A Machine Vision-Based Method for Detecting Surface Hollow Defect of Hot-State Shaft in Cross Wedge Rolling

Abstract

In order to solve the problems of low detection efficiency and safety of artificial surface defects in hot-state cross wedge rolling shaft production line, a machine vision-based method for detecting surface hollow defect of hot-state shafts is proposed. Firstly, by analyzing the high reflective properties of the metal shaft surface, the best lighting method was obtained. And by analyzing the image contrast between image foreground and image background, the most suitable optical filter type in image acquisition was determined. Then, Fourier Gaussian low-pass filtering method is used to remove the interference noise of rolled shafts surface in frequency domain, such as high-light, oxide skin and surface texture. Finally, by analyzing the characteristics of the surface hollow defect area, a defect identification method combining the Otsu threshold method and the adaptive threshold method is proposed to realize the effective extraction of surface hollow defect of rolled shafts. The test results show that the average recognition rate of the method based on machine vision is 95.7%. The results of this paper provide technical support to meet the production requirements of high quality and high performance of cross wedge rolling.