1
Dissertation Defense
System Reliability-Based Design and
Multiresolution Topology Optimization
Tam H. Nguyen
Advisors: Glaucio H. Paulino & Junho Song
Department of Civil and Environmental Engineering
University of Illinois at Urbana-Champaign
07/16/2010
2
Introduction
Multiresolution Topology Optimization (MTOP)
Improving Multiresolution Topology Optimization (iMTOP)
System Reliability-based Design Optimization (SRBDO)
System Reliability-based Topology Optimization (SRBTO)
Summary and Conclusions
Contents
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
3
Topology Optimization
Classical structural design optimization: the optimal sizes or shapes
for a given layout and connectivity
Topology optimization: the best topology, shape, size under a given
domain and boundary conditions
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
size optimization
shape optimization
4
Topology Optimization Applications
Skidmore, Owings & Merrill, LLP (SOM)
(www.altair.com)
Airbus Wing box rib
500 kg reduction/wing
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
5
F1 F2F3
30×10×10 B8180×60×60 B8
Coarse mesh Fine mesh
~ 1.0 mil. unknowns
Fast solver, PC, C++
Run time: ~ 45.7 hours
Question 1: How to obtain high resolution with affordable computational cost?
Large-scale Topology Optimization
Wang, de Stuler, and Paulino, (2007), IJNME
Computationally
expensive
An example using Matlab code
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Low
resolution
6
Reliability-Based Design Optimization
High probability
of failure
Low probability
of failure
( , )f Xd μ
Safe
Objective function increase
B
A Unsafe
Deterministic Optimization
Reliability-Based Design
Optimization (RBDO)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
7
System Reliability-Based Design Optimization
Component RBDO
,min ( , )
. . ( , ) 0 =1,...,
,
t
i i
L U L U
f
s t P g P i n
X
Xd μ
X X X
d μ
d X
d d d μ μ μ
System RBDO
,
sys
min ( , )
. . ( )= g ( , ) 0
,
k
t
i sys
k i C
L U L U
f
s t P E P P
X
Xd μ
X X X
d μ
d X
d d d μ μ μ
?
Question 2: How to handle system probability in RBDO?
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
8
1. To obtain high resolution with affordable
computational cost in topology optimization.
2. To handle system probability in Reliability-
Based Design Optimization (RBDO).
3. To apply RBDO framework in topology
optimization (RBTO).
Objectives
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
9
Multiresolution Topology
Optimization
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
10
Topology Optimization Procedure
Problem formulation
Solid and Isotropic Material
with Penalization (SIMP)
0( ) ρ( ) pE Eψ ψ
T
ρ
min
min (ρ, )
. . : (ρ)
(ρ) ρ( )
0 ρ ρ( ) 1
d d
d
s
C
s t
V dV V
u f u
K u f
ψ
ψ
Optimizers
Optimality Criteria (OC)
Method of Moving Asymptotes
(MMA)
Finite Element Analysis
Objective Function & Constraints
Converged?
Result
Sensitivities Analysis
Update Material Distribution
Initial guess
Yes
No
Filtering (Projection) Technique
P
computationally
expensive
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
11
Large-scale (high resolution) TOP
Large number of finite elements
Computationally expensive
Existing high resolution TOP
Parallel computing (Borrvall and Petersson, 2000)
Fast solvers (Wang et al. 2007)
Approximate reanalysis (Amir et al. 2009)
Adaptive mesh refinement (de Stuler et al. 2008)
High Resolution Topology Optimization
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
12
Parallel computing:
Domain decomposition
TOP (1): Parallel Computing
A stool (884,736 B8/U)
96x96x96
40x120x120
A cross-shaped section (320,000 B8/U)
Borrvall and Petersson, (2000), IJNME
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
13
Fast iterative solvers
Use precondition Krylov subspace methods with recycling
Reduce computational time for FEA
TOP (2): Fast Solvers
Coarse mesh: 32x12x12
Fine mesh: 180x60x60
Configuration
Solution on a PC with approx. 1 million unknowns
Wang, de Stuler, and Paulino, (2007), IJNME
p
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
14
Reduce the number of FEA solutions
FEA at an interval of iterations
Approximate at other iterations
Efficiency factor: 1 ~ 5 times
TOP (3): Approximate Reanalysis
MBB: 60x20 Q4/UCantilever: 48x16x16 B8/U
Finite Element Analysis
or
Approximate the Displacement
Objective Function & Constraints
Converged?
