This study employs a Gaussian Process (GP) model to model the responses of the direct laser deposition (DLD) process, focusing on parameters such as ‘Width’, ‘Height’, ‘Depth’, ‘Dilution’ and ‘Cracks’. GP models, known for capturing complex relationships and providing uncertainty estimates, were trained on two datasets: one with preheating (PHT) and the other without PHT. This division enabled comparison of PHT’s influence on modelling accuracy and response characteristics. Interpretative analysis of the model predictions was performed using SHAP values (SHapley Additive exPlanations), which measure the importance of the input features (‘P’, ‘SS’, ‘FR’) and their influence on the model responses. The integrated approach of modelling with Gaussian Processes and interpretability analysis with SHAP allowed us to identify the most influential features and better understand the process dynamics. The results indicate that preheating (PHT) significantly affects cladding properties such as depth and dilution due to Marangoni convection and increased energy absorption by the substrate. The combination of GP models and SHAP values provided accurate predictions and information on the most influential factors in the process responses, allowing the optimization of the DLD process. This integrated methodology improved the accuracy of predictions and facilitated a more informed decision-making process, contributing to the optimization of laser processing.
Ferreira et al. (Mon,) studied this question.