Prediction of Melt Pool Temperature for Directed Energy Deposition Using Supervised Learning Methods on Optical Measurement Data

Abstract

One of today's ongoing challenges in directed energy deposition (DED) is controlling the geometry and material properties of parts. The objective of this paper is to investigate the relationship between several printing parameters of DED (laser power, laser speed, powder feed rate) and the melt pool temperature. Because DED is a complex and nonlinear process, well-established supervised-learning models such as support vector regression and artificial neural networks are particularly well suited to represent it. The MiCLAD machine, designed at the Vrije Universiteit Brussel, is equipped with a hyperspectral camera that monitors the light emitted at several wavelengths by the melt pool during the building process. A steady-state data set produced by the hyperspectral camera is postprocessed by an advanced temperature estimation method, and the limitations of the temperature estimation method are identified and discussed. The temperature data are used as training data for supervised-learning methods, and a study is conducted to compare the performance of the considered methods using the measured optical data. This study demonstrates that the melt pool temperature of the DED process can be effectively modeled through the printing parameters thanks to supervised-learning methods.