Result
Sensitivities Analysis
Update Material Distribution
Initial guess
Yes
No
Filtering (Projection) Technique
Amir, Bendsoe, and Sigmund, (2009), IJNME
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
15
AMR TOP
Refine the solid and surface regions
Reduce the total number of FEs
Obtain resolution as fine uniform mesh (efficiency factor 3)
TOP (4): Adaptive Mesh Refinement
Configuration: 2:1:1 Uniform mesh: 128x64x64
524,288 B8/U
AMR: initial mesh 64x32x32
initial 65,520 B8/U
final 228,692 B8/U
de Stuler, Wang, and Paulino, (2008), IASS-IACM
P
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
16
Large-scale TOP
Fine mesh: Large number of finite elements
FEA cost increases
Remarks on High Resolution TOP
Existing approaches:
Powerful computing resources: many processors
Reduce cost associated with FEA:
• Fast solvers
• Approximate reanalysis
• Adaptive mesh refinement
Same discretization for analysis and design
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
17
Conventional element-based approach (Q4/U)
Same discretization for displacement and density
Proposed Multiresolution TOP (MTOP)
Displacement Density/design variable
Proposed MTOP approach (Q4/n25)
Different discretizations for displacement and density/design variables
Displacement Density Design variable
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
18
Stiffness matrix
MTOP: Integration of Stiffness Matrix
Te
e
dK B DB
Numerical integration
1
nT
e ii
i
AK B DB
SIMP model
T
0
1 1
( ) ( )n nN N
p p
e i e e i i iiii i
AK B D B I
i
iA
Vi
i
Sensitivity
1 1
(ρ )ρ ρ ρ
(ρ )ρ ρ
nNp
j j
j pe e i i ii i
n i n i n n
pd d d d
IK K
I
Q4/n25
B8/n125
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
19
Compute density from design variables
Minimum length-scale (Guest et al. 2004, Almeida et al. 2009)
MTOP: Projection (filtering)
( )i nf d
( )
( )
i
i
n nin S
i
mim S
d w r
w r
minmin
min
if ( )
0 otherwise
mimi
mi
r rr r
rw r
ρ ( )
( )i
i ni
n mim S
w r
d w r
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
20
MTOP Examples: 2D Cantilever Beam
Configuration MTOP Q4/n25 FE mesh 48x16
(C=208.23)
Q4/U FE mesh 48x16
(C=205.57)
Q4/U FE mesh 240x80
(C=210.68)
Convergence history
Objective: minimum compliance
Constraint: volfrac = 0.5
Length scale: rmin = 1.2
16
48
Nguyen, Paulino, Song, and Le, (2010), JSMO
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
21
MTOP Examples: 2D Michell Truss
Domain (3:2) Analytical solution
Sigmund solution (Sigmund, 2000) MTOP: 180x120 Q4/n25 elements
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
22
MTOP: 3D Cross-shaped Section
Borrvall & Petersson (2000)
320,000 B8/U elements
MTOP
5,000 B8/n125 elements
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
23
MTOP: 3D Bridge Design
Configuration
MTOP B8/n125
36,000 elements
10x120x30
Existing design
(http://www.sellwoodbridge.org)
6L
q
L
2L/3
L
L
L
non-designable layer Non-designable layer
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
24
Can MTOP’s efficiency be improved?
