A deep learning approach for low-cycle fatigue life prediction under thermal–mechanical loading based on a novel neural network model

Abstract

In this study, a novel neural network model is designed to predict the low-cycle fatigue (LCF) life under thermal–mechanical loading. The model is composed of three sub-networks: (1) a Convolutional Neural Network (CNN), (2) a Transformer, and (3) a Fully Connected Neural Network (FCNN). These three sub-networks function as the Backbone, Neck, and Head of the model, respectively. The model, named ConTrans (an abbreviation for Convolutional Neural Network-Transformer), utilizes binarized hysteresis images as input. The predictive capability of the ConTrans model for LCF life has been validated using experimental data from four different materials. The results demonstrate that almost all predictions fall within the 2-factor band, confirming the model’s accuracy and effectiveness.