PCA-Enhanced Geometric Inspection: Robust Anomaly Detection in Industrial Settings

This paper introduces an intelligent visual inspection method for Industry 4.0, integrating Principal Component Analysis (PCA) for shape alignment with a Geometric Center Shift-based Anomaly Detection (GCS-AD). This Computer Vision-based approach is focused on external contours and is highly robust to lighting differences, rotations, and deformations in the part that can occur in industrial contexts. The experimental results presented here showed that the method achieved an average 98% classification accuracy (95% CI: [93.0%, 99.8%]) on a controlled screw dataset of 100 samples, which is better than K-Nearest Neighbors (87%; 95% CI: [78.8%, 92.7%]) and Support Vector Machines (89%; 95% CI: [81.0%, 94.3%]). The descriptors formed from PCA aligned contours proved highly interpretable, providing a straightforward visualisation for quality control and decision-making. With an average processing time of approximately 0.05 s per image, the method shows potential for near-real-time use in resource-constrained industrial environments. Here, the application in screw inspection is showcased as a proof-of-concept study under controlled laboratory conditions. The method was evaluated on screws and qualitatively assessed on nuts and gears, demonstrating robustness to lighting and alignment changes in controlled tests. Despite the limited dataset, the results indicate that GCS-AD maintains high interpretability and reliability for near-real-time quality inspection. Broader industrial validation on larger and more diverse datasets is required to confirm generalizability.