Machine Learning for The Qualification of Direct Energy Deposition Processes
Please login to view abstract download link
The objective of this work consists in the development and implementation of a Machine Learning framework for the process optimization metal Additive Manufacturing (AM) by Direct Energy Deposition (DED). The in-house software Add2Man developed at CIMNE for the numerical simulation of the AM-DED thermomechanical process is used to analyse the temperature evolution, the melt-pool morphology and the microstructure features during manufacturing and the following cooling phase(1). The software is prepared for parallel computing in distributed memory (clusters) and makes use of the most advanced techniques of adaptive meshing (AMR) to ensure the best performance and the highest accuracy. The novelty proposed in this work consists of taking advantage from Machine Learning to provide artificial intelligence (AI) to the DED manufacturing process, allowing for the optimization of the process parameters such as the power input, the printing speed or the dwell time among layers. By adopting for the software exactly the same input as for the DED machines (G-code format), it is possible to faithfully reproduce the power delivery of the laser along its path, as well as the cooling during repositioning pauses, waiting times, etc. The melt-pool volume is used to continuously monitor and modulate the process parameters. In this way, it is intended to add an active and automated control to AM manufacturing, qualifying this technology for its adoption and integration in the industrial manufacturing chain.
