Centroid growth selective clustering method for surface defect detection in silicon nitride ceramic bearing rollers

Surface defects on silicon nitride ceramic bearing rollers typically exhibit fuzzy edge characteristics and gradient plunge features, which present significant challenges in image segmentation, including contour anomalies, incomplete segmentation, and notch misidentification. To address these challenges, this paper proposes the Centroid Growth Selective Clustering Method for the accurate detection and segmentation of fuzzy surface defect features. The method first analyzes the discontinuities in the notch regions associated with fuzzy edges, determining the image centroid based on Euclidean distance probabilities. Hierarchical clustering is then applied to effectively separate and cluster image content, enabling precise detection of feature connectivity. This approach ensures high-precision overlay and extraction of target features. Experimental results show that the proposed method achieves a segmentation accuracy of 97.85%, an edge coverage rate of 95.71%, and intersection and concurrency ratios exceeding 85% for three distinct types of surface defects. These findings significantly mitigate the impact of fuzzy edge features, enhancing defect detection accuracy and enabling complete extraction of surface defects. The proposed method enhances the reliability and performance of silicon nitride ceramic bearing rollers in high-precision applications.