关键词: 可靠性分析, Kriging模型, 学习函数, 极限状态曲面, 主动学习

Abstract:

For the problems of low computational accuracy and efficiency in the reliability analysis of complex automotive structures， an active learning Kriging reliability analysis method is proposed based on the limit state approximation and normalized response. Firstly， based on the probability density function under the assumption of Gaussian distribution and the Expected improvement criterion， the expectation of improved distance is derived when the degree of limit state surface （LSS） approximation improves， and the expected improvement index is proposed in combination with the current maximum distance. Then， considering the uncertainty of prediction， a learning function named expectation of redefined improvement （ERI） is developed based on the Kriging standard variance， which is used to adaptively increase the training samples. Next， a normalized response method is introduced to optimize the convergence threshold， which effectively alleviates the instability of dimensional convergence criteria. Finally， the effectiveness of the proposed method is verified through four application examples. The results show that the proposed method can achieve strong fitting of LSS with fewer function calls， with good applicability in multiple reliability analysis scenarios.

Key words: reliability analysis, Kriging model, learning function, limit state surface, active learning