The mechanistic model of a dynamic system is often so complex that it is not conducive to probability-based design optimization. This is so because the common method to evaluate probability is the Monte Carlo method which requires thousands of lifetime simulations to provide probability distributions. This paper presents a methodology that (1) replaces the implicit mechanistic model with a simple explicit model, and (2), transforms the dynamic, probabilistic, problem into a time invariant probability problem. Probabilities may be evaluated by any convenient method, although the first-order reliability method is particularly attractive because of its speed and accuracy. A part of the methodology invokes design of computer experiments and approximating functions. Training sets of the design variables are selected, a few computer experiments are run to produce a matrix of corresponding responses at discrete times, and then the matrix is replaced with a vector of so-called metamodels. Responses at an arbitrary design set and at any time are easily calculated and then used to formulate common, time-invariant, performance measures. Design variables are treated as random variables and limit-state functions are formed in standard normal probability space. Probability-based design is now straightforward and optimization determines the best set of distribution parameters. Systems reliability methods may be invoked for multiple competing performance measures. Further, singular value decomposition may be used to reduce greatly the number of metamodels needed by transforming the response matrix into two smaller matrices: One containing the design variable-specific information and the other the time-specific information. An error analysis is presented. A case study of a servo-control mechanism shows the new methodology provides controllable accuracy and a substantial time reduction when compared to the traditional mechanistic model with Monte Carlo sampling.
[1]
Achille Messac,et al.
Metamodeling using extended radial basis functions: a comparative approach
,
2006,
Engineering with Computers.
[2]
Margaret J. Robertson,et al.
Design and Analysis of Experiments
,
2006,
Handbook of statistics.
[3]
Bruno Sudret,et al.
The PHI2 method: a way to compute time-variant reliability
,
2004,
Reliab. Eng. Syst. Saf..
[4]
Sanjay B. Joshi,et al.
Metamodeling: Radial basis functions, versus polynomials
,
2002,
Eur. J. Oper. Res..
[5]
Jian Deng,et al.
Structural reliability analysis for implicit performance function using radial basis function network
,
2006
.
[6]
Gordon J. Savage,et al.
Optimal Mean and Tolerance Allocation Using Conformance-based Design
,
2006,
Qual. Reliab. Eng. Int..
[7]
L. Schueremans,et al.
Benefit of splines and neural networks in simulation based structural reliability analysis
,
2005
.
[8]
Zissimos P. Mourelatos,et al.
Reliability-based design optimisation of vehicle drivetrain dynamic performance
,
2008
.
[9]
Farrokh Mistree,et al.
Kriging Models for Global Approximation in Simulation-Based Multidisciplinary Design Optimization
,
2001
.
[10]
T. Simpson,et al.
Use of Kriging Models to Approximate Deterministic Computer Models
,
2005
.
[11]
T. Simpson,et al.
Comparative studies of metamodelling techniques under multiple modelling criteria
,
2001
.