REAL-TIME TEMPERATURE PREDICTION OF A MOVING HEAT SOURCE PROBLEM USING MACHINE LEARNING

Moving heat source problem is commonly seen in many manufacturing applications, such as machining, laser cutting, welding, and additive manufacturing processes, while numerical modeling often takes time to analyze. This paper presents a neural network (NN) and linear-time invariant (LTI) system-based framework, aiming at real-time temperature prediction both spatially and temporally. Training data are generated from finite element analysis (FEA) and processed with convolution neural network (CNN) to form a surrogate model for location-dependent thermal response. LTI is used to superimpose thermal responses based on the heat source's path and magnitude. The suitability of this framework is evaluated for materials of both low and high thermal diffusivities as well as adiabatic and nonadiabatic cases. In the training of the model, the low thermal diffusivity and high thermal diffusivity cases both showed training and testing correlations of over 99%. Overall, all validation studies show good agreement between the predicted temperature and the ground truth. More errors are seen when the material has a high thermal diffusivity (< 21.7 %), and the heat is applied adjacent to the boundaries (< 23.6 %).