Melt Pool Depth Prediction in Directed Energy Deposition Single-Track Prints Using Point Cloud Analysis

Abstract Traditional assessment methods of melt pool size in metal additive manufacturing (AM) are known for their time-consuming and expensive nature, involving tasks such as cutting, polishing, and detailed microscopy. In this paper, we present a novel approach to predict the melt pool depth in directed energy deposition (DED) AM process, by exclusively relying on point cloud data acquired through a laser scanner coupled with machine learning (ML) tools. Our method streamlines the process of identifying the size of a melt pool by automating the entire procedure, negating the need for any manual intervention starting from the initial data collection stage. This improved analysis opens up possibilities for point-cloud-based real-time predictions and control, making it a highly efficient and adaptable solution. By training multiple ML models using track width and height data extracted from the point cloud, we surpass the limitations of relying solely on process parameters to accurately predict melt pool depth — a parameter not visible during print inspection, especially using a laser scanner. Among the various regression models utilized, the Gaussian Process Regression model demonstrates superior performance, yielding a Mean Absolute Error (MAE) of 18.89 μm and a Root Mean Squared Error (RMSE) of 25.5 μm. The results indicate a promising potential for the developed methods to eventually subside the requirement for external post-processing of print tracks in order to characterize the melt pool.