|About this Abstract
||2022 TMS Annual Meeting & Exhibition
||Additive Manufacturing: Advanced Characterization with Synchrotron, Neutron, and In Situ Laboratory-scale Techniques II
||Deep Learning for Real-time Non-destructive Inter-layer Quality Control during Additive Manufacturing Process
||Steven Hespeler, Michael Juhasz, Ehsan Dehghan-Niri, Jeffrey Riemann
|On-Site Speaker (Planned)
Defects are a leading issue for rejection of parts manufactured through Additive Manufacturing (AM). Typically, defect measurements are gathered post-production with computerized tomography (CT) scans which are difficult to implement into real-time monitoring. In-situ monitoring of AM produced parts was employed to collect real-time data during a Directed Energy Deposition (DED) build. Machine Learning (ML) methods were utilized on time sensitive parameters during the build to provide a defect classification system. Statistical analysis performed evaluated feature impacts on inter-layer classification and investigates the relationships between these features. A threshold was applied to detect porosity sizes and quantities that render individual layers as acceptable or unacceptable. We demonstrate the effectiveness of using a Deep Learning (DL) technique for the purpose defect monitoring. The customized DL technique is shown to be highly effective at classifying acceptable/unacceptable layers real-time during build process.
||Additive Manufacturing, Machine Learning, Characterization