The design of dynamic systems in terms of dependability and sustainability is a computationally intensive task - especially when probabilities are needed. This paper presents a methodology that greatly reduces the design time with a controlled loss of accuracy. The proposed approach combines design of computer experiments, a judicious application of linear metamodels and an explicit evaluation of probabilities. This approach is much faster than traditional Monte Carlo simulation. A further reduction in time can be accomplished through the application of singular value decomposition. The application of the methodology for two popular forms of metamodels (i.e., response surface methods via Least-squares fit and Kriging via maximum likelihood estimation) is presented. Parameter design is provided through robust design principles and constrained optimization. A detailed case study of a system with multiple performance measures and multiple design variables shows the efficacy and practicality of the methodology.
|Original language||English (US)|
|Number of pages||14|
|Journal||International Journal of Performability Engineering|
|State||Published - Mar 1 2010|
- Limit-state functions
- Robust design
- Singular value decomposition