MTOP approach
Density/design variable:
same fine mesh
FE mesh: coarse
Reduce cost (ρ) d K u f
Finite Element Analysis
Objective Function & Constraints
Converged?
Result
Sensitivities Analysis
Update Material Distribution
Initial guess
Yes
No
Filtering (Projection) Technique
MTOP
???Improving MTOP efficiency?
Different discretizations for
density & design variable?
Q4/n25
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
25
MTOP approach (Q4/n25) or (Q4/n25/d25)Displacement Density Design variable
Proposed iMTOP approach (Q4/n25/d9) and (Q4/n25/d16)
Displacement Density Design variable
Q4/n25/d16Q4/n25/d9
Improving Multiresolution Topology Optimization (iMTOP)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
26
ρi
iV
jd jd
B8/n125/d8 B8/n125/d15
jd
jd
ρi
iV
TET4/n64/d8 TET4/n64/d10
Displacement Density Design variable
ρi
iV
di
B8/n125
MTOP iMTOP
ρi
iV
di
TET4/n64
Improving Multiresolution Topology Optimization (iMTOP)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
27
Compute density from design variables
Minimum length-scale (Guest et al. 2004, Almeida et al. 2009)
iMTOP: Projection (Q4/n25/d9)
( )i nf d
( )
( )
i
i
n nin S
i
mim S
d w r
w r
minmin
min
if ( )
0 otherwise
mimi
mi
r rr r
rw r
ρ ( )
( )i
i ni
n mim S
w r
d w r
Displacement
Design variable
Density
rmin
rni
in
Si
w(r)
1
rni
rmin
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
28
iMTOP: MBB Beam
300x100 Q4/U 60x20 Q4/U
60x20 Q4/n25/d25 60x20 Q4/n25/d16
60x20 Q4/n25/d9 60x20 Q4/n25/d4
20 40 60 80 1000
200
400
600
800
1000
Iteration
Com
pli
ance
conventional (60x20 Q4/U)
iMTOP (60x20 Q4/n25/d4)
iMTOP (60x20 Q4/n25/d9)
iMTOP (60x20 Q4/n25/d16)
MTOP (60x20 Q4/n25)
conventional (300x100 Q4/U)
1 2 3 4 5 60
1
2
3
4
5
6
7
8
9
10
11
norm
aliz
ed c
om
puta
tional
tim
e
1: conventional (60x20 Q4/U)
2: iMTOP (60x20 Q4/n25/d4)
3: iMTOP (60x20 Q4/n25/d9)
4: iMTOP (60x20 Q4/n25/d16)
5: MTOP (60x20 Q4/n25)
6: conventional (300x100 Q4/U)
convergence
Efficiency
?
60x20
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Messerschmitt-Bolkow-Blohm (MBB) beam
29
iMTOP: A Cube with Concentrated Load
MTOP B8/n125/d125
(C=29.04)
iMTOP B8/n125/d64
(C=29.06)
iMTOP B8/n125/d27
(C=29.08)
iMTOP B8/n125/d8
(C=29.33)
LP
L
L=24
24x24x24
convergence
1 2 3 40
1
2
3
4
5
6
7
8
9
10
11
no
rmal
ized
co
mp
uta
tio
nal
tim
e
1: iMTOP B8/n125/d8
2: iMTOP B8/n125/d27
3: iMTOP B8/n125/d64
4: MTOP B8/n125
10 20 30 40 500
50
100
150
200
Iteration
Co
mp
lian
ce
iMTOP B8/n125/d8
iMTOP B8/n125/d27)
iMTOP B8/n125/d64)
MTOP B8/n125)
Efficiency
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Nguyen, Paulino, Song, and Le, (submitted), IJNME
30
Why Adaptive MTOP?
Further improve the efficiency
Reduce the number of density elements and design variables
during optimization process?
