A Novel Single-Loop Kriging Modeling-Based Method for Time-Variant Reliability Analysis

Abstract

Time-variant reliability analysis is crucial for enhancing the safety and reliability of products throughout their service life. However, the performance analysis of a complicated mechanical structure typically relies on a time-consuming simulation process, and performing time-variant reliability analysis requires a large number of performance function evaluations. To reduce the number of evaluations and improve computational efficiency, a novel Kriging modeling-based method is developed for time-variant reliability analysis. Firstly, the stochastic process can be converted into a set of random variables through spectral decomposition, which allows the well-trained Kriging model to perform reliability analysis using the extreme value method. Then, an improved adaptive sampling strategy is derived based on the error of Kriging-based estimation. Meanwhile, the Kriging model is sequentially refined through the error-guided active sampling process. To construct an accurate Kriging model, an error-guided stopping criterion is introduced to terminate the adaptive sampling process. Finally, three case studies involving engineering problems are employed to check the capability and applicability of the proposed method. The comparison results indicate the proposed strategy achieves a satisfactory balance between the function calls and computational accuracy.

Keywords

• Spectral decomposition
• Kriging model
• time-variant reliability
• adaptive sampling strategy