Аналіз вхідних параметрів штучної нейронної мережі у прогнозуванні довговічності за багатоосьового навантажування

The research focuses on optimizing the input parameters of artificial neural networks (ANN) for predicting fatigue life under multiaxial loading conditions. Traditional methods for predicting fatigue life often fail to provide the necessary accuracy, especially under complex multiaxial loading. Therefore, it is essential to explore new approaches based on machine learning methods, particularly using artificial neural networks. In this study, 30 different parameters commonly used in multiaxial fatigue models were analyzed by including them as input parameters for the ANN. The application of the SHapley Additive exPlanations (SHAP) method enabled filtering out irrelevant parameters, leaving only the 10 most influential ones, such as the coefficients of the Coffin-Manson equation and critical plane parameters. The resulting model demonstrated better predictive capabilities compared to widely used multiaxial fatigue models, particularly the Fatemi–Socie model.