Precise localization and semantic segmentation detection of printing conditions in fused filament fabrication technologies using machine learning

Abstract Although a vast array of anomaly detection methods has been developed in fused filament fabrication, a widely-applied additive manufacturing technology, acquiring in-situ detailed spatial information of the defects within the detection field remains a significant challenge in actual processing. In this paper, machine learning algorithms are proposed to realize precise localization and semantic segmentation detection of the in-plane printing conditions including over-extrusion and under-extrusion in both local and global frameworks. Results visualization and evaluation methods are conducted to demonstrate the high performance of the models. Our results show that detection latency is also improved by successfully recognizing the transitions between print quality conditions within a single raster. This advanced detection system is able to provide comprehensive defect information for real-time assessment and has great potential for further automated control as well as correction of additive manufacturing systems.