Defect detection: Defect Classification and Localization for Additive Manufacturing using Deep Learning Method

In this paper, a novel visual-based defect detection method using deep learning algorithms is reported to classify and localize the defect derived from the process of Additive Manufacturing (AM). The processing conditions and costs make it impossible to manually inspect the built parts defect. Vision-based in situ real-time defect monitoring method was conducted to acquire image data and then recognize the defect class and locality in the manufacturing process. Nevertheless, the traditional image processing technology is of poor performance when dealing with defect detection in sophisticated fabrication scenarios. Therefore, we research the viable technique, using deep learning algorithms to extract the defect image feature in continuous frame images and trained our network to learn the defect class label as well as its position. In order to test this method, we also report a defect dataset called AM Defect Recognition Benchmark Dataset, we proved the method can achieve the state-of-art defect detection performance in actual scene dataset.