Abstract: As a core practical course for equipment manufacturing majors, metalworking training plays a vital role in cultivating students' metal cutting capabilities. Traditional lathe operation training relies on manual vernier caliper measurements, which suffer from low efficiency, significant errors, and inconsistent evaluation criteria. To address these issues, this paper develops a machine vision-based intelligent scoring system for circular components, significantly enhancing detection efficiency and accuracy. The system hardware comprises a Daheng industrial camera, ring light source, and workstation frame, while the software leverages Halcon algorithm library with a Qt Creator-built interactive interface. The detection process involves four key stages: 1） High-precision image acquisition via industrial camera, ensuring image quality through camera calibration and distortion correction; 2） Image preprocessing using grayscale conversion, filtering, and noise reduction to eliminate surface scratch interference; 3） Precise geometric feature extraction through Canny operator edge detection combined with least squares contour fitting; 4） Dual-criteria automatic scoring: initial shape defect screening via circularity （Cr<0.75）, followed by tolerance detection through subpixel-to-actual dimension conversion. Verified with 400 component tests, the system achieves dimensional errors below 0.06mm and reduces detection time to 1/5 of conventional methods. Real-time results display through the interface, with automatic Excel report generation for qualification rate statistics and grading. This system not only resolves subjective manual evaluation issues but also improves detection efficiency by 83% while establishing standardized assessment protocols. Future enhancements may incorporate multi-angle imaging for 3D detection accuracy and extend to complex features like threads and curved surfaces, providing comprehensive technical support for intelligent manufacturing education.