On the generalization capability of artificial neural networks used to estimate fretting fatigue life

Abstract

This study assesses the generalization capacity of artificial neural networks (ANNs) for predicting fretting fatigue of mechanical contacts using different materials and geometries. These ANN utilize as inputs, material properties and stress quantities that have been physically related to the fatigue crack initiation mechanism under multiaxial loading. Initially trained and validated on aeronautical aluminum alloys data, one tests their generalization performance by applying them to fretting fatigue data for Ti‐6Al‐4V, ASTM A743 CA6NM steel, and Al 7050-T745, employing both cylindrical and spherical pads under various loading conditions, including out-of-phase loading. The ANNs adeptly predict fatigue lives across this extensive dataset, surpassing classical multiaxial fatigue criteria in accuracy. This underscores the effectiveness of ANN-based methodologies in diverse fretting fatigue scenarios.