Key Technology And Application Of Machine Vision Online Detection For Magnetic Tile Surface Defects

The magnetic tiles made of ferrite permanent magnet material are the key component of permanent magnet motor.Their function is to generate magnetic field instead of excitation windings.In the manufacturing process of magnetic tiles,it is easy to produce many defects such as crack,collapse of block,lack of grinding,eccentric wear,peeling and interlayer.It is very difficult to detect the surface defect of magnetic tiles because of their multiple specifications,large shape difference,space surfaces,low contrast between the defects and the magnetic tile itself,many inspection faces,and fast production rhythm.At present,the automatic detection equipment for surface defect detection of magnetic tiles(SDDMT)that can meet the requirements of production site has not been successfully developed at home and abroad.For a long time,the detection of the SDDMT is mainly based on the artificial visual inspection.The efficiency of this detection method is low,and its rate of false detection and missed detection is high,which leads to the unstable quality of products.Therefore,for the magnetic tiles of production characteristic belongs to small profit per piece and mass production,it is urgent to develop a fast,economical,non-destructive,non-sampling,real-time,on-line detection equipment to ensure the best products.Machine vision inspection has the outstanding advantages of high-speed,accuracy,reliability and non-contact,and plays an important role in improving product quality.In this paper,the key technology of machine vision on-line detection of magnetic tile surface defects is studied according to the technical requirements of SDDMT,and the flexible intelligent detection equipment for surface defects of magnetic tiles(FIDESDMT)is developed.The main research contents and completed work of this paper are summarized as follows:1.The current situation of production technology and equipment technology of domestic magnetic tile manufacturing enterprises was investigated,and the actual situation that the inspection of magnetic tile surface defects was still dominated by manual visual inspection was also investigated.It summarizes the progress and technology trends of machine vision inspection technology,as well as the research status of SDDMT technology.The emphasis and direction of research work is pointed out by hackling the technical difficulties of SDDMT.2.According to the requirements of technical indicators,two prototypes of surface defect detection equipment for magnetic tiles,belt conveyor type and clamping rotary type,were designed and processed successively,and improvement measures were put forward from the aspects of mechanical structure,electrical control system and image acquisition system.The key point of improvement is to improve and design the flexible magnetic tile imaging device and the uniform magnetic tile end surface imaging device by using a flexible magnetic tile imaging light source(FMTILS)and a uniform magnetic tile end surface imaging light source(UMTESILS)designed independently and innovatively.These two imaging devices increase the flexibility of the image acquisition system and significantly improve the imaging quality of magnetic tiles.In addition,illumination mode of light source is analyzed;the focal length and depth of field required by the system is calculated and high-speed image acquisition network and computer control system is built.3.Three kinds of new imaging light sources,such as a uniform magnetic tile imaging light source(UMTILS),are designed and manufactured.Three-dimensional model of the UMTILS is designed through Solid Works.The design idea is that the light emitted from the LED at the bottom of the light source diffuses through the inner surface of the hemispherical reflector and then radiates to the detection plane at the bottom of the light source to form indirect illumination.The light shielding plate can block and absorb part of the light from the upper part of the light source.The light shielding ring can prevent the light emitted by the LED from directly irradiating the magnetic tile and entering the lens.By adjusting the distance between the light source and the magnetic tile and the upper and lower position of the light shield plate,the intensity ratio of the incident light on the chamfer surface and the arc surface of the magnetic tile can be controlled,so that the brightness of imaging surfaces on the magnetic tile is basically the same,and then the uniform brightness image of the magnetic tile can be obtained.Because the light from the upper part of the reflector is blocked and absorbed by the shading plate,a low angle uniform illumination is formed for the magnetic tile.This illumination method accentuates the surface defect feature of magnetic tile,increases the contrast between the defects and the magnetic tile and is conducive to the extraction of defect features.In order to avoid motion blurring,a series of optical calculations,such as the irradiance distribution of the LED ring array,the illumination needed for detection plane and the CCD image plane are carried out to obtain the input