A nondestructive detection method for Si₃N₄ bearing roller microcrack

Abstract

To address the challenges of traditional single algorithms in segmenting defect images with low contrast and high noise during the segmentation process of microcrack defects in Si₃N₄ bearing rollers, a coupled method based on weighted histogram equalization and adaptive Otsu is proposed. This method achieves high‐precision extraction of microcracks in Si₃N₄ bearing rollers. By analyzing the texture characteristics of microcrack formation, a distribution cumulative function filter is applied to map gray values, enabling inner loop nested gray value accumulation. The maximum interclass variance of the background before and after applying Otsu is used to calculate the optimal adaptive threshold for binary segmentation. The structural similarity (SSIM) index of the enhanced image is 0.924862, and the coupling algorithm achieves a segmentation accuracy of 97% for the microcrack feature images of Si₃N₄ bearing rollers. This approach effectively reduces the impact of low contrast and background noise on defect segmentation, thereby improving the effectiveness and accuracy of microcrack feature extraction.