The Laser Metal Deposition (LMD) process involves rapid thermal cycles and complex thermo-material interactions, making accurate prediction of melt pool characteristics challenging. This study presents a physics-informed, data-driven approach to accurately predict peak melt pool temperature and bead geometry across distinctive substrate–powder combinations using real-time thermal data from an integrated pyrometer and infrared camera. Experiments have been performed on three substrates (Al3002, Ti, and Mild Steel) and five powders (Ti6Al4V, AA2024, SS316L, Cu, and Nitinol) to develop framework using Physics-Informed Neural Networks (PINNs) with the inclusion of heat conduction physics in the learning process. The developed PINN model outperformed conventional machine learning models, obtaining an R 2 of 0.97 and MSE of 417.67 K 2 in predicting temperature. For predicting the bead geometry, the model yielded R 2 values above 0.96. SHAP analysis determined scanning speed and beam power as influential parameters. The results demonstrates the effectiveness of integrating physical laws with machine learning for reliable prediction in data-limited LMD environments.
Chowdhury et al. (Sun,) studied this question.