Progress on fatigue life prediction of aeronautical materials based on machine learning

Aerospace equipment materials demand an ultra-high level of safety and reliability, with fatigue performance being one of their core performance metrics. Traditional fatigue prediction methods rely heavily on extensive experimental tests, which are associated with high costs and long development cycles, thus failing to meet the requirements of modern aerospace engineering for efficient and accurate performance evaluation. In recent years, machine learning has exhibited remarkable potential in the fatigue life prediction of aerospace materials. This work presents a systematic review of the research progress in this field, with a focus on mainstream models and modeling workflows. It clarifies the core ideas and key research findings of both pure data-driven methods and physics-integrated approaches, and centers on the role of physical information embedding in enhancing model accuracy, credibility, and interpretability. Moreover, the paper critically discusses the existing limitations, including insufficient information mining in terms of data dimensions and complex failure mechanisms, inadequate model interpretability and low trustworthiness for engineering applications, as well as poor adaptability to complex service conditions. Finally, key research directions for addressing these limitations are highlighted, such as constructing standardized and highly reliable fatigue datasets, establishing a task-oriented automatic fusion mechanism for physical knowledge, and advancing fatigue life prediction at the level of structural components under complex service conditions.