Height prediction in Directed Metal Deposition with Artificial Neural Networks

Abstract

Directed Metal Deposition (DMD) is a promising metal additive manufacturing technology, where parts are manufactured by fusing injected metal powder particles with a laser beam moving along a predefined trajectory. A toolpath typically includes sections as curves or edges, where machine axes need to decelerate and accelerate accordingly. As a result, the locally applied laser energy and powder density vary during the deposition process, leading to local over-deposition and over-heating. These deviations are additionally influenced by the toolpath geometry and process duration: previous depositions can influence close toolpath segments, in terms of time and space, resulting in local heat accumulations and develop profiles and microstructures that are different from the ones generated in other segments deposited with the same parameters due to geometry- and temperature dependent catchment profiles. To prevent these phenomena, lightweight and scalable models are required to predict the process behaviour for variable toolpaths.In this work, an artificial intelligence-based approach is presented to handle the process complexity and the multitude of toolpath variations for Inconel 718. Artificial neural networks (ANN) are used to predict the height of the deposition considering the previously defined toolpath. Training data have been generated by printing a randomized toolpath containing multiple curvatures and geometries. Based on the trained models, significant local geometric deviations are successfully predicted for the complete toolpath and could be anticipated by adapting process parameters accordingly