
0
1 11 1
T 0
0
ξ η (ρ ) (ρ )n n
ie
i
N Np i p
e i i i
i i
i
i
d Jd d Jd
Jd
K B DB B DB B D B I
I B D B
(12)
From these formulations, the sensitivities of the objective function and the constraint in Eq. (9) can be derived.
Figure 2. Element-based and MTOP elements: (a) Q4/U, and (b) MTOP Q4/n25
Displacement
Density
Design variable
(a) (b)
American Institute of Aeronautics and Astronautics
7
VI. Numerical Example
This numerical example is to demonstrate the improved accuracy of the proposed SORM-based RBTO methods.
The objective of optimization is to minimize the volume in a cube domain shown in Fig. 3a while satisfying
deterministic or probabilistic constraints on the compliances for multiple load cases. One corner is fixed in all three
directions while the other corners are restricted in the vertical direction only. The isotropic material is assumed to
have Young’s modulus of E0=1,000 and Poisson’s ratio of =0.3. A cube of edge length L=24 is divided into
12×12×12 B8/n125/d125 MTOP elements. The minimum length scale rmin=L/10, and penalization parameter p=3 are
employed. The structure is subjected to three random loads applied at five locations as shown in Fig. 3a. F1 denotes
the magnitude of the force at the center while F2 and F3 represent the loads at the midpoints between the center and
the four corner points of the top face. F1, F2 and F3 are assumed to be normal random variables with the mean values
100, 0 and 0, and with the standard deviations 10, 30 and 40, respectively.
Limit-states are defined on the compliances caused by two load combinations1 1 2( , )F FF and
2 1 3( , )F FF as
follows.
T( , ) ( , ) , 1,2t t
i i i i i i ig C C C iρ F ρ F u F (13)
where ( 120)t
iC is the constraint on the compliance; ( , )i iC ρ F is the compliance corresponding to the load case iF ;
and iF is the global force vector assembled based on the load case iF . The following three topology optimization
methods are investigated: (1) Deterministic Topology Optimization (DTO) using the mean values of the loads with
Figure 3. Topology optimization of the cube: (a) configuration, (b) DTO (volfrac=6.3%), (c) SORM-based
CRBTO ( (F1)=10, volfrac=24.4%), (d) SORM-based SRBTO ( (F1)=10, volfrac=22.3%), and (e) SORM-
based SRBTO ( (F1) = 20, volfrac = 23.9%)
(b)
(c) (d) (e)
F
1
L
F1F3F3F2
F2
L
L
(a)
American Institute of Aeronautics and Astronautics
8
deterministic constraints ( , ) 0i ig ρ F ; (2) CRBTO with probability constraints 1 2 0.02275t tP P , i.e. reliability
indexes 1 2 2.0t t; and (3) SRBTO with the system limit-state 1 1 2 2{( ( , ) 0) ( ( , ) 0)}sysE g gρ F ρ F with
0.04493tsysP , which is given so as to match the
system failure probability of the optimal
topology of the CRBTO.
Fig. 3 shows the optimal topologies by DTO
(Fig. 3b), SORM-based CRBTO (Fig. 3c), and
SORM-based SRBTO (Fig. 3d). The volume
fraction of DTO is lower than CRBTO and
SRBTO because the risk of high compliance
caused by the load uncertainties is ignored. It is
noteworthy that, with the same system failure
probabilities, the volume fraction of CRBTO is
10% higher than SRBTO. This is because
CRBTO approach (assigning fixed constraints
on individual components) is generally more
constrained than SRBTOs (assigning a
constraint on system event, not on the individual
components) at the same level of system failure
probability11
. Fig. 3e shows the result of
SRBTO with the standard deviation of F1
increased to 20 in order to see the impact of the load variability on the optimal topology. In summary, it is seen from
Fig. 3 that the optimal topology is affected significantly by the load variability and the failure event definitions on
the optimal topology of a structure.
The convergence histories of the optimizations are shown in Fig. 4. The proposed single-loop SORM-based
CRBTO and SRBTO show similar rates of convergence rate, which are also comparable to that of DTO. The system
failure probability of the optimal topology found by SORM-based SRBTO/MSR, Psys= 0.04493 is verified by a
fairly close estimate of Monte Carlo simulations (MCS), Psys= 0.04515 (106 times, c.o.v = 0.005).
