Application of a physics-informed hybrid model with additional output constraints for the prediction of the threshold of deep-penetration laser welding

Abstract

The quantitative prediction of process constraints, such as the threshold of deep-penetration laser welding, plays a crucial role for the fast and reliable development of robust process windows for laser manufacturing processes. A physics-informed hybrid model with additional output constraints for the prediction of the threshold of deep-penetration laser welding is presented. A “residual model” approach is used, where a machine learning model, employing Gaussian processes, is used to model and compensate for the deviations between experiments and a physical model, and output warping is used to incorporate additional output constraints into the model. The main benefits that result from applying such a model are found to be (1) an increased prediction accuracy compared to only using the physical model, leading to a reduction of the mean relative error of about 76%; (2) a reduction of the number of required training data compared to only using a black-box machine learning model; (3) an increased prediction accuracy compared to only using a black-box machine learning model; (4) and an increased compliance with physical boundary conditions by applying the additional output constraints.