Research On Non-probabilistic Reliability Analysis Method Of Structure Based On Kriging Surrogate Model

Structural reliability analysis has attracted the interest of scholars from academia and industry fields.Compared with the several existing probabilistic reliability analysis methods,the non-probabilistic reliability analysis method has gained much attention for mechanical liability analysis,especially for poor information and small sample size data.In practical engineering,the performance functions of structural components or systems are often implicit and high nonlinear,and require much time to perform finite element simulation for non-probabilistic structural reliability.Thus,performing traditional nonprobabilistic reliability analysis is difficult.To improve the computational efficiency,surrogate model can be considered.This study focuses on structural non-probabilistic reliability analysis methods based on Kriging models.The thesis is organized as follows:(1)Non-probabilistic reliability analysis method for structures with single failure mode based on Kriging model.Existing non-probabilistic reliability analysis methods mostly assumed independence among interval variables,but the correlations among variables are commonly existing in engineering.Several existing methods can be effectively applied for structure with explicit performance functions,but are difficult for the implicit performance functions.Therefore,this paper utilizes the non-probabilistic convex set model to consider the correlation among interval variables,combined with the parallel strategy and the newly active learning function,a non-probabilistic reliability analysis method is proposed for structures with a single failure mode based on Kriging surrogate model.In addition,multiple samples can be selected in the refining process of Kriging model,the parallel strategy is also adopted to improve the computing efficiency.(2)Non-probabilistic reliability analysis method for structural systems with multiple failure modes based on Kriging model.Several traditional non-probabilistic reliability analysis methods focus on the calculation of non-probabilistic reliability index with single failure mode,then the reliability index of the system can be obtained by the logic operation.However,all the failure modes are treated as one combined failure mode in these methods,which require a lot of computing resources and time costs.To address the issue,a non-probabilistic reliability analysis method for structural system with multiple failure modes based on Kriging model is proposed,and a learning function for selecting the critical failure mode of system is proposed,which can reduce the useless evaluation of failure modes that contribute less to the non-probabilistic reliability index.Subsequently,the computational efficiency is greatly improved.(3)Structural non-probabilistic reliability sensitivity analysis method based on Kriging model under considering correlation.With the implicit performance function,the reliability sensitivity analysis for implicit performance functions is computationally expensive.In addition,most of probabilistic reliability sensitivity analysis methods based on partial derivation or finite difference methods,so the global sensitivity indicators are not accurate.Furthermore,existing non-probabilistic reliability sensitivity analysis methods do not consider the correlations among interval variables.To address the issures,the traditional sensitivity analysis method is further improved by considering the correlation among variables and establishing a structural non-probabilistic reliability sensitivity analysis method based on Kriging model,the effectiveness of the proposed method is verified by two examples.