Surface pitting detection of ball screw based on deep learning

In view of the fact that most of the existing research on the pitting failure of ball screw focuses on the vibration signal experiment and the establishment of the degradation model, and the use of more intuitive visual aspects is less, the deep learning is studied in the pitting detection of the surface of the ball screw. The application of Faster R-CNN and Mask R-CNN two network models were built, and the two were compared and analyzed through experiments. The results show that both Faster R-CNN and Mask R-CNN can guarantee high classification accuracy under different learning rates, and both can excellently complete the detection task of pitting on the surface of the ball screw. While locating the eclipse, the mask of the pitting is output synchronously, which is helpful in the face of the subsequent task of describing the size of the pitting, and has more advantages in the face of small-area pitting, and there are fewer missed detections.