A Novel Deep Learning Approach Combined With Loading Path Recognition and Physical Information Guidance to Predict Multiaxial Fatigue Life

ABSTRACT Multiaxial fatigue represents one of the most common and complex failure modes in engineering. Accurate prediction of multiaxial fatigue life remains challenging. Currently, machine learning methods have attracted considerable attention in multiaxial fatigue prediction, though insufficient fatigue data somewhat restricts their application. Given the direct and significant influence of multiaxial loading paths and material physical information on fatigue life prediction, this paper presents a machine learning framework incorporating loading paths and material physical information to address the data scarcity issue inherent in traditional machine learning approaches. The framework utilizes end‐to‐end learning to extract discriminative latent features for distinct load paths from material properties and original stress‐strain response features. These features are fused with Fatemi‐Socie (FS) model predictions as multi‐source inputs to achieve accurate fatigue life prediction. Prediction results demonstrate that the new method exhibits enhanced physical consistency and superior predictive performance in multiaxial fatigue life prediction of materials.