Neural Network Based Low Cycle Fatigue Life Prediction

The existence of low cycle fatigue has always been a key concern in engineering practice.This paper proposes a new method for predicting low cycle fatigue life,based on the model of neural network and local stress strain method,and explores the research problem of the prediction method for low cycle fatigue life.The model addresses the difficulty of calculating theoretical stress concentration coefficients and uses artificial neural network methods to predict the theoretical stress concentration coefficients of commonly used standard parts,starting with an analysis of the factors influencing the stress concentration to derive the input characteristics and processing the theoretical stress concentration coefficients into a sequence of data that varies proportionally with size.Different neural network algorithms such as BP neural network,recurrent neural network and long and short term memory neural network are selected,and different hyperparameters and optimisation methods are analysed and compared in simulations.It was concluded that the prediction of the theoretical stress concentration coefficients for different standard parts under various loading conditions into serial data was feasible and that the prediction of the theoretical stress concentration coefficients by the long and short-term memory neural network could achieve good results.On this basis,the fatigue life of the component is predicted by combining the local stress strain method of low cycle fatigue life analysis.To further solve the problems of cumbersome data query and a lot of time-consuming manual calculation in traditional fatigue life.This paper relies on modern computer technology,using Python to build a neural network,Matlab to analyze the background of the local stress strain method,My SQL to build a database of fatigue properties of metal materials,and Py Qt to design the user interface,and finally composed of a fatigue life prediction platform with certain accuracy and ease of use.After the actual experimental example to support,the low circumference fatigue life prediction system based on recurrent neural network can predict the fatigue damage life within a reasonable expectation range for the fatigue damage caused by the excessive local stress of a specific standard part that can be simplified.The result is a significant reduction in unnecessary labour and time consuming fatigue life estimation and prediction.