ANOMALY DETECTION METHOD OF ROLLER BASED ON VIDEO

Aimed at the problems of low detection accuracy and long manual detection time for abnormal phenomena such as stuck rollers, blockages, and bearing failures of the rollers, an image-based abnormal detection method of the idler is proposed to realize the real-time abnormal detection and early warning of the idler. We analyzed the causes of abnormal rollers, proposed a video based abnormal roller detection scheme, and designed an abnormal roller detection network. The experiments show that the abnormal roller detection accuracy rate of 89.53%, and the detection speed of 103 FPS. Five evaluation criteria and comparative experiments were used to verify the pros and cons of the C3D network abnormal roller detection scheme. Compared with the roller anomaly detection scheme based on sound signals, the detection method based on the C3D network reduces the difficulty and cost of sensor installation and maintenance, and increases the accuracy of stuck roller detection to 94.57%.