A deep neural network approach combined with Findley parameter to predict fretting fatigue crack initiation lifetime

Abstract

In the aviation field, the crack initiation stage could be the dominant one of the whole damage failure process of components subjecting to fretting fatigue. Thus, getting an accurate prediction of the crack initiation lifetime become an important topic in research of fretting fatigue problems. In this work, a new prediction method for fretting fatigue crack initiation lifetime is proposed, which is based on Deep Neural Network (DNN), Findley Parameter (FP) and Finite Element Method (FEM). Input features are shear stress amplitude and maximum normal stress in the critical plane, and crack initiation lifetime is set as the output target. Dropout and bootstrapping are considered to improve the performance of DNN. Through comparative analysis, it is shown that this new proposed approach can effectively predict the fretting fatigue crack initiation lifetime. The prediction accuracy and stability are greatly improved compared with the theoretical formulation based on FP.