Grace Gu1
University of California, Berkeley1
Grace Gu1
University of California, Berkeley1
Additive manufacturing (AM) has made huge strides in the past decades, enabling the fabrication of multiscale and multi-material designs previously deemed infeasible. As AM spreads to the fields of aerospace, automotive, and medical fields, part quality and reliability become increasingly important. Improper settings of process parameters can lead to imperfections in the part; small imperfections in one layer can propagate into the rest of the build. In this work, we present a real-time monitoring and autonomous correction system enabled by computer vision and machine learning to diagnose the quality of parts and modify 3D-printing parameters iteratively and adaptively in real-time. Sensor technologies embedded in our platform are used to capture images during the printing process to train a machine learning model. Precise localization and semantic segmentation detection algorithms are implemented to shed light on the evolution of defects during real-time printing, specifying a quality profile at each layer of the print. This advanced detection system is capable of providing defect information for quality assessment and has the potential for further automated control as well as correction of AM systems.