A physics-guided artificial neural network framework for fretting fatigue life prediction incorporating stress gradient effects

Fretting fatigue (FF) involves highly localized stress fields characterized by severe gradients, multiaxial and non-proportional behaviour, posing significant challenges for life prediction. In this work, a simple physics-guided artificial neural network (ANN) framework is proposed to predict FF life while inherently accounting for stress gradient and multiaxial effects, without the use of a material critical distance parameter. The model employs fatigue parameters derived from finite element analyses representing both cylindrical and spherical contact configurations and introduces physical constraints to enforce mechanically consistent responses. A total of 214 different FF experimental data points were compiled from the literature to train and validate the model. The proposed ANN achieved accurate life predictions across diverse materials and loading conditions, with mean errors within a 1.5 × scatter band. The inclusion of physical constraints significantly reduced nonphysical trends, demonstrating the robustness and generalization capability of the proposed framework for fretting fatigue life prediction.