Data-Free Prediction of Aluminum Laser Welding Thermal Fields Using a Physics-Informed Neural Network

Abstract

Laser welding involves complex molten pool dynamics that critically affect weld quality and defect formation. In this work, a physics-informed neural network (PINN) model is developed to predict the temperature distribution and molten pool geometry in laser welding. The PINN incorporates the heat conduction equation into its loss function, ensuring that the predicted temperature field follows physical laws without relying on experimental training data. This model accepts spatial and temporal coordinates, material properties, and laser parameters as inputs and outputs the predicted temperature field. Validation against X-ray observations demonstrates that the predicted molten pool depth and width agree well with experiments, with relative error for penetration depth remained as low as 1.0%, while the deviation in the top weld width remained below 10%. The proposed PINN approach provides a fast, interpretable, and data-free framework for simulating temperature evolution and weld pool formation in laser welding.