Professional Documents
Culture Documents
Reliability Based Design Optimization For Multidisciplinary Systems Using Response Surfaces
Reliability Based Design Optimization For Multidisciplinary Systems Using Response Surfaces
AIAA-2002-1755
Abstract Nomenclature
x Design Variables
This paper investigates reliability based design op- z Random Variables
timization (RBDO) using response surface approxima- u Independent Standard Normal Random
tions (RSA) 1,2 for multidisciplinary design optimization Variables
(MDO). In RBDO the constraints are variational (reli-
yd Deterministic State Variables
ability based) since the design variables and the system
yr Random State Variables
parameters can have variation and can be subjected to un-
px Traditional Optimization Parameters
certainties 18 . For these problems the objective is to min-
gR Reliability Based Constraints
imize a cost function while satisfying reliability based
pz Reliability Based Constraint Parameters
constraints. This class of problems is referred to as relia-
gD Deterministic Constraints
bility based multidisciplinary design optimization (RB-
xl , xu Lower and Upper bounds on design space
MDO) problems 5 . The reliability constraints, which
fZ (z) Joint Probability Density Function of the
can be formulated in terms of the reliability indices or
random variables
in terms of the probability of failure, themselves repre-
Φ(u) Standard Normal Cumulative Distribution
sent an optimization problem and can be very expensive
Function (CDF)
to evaluate for large scale multidisciplinary problems.
g(z) Actual Limit State Function;
Response surface approximations of the constraints are
Safe : g(z) > 0, Fail : g(z) < 0
used in estimating the reliability indices or probability of
βreqd Required Value of Reliability Index
failure when solving an approximate optimization prob-
g̃(u) Approximation of the Limit State in
lem using FORM. In this research RSAs are integrated
Standard Normal Space
within RBDO to significantly reduce the computational
cost of traditional RBDO. The proposed methodology is
compared to traditional RBDO in application to multidis- Introduction
ciplinary test problems, and the computational savings
and benefits are discussed.
In deterministic multidisciplinary design optimiza-
tion, the designs are often driven to the limit of the design
constraints (active constraints at the optimum). These
designs may be subject to failure due to inherent uncer-
tainties that exist both in the mathematical modeling and
∗ Graduate Research Assistant, Student Member AIAA simulation tools and the variability in physical quanti-
† Associate Professor, Associate Fellow AIAA ties of the actual artifact. Optimized designs determined
Copyright 2001 by John E. Renaud. Published by the American In-
without due consideration of variability can be unreliable
stitute of Aeronautics and Astronautics, Inc. with permission. leading to catastrophic failure. However, the existence of
physical uncertainty and model uncertainty requires a re-
1
American Institute of Aeronautics and Astronautics
Copyright © 2002 by the author(s). Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
liability based design optimization (RBDO) to be taken sub ject to : gi (x, px , yd ) ≥ 0, i = 1..m (1)
into account. The uncertainties include variational un- x ≤x≤x
l u
certainty and simulation based uncertainty. Variational
uncertainty is mainly associated with the randomness of
physical quantities and can be easily modeled mathe- In today’s competitive marketplace, it is very important
matically by statistical means i.e., by using probability that the resulting designs are reliable. Optimized designs
and cumulative density functions. Model and simulation without considering the variability of design variables
uncertainties are difficult to characterize and have to be and parameters can be subjected to failure in service.
modeled using other means such as possibility theory or In RBDO, the constraints are reliability based and the
fuzzy sets etc 18 because of a lack of knowledge. In this objective function is performance based. The reliability
paper only variational uncertainty is treated within the based multidisciplinary optimization problem in terms of
RBDO framework. In future we plan to incorporate the RIA can be formulated as follows.
effects of other types of uncertainty in RBDO.
minimize : f (x, px , yd )
sub ject to : gRi = βi (z, pz , yr ) − βrequired ≥ 0, i = 1..l (2)
Literature Survey
i (x, px , y )
gD ≥ 0, i = l + 1, .., m
d
2
American Institute of Aeronautics and Astronautics
Starting Point based on
Engineering Knowledge Modified Barnes Problem
Deterministic Optimization
This is a purely analytical two-dimensional problem
and it was originally formulated by G.K. Barnes as part
of his Master’s Thesis 22 . We have chosen this as a
Use Deterministic Optima
for RBDO
test problem to illustrate the usefulness of the proposed
methodology. This is a highly nonlinear problem even
though it is just a two-dimensional problem. The prob-
Identify deterministic and Do RBDO with
random variables and exact constraints lem is stated as follows.
