Automatic identification and location method for shaft-hole interference fit crack based on deep learning

Cracking on the shaft of the interference fit joint surface is a common failure form for the shaft-hole interference fit. Because the crack is small and is located at the shaft edge, the crack signal is affected by echo and reflection signals and noise when traditional ultrasonic guided wave methods are used to detect the crack. When the crack is small-sized, the crack signal may even be submerged completely in the echo and reflection signals. Therefore, the traditional detection method is unable to obtain accurate information about the crack locations easily and thus cannot evaluate the overall safety performance of the interference fit accurately. To solve these problems, this paper presents an on-axis crack monitoring system for interference fit joints using laser ul-trasound and an improved YOLOv5 network that integrates early warning, detection, positioning, and analysis functions. The system detects axial cracks at the shaft-hole interface via laser ul-trasonic guided waves on the workpiece, and captures the crack information by deep learning; then, the captured crack information is analyzed to extract the essential crack location infor-mation. Finally, the method is verified experimentally. The experimental results show that this method can not only identify axial cracks at the shaft-hole interference fit surface automatically, but also can locate these cracks accurately and obtain both the number of cracks and the distance between pixels.