Physics-Informed Machine Learning Assisted Laser Additive Manufacturing Process Optimization

Laser additive manufacturing (LAM) has revolutionized multiple industries by leveraging its unique capabilities. In the oil and gas sector, LAM has demonstrated remarkable efficacy in repairing critical components of logging and drilling equipment. However, optimizing repair quality remains challenging due to the process’s inherent complexity and the multitude of interacting factors involved. In this study, a physics-informed data-driven framework was proposed to correlate LAM process parameters with two key physical descriptors—volumetric energy density (VED) and cooling rate—for predicting the density of 316 stainless steel components fabricated via LAM. Four machine learning (ML) algorithms—Support Vector Regression (SVR), Artificial Neural Networks (ANN), Gaussian Process Regression (GPR), and eXtreme Gradient Boosting (XGBoost)—were systematically compared to evaluate their predictive performance. The results indicate that physics-informed ML models outperform purely data-driven counterparts in terms of prediction accuracy, highlighting the model’s superiority in characterizing part performance. Furthermore, the predicted results were utilized to generate process parameter maps, enabling the identification of optimized parameter combinations that achieve full-density fabrication. This work advances the mechanistic understanding of LAM by incorporating physical insights into data-driven modeling. The proposed physics-informed ML approach exhibits substantial promise for enhancing the quality and reliability of LAM-produced components, offering a pathway toward process optimization in industrial applications.