Real-Time Data-Driven Method for Bolt Defect Detection and Size Measurement in Industrial Production

To enhance the automatic quality monitoring of bolt production, YOLOv10, Intel RealSense D435, and OpenCV were integrated to leverage GPU parallel computing capabilities for defect recognition and size measurement. To improve the model’s effectiveness across various industrial production environments, data augmentation techniques were employed, resulting in a trained model with notable precision, accuracy, and robustness. A high-precision camera calibration method was used, and image processing was accelerated through GPU parallel computing to ensure efficient and real-time target size measurement. In the real-time monitoring system, the average defect prediction time was 0.009241 s, achieving an accuracy of 99% and demonstrating high stability under varying lighting conditions. The average size measurement time was 0.021616 s, and increasing the light intensity could reduce the maximum error rate to 1%. These results demonstrated that the system excelled in real-time performance, accuracy, robustness, and efficiency, effectively addressing the demands of industrial production lines for rapid and precise defect detection and size measurement. In the dynamic and variable context of industrial applications, the system can be optimized and adjusted according to specific production environments and requirements, further enhancing the accuracy of defect detection and size measurement tasks.