Adaptive MTOP
Adaptive MTOP (e.g. Q4/U & Q4/n25/d4)
Q4/n25/d4 requires more computational cost than Q4/U
Q4/n25/d4 provides higher resolution
Use Q4/n25/d4 where and when needed only, otherwise Q4/U
Unchanged the Finite Element Mesh during optimization process
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
31
Adaptive MTOP Procedure
No
Element type array
A(i,j) = 1
(all Q4/U elements)
if k < L or U < k , k=1,25
Q4/U Q4/n25/d4
A(i,j) = 1
if L < < U
A(i,j) = 25
Q4/U Q4/n25/d4
A(i,j) = 25 A(i,j) = 1
Q4/U
Q4/n25/d4
A(i,j) = 1
A(i,j) = 25
Finite Element Analysis
Objective Function & Constraints
Converged?
Result
Sensitivities Analysis
Update Material Distribution
Initial guess
Yes
Filtering (Projection) Technique
Check & update element type array
Element type
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
32
Optimal topologies by iMTOP and adaptive MTOP
Adaptive MTOP: 2D Cantilever
2L
L
P
Configuration 160x80 Q4/U 32x16 Q4/U
initial iteration intermediate iteration final iteration
Q4/U Q4/n25/d4
Mesh (element types)
Topology
32x16 Q4/n25/d4 32x16 - adaptive
Q4/n25/d4 & Q4/U
Adaptive MTOP optimization process
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
33
Adaptive MTOP: 3D Cantilever Beam
Configuration 24x12x12
B8/U
24x12x12
B8/n125/d8
L
2L
L
P
24x12x12
B8/n125/d8 & B8/U
Adaptive MTOP optimization process
Initial iteration
(3,456 B8/U)Intermediate iteration
(2,288 B8/U & 1,168 B8/n125/d8)
Final iteration
(2,072 B8/U & 1,384 B8/n125/d8)
(FE mesh : 24x12x12 unchanged)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
34
System Reliability-Based
Design/Topology Optimization
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
35
RBDO Formulation
Component RBDO
,min ( , )
. . ( , ) 0 =1,...,
,
t
i i
L U L U
f
s t P g P i n
X
Xd μ
X X X
d μ
d X
d d d μ μ μ
System RBDO
,
sys
min ( , )
. . ( )= g ( , ) 0
,
k
t
i sys
k i C
L U L U
f
s t P E P P
X
Xd μ
X X X
d μ
d X
d d d μ μ μ
?
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
36
Matrix-based System Reliability (MSR) Method
Convenient: matrix-based procedures for c and p; easy SRA calculation (inner product)
General: uniform application to series, parallel, and any general systems
Flexible: inequality-type information; incomplete information (“LP bounds” method)
Efficient: no need to re-compute “p”; replace “c” for SRA of a new event
Common Source Effect: can account for statistical dependence between components
Decision Support: parameter sensitivities, component importance measure; inferences
Song and Kang, (2009); Structural Safety
e1e3
e8
e4e6
e5
e2
E3
E2
E1
e7
mutually exclusive and collectively
exhaustive events (MECE)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
37
Proposed approach: SRBDO using MSR
Adopt a single-loop RBDO (Liang et al. 2007)
Use MSR method to compute Psys and its gradients
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Single-loop RBDO Double-loop RBDO
Approx. MPP
1st constraint
Karush–Kuhn–Tucker (KKT) optimality conditions
Objective
function
Approx. MPP
nth constraintReliability eval.
1st constraint
Objective
function
Reliability eval.
nth constraint
MSR method
Single-loop PMA
1, , ,...,
T
T
min ( , )
. . ( , ( )) 0 =1,...,
( ) ( ) dependent
indepdendent
t tnP P
t
i i
t
sys
syst
sys
f
s t g i n
f d PP
P
X
Xd μ
S
s
d μ
d X U
c p s s s
c p
38
Proposed approach: SRBDO using MSR (contd.)