define reliability based
using the present
constraints
design point
No
Minimize :
converged yes
f (x, y) = a1 + a2 x1 + a3y4 x1 + a5 y24 + a6x2 +
Actual reliability FINAL
RBDO constraints satisfied DESIGN
Test Problems and Results Note the mapping that the variables and parameters have
to undergo as we move from the design space to the
The proposed methodology is implemented in appli- random variable space. In the design space the terms
cation to test problems. A small analytic problem and a zi , i = 1, .., 4 are the constant parameters. They are repre-
multidisciplinary structural design test problem are used. sented as px in the nomenclature. In the random variable
3
American Institute of Aeronautics and Astronautics
space the terms xi , i = 1, 2 are the constant parameters. 49
28
26
They are represented as pz in the nomenclature. In this 48
x2
24
problem all the states are deterministic (yd ). In general
x
47
22
the states can be random (yr ) too. The set of basic ran- 46
20
dom variables and their assumed distribution are shown 1 2 3 4 5 1 2 3 4 5
−20 0
in Table 1. All random variables have been assumed to
f = merit function
be statistically independent. −25
−0.5
gr1
−1
−30 −1.5
Z Distribution Mean Standard Deviation
1 2 3 4 5 1 2 3 4 5
z1 Normal 700 1
0 0.35
z2 Normal 1 0.3 −1 0.3
g2
gd
3
r
z3 Normal 5 1 −2 0.25
z4 Normal 25 0.3 −3
1 2 3 4 5 1 2 3 4 5
Table 1. Random Variables (Modified Barnes Problem) Figure 2. Convergence Plots for Modified Barnes Problem Us-
ing Actual Constraints
46
The deterministic optimum is chosen as the start-
1
x2
25
x
44
ing point for the RBDO. A performance comparison of 42
using RBDO with the actual constraints and using re- 1 2 3 4 5 6
20
1 2 3 4 5 6
sponse surface approximations of the constraints in stan- 0
−20 −0.5
the reliability index for each of the reliability based con-
gr1
−25 −1
straints is chosen as 3.0. The results are outlined below.
−30 −1.5
1 2 3 4 5 6 1 2 3 4 5 6
10
0.35
Deterministic RBDO with RBDO with
5 0.3
2
gd
3
gr
4
American Institute of Aeronautics and Astronautics
reliable design that is obtained from RBDO using the such large dimensionality is chosen to test the effective-
RSA of constraints does not violate the actual reliabil- ness of the suggested approach, to investigate the com-
ity based constraints. In the case where they are violated putational savings and to see whether the reliable optima
due to approximation error, an RBDO using the actual obtained using RSA is practical.
constraints, could be initiated at the reliable RSA optima The objective is to find the size and shape of the truss
xrsa . such that the weight (Wtot ) of the structure is a mini-
mum (low cost) and the loads (Pi ) it is capable of sus-
taining and the payload (Mi ) it carries are a maximum
HPLC Structure (high performance). This multi-objective problem can
be formulated in a single objective problem by defining
a cost-performance index (CPI) which includes each of
A high performance low cost (HPLC) structural de-
the objectives. The design is subject to minimum pay-
sign problem that was first introduced as a multidisci-
load and load requirements as well as yield stress and
plinary design optimization (MDO) test problem in Wu-
first natural frequency constraints. A total of 13 inequal-
jek et al15 is used for additional testing of the RBDO
ity constraints are defined for this problem. In standard
method. The system is illustrated in figure 4 (Tenbar
form the deterministic system optimization problem is:
Truss). The deterministic optimum design point is ob-
Minimize :
P1 P2 w2 w3
f (x) = CPI = w1Wtot + + (5)
∑ i ∑ Mi
P
L1 L2
M1 M2 Subject to :
A4 A9 (Mtot )min
g1 = 1.0 − ≥ 0.0 (6)
A3 ∑ Mi
A8 (Ptot )min
A5 A10 g2 = 1.0 − ≥ 0.0 (7)
A7 ∑ Pi
A2 ω1,min
g3 = 1.0 − ≥ 0.0 (8)
A6 ω1
A1
|σ1−10 |
M3 M4 g4−13 = 1.0 − ≥ 0.0 (9)
σyield
L3 (l) (u)
xi ≤ xi ≤ xi (10)
P3 P4
Figure 4. Schematic of HPLC Structure
where : w1 = .003, w2 = 106 , w3 = 3.5X106
(Mtot )min = 5000 lbs, (Ptot )min = 100, 000 lbs
ω1,min = 2.0 Hz, σyield = 14, 000 psi
tained first. It is then used as the starting point for RBDO
in order to obtain the reliable design point.
