Machine learning-based fatigue life prediction of metal materials: Perspectives of physics-informed and data-driven hybrid methods

•The application of machine learning in fatigue life prediction is reviewed.•The shift from data-driven to physics-based hybrid approaches is discussed.•The approaches of combining physics-based and data-driven models are described.•The future challenges and development directions of fatigue life prediction are discussed. Fatigue life prediction is critical for ensuring the safe service and the structural integrity of mechanical structures. Although data-driven approaches have been proven effective in predicting fatigue life, the lack of physical interpretation hinders their widespread applications. To satisfy the requirements of physical consistency, hybrid physics-informed and data-driven models (HPDM) have become an emerging research paradigm, combining physical theory and data-driven models to realize the complementary advantages and synergistic integration of physics-based and data-driven approaches. This paper provides a comprehensive overview of data-driven approaches and their modeling process, and elaborates the HPDM according to the combination of physical and data-driven models, then systematically reviews its application in fatigue life prediction. Additionally, the future challenges and development directions of fatigue life prediction are discussed.