Sensitivity w.r.t. design variables d, x}
sys T
1 2
T
1 2
( )= ( )
θ θ
( ) ( ) ( )ˆ= ( ) ( ) ... ( ) ( )θ θ θ
( ) [ ( ) ( ) ( )]
n
n
Pf d
P P P
S
s
p sc s s
p s P s P sp s p s p s P s
P s s s s
Sensitivity w.r.t. component failure probability Pit
( ) ( ) β ( ) 1
β β φ( β )
i i i i
i i i i i
P P P
P P
s s s
Use probabilities and sensitivities by component reliability analysis (FORM)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
39
SRBDO of Truss System
1 6
1 2 3 4 5 6 { ,..., }
15
sys
1
1 2 3 4 5 6
min ( ) 2( )
. . = g ( , ) 0 0.001
( , ) 0.707 1, 2
0.500 3,...,6
, , , , ,
k
A A
t
i sys
k i C
i i i A
i i A
f A A A A A A
s t P P P
g A F F i
A F F i
A A A A A A
dd
d X
d X
0
Minimize total weight of the
system
Definition of system failure: at
least two members fail (cut-set
systems): effects of load re-
distributions NOT considered
L
FA
L
4
1 2
53
6
Random Variables
(Gaussian distribution)Mean Std Dev
Member strength Fi , i=1,..6 (Mpa) 745 62Applied load FA (kN) 4450 45
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
40
SRBDO of Truss System (contd.)
Members
Area: Ai (×103 mm2) Reliability Index: i
McDonald &
MahadevanSRBDO/MSR
McDonald &
MahadevanSRBDO/MSR
1 18.43 17.89 2.89 2.67
2 18.27 17.89 2.83 2.67
3 13.51 13.20 3.16 2.99
4 13.44 13.20 3.12 2.99
5 13.33 13.20 3.06 2.99
6 13.09 13.20 2.92 2.99
f(x) 105.24 103.36
Better optimal design (i.e. less total weight) and symmetric results
Monte Carlo simulations (c.o.v. = 0.03, 106 times) on the system
failure probability: Psys = 0.00107 (cf. MSR gives 0.001)
4
1 2
53
6
>
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
41
SRBDO of Truss System (contd.)
T
T
( ) 'CIM = ( | )=
( )
i sys
i i sys
sys
P E EP E E
P E
c p
c p
0.00
0.25
0.50
0.75
1.00
1 2 3 4 5 6
CIM
Components
4
1 2
53
6
Conditional probability Importance Measure (CIM)
Relative contribution of components to the system failure
probability (can be computed efficiently by MSR method)
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
42
SRBDO of Truss System (contd.)
6
1 2 2 4
1 2
5 3
2 30
27 28
29 1 2 2
35
32 33
34
11
1 2 9
7
8
16
1 2 2 14
12
10 15 20
26
1 2 2
22 23
25
5
13
17 18
19
21
31
36 24
Effects of load re-distributions (sequential failures)
Effects of correlation between random variables and between
components
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
43
Discrete structures
Mogami et al. (2006)
Truss examples
Existing SRBTO Approaches
11
1 2
1 1 11
( ) ( ) ( ) ( 1) ( )n n n n
n
sys i i i j n
i i j ii
P E P E P E P E E P E E E
1 2 3 1 2 3( ) ( ) ( ) ( )P E E E P E P E P E
Continuum structures
Silvia et al. (2010)
Limit-states: statistically independent
DTO SRBTO
Ground structure Optimal structure
Objective: SRBTO for continuum structures with dependent limit-states?