In order to demonstrate the improved accuracy of the SORM-based single-loop CRBTO method, the results are
compared with those by the FORM-based CRBTO with the component probability targets 1 2 0.02275.t tP P
Figure 5a shows the differences in the volume fractions of the optimal designs. Monte Carlo simulations (MCS: 106
times, c.o.v=0.005) are performed to find the component failure probabilities of the optimal topologies by the
FORM-based and SORM-based CRBTOs. The results in Fig. 5b and Fig. 5c show that the component probabilities
of SORM-based CRBTOs are fairly close to the target probabilities while the FORM-based CRBTOs show
significant errors especially when the random loads have large variability.
The accuracy of the SORM-based single-loop SRBTO method is also investigated. The FORM-based and
SORM-based SRBTO are performed with the system probability target of 0.04493 while standard deviation of load
F1 is varied from 10 to 60. Fig. 6a compares the volume fractions by the FORM and SORM-based SRBTOs. It is
seen that the FORM-based SRBTO provides unconservative designs due to the inaccuracy in reliability calculations.
The results of Monte Carlo simulations (MCS: 106 times; c.o.v=0.005) in Fig. 6b show that the proposed SORM-
based SRBTO provides improved accuracy in predicting the system failure probability.
The results in Fig. 5a and Fig. 6a show the volume fractions of the optimal designs increase significantly as the
load variability increases. It is because the variability of random load increases the uncertainty of the compliance
and thus the probability of violating given constraints.
Figure 4. Convergence history
10 20 30 40 500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
iterationvolu
me
frac
tion
DTO
CRBTO
SRBTO
American Institute of Aeronautics and Astronautics
9
Figure 5. CRBTOs with standard deviation of load F1: (a) volume fraction of optimal designs, (b) failure
probabilities on the first limit-state, and (c) failure probabilities on the second limit-states
Figure 6. SRBTOs with standard deviation of load F1: (a) volume fractions of optimal designs, and (b) system
failure probabilities
(a)
10 20 30 40 50 600.2
0.3
0.4
0.5
0.6
0.7
standard deviation (F1)
opti
mal
volu
me
frac
tion
FORM-based CRBTO
SORM-based CRBTO
(b)
10 20 30 40 50 600.022
0.024
0.026
0.028
0.030
standard deviation (F1)
com
ponen
t pro
bab
ilit
y
MCS on FORM-based
MCS on SORM-based
constraint on P1(C
1)
(c)
10 20 30 40 50 600.022
0.024
0.026
0.028
0.030
standard deviation (F1)
com
ponen
t pro
bab
ilit
y
MCS on FORM-based
MCS on SORM-based
constraint on P2(C
2)
(a)
10 20 30 40 50 600.2
0.3
0.4
0.5
standard deviation (F1)
volu
me
frac
tion
FORM-based SRBTO
SORM-based SRBTO
(b)
10 20 30 40 50 600.044
0.046
0.048
0.050
0.052
standard deviation (F1)
syst
em p
robab
ilit
y
MCS on FORM-based
MCS on SORM-based
constraint on Psys
American Institute of Aeronautics and Astronautics
10
VII. Conclusion
This paper presents three research developments for enhancing the theories and applications of component and
system reliability-based topology optimization (CRBTO and SRBTO): (1) introducing a single-loop SRBTO
approach that employs the matrix-based system reliability (MSR) method to handle the statistical dependence
between multiple limit-states; (2) developing SORM-based single-loop approaches for CRBTO and SRBTO to
improve the accuracy in evaluating probabilistic constraints; and (3) incorporating multiresolution topology
optimization (MTOP) approach to CRBTO and SRBTO in order to obtain high resolution design with a relatively
low computation cost. A computational demanding numerical example demonstrates the importance of uncertainties
on structural topologies and the improvement of the accuracy by the proposed approaches.
Acknowledgments
This research was funded in part by a grant from the Vietnam Education Foundation (VEF) and the National
Science Foundation (NSF). The supports are gratefully acknowledged. The opinions, findings, and conclusions
stated herein are those of the authors and do not necessarily reflect those of sponsors.
