Loosening Bolt Detection of Sling Cars Based on Deep Learning and Feature Matching

In the automobile production line, a significant number of sling cars are used to facilitate the transfer of processing parts across various stations. The bolts may be loose after prolonged vibrations and impacts during the sling car’s operation. To address these issues, a bolt loosening detection technology based on deep learning and feature matching is introduced, which makes up for the drawbacks of manual detection inefficiency and the limited application of machine vision. The improved YOLOv5 is used to locate the bolts, which improves the detection accuracy and reducing the numbers of misidentified bolts with the same performance as YOLOv5. Furthermore, the YOLOv8OBB model is used to accurately extract the character area within the bolts for subsequent detection. The images to be detected and the template images are matched by the Scale Invariant Feature Transform (SIFT), followed by feature point screening conducted by the Random Sample Consensus (RANSAC). The loosening angle of the bolt is obtained by analyzing the homography matrix of the RANSAC. The proposed method is applied to the sling cars in the manufacturing line and results show that the proposed method has a higher accuracy in detecting the bolts loosening, which can guarantee the reliable operation of the sling cars.