Research on Automatic Crack Recognition Method for Magnetic Particle Inspection Robot Based on YOLOv8

Large pressure-bearing equipment, represented by spherical tanks, primarily relies on manual magnetic particle inspection to detect cracks. This requires inspectors to work at heights in confined spaces while maintaining continuous visual focus. Utilizing magnetic particle inspection robot can automatically and continuously collect video information of magnetic particle results without the need for scaffolding. However, the issue of extreme fatigue caused by manual defect recognition persists. This paper employs the open-source object detection algorithm YOLOv8 to annotate and train crack defects discovered during the actual inspection and testing of various types of special equipment. The training effect is enhanced through manual image cropping and preprocessing. Applying the trained model to the real-time video collected by the magnetic particle inspection robot can quickly generate a list of crack target detection results, significantly reducing manual labor intensity and offering substantial practical engineering application value.