Heat Source Parameter Identification Based on Attention-Enhanced Residual Convolutional Neural Network

Heat source parameters are critical input variables in welding thermal analysis, directly and significantly affecting the accuracy of the temperature field distribution, welding distortion, and residual stress prediction. This is particularly important in safety-critical welded structures, where high-precision heat source parameter identification is essential for ensuring the thermal simulation accuracy and mechanical performance reliability. Traditional parameter identification methods based on finite element simulations or experiments have limitations in adapting to complex working conditions and variable environments. To address this, this paper proposes the Heat Source Parameter Identification Network (HSPINet) model based on a residual convolutional neural network (ResNet) architecture with an attention mechanism capable of extracting key features from the weld morphology of T-joint structures, while accounting for the influence of process parameters and joint dimensions to achieve efficient and accurate identification of heat source parameters. This study not only enhances the intelligence level of heat source parameter identification but also provides a practical, intelligent tool for welding simulation and thermal field evaluation in complex industrial applications, demonstrating significant theoretical value and broad applicability in laser processing and manufacturing scenarios.