SurfaceVision

In the era of Industry 4.0, the advent of 3D printing has revolutionized manufacturing by significantly reducing financial and time efforts. 3D printed products are created layer-by-layer based on digital Computer Aided Design (CAD) inputs, yet they remain susceptible to defects that can compromise quality. Detecting these layerwise faults is important to ensure high quality outputs. The traditional method currently requires visual information processing devices or continuous monitoring of the process via a camera, which is very resource consuming and costly. Machine learning techniques being used for automatic detection of the faults suffer in real time conditions with inefficient fault detection due to the inability of adaptation to real time changes in the printing process. Along with the inability to assess layer by layer protrusion development, the current ML techniques are lacing in 3D printing fault detection. This paper introduces SurfaceVision, an automated system for surface fault detection in 3D printed products, leveraging the ResNet-18 architecture as the backbone. Our framework utilizes a combination of contrastive learning and multi domain loss function to identify and classify defects with high accuracy. Comparative experiments demonstrate that the ResNet-18 based SurfaceVision outperforms the baseline.