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Mogami et al. (2006), JSMO
Silvia et al. (2010), JSMO
44
Proposed Approach: SORM-based RBTO
SORM-based CRBTO g , ( ) 0d X U
KKT
1U
2U
iU
ˆ tiα
ˆβt ti iα
βtit
iU
*
iU
( ) t k ti i
T ( )
( )
T ( )
( ) ( )( ; )
t t
syst
sys t t
SORM
SORM
SORM
sys
f d PP E
P
Ssc p s s s
Pc p
( ) ( 1)
( 1)( )
SORM
t
t k t ki
i it k
i
T
T
( ) ( )( ; )
t t
syst
sys t t
sys
f d PP E
P
Ssc p s s s
Pc p
At the k-th step
SORM-based SRBTO
Enhance the accuracy in RBTO
First-Order Reliability Method (FORM) inaccurate for nonlinear limit-states
Propose to use Second-Order Reliability Method (SORM) to improve the accuracy
At the k-th step
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
45
SRBTO of a Stool
F1
L
F1F3
F3F2
F2
L=12
L
Objective: minimize volume V( )
Limit-states:
Random loads: : F ~ (F1,F2,F3) ~
N(100,10), N(0,30), N(0,40)
Load cases:
( , ) 120 ( , ), 1,2i i i ig C iρ F ρ F
1 1 2( , ),F FF 2 1 3( , )F FF
Constraints
Deterministic TO (DTO):
Component RBTO (CRBTO):
System RBTO (SRBTO):
( , ) 0, 1,2ig iρ f
( ( , ) 0) , 1,2ti i iP g P iρ F
( ( , ) 0) ti i sysP g Pρ F
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
46
Optimal Topologies
DTO CRBTO
volfrac = 24.4%
( F=10)
SRBTO
volfrac =22.3%
( F1=10)
SRBTO
volfrac =23.9%
( F1=20)
volfrac = 6.3%
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
47
Improve Accuracy by Second-Order Reliability Method
10 20 30 40 50 600.2
0.3
0.4
0.5
0.6
0.7
standard deviation (F1)
op
tim
al v
olu
me
frac
tio
n
FORM-based CRBTO
SORM-based CRBTO
10 20 30 40 50 600.022
0.024
0.026
0.028
0.030
standard deviation (F1)
com
po
nen
t p
rob
abil
ity
MCS on FORM-based
MCS on SORM-based
constraint on P1(C
1)
10 20 30 40 50 600.022
0.024
0.026
0.028
0.030
standard deviation (F1)
com
po
nen
t p
rob
abil
ity
MCS on FORM-based
MCS on SORM-based
constraint on P2(C
2)
10 20 30 40 50 600.2
0.3
0.4
0.5
standard deviation (F1)
vo
lum
e fr
acti
on
FORM-based SRBTO
SORM-based SRBTO
10 20 30 40 50 600.044
0.046
0.048
0.050
0.052
standard deviation (F1)
syst
em p
rob
abil
ity
MCS on FORM-based
MCS on SORM-based
constraint on Psys
Component RBTO
System RBTO
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Nguyen, Song, and Paulino, (submitted), JSMO
48
SRBTO of a Building Core
Objective: minimize volume V( )
Limit-states:
Random loads: F ~ (P1,P2,P3,q1 ,q2,q3)
Load cases:
0( , ) ( , )i i i i ig C Cρ F ρ F
( , )i i iP qF5L
LL
P
L
L
P
q/2
q
P2,q2 Load
case 2
P2,q2 P2,q2
P2,q2
4 symmetry axes
P1,q1 P1,q1
P1,q1P1,q1
Load
case 1
P3,q3
Load
case 3
P3,q3 P3,q3
P3,q3
L/12L/12 10L/12
Load
Cases
P q (at top)
mean c.o.v mean c.o.v
Case 1 70.71 0.30 2.82 0.15 250
Case 2 50.00 0.15 2.00 0.30 125
Case 3 50.00 0.20 2.00 0.15 125
t
iC
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
49
Optimal Topologies of the Building Core
DTO
volfrac = 21.93%
SRBTO
volfrac =28.15%
(Psys=0.05)
0 0.2 0.4 0.6 0.8 10.200
0.225
0.250
0.275
0.300
0.325
target system failure probability
opti
mal
volu
me
frac
tion
SRBTO (same
=0.50, diff
=0.25)
DTO
SRBTO
volfrac =22.25%
(Psys=0.85)
volfrac v.s Psys
P1 P2 P3 Psys
same = 0.5
diff = 0.25
SRBTO 0.02731 0.02088 0.00539 0.05000
MCS 0.02747 0.02101 0.00542 0.05023
same = 0.5
diff = 0.25
SRBTO 0.26940 0.25973 0.20818 0.50000
