Probabilistic Fatigue Assessment for Complex Engineering Structures with Time-dependent Surrogate Modelling

Abstract

The dynamic probabilistic analysis of complex engineering structures requires thousands of simulations with nonlinear characteristics and hyperparameters, indicating that unacceptable computational loads exist. In this case, the performance of model directly determines the efficiency and accuracy of complex structural dynamic probabilistic analysis. To improve the computational efficiency and accuracy of probabilistic fatigue life prediction for complex structure, we introduce a time-dependent particle swarm optimization (PSO)-based general regression neural network (GRNN) surrogate model (called as TD/PSO-GRNN) by integrating the proposed space-filling Latin hypercube sampling technique and PSO-GRNN regression function. The related theory and method of TD/PSO-GRNN model is first investigated in details. Then the probabilistic fatigue life prediction framework is illustrated in respect of the TD/PSO-GRNN surrogate model. Moreover, the probabilistic fatigue life prediction of an aircraft turbine blisk under multi-physics interaction is performed to validate the TD/PSO-GRNN model. We obtain the distributional characteristics, reliability degree and sensitivity degree of fatigue failure cycle, which are useful for the turbine blisk design. By comparing the direct simulation (FE/FV model), RSM, GRNN, PSO-GRNN and TD/PSO-GRNN, it is revealed that the TD/PSO-GRNN surrogate model is promising to perform the probabilistic fatigue life prediction of the turbine blisk with high computational efficiency and acceptable computational accuracy. The efforts of this study offer a useful insight for the probabilistic design optimization of complex structure and enrich mechanical reliability theory and methods.