High Precision Appearance Quality Inspection Of Bearing Inner Ring Based On Machine Vision

Bearing is the basic component of mechanical equipment.With the continuous development of the bearing industry and the improvement of industrial standards,the bearing industry urgently needs to be transformed and upgraded,which puts forward higher requirements for the speed and accuracy of bearing product quality inspection.Currently,the quality inspection of bearing products mainly relies on manual visual inspection methods to complete,and there are problems with the reliability and speed of the inspection.The machine vision systems can replace the human eye to complete the product appearance quality inspection,which have been widely used in machine tools,agricultural machinery,aerospace and other industries.Based on machine vision inspection technology and super-resolution reconstruction ideas,this thesis designed an appearance quality inspection system for the side dimensions and end face black skin defect of bearing inner ring parts,which could effectively improve the inspection accuracy and speed as well as reduce the cost of hardware equipment.The main research contents of this thesis are as follows:(1)The hardware design of a visual inspection system for bearing inner ring was researched.According to the different detection requirements and characteristics of the black skin defect on the end face and the side dimension of the bearing inner ring,a debuggable machine vision detection system was constructed,and the hardware selection such as cameras,lenses,and light sources was completed.According to the different detection requirements,the imaging mode and lighting mode were determined,and the experimental verification was carried out for different imaging principles.Aiming at the distortion problem of the black skin end face image,the camera calibration principle was analyzed,the camera’s internal parameters were obtained,and the distortion correction was completed.Aiming at the requirement of high accuracy of dimension detection,the principle of dual camera mode and telecentric lens were analyzed,and the corresponding positioning plate was designed to facilitate obtaining complete contour images of the bearing inner ring.(2)The algorithm for detecting black skin defects on the end face of the bearing inner ring was studied.The image preprocessing algorithms,including image graying,filtering smoothing,threshold segmentation and morphological processing,were researched and implemented to eliminate the information unrelated to the end face characteristics of the bearing inner ring.The algorithm for extracting the region of interest(ROI)from the end face of the bearing inner ring was studied in detail.Firstly,the contour tracking algorithm based on the connected domain was used to segment the inner wall and the end face of the bearing inner ring,and then the polar coordinate transformation algorithm was used to extract the ROI.On this basis,the extraction and localization algorithms of the black skin defect were studied so that the non-defect features were eliminated by locating and quantitatively analyzing the defect features.The experimental results showed that the proposed black skin detection algorithm based on 5million pixels has the correct rate of 96%,the false detection rate less than 5%,and the missed detection rate basically close to 0%,which can meet the requirements of industrial detection.(3)The sub-pixel level bearing inner ring side dimension detection algorithm was researched.Firstly,the imaging image of the bearing inner ring side was preprocessed.Then,the ROI partition of the side image of the bearing inner ring was completed,and an adaptive ROI extraction algorithm based on improved corner point detection was proposed,which effectively avoided the problem of misjudgment in the edge detection of the bearing inner ring.After then,the pixel level edge extraction algorithm was studied where the pre-experiment was used for comparative analysis and the Canny operator was selected to extract the pixel-level edge of the bearing inner ring.On this basis,a sub-pixel edge extraction algorithm based on Zernike moments was proposed to improve the accuracy of edge localization.Finally,the sub-pixel edge point fitting algorithm based on the Least Squares Fitting was studied,and the methods of the pixel calibration and the interval calculation of field view were completed to achieve high-precision size detection.The final experimental results showed that the standard uncertainty of the dimensional inspection system based on 5 million pixels is less than 0.005 mm.(4)The super-resolution reconstruction algorithm of bearing inner ring image based on deep learning was studied.In order to effectively reduce the hardware cost of machine vision system,the model based on Fast Super-Resolution Convolution Neural Network(FSRCNN)was studied.Firstly,the FSRCNN super-resolution reconstruction model adapted to the image of the bearing inner ring was constructed.Secondly,the highresolution(5 million pixels)and low-resolution(2 million pixels)bearing inner ring image datasets were constructed,and the FSRCNN model was trained.The Peak Signalto-Noise Ratio(PSNR)and Structural Similarity(SSIM)were used to evaluate the model and determine the model parameters.Finally,the images after super-resolution reconstruction were used for experimental verification for black skin defect and dimensional detection.The experimental results showed that the detection accuracy of the 2 million pixels super-resolution reconstruction system is more than 95% in the black skin defect detection of the inner ring end face of the bearing.In the lateral dimensions inspection of the bearing inner ring,the standard uncertainty of the each dimension based on the 2 million pixels super-resolution reconstruction system is less than 0.005 mm.The detection result of the 2 million pixels super-resolution reconstruction system is close to the 5 million pixels vision system,and meets the detection requirements.Therefore,the method can effectively reduce the cost of the system equipment.The bearing appearance quality inspection system designed in this thesis can achieve high-precision and rapid detection of the quality of the bearing inner ring at a lower cost,which can meet the requirements of industrial inspection,and provide a reliable solution for the bearing appearance quality inspection.