A fatigue life prediction method based on multi-signal fusion deep attention residual convolutional neural network

• A multi-signal fusion deep attention residual convolutional neural network model is proposed for fatigue life prediction . • The model leverages both channel and temporal attention mechanisms to enhance its feature sensitivity. • The residual convolutional structures are employed to enhance its feature extraction ability. • Acoustic emission and temperature signals are integrated into the model to learn degradation process at multiple scales. The traditional fatigue life prediction method is to construct a mathematical model of the material fatigue degradation process to achieve material fatigue life prediction. However, this method heavily relies on the prior knowledge and expertise of researchers, which limits its generalization. In order to address this limitation, a multi-signal fusion deep attention residual convolutional neural network (MSF-DARCN) model is proposed in this paper for fatigue life prediction. The acoustic emission signal and temperature signal of metallic materials throughout its entire life cycle are integrated into the MSF-DARCN model to learn the fatigue degradation process of the material from multiple information sources and improve its fatigue life prediction accuracy. At the same time, the MSF-DARCN model leverages both channel attention and temporal attention mechanisms to learn important feature information in input data and enhance its sensitivity to the feature information of material degradation process. Additionally, the stacked residual convolutional structures of the MSF-DARCN model are employed to extract the spatial features of input data to enhance its feature extraction ability. Finally, based on the fatigue life experiment of 304 stainless steel specimens, the accuracy and effectiveness of the MSF-DARCN model are analyzed. The results indicate that the MSF-DARCN model exhibits high accuracy in predicting fatigue life.