MCS 0.26977 0.26006 0.20800 0.50008
same = 0.9
diff = 0.45
SRBTO 0.02812 0.02227 0.00625 0.05000
MCS 0.02816 0.02242 0.00638 0.05017
Probabilities: SRBTO/MSR v.s MCS
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
Nguyen, Song, and Paulino, (submitted), JSMO
50
pattern repetition
pattern symmetry
Building Core with Pattern Repetition
m=3 m=12m=6 m=10
DTO
1 3 6 10 120.20
0.25
0.30
0.35
0.40
0.45
number of pattern repetitions, m
opti
mal
volu
me
frac
tion
DTO
SRBTO (same
=0.50, diff
=0.25)
SRBTO (same
=0.90, diff
=0.45)
SRBTO
LL
http://www.som.com
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
m
51
MTOP & iMTOP:
Use three distinct displacement, density, and design variable fields
Improve efficiency, apply to large-scale problems
Adaptive MTOP:
Use MTOP and iMTOP elements where and when needed
Reduce the number of density elements and design variables
SRBDO/MSR:
Apply to general system including link-set, cut-set
Address dependence between limit-states, provide sensitivity
SRBTO
Propose accurate single-loop SORM-based CRBTO & SRBTO approaches
Include pattern repetition constraints
Summary & Conclusions
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
52
Optimal locations of design variables in MTOP
MTOP approach in nonlinear and stress-based problems
MTOP using Krylov subspace methods and recycling
SRBDO with multi-scale MSR approach
SRBDO with mixed continuous-discrete random variables
Future Research Topics
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
53
Nguyen, T. H., Paulino, G. H., Song, J., Le, C. H., (2010). "A computational paradigm for
multiresolution topology optimization (MTOP)." Structural and Multidisciplinary Optimization
41(4): 525-539.
Nguyen, T. H., Song, J., Paulino, G. H., (2010). "Single-loop system reliability-based design
optimization using matrix-based system reliability method: theory and applications." Journal
of Mechanical Design 132(1): 011005.
Sutradhar, A., Paulino, G. H., Miller, M. J., Nguyen, T. H., (2010). “Topological optimization
for designing patient-specific large craniofacial segmental bone replacements.” Proceedings
of the National Academy of Sciences 107(30) 13222-13227.
Nguyen, T. H., Paulino, G. H., Song, J., Le, C. H., "Improving multiresolution topology
optimization via multiple discretizations." International Journal for Numerical Methods in
Engineering (submitted).
Nguyen, T. H., Song, J., Paulino, G. H., "Single-loop system reliability-based topology
optimization considering statistical dependence between limit states." Structural and
Multidisciplinary Optimization (submitted).
Contributions
Intro. MTOP iMTOP SRBDO SRBTO Conclusions
TOP RBDO Reviews MTOP Examples improving Examples Adaptive Examples MSR SRBDO//M Examples Existing Improved Examples Summary Future
54
Advisors:
Glaucio H. Paulino & Junho Song
Committee:
Glaucio H. Paulino, Junho Song, Jerome F. Hajjar, C. Armando
Duarte, Ravi C. Penmetsa, William F. Baker, Alessandro
Beghini, Alok Sutradhar
Financial Support:
Vietnam Education Foundation
National Science Foundation
Computational Mechanics Group, Structural System Reliability Group
and colleagues at UIUC
Special thanks to my family
Acknowledgements
55
Thank you for your attention !