The design variables in this problem come from three The coefficients wi in the objective function are intro-
different disciplines. The configuration of the structure is duced to scale the separate components so that no one
varied in order to explore different topologies. Thus, the component dominates the others in driving the optimiza-
length of the rectangular first bay (L1 ) and the top and tion. The yield stress of 14,000 psi is based on the choice
bottom lengths of the outer bay (L2 , L3 ) are selected as of aluminum as the material for the structure. The loads
geometric design variables. The masses placed on all of (Pi ) applied to the structure are defined to be a function
the unconstrained nodes (M1 − M4 ) are structural design of the lengths of the bays (Li ) and the payload masses
variables representing the system payload. The areas of (Mi ) placed on the structure as shown in Equation 11.
the truss members (A1 − A10 ) make up the final category
of design variables since sizing is one of the main de- bk i 4 i
3
Lk Mj dj
∑ ak + ∑ cj
sign considerations (see figure 4). In all seventeen design
Pi = i i
(11)
variables are defined for this problem. A problem with k=1 Lre f j=1 Mre f
5
American Institute of Aeronautics and Astronautics
This is similar to an aeroelastic structure in which the RV Description Mean Coeff. of Var.
loads incurred are dependent on the size and shape of the
structure. The coefficients (a, b, c and d) in equation 11 ρ Density 0.1 0.01
7
are chosen so as to apply greater emphasis to the effect E Young’s Modulus 10 0.05
that certain lengths or masses have on the given loads. σyield Yield Stress 14000 0.05
They are listed in the Appendix.
Table 4. Random Variables (HPLC Structure)
The deterministic optimization was run with the ini- are found, the reliability constraints are evaluated using
tial design of Li = 360 in., Mi = 1500 lbs and Ai = 12 Equation (2). The results are listed in Table 5
in2 (for all i). The results are tabulated in Table 3. Note MATLAB’s optimizer is used for both upper and
that the constraints g3 , g4 , g5 and g6 are active at the solu- lower level optimization. The reliable optima is higher
tion. To carry out the RBDO, the deterministic optimum than the deterministic optima. It is observed that the
design point is chosen as the starting point. value of the payloads have gone down and that of the
areas have gone up. This is expected for a more reliable
structure in which the areas of elements are larger and the
Reliability Based Design Optimization payloads are reduced so that the structure is subjected to
less stress and hence is more reliable. The reliable op-
In reliability based design optimization, it is imprac- tima obtained using RSA is better than the one obtained
tical to carry out the sub level optimization (Equation 3) using actual RBDO. This design is used to evaluate the
for all the constraints. This is illustrated in the two cases actual RBDO constraints. They are listed in the last col-
that follow. We actually need to consider only the con- umn of Table 5. We observe that the actual constraints
straints that are active or nearly active and evaluate the are satisfied. Note the significant savings is computa-
reliability index for only those constraints. In addition tional time. The time taken in RBDO using RSA of con-
significant time is saved if the deterministic optimum de- straints is one-eighth of the time required for the actual
sign is used as starting point for RBDO. RBDO.
Test Case : 1
Test Case : 2
Material properties such as density, young’s modulus
and yield stress (parameters in actual design space Px )
are chosen to have randomness. The set of basic random The random variables are same as in Case 1. Now all
variables and their mean and coefficient of variation are the constraints that actually depend on the random pa-
shown in Table 4. The coefficient of variation have been rameters are chosen as reliability based constraints. So
chosen arbitrarily for this test case. All random variables there are 11 reliability based constraints in this test case
have been assumed to be lognormally distributed and are i.e., the first frequency constraint and all the stress con-
statistically independent. The second order response sur- straints. The results are listed in Table 6.
