Real-time prediction model of welding temperature based on LSTM

Temperature variations during the welding process directly affect the melting and solidification process of welding, which has a significant impact on the quality and mechanical properties of the weld. Real-time prediction of the welding temperature field plays a crucial role in solving the problems posed by the instability and complexity of the welding process, and has developed into an important direction in welding technology. In this paper, the real-time temperature variations during the three-layer, three-pass MIG (melt inert-gas) welding of aluminum alloys were monitored in real time. On this basis, a welding temperature prediction model based on the long-short-term memory neural network (LSTM) was proposed to predict the temperature changes in the unwelded region by taking the temperature data of the torch-passed region during the welding process , by analyzing the prediction results in comparison with the actual measured temperature changes, the results show that the method can predict the welding temperature changes better. This method of over-prediction is important for welding quality control, defect prevention and process optimization.