Reliability analysis for highly non-linear and complex model using ANN-MCM simulation

Abstract

To predict reliability for highly non-linear and complex model efficiently and accurately, an artificial neural network (ANN) based Monte Carlo method (MCM) is proposed. In this method, ANN trained by backpropagation (BP) algorithm is developed to establish the relationship between inputs and outputs. The trained ANN model is then connected to MCM to predict reliability and reliability sensitivity of structure. To examine the accuracy and efficiency of the proposed method, results obtained by polynomial-FOSM (first-order second moment) method and ANN-AFOSM (advanced first-order second moment) method are provided. Compared with the traditional Monte Carlo method, the time consumed by ANN-MCM, polynomial-FOSM and ANN-AFOSM methods only accounts for 0.14% of the traditional MCM. The reliability analysis and reliability sensitivity analysis results of ANN-MCM method are very close to traditional MCM, polynomial-FOSM method yields large errors, ANN-AFOSM is in between. It shows that the proposed method is of high accuracy and efficiency.