face is constructed in the standard normal random vari- We observe that the reliable optima obtained using
able space at each step of the upper level optimization actual RBDO and using RSA in RBDO has a higher
iteration. In the reliability space the design variables x1 merit function value than that in case 1. This is expected
to x17 are constant parameters pz . since there are more reliability based constraints in case
It is observed in the deterministic optimization that 2. The time taken in case 2 for actual RBDO is about
the constraints 3,4,5 and 6 are critical (active). So in this 3.3 times more than that in case 1. We do not observe
test case they are treated as the failure driven constraints. significant difference in time for RSA in RBDO in two
The reliability indices are calculated using Equation (3) cases because the system analysis gives the value for all
for the active constraints (see Table 3). The active con- the constraints at any given sampling point. The reliable
straints in this case are the first frequency constraint and design that is obtained from RBDO using RSA doesn’t
the first 3 stress constraints. Once the reliability indices violate the actual reliability based constraints.
6
American Institute of Aeronautics and Astronautics
Initial Design Optimal Design Constraints Optimal Design
f 39.93 11.5934
Design Variables
No of Iterations 99
Function Calls 1883
Summary and Conclusions as the HPLC structure where the design variables and
the states are part of three different disciplines (config-
uration, structures, dynamics). The results provided for
In this study a new methodology for the reliability the test problems give reliable optimas that do not vio-
based design optimization (RBDO) of multidisciplinary late the actual reliability based constraints. Future stud-
systems is proposed. The methodology makes use of ies will address other types of uncertainty such as the
response surface approximations (RSA) to significantly epistemic uncertainty in the reliability based design op-
reduce the computational cost of RBDO. The evalua- timization framework. Mathematical theories available
tion of reliability based constraints requires solution of for representation of uncertainty are, for example, evi-
a sublevel optimization for each constraint which makes dence(Dempster/Shafer) theory, possibility theory, fuzzy
implementation of RBDO impractical for multidisci- set theory and imprecise probability theory.
plinary systems. The use of response surface approxi-
mations within the sublevel optimization of RBDO can
significantly reduce the computational cost of perform-
ing RBDO. The proposed methodology (Figure 1) of
Acknowledgements
building response surface approximations of constraints
in standard normal space has been implemented in ap-
plication to two test problems and the results are pre- This multidisciplinary research effort is supported in
sented. Significant savings in computational time is ob- part by the following grants; NSF grants DMI98-12857
served in application to a multidisciplinary problem such and DM101-14975 and NASA grant NAG1-2240.
7
American Institute of Aeronautics and Astronautics
Starting with Reliable optima Reliable optima Constraints RBDO with RBDO with
Optimal Design (Actual) (RSA) Actual constraints (RSA)
Design Variables
8
American Institute of Aeronautics and Astronautics
Starting with Reliable optima Reliable optima Constraints RBDO with RBDO with
Optimal Design (Actual) (RSA) Actual constraints (RSA)
Design Variables
Probabilistic Mechanics and Structural Reliability, H., “Large Scale Multidisciplinary Optimization
PMC2000. of a Vehicle System in a Scalable, High Per-
8. Qu, X. S., Venkataraman, S., Haftka, R. T., formance Computing Environment”,Proceedings of
“Response Surface Options For Reliability-Based DETC’01, ASME Design Engineering Technical
Optimization Of Composite Laminate”, Proceed- Conference and Computers and Information in En-
ings of the 8th ASCE Speciality Conference on gineering Conference, 2001.
Probabilistic Mechanice and Structural Reliability, 12. Koch, P. N., Kodiyalam, S., “Variable Complexity
PMC2000. Structural Reliability Analysis For Efficient Relia-
9. Battaini, M., Breitung, K., Casciati, F., Faravelli, L., bility Based Design Optimization”, Proceedings of
“Active Control And Reliability Of a Structure Un- the 40th AIAA/ASME/ASCE/AHS/ASC Structures,
der Wind Excitation”, Journal of Wind Engineering Structural Dynamics, and Materials Conference &
and Industrial Aerodynamics, 1998. Exhibit, 1999.
