Data-Driven Prediction of Metal Fatigue Life

With the rapid development of additive manufacturing (AM) technology, its applications in aerospace, medical, and automotive industries are becoming increasingly widespread. This paper investigates the fatigue life prediction of additively manufactured metallic materials and proposes a parallel deep learning architecture combining Gated Recurrent Unit (GRU) and Transformer models, optimized using Hybrid Leader-Based Optimization (HLBO) to improve prediction accuracy. The model simultaneously processes the outputs of both GRU and Transformer models, leveraging their respective strengths to learn features from both local and global perspectives, thereby enhancing the accuracy of fatigue life prediction. The paper first reviews the theoretical background of GRU and Transformer, then presents the proposed parallel model architecture in detail. The core principles of the HLBO optimization algorithm are introduced, along with its application in hyperparameter optimization. Experimental results show that the proposed parallel deep learning model significantly outperforms traditional single deep learning models in terms of prediction accuracy and generalization ability. Furthermore, the performance of the HLBO optimization algorithm is evaluated using benchmark test functions such as the Ackley and Rosenbrock functions to validate its global and local search capabilities. Experimental results demonstrate that HLBO performs excellently in handling high-dimensional, multi-modal optimization problems and effectively enhances the prediction capability of the model. Finally, the model is trained and validated using the FatigueData-AM2022 dataset. The results show that the proposed model exhibits strong adaptability and high precision in predicting the fatigue life of different additively manufactured metallic materials, providing an effective tool and method for the fatigue life assessment of additively manufactured components.