A defect detection method of gear end-face based on modified YOLO-V3

Due to the continuous improvement of gear machining quality requirements, the detection of gear face defects has been paid more attention. The traditional sampling method based on artificial experience has been unable to meet the detection needs of manufacturing enterprises. Manufacturing enterprises urgently need an accurate, fast and low-cost detection method. Due to the small surface defects of gear end-face, the complex shape of the gear end-face, the complex processing environment in the workshop and the high requirements of real-time detection, there are great challenges in the detection of gear end-face defects. Therefore, this paper proposes a method for gear end-face defect detection based on improved YOLO-V3. Firstly, this paper proposed a gear image preprocessing method based on gray histogram, it's used to eliminate the unfavorable and unobvious defects caused by the workshop processing environment and non-machined areas. Then, based on the newly designed feature extraction backbone network DWPW-ResNet34 that takes into account both accuracy and efficiency, an improved YOLO-V3 defect detection network is proposed to realize the real-time detection of metal gear end-face defects in the production process.