Physics-Informed Machine Learning of Thermal Stress Evolution in Laser Metal Deposition

Rapid laser scanning generates a complex heat-affected zone with steep temperature gradients in laser additive manufacturing, including laser metal deposi tion (LMD) and laser powder bed fusion process (LPBF). The complex thermal history and severe gradients lead to very high thermal stresses that evolve into residual stresses after the component cools down. Data-driven methods, such as machine learning (ML), offer an alternative to traditional physics-based simulations for calculating the thermal stress evolution. However, ML often requires a large, labeled training dataset, which is computationally inefficient. In addition, the “black box” nature of data-driven ML methods makes it difficult to interpret the results. Additionally, data-driven ML methods do not use governing physical laws under pinning laser additive manufacturing to make them data-efficient. This study aims to develop a physics-informed ML (PIML) model that can predict thermal stresses during laser scanning without requiring any labeled training dataset. A case study has beenconducted todemonstrate the predictive capability of the PIML method and examine the evolution of thermal stresses in an LMD process.