A Mask R-CNN based process monitoring system for fabricating high density ceramic parts using photo-polymerization

Traditional fabrication of ceramic parts face limitations due to hardness and brittleness, despite of having good mechanical properties. Digital light processing (DLP) additive manufacturing technology offers promising way to fabricate intricate geometry of ceramic parts. To fabricate high performance, maximizing solid loading of ceramic powder is important to reduce the shrinkage and distortion during post-processing. However, it increases viscosity dramatically and makes difficult with material supply during printing process. Therefore, not only increasing the solid loading of ceramic powder but also minimizing random defects during printing process is essential for us to achieve high-quality ceramic parts. In this study, vision-based defect monitoring system using Mask R-CNN model was employed. We classified two types of defects called pinhole and un-even paste and quantify the defect characteristics such as number and size with pixel level. This method provides us the basis of a feed-back system for controlling the process parameters in real time.