Research On Bearing Roller Blemish Detection Technology Based On Improved Retinex Algorithm

During the “12th Five-Year Plan” period,the state clearly pointed out bearings are the cornerstone of the 16 key fields to boost the revitalization of the equipment manufacturing industry,it has a wide range of applications in a number of key fields.As one of the bearing core parts,the surface quality directly determines the quality of the bearing.In recent years,with the vigorous development of machine vision technology,scholars have applied machine vision technology for the detection of bearing roller surface blemish.However,in the process of detecting bearing roller blemish,the uneven illumination leads to false or missed detection which further causes a reduction in the accuracy of the system detection.In response to this problem,the thesis presents a model to build an image acquisition module of the blemish detection system through the selection of industrial cameras,lenses and illumination sources.The outline of the research which is carried out based on the actual project is given as follows:(1)This thesis analyzes the traditional Retinex illumination correction algorithm and proposes an improved Retinex illumination correction algorithm based on median filtering for the detail loss in the detection of bearing roller blemish.The improved algorithm uses median filtering instead of Gaussian surround function to estimate the illumination component of the image,and information entropy function to detect the illumination uniformity of the original image,in order to reduce the redundancy of the algorithm in practical applications.The simulation results show that compared with the conventional Retinex illumination correction algorithm,the modified algorithm greatly improves the detail loss of the traditional Retinex illumination correction algorithm on the premise of the similar processing time.(2)Based on the illumination correction algorithm,a set of bearing roller blemish detection algorithm is designed,and the algorithm is implemented by multi-threading technology and multi-station result fusion.The bearing roller blemish detection algorithm has five steps,ROI positioning,illumination correction,pre-processing,suspected blemish positioning and blemish confirmation.In addition,thesis performed CUDA acceleration on the anisotropic filtering algorithm that affects the processing speed and improved the algorithm speed by nearly 20 times.Finally,the detection speed of the bearing roller blemish detection system reached 400~500ms/piece,which is satisfied in the actual project for the requirements of bearing roller detection speed.Through field testing,the bearing roller blemish detection system based on the illumination correction algorithm can avoid false and miss detection compared to the system without illumination correction.Furthermore,this thesis proposed model has faster detection speed and more stable of accuracy than the manual detection method.