Grid Search Optimized LSTM Method for Multi-Axial Fatigue Life Prediction

Traditional models for predicting multi-axial fatigue life are typically limited to specific materials and loading conditions. To address this limitation, this study adopts a deep learning approach to handle different multi-axial fatigue loading conditions. By using a grid search optimization algorithm, the optimal combination for three hyperparameters of the LSTM deep learning model-learning rate, number of hidden layers, and number of iterations are determined to achieve the best predictive performance. The method was applied to predict the fatigue life of three materials including pure titanium, SS 316L, and TC4 aluminum alloy. The results were all within the scatter band of two times the standard deviation, with a significant portion of predictions falling within the scatter band of 1.5 times. Furthermore, the extrapolation ability of the proposed method was validated using two sets of adjusted datasets, demonstrating good extrapolation performance for unknown loading paths. Therefore, the grid search optimized LSTM method for multi-axial fatigue life prediction can be applied to various materials and loading paths for multi-axial fatigue life prediction.