Towards Unsupervised Anomaly Detection and Clustering of Defects in Electrode Surface Inspection

Pouls, Kevin; Hamid, Sagad; Sensmeier, Leonard; Eiling, Fabian; Menne, Jan; Oehme, Mathias; Pfeifer, Max; Ridder, Fabian; Tepe, Mika; Braun, Tanya

Abstract

Large, diverse datasets are necessary to advance automatic surface inspection systems for electrode sheet coatings. Creating such datasets is time-consuming and costly because defects are sparse and not easily separable. To address this challenge, we propose a novel architecture for unsupervised anomaly detection and clustering of defects in image-based surface inspection for electrode manufacturing. Our approach leverages a modular and flexible design, enabling independent optimization of anomaly detection and clustering. We evaluate the architecture on a small dataset from a real battery production line. Our research facilitates autonomous defect detection, root cause analysis, and the efficient generation of new datasets.