Anomal Detection for In-situ Quality Control of Directed Energy Deposition (DED) Additive Manufacturing
One common cause for the rejection of parts produced during metal Additive Manufacturing (AM) is the presence of unacceptable defects within the part. While powerful, post-processing nondestructive techniques can be unapproachable due to time constraints or simply impractical for certain inspection and quality control applications of the AM, especially with parts of high complexity. The AM process requires a layer-by- layer execution to build parts, allowing for a unique opportunity to collect data and monitor the process in real/semi-time. The incipient phase of AM monitoring and control typically consists of developing an automated unsupervised statistical anomaly detection algorithm that is capable of detecting irregularities through parameter measurement and sensing features. In this paper, we develop a simple and effective method for detecting anomalies through use of statistical distances from data collected during the laser-based Directed Energy Deposition (DED) AM process.