A Two-Scale Model of Fretting Fatigue Crack Initiation Life Based on Long Short-Term Memory Networks Improved by Genetic Algorithm

Abstract

Fretting fatigue is a common engineering failure phenomenon, often resulting in a shorter lifespan compared to plain fatigue. To consider the interaction between different scales, this study proposes a fully coupled two-scale model based on continuum damage mechanics (CDM) and the crystal plastic finite element method (CPFEM) for the fretting fatigue crack initiation. Furthermore, the life data series are generated by employing feature engineering and long short-term memory (LSTM) networks optimized with a genetic algorithm, ensuring the minimization of redundancy. Additionally, the genetic algorithm, enhanced by the Markov chain Monte Carlo method, was used to optimize the hyperparameters of the LSTM network. Simulation results indicate that the two-scale model offers improved accuracy in predicting crack initiation life and provides physical information of crack initiation from different scales simultaneously.