Real-time prediction of temperature field during welding by data-mechanism driving

The temperature field during welding (TFW) is a crucial factor that significantly influences the weld seam's shape and overall performance. The accurate prediction of the TFW is crucial for optimizing welding process parameters and achieving high-precision control during welding. This study proposes a real-time prediction method for the TFW, driven by a combination of data and physical mechanisms. By defining the heat transfer mechanisms, welding methods, material properties, and process parameters, TFW finite element simulation data is obtained for training a data-driven neural network. Real-time images of the welding pool are used to extract the weld pool surface width (WPSW) by image processing techniques, and a Long Short-Term Memory model is employed to extract heat source (HS) parameters from the continuously changing WPSW. The HS function is updated using real-time welding current, arc voltage, and HS parameters to calculate the real-time heat flux density at various locations in the welded workpiece. Finally, the DeepONet neural operator model predicts the temperature values at these locations by solving for the real-time heat flux density, thereby achieving TFW prediction. This method has high flexibility and real-time performance, which provides an effective and practical solution for real-time monitoring of the TFW, and lays a foundation for the high-precision control during welding.