Defect Detection and Grain Size Evaluation based on Metallographic Images of Long Shafts

Abstract

Aiming at the difficulties in identifying tiny surface defects and automatic grain size evaluation of long shaft parts, this paper proposes an automatic detection and grading method based on robot visual guidance and adaptive image segmentation. The YOLOv8 algorithm is adopted to identify surface defects on long shafts, and the defect coordinates are mapped to the base coordinate system through the hand-eye calibration matrix with a positioning error within 0.9 mm, which guides the end effector to complete fixed-point grinding and etching. This paper focuses on developing an OpenCV-based grain size calculation algorithm: CLAHE contrast enhancement, adaptive Gaussian threshold segmentation and watershed algorithm are applied to metallographic images after etching to accurately separate adhered grains. Pseudo-boundary interferences are eliminated via contour morphological filtering, and the average grain area method is finally used to calculate and grade grain sizes. Experiments verify that the proposed system realizes integrated operations from defect localization to metallographic grading for long shafts, with a repeatability error of grain size grading less than 0.5 grades, satisfying the high-precision detection requirements of industrial sites.