Research on double-sided detection method of bolt looseness based on machine vision

As the key connecting parts at the bottom of the train, bolt looseness has a crucial impact on the safe operation of the train. Among the current train bolt state detection methods, the manual method is prone to false detection and missed detection, and the maintenance quality cannot be guaranteed; while the existing intelligent detection methods are mainly aimed at the single looseness feature on the top or side of the bolt, which cannot fully reflect the actual situation of the detected bolt. In this study, a bolt looseness detection method combining the top and the side features of the bolts is proposed, which mainly includes: (1) building an experimental platform for double-sided detection of bolts, simulating the looseness state and the tightening state of the bolts, processing images and generating dataset; (2) in order to solve the problem that the looseness features of bolts are relatively small, adding attention mechanism (CBAM) to the YOLOv3 network structure, and training an accurate and efficient convolutional neural network (CNN) for detection. Experiments show that the detection accuracy of the algorithm reaches more than 99.92%; and (3) the trained model is tested for different conditions (different lighting environments, different detection angles under different observation directions, and the coincidence of bolt looseness marks). The experimental results show that the proposed double-sided detection scheme and the improved YOLOv3 algorithm have good detection results.