Research On Structural Reliability Analysis Method Based On Adaptive Kriging Surrogate Model

With the increasing size and complexity of equipment or structural systems,reliability analysis is being increasingly complex.Reliability analysis of many engineering problems depends on the calculation of implicit performance functions.Generally,the implicit performance function is a highly nonlinear function with high dimensions in engineering,causing reliability analysis needs to use a finite element model and other time-consuming calculations.Therefore,to seek a better balance between accuracy and efficiency,reliability analysis methods based on adaptive Kriging modelare proposed,whereas the accuracy can be guaranteed.The proposed methods can reduce the number of calls to performance functions or finite element models,which will improve computational efficiency.The research results of this thsis can be summarized as follows:(1)A hybrid reliability analysis method based on adaptive Kriging for single failure mode is proposed.Most existing reliability analysis methods based on Kriging model only consider alatory ucnertainty,little attention fousing on mixed uncertainties.Moreover,most existing methods only focus on whether the sign of the responses are correctly,and they ignore the accuracy of the model itself.Therefore,the above two points are considered comprehensively to construct the learning function,which can be used as the guideline to select new training sample points for updating the Kriging model.(2)A reliability analysis method for complex structual systems based on adaptive Kriging model is proposed.Existing Kriging-based system reliability analysis methods,such as AK-SYS and ALK-TCR,are only suitable for simple series or parallel and seriesparallel hybrid systems.To address to problem,the paper proposes a Kriging-based reliability analysis method for complex systems,such as bridge-network structure systems.The proposed method combines Kriging model,system structure-function,the minimal path(cut)set,and learning function U to address the limitations of the existing methods.Besides,the magnitude of responses between components has no effect on the number of calls to performance function or finite element model during building Kriging model.Moreover,the proposed method is also easy to code for complex series-parallel systems.(3)A reliability analysis method of complex system based on adaptive dependent Kriging model is proposed.Note that the existing methods only used the mean value and variance information of Kriging predictions to construct the learning function.It means that these methods ignored the dependence information among samples.Thus,to further improve the efficiency,all information from Kriging,including the mean,the variance,and covariance,are used to construct the learning function,which can reduce the calls to performance functions or finite element models.Furthermore,to reduce overall computational time,a parallel learning strategy with adding multiple sample points in a single iteration process is developed to update the model.