Abstract
Bolted joints are widely used in the field of aerospace, civil and mechanical engineering. During their service life, extreme loading or environmental factors can cause the loosening of bolts. In this paper, a bolt loosening detection method based on computer vision and image processing is developed to identify bolt rotation angle in a steel multi-story frame structure. The experimental results show that the bolt target detection accuracy can reach 100% by using the Yolo-V5s deep learning model trained with a self-developed bolt object dataset. The dataset consists of 337 bolt images captured in nature scenes. For the angle calculation, the final result shows that the identification error is less than 5.8°, and at a slight camera angle (0∼20°), the maximum error even does not exceed 2.8°. Thus, the effectiveness of this method for detecting rotary loosening of bolts is well validated.
