A novel image feature based self-supervised learning model for effective quality inspection in additive manufacturing

With the rapid development of additive manufacturing (AM) technology, quality inspection has become one of the most crucial research topics in additive manufacturing. Although numerous image-based deep learning methods have been successfully developed to monitor and inspect AM product quality effectively, many require substantial labels in order to achieve satisfactory training, which is often impractical in real-life AM processes. In this article, a novel image feature-based self-supervised learning (IFSSL) model is proposed for effective quality inspection in AM. Through a feature-based image fusion approach based on defect-relevant feature extraction, the IFSSL model is able to guide machine vision to focus on highlighted defect-relevant regions in the AM product image. In addition, the defect-relevant features are used to generate pseudo-labels for self-supervised learning. With self-supervision, the IFSSL model leverages the advantages of supervised learning and unsupervised learning by requiring no sample label while retaining defect-relevant information. The effectiveness of the proposed IFSSL method is demonstrated through a real case study of fused deposition modeling product image dataset. Results show that the IFSSL model can guide machine vision to pay more attention to potential defective regions, enabling it to detect and locate faults effectively and automatically for machine vision guided quality inspection.

Automated summary

With the development of AM technology, quality inspection has become a crucial research topic. This article proposes a novel IFSSL model that utilizes self-supervised learning to effectively inspect the quality of AM products. The model leverages defect-relevant feature extraction and image fusion to guide machine vision towards relevant regions and detect faults automatically.