Unsupervised Defect Segmentation in Selective Laser Melting

Obtaining good surface quality in the additive manufacturing process is very challenging as complex physical processes are involved. Although there are deep learning methods proposed to detect defects on surfaces because of the ability of the process to extract features, many of them require a large dataset and hand-labeling work. In this article, we propose an unsupervised segmentation method to detect defects on additive manufactured surfaces, which requires only a single scanned image. The proposed method has three modules, including feature learning, self-attention, and clustering modules, which are responsible for extracting features, capturing global features, and assigning cluster labels. To validate the segmentation performance of the proposed method, experiments were conducted on surfaces fabricated using different materials and different process parameters. The results show that the proposed method segments surface defects effectively compared with other state-of-the-art models.