Active learning analysis method for structural reliability of systems with high dimensional performance factors
In the structural reliability analysis of the system,the learning function in the active learning Kriging(AK)method does not cover comprehensive information and the termination criterion is too conservative,resulting in too many points added in the solution process of the high-dimensional factor system.To solve the problems of high cost and low efficiency,an active learning analysis method for the structural reliability of systems with high-dimensional performance factors is proposed.First,the Kriging model is constructed based on the initial samples,and a new learning function is proposed,which considers the uncertainty measured near the limit state surface,the variance,and the probability density of the candidate points themselves,making the added learning points more representative and updating the model.Then,the maximum relative error of the predicted value is used as the termination criterion,and the failure probability of the system is estimated.Finally,based on the verification of three numerical function examples,the proposed method is applied in an instability problem of the connecting rod in an 8-dimensional crank slider mechanical structure.The experimental results show that the proposed method reduces the number of addition points while ensuring the prediction accuracy,and can achieve accurate and efficient reliability analysis compared with the typical learning functions.
reliability analysisactive learning Kriginglearning functiontermination criterionfailure probability
NETL
NSTL
万方数据