power of the light source and the number of LEDs.According to the reserved redundancy and control needs,the number of LEDs is adjusted and the LED light source circuit boards are designed and processed.Lighttools and Zemax are used for illumination simulation and magnetic tile imaging simulation,and the light source is optimized according to the simulation results.Because the heat dissipation area of the existing structure of the light source is not enough,the trigger control mode is adopted to reduce the heating power of the light source.The magnetic tile image acquired by image acquisition is of high quality,the light source works steadily and reliably,and there is no overheated condition.All these fully demonstrate the rationality of the design scheme of the light source.According to the needs of actual detection situations,on the basis of the uniform magnetic tile imaging light source,the FMTILS and the UMTESILS are improved and designed,and good imaging effect has achieved.4.Two kinds of algorithms for the SDDMT based on frequency domain analysis are proposed in this paper.One is the defect detection based on nonsubsampled contourlet transform(NSCT)and adaptive threshold surface,and the other is the defect detection based on nonsubsampled shearlet transform(NSST)and Gaussian scale space(GSS).In the first algorithm,the magnetic tile image is decomposed by nonsubsampled contourlet transform to obtain the decomposition coefficients of each layer,then the decomposition coefficients are normalized in each decomposition subband,and a normalized adaptive threshold surface is constructed to eliminate the decomposition coefficients representing texture and noise in NSCT.Finally,the modified coefficients are reconstructed to obtain the image which eliminates texture features and retains the magnetic tile defects.The experimental results show that the accuracy of the method reaches 95%,which has a good detection effect,but the detection speed is low.In the second algorithm,the original image of the magnetic tile is decomposed by nonsubsampled shearlet transform to obtain the low frequency decomposition coefficient and the high frequency decomposition coefficient.The noise and texture in the two coefficients are suppressed by exponential low-pass filtering and adaptive soft shrinkage criterion respectively.The magnetic tile image after eliminating noise and texture is reconstructed,then the non-uniform background is eliminated by the Gaussian scale space(GSS).Finally,the adaptive threshold method is used to completely extract the defects from the image with uniform background.The experimental results show that the accuracy of the detection method reaches 94.53%,and the average running time is 0.383 s,which meets the actual production inspection requirements of the magnetic tiles.5.The algorithm of magnetic tile surface defect detection based on multi-branch feature fusion convolution neural network is studied.Based on VGG-16,a novel deep convolution classification network IVGG is constructed.The Accuracy of surface defect detection and classification of magnetic tiles for IVGG is 0.9952,which is better than VGG-16,Alex Net,Res Net18 and Goog Le Net.An 8-branch MBIVGG convolutional neural network is constructed based on IVGG network.Each branch extracts the features of the corresponding input image,and the important features are highlighted through Convolutional Block Attention Module(CBAM);At the same time,the correlation features between multiple images are obtained to improve the accuracy of surface defect detection through the correlation analysis of each branch feature.In the prediction experiment,the correct rejection rate of MBIVGG model is0.33%,the false acceptance rate is 0.83%,and the classification accuracy rate is99.33%.6.The software system of FIDESDMT is developed and improved.The software system mainly includes control module,image processing module and cloud function module.The control module mainly controls the light sources,cameras,camera motion control mechanisms,conveyor belt and sorting device.In the image processing module,non subsampled shearlet transform and Gaussian scale space(NSSTGSS)algorithm and MBIVGG deep convolution neural network are used to complete the detection and classification of magnetic tile surface defects.The cloud computing architecture of the detection system is builded in the cloud function module,the cloud service functions of the devices and an application based on Android system is developed.The Operators can remotely manage,debug and maintain the production equipment on site through mobile phone client.During the development of the FIDESDMT,from scheme design to experimental prototype,and then through improvement to trial products,five invention patents with independent intellectual property rights have been obtained through the research on the key technology of the SDDMT based on machine vision.The field test results show that the FIDESDMT can adapt to the surface defect detections of various types of magnetic tiles.The detection results are accurate and the speed of detection meets production requirements,which verifies the feasibility and practicability of the detection equipment.