The defect detection of 3D-printed ceramic curved surface parts with low contrast based on deep learning

Some defects were produced during 3D printing ceramic parts and it is difficult to detect these defects, especially for the ceramic curved surface parts, because of low contrast between defects and ceramic background, blurred edges and low efficiency. In this word, a defect detection method with low contrast based on deep learning is studied. Firstly, the difficulties and adverse effects that occur in the detection of ceramic curved surface parts are analyzed qualitatively. Then, a blurry inpainting network model is proposed to effectively reduce the degree of blurring on the curved surface. A multi-scale detail contrast enhancement algorithm is also established to solve the problems low contrast among defective and the background regions, which can highlight the characteristic information of the defect regions. On this basis, two types of defects in the curved surface parts are detected by using our constructed ECANet-Mobilenet SSD network model. The results show that the prediction accuracy for crack and bulge defects recognition can reach 94.35%, and 96.72%, respectively, and meanwhile, their average detection time of a single image is 0.78s. Therefore, this study on the defect detection of 3D-printed ceramic curved surface parts can contribute to the intelligent and 3D printing development of advanced ceramic industry.