Data-driven algorithm for real-time fatigue life prediction of structures with stochastic parameters

Abstract Fatigue crack growth analysis using extended finite element method (XFEM) is an efficient way to predict the residual life of structures; however, when the structure parameters vary stochastically, it will be very hard to make accurate predictions. To bridge this research gap, this work proposed a data-driven learning algorithm to improve the prediction capacity of fatigue life by considering stochastic parameters of structures. In this new algorithm, the XFEM was firstly employed to generate a large amount of datasets that pair the structural responses with remaining fatigue life. Then, the back propagation neural network (BPNN) was employed to construct a fatigue life prediction model based on the XFEM datasets. Real-time prediction for the structural fatigue life was achieved using the constructed BPNN model without knowing the exact distribution functions of stochastic parameters. Several numerical examples were performed to evaluate the performance of the proposed algorithm. The analysis results demonstrate that the proposed data-driven algorithm can accurately predict the fatigue life of the structures with stochastic parameters.