Professional Documents
Culture Documents
Jian Tu, K.K. Choi, Young Ho Park, and Byengdong Youn Center for Computer Aided Design College of Engineering The University of Iowa Iowa City, IA 52242
arc
MOTIVATION
z Due to increasing global competitive market, engineering designs
z z
z z
are pushed to the limit of the design constraint boundaries using design optimization, leaving very little or no room for tolerances in modeling and simulation uncertainties and/or manufacturing imperfections. Optimum designs obtained without consideration of uncertainty could lead unreliable or even catastrophic designs. The Reliability-Based Design Optimization (RBDO) methodology will provide not only improved designs but also a confidence range of the simulation-based optimum designs -- Robust Designs. The fatigue life is very much sensitive to uncertainties in material properties, empirical fatigue models, and external loads. The proposed method is applicable to general RBDO problems.
arc
MOTIVATION (cont.)
In ARC Phase I research, a RBDO method was developed using the conventional Reliability Index Approach (RIA) and successfully applied to a shape design of road arm fatigue of an M1A1 tank. However, the required computational time was extremely large -- optimization of optimization(s).
V olum e (in 3 ) Fatigu e Life (H rs) Initial D esign 51 5.1 21 89 O ptim al D esign 52 2.1 (+1.4% ) 69 623 (+30 80% )
Description Volume Life at node 1216 Life at node 926 Life at node 1544 Life at node 1519 Life at node 1433
Pf = () Changes 2 RBDO Iterations 447.691 in3 2.5% 0.532% 0.056 0.992% 2.2 0.998% 2.2 0.721% 0.11 0.018% 0.005
arc
RBDO PROBLEM
Design variables: mean d = [1, 2, , n]T of non-normally distributed random parameter X = [X1, X2, , Xn]T min s.t. cost f(d) P(Gi(X)0) Pfi, i=1-m dL d dU
Gi (X) : performance function Pfi = (ti): prescribed failure probability ti: target safety reliability index for Gi (X) (): monotonically increasing Cumulative Distribution Function (CDF)
arc
RBDO METHODOLOGIES
z Methods Requiring Probability and Statistical Analyses
Probabilistic Feasibility Formulation -- Distributional input parameters Distributional output responses Moment matching Formulation -- Simplistic approach to reduce the computational cost Worst Case Analysis Corner Space Evaluation Variation Patterns Formulation
Xiaoping Du, U of Illinois at Chicago, ASME, DETC99/DAC-9565. Concluded that the probabilistic feasibility formulation such as RIA is the best method.
arc
FG (g) = P(G(X) g) =
x iL x i x iU
FG(0) (t)
arc
Probabilistic Constraint can be described in two forms as: G(0) = 1(FG(0)) t: Reliability Index Approach (RIA) g(t) = FG1((t)) 0: Performance Measure Approach (PMA)
arc
Automotive Research Center
Center for Computer Aided Design
G(x) 0
G( x ) g *
These integrations are difficult to evaluate. FORM : transform random system vector X to an independent, standardized normal vector u: ui = 1(FX(xi))
G ( g) FORM (g) = u * g or g( G ) g( FORM ) = G( u * )
u * : Most Probable Point (MPP) for a given g -- RIA g u * : Most Probable Point (MPP) for a given -- PMA
arc
minimize ||u|| subject to G(u) = g z Inverse Reliability Analysis minimize G(u) subject to ||u|| =
arc
T G k ( u *=0 ) g d u G k (u *=0 ) g
(d d k ) t
arc
RBDO EXAMPLE
Random variable X = [X1, X2]T, Xi~uniform [i1, i+1], i=1,2 Design variable d = [1, 2]T RBDO Problem: minimize Cost(d) = 1 + 2 subject to P(Ga(X) = X1 + 2X2 10 0) 15.87% = (1) P(Gb(X) = 2X1 + X2 10 0) 2.275% = (2) Transform to u: X1 = 1 1 + 2(u1) and X2 = 2 1 + 2(u2) Ga(u) = 2(u1) + 4(u2) + (1 + 22 13) Gb(u) = 4(u1) + 2(u2) + (21 + 2 13): both are nonlinear
arc
arc
Using solutions of two inverse reliability analyses, RBDO becomes Minimize Cost(d) = 1 + 2
* Subject to ga (d) = 1 + 2 2 11.611 0 * g b (d) = 2 1 + 2 12.552 0 : linear constraints
arc
1 4 1
z It is easier to solve inverse reliability analysis than reliability analysis. z PMA yields linear probabilistic constraints if the performance
functions of non-normally distributed random parameter X are linear, whereas RIA converts them to nonlinear probabilistic constraints.
z If the the current design is feasible
with a large safety margin, reliability analysis may not converge for RIA.
arc
z (s, 0): RIA and (t, g*): PMA z Required sample size L=10/(a) for Monte Carlo method
increases exponentially as |a| increases z Computational effort for RIA is less if s< t (infeasible) and PMA is less if s> t (feasible)
arc
Automotive Research Center
Center for Computer Aided Design
d a ga t , for given ga = g * RIA if = 0 Adaptive RIA: G (0) = a dg dg g( t ) = ga + a ( t a ) 0, for given Adaptive PMA: d G
a = (1 ) s + t PMA if = 1
arc
If s>0, then choose a= s and use RIA (MPP u * is closest to the g=0 origin) If s0, then choose a= 0 and select the origin as MPP * ( u = 0 = 0 )
arc
z z
probabilistic constraint. Significant differences of RIA and PMA in solving RBDO is shown for a non-normally distributed system. PMA provides better convergence. General probabilistic constraint for a linear performance function of the non-normally distributed random system parameters yields a linear constraint in the proposed PMA, but it becomes a nonlinear constraint in the conventional RIA. Adaptive approach that considers both the convergence rate and computational effort will be developed. Design sensitivities of a probabilistic constraint of PMA and RIA are different since they are approximate sensitivities. In the unified system, the exact sensitivities can be defined and computed. This exact sensitivities will be derived to provide very efficient RBDO.
arc