A Literature Review on Advances in Data‐Driven Research on Very High Cycle Fatigue: From Experiments to Prediction

ABSTRACT With the increasing demand for ultra‐long service life design of engineering components, research on very high cycle fatigue (VHCF) is confronted with critical challenges, including time‐consuming fatigue tests, complex failure mechanisms, and sparse experimental data. This paper presents a systematic review of research advances in VHCF from a data‐driven perspective. In terms of experimental methodologies, the evolution of ultrasonic fatigue testing systems is reviewed, highlighting advanced in situ monitoring techniques, linear active disturbance rejection control (LADRC) strategies, multiaxial proportional/nonproportional loading, and testing methods under extreme temperature environments. Regarding specimen design, size effect and adaptive specimen configurations are summarized, and a closed‐loop numerical optimization framework for precise resonant volume matching is introduced. In terms of mechanistic analysis, the dynamic evolution characteristics of subsurface crack initiation under VHCF are revealed for various metallic materials, and the frequency strengthening effect under multiphysics coupling is elucidated. Finally, regarding fatigue life prediction, the multidimensional theoretical evolution is systematically reviewed, ranging from macroscopic phenomenological models and energy‐based methods to defect/fracture mechanics models and ultimately to probabilistic statistical and extreme value models. To tackle the challenges posed by heterogeneous and limited fatigue data, the advantages of machine learning in processing complex fatigue datasets are evaluated, with a particular focus on the high predictive accuracy of physics‐informed neural networks (PINNs). By embedding fundamental fatigue physics and constraints, PINNs achieve a profound bidirectional synergy between physical laws and data‐driven modeling, significantly enhancing the model's generalization capability and extrapolative robustness.