Research On Surface Image Reconstruction Of Mechanical Parts Based On Machine Vision

In recent years,China’s railway construction is very rapid,operating mileage and network density has been greatly increased.Among them,the train bearing is an important part related to the safety of transportation,and the railway accident caused by the bearing fault is one of the important factors threatening the safety of the train,so the railway department has been paying great attention to the maintenance of the bearing fault.At present,the railway department’s maintenance methods for bearings are as follows:Artificial will be half bearing dismantling and cleaning,and then through the eyes,hands and experience to determine degree of defects and defect on the surface of the bearing type,the test results to rely too much on maintenance personnel’s experience and sense of responsibility,and repair work intensity,long operation time,easy to cause visual and physical fatigue,human factors in the test results is more,lead to the result is not accurate.Therefore,in this paper,in order to achieve high quality automatic detection of bearing maintenance work,so as to conduct research on machine vision detection,the main research contents and steps are as follows:(1)Firstly,the hardware of the machine vision detection system is selected,and the gray scale transformation of the collected images is carried out to improve the calculation speed.The bearing defect image containing noise was decomposed into finite intrinsic mode function(IMF)components by 2D-VMD algorithm.Then,the IMF components were screened by using fuzzy linear index and standard deviation.The effective items were selected and the noise items were eliminated to realize image denoising.The mean square error and peak signal-to-noise ratio are used to verify the better denoising image quality obtained by the 2D-VMD algorithm.(2)Secondly,the de-noised image is transformed by cylindrical back projection to realize image distortion correction,which provides a guarantee for subsequent image reconstruction.SIFT algorithm has good stability,large amount of information can achieve fast matching and has good compatibility.Through the analysis of the principle of SIFT algorithm,SIFT feature point extraction and feature matching of image can be realized.However,due to the low matching accuracy of SIFT,only rough matching of image features can be realized.In order to further optimize the matching accuracy and eliminate mismatched pairs,RANSAC algorithm is introduced to achieve accurate matching of image features.(3)Finally,the linear fusion algorithm is used to Mosaic bearing images to realize surface image reconstruction,and the Faster R-CNN algorithm is used to realize surface defect detection of bearing images.The convolutional neural network adopts the ZF NET model,expands the image,establishes the artificial data set BSD and trains the data set to realize the defect detection of bearing image.Compared with the detection results of the traditional detection method Canny algorithm,it is found that the Faster R-CNN algorithm is superior to the Canny algorithm in the detection ability of bearing defects,and has a significant improvement in the detection precision,precision rate,detection time and other indicators of bearing defects.The research results show that the automatic detection system based on machine vision can realize the non-contact detection of train bearings,can efficiently and accurately detect the types of defects,and can intuitively judge the degree of defect damage according to the detection results,which provides a reliable basis for the maintenance work of railway departments.