AI based monitoring system for DED part quality evaluation

Abstract

This paper describes an Artificial Intelligence based approach to analyze data collected during the manufacturing of metal parts through Direct Energy Deposition (DED) process. A dedicated monitoring system has been integrated into the DED machine with the aim of gathering data while the manufacturing is running. In particular, a camera and a bi-chromatic pyrometer have been added to the optical chain: images and temperatures of the melted area are stored synchronously - as well as the machine axis positions and process parameters. Images represent the process signature, for example the frame brightness has been proved to be related to the local energy or spikes visible in the picture are indicator of process stability. Temperatures, on the other hand, are key in such manufacturing processes as they can trigger residual stresses or distortions in the final parts. In the proposed approach both signals have been considered to pursue a robust and complete solution. A number of tests have been conducted by varying tool path strategies and built parts have been geometrically evaluated. An AI based classifier has been trained starting from this dataset and validated consequently: it receives the signals coming from the monitoring system and provides an indication about the expected quality of the deposited track. In particular, it classifies a track as "good" if no material under/over-deposition is estimated and labels it as "bad" in the opposite situation. The benefits of this solution have been assessed with regard to Inconel 718 parts. Finally, it has been tested on a real use case consisting of deposition of materials on turbine blade tips for repairing actions.