10. Haldar, A., Mahadevan, S., Probability, Reliability 13. Ang, A. H. S., and Tang, W. H., Probability Con-
and Statistical Methods in Engineering Design, J. cepts in Engineering Planning and Design, Vol.1
Wiley & Sons, New York, 2000. and 2, J. Wiley & Sons, New York, 1984
11. Kodiyalam, S., Yang, R. J., Gu, L., Tho, C. 14. Rackwitz, R., “Reliability Analysis-Past, Present
9
American Institute of Aeronautics and Astronautics
and Future”, Proceedings of the 8th ASCE Spe- a3 3 = 20.0 b3 3 = 3.7 c3 3 = 50.0 d3 3 = 4.0
ciality Conference on Probabilistic Mechanics and c4 4 = 37.0 d4 4 = 2.9
Structural Reliability, PMC2000. Coefficients for P4
15. Wujek, B. A., Renaud, J. E., Batill, S. M., “A a1 4 = 17.0 b1 4 = 3.5 c1 4 = 25.0 d1 4 = 2.7
Concurrent Engineering Approach for Multidisci- a2 4 = 15.0 b2 4 = 3.0 c2 4 = 27.0 d2 4 = 2.7
plinary Design in a Distributed Computing Environ- a3 4 = 19.0 b3 4 = 3.8 c3 4 = 25.0 d3 4 = 2.7
ment”,Proceedings of the ICASE/LaRC Workhop c4 4 = 27.0 d4 4 = 2.7
on Multidisciplinary Design Optimization, 1995.
16. Wujek, B. A., Renaud, J. E., Batill, S. M., Brock-
man, J. B., “Concurrent Subspace Optimization us-
a1 75.196 a2 -3.8112 a3 0.12694
ing design variable sharing in a distributed com-
puting environment”, Design Engineering Techni- a4 -2.0567e-3 a5 1.0345e-5 a6 -6.8306
cal Conference, ASME 1995. a7 0.030234 a8 -1.28134e-3 a9 3.5256e-5
17. Du, X., Chen, W., “A most probable point based
a10 -2.266e-7 a11 0.25645 a12 -3.4604e-3
method for uncertainty analysis”, Design Engineer-
ing Technical Conferences, ASME 2000. a13 1.3514e-5 a14 -28.106 a15 -5.2375e-6
18. Oberkampf, W. L., Helton, J. C., “Mathemati- a16 -6.3e-8 a17 7.0e-10 a18 3.4054e-4
cal Representation of Uncertainty”, Proceedings of a19 -1.6638e-6 a20 -2.8673 a21 0.0005
the 42nd AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference & Table 7. Coefficients for the Modified Barnes problem
Exhibit, 2001.
19. Sundaresan, S., Ishii, K., Houser, D. R., “A Robust
Optimization Procedure with Variations on Design
Variables and Constraints”, Advances in Design Au-
tomation - Volume 1, ASME 1993.
20. Haftka, R., Gurdal, Z., and Kamat, M. P., Elements
of Structural Optimization, Kluwer Academic Pub-
lishers, Dordrecht, Netherlands, Second Edition,
1990.
21. Kiureghian, A. D., Liu, P. L., “Optimization Algo-
rithms For Structural Reliability”, Journal of Struc-
tural Safety, Vol. 9, pp. 161-177.
22. Barnes, G. K.,1967 M.S. Thesis, The University of
Texas, Austin, Texas.
Appendix
Load Coefficients
Coefficients for P1
a11 = 25.0 b1 1 = 4.0 c1 1 = 50.0 d1 1 = 4.0
a2 1 = 20.0 b2 1 = 3.7 c2 1 = 37.0 d2 1 = 2.9
a3 1 = 20.0 b3 1 = 3.7 c3 1 = 35.0 d3 1 = 2.9
c4 1 = 37.0 d4 1 = 2.9
Coefficients for P2
a1 2 = 17.0 b1 2 = 3.5 c1 2 = 25.0 d1 2 = 2.7
a2 2 = 19.0 b2 2 = 3.8 c2 2 = 27.0 d2 2 = 2.7
a3 2 = 15.0 b3 2 = 3.0 c3 2 = 25.0 d3 2 = 2.7
c4 2 = 27.0 d4 2 = 2.7
Coefficients for P3
a1 3 = 25.0 b1 3 = 4.0 c1 3 = 35.0 d1 3 = 2.9
a2 3 = 20.0 b2 3 = 3.7 c2 3 = 37.0 d2 3 = 2.9
10
American Institute of Aeronautics and Astronautics