References 1Rozvany, G. I. N., "Aims, Scope, Methods, History and Unified Terminology of Computer-Aided Topology Optimization in
Structural Mechanics," Structural and Multidisciplinary Optimization, Vol. 21, No. 2, 2001, pp. 90-108. 2Rozvany, G. I. N., "A Critical Review of Established Methods of Structural Topology Optimization," Structural and
Multidisciplinary Optimization, Vol. 37, No. 3, 2009, pp. 217-237. 3Enevoldsen, I., Sørensen, J. D., "Reliability-Based Optimization in Structural Engineering," Structural Safety, Vol. 15, No.
3, 1994, pp. 169-196. 4Tu, J., Choi, K. K., Park, Y. H., "A New Study on Reliability-Based Design Optimization," Journal of Mechanical Design,
Vol. 121, No. 4, 1999, pp. 557-564. 5Silva, M., Tortorelli, D. A., Norato, J. A., Ha, C., Bae, H. R., "Component and System Reliability-Based Topology
Optimization Using a Single-Loop Method," Structural and Multidisciplinary Optimization, Vol. 41, No. 1, 2010, pp. 87-106. 6Liang, J., Mourelatos, Z. P., Tu, J., "A Single-Loop Method for Reliability-Based Design Optimisation," International
Journal of Product Development, Vol. 5, No. 1, 2008, pp. 76-92. 7Nguyen, T. H., "System Reliability-Based Design and Multiresolution Topology Optimization," Ph.D thesis, Department of
Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, 2010. 8Royset, J. O., Der Kiureghian, A., Polak, E., "Optimal Design with Probabilistic Objective and Constraints," Journal of
Engineering Mechanics, Vol. 132, No. 1, 2006, pp. 107-118. 9Rahman, S., Wei, D., "Design Sensitivity and Reliability-Based Structural Optimization by Univariate Decomposition,"
Structural and Multidisciplinary Optimization, Vol. 35, No. 3, 2008, pp. 245-261. 10Lee, I., Choi, K. K., Gorsich, D., "System Reliability-Based Design Optimization Using the MPP-Based Dimension
Reduction Method," Structural and Multidisciplinary Optimization, Vol. 41, No. 6, 2010a, pp. 823-839. 11Nguyen, T. H., Song, J., Paulino, G. H., "Single-Loop System Reliability-Based Design Optimization Using Matrix-Based
System Reliability Method: Theory and Applications," Journal of Mechanical Design, Vol. 132, No. 1, 2010, pp. 011005-1-11. 12Song, J., Kang, W. H., "System Reliability and Sensitivity under Statistical Dependence by Matrix-Based System
Reliability Method," Structural Safety, Vol. 31, No. 2, 2009, pp. 148-156. 13Liang, J., Mourelatos, Z. P., Nikolaidis, E., "A Single-Loop Approach for System Reliability-Based Design Optimization,"
Journal of Mechanical Design, Vol. 129, No. 12, 2007, pp. 1215-1224. 14Borrvall, T., Petersson, J., "Large-Scale Topology Optimization in 3D Using Parallel Computing," Computer Methods in
Applied Mechanics and Engineering, Vol. 190, No. 46-47, 2001, pp. 6201-6229. 15Wang, S., de Sturler, E., Paulino, G. H., "Large-Scale Topology Optimization Using Preconditioned Krylov Subspace
Methods with Recycling," International Journal for Numerical Methods in Engineering, Vol. 69, No. 12, 2007, pp. 2441-2468. 16Nguyen, T. H., Paulino, G. H., Song, J., Le, C. H., "A Computational Paradigm for Multiresolution Topology Optimization
(MTOP)," Structural and Multidisciplinary Optimization, Vol. 41, No. 4, 2010, pp. 525-539. 17Bendsøe, M. P., "Optimal Shape Design as a Material Distribution Problem," Structural and Multidisciplinary
Optimization, Vol. 1, No. 4, 1989, pp. 193-202. 18Rozvany, G. I. N., Zhou, M., Birker, T., "Generalized Shape Optimization without Homogenization," Structural and
Multidisciplinary Optimization, Vol. 4, No. 3, 1992, pp. 250-252.