The two major components of limit state-based reliability analysis are the search for the MPP and the subsequent estimation of the probability of failure. The proposed method first searches for the MPP using adaptive response surfaces with trust region updates. During the MPP search, it finds the MPP successfully with adequate efficiency for a numerical example and a vehicle side impact problem both of which have highly non-linear limit state functions. Following the MPP search is the reliability estimate, in which different methods can be applied. It is suggested that FORM and SORM be applied first using the MPP information and the adaptive response surface, and if high non-linearity is identified, adaptive importance sampling be used to improve the reliability estimate. The adaptive importance sampling can be started by using only the MPP information or taking advantage of the multiple adaptive response surfaces created during the MPP search. By applying the reliability methods mentioned above, it is observed that the reliability estimation is improved progressively for both examples. The proposed method is general and can be, therefore, applied to a wide variety of reliability estimation problems with non-linear limit state functions. For problems with a large number of random variables, the OSLH design is used as the DOE technique, allowing efficient and accurate local response surface approximations. A potential improvement of the proposed method is to investigate the effectiveness of other response surface construction methods, such as kriging and moving least squares (Lancaster and Salkauskas, 1986), in constructing the local approximations of the limit state.
Adaptive trust regions and response surfaces for reliability analysis
International Journal of Materials and Product Technology ; 25 , 1-3 ; 84-98
2006
15 Seiten, 3 Bilder, 5 Tabellen, 20 Quellen
Aufsatz (Zeitschrift)
Englisch
Reliability Analysis with Adaptive Response Surfaces
AIAA | 2003
|Reliability Analysis with Adaptive Response Surfaces
British Library Conference Proceedings | 2003
|