Research On Two-dimensional Detection Method Of Powder Metallurgy Gear Defects Based On Machine Vision

Gears are widely used components in mechanical products,and their machining accuracy is one of the key factors that determine the accuracy of equipment operation.At present,the processing and manufacturing processes of gears include:milling,grinding,hobbing,shaving,powder metallurgy and other processes.The gears produced by the first four manufacturing processes are generally used in high-precision high-end equipment,and the required manufacturing equipment is relatively expensive,generally in small and medium-sized batch production.However,the latter powder metallurgy process gears,because of their low production cost,and the production scale are mostly mass-produced,and the products in the fields of automobiles,motorcycles and the like are widely used,and the quantity far exceeds the number of gears produced by other processes.However,the powder metallurgy process gear has low cost and simple processing technology.Compared with other high-end gears,it has low precision and lacks high-efficiency detection means.This will undoubtedly reduce the service life and increase the risk of use for many operating equipment using powder metallurgy process gears.Therefore,this thesis takes the three defects of powder metallurgy gear as the research object,and studies the two-dimensional detection method of powder metallurgy gear based on machine vision.The main contents of this thesis are as follows:1.Analyze and build the corresponding machine vision software and hardware environment.Mainly from the choice of light source,lighting angle,camera lens selection and other aspects have been considered.The corresponding filtering methods are studied,mainly from the median filtering,Gaussian filtering,mean filtering and bilateral filtering processing methods.The conclusion that the bilateral filtering effect is more significant is obtained.This is used to detect noise and other interference problems caused by the gear production environment or the shooting environment.2.Analyze the effect of global equalization and partial equalization.The sample was enhanced with histogram equalization.Two kinds of segmentation algorithms based on threshold and gradient edge are studied.Among them,eight common segmentation methods are mainly explored.The comprehensive time efficiency and effect show that the Sobel operator has better segmentation effect.Finally,the morphological corrosion operation of the segmented gear image is carried out,and the ideal segmentation effect is further obtained.3.Feature extraction is performed on the segmented image.In this thesis,geometrical features(perimeter,area,inner hole area,circumscribed width,height)are extracted,and color features(H,I,S,Y,Cr,Cb)are extracted,and texture features(entropy(ENT),energy(ASM),moment of inertia(CON)and correlation(COR),and show the data extracted by some samples.A PCA-BP principal component analysis method is proposed for color texture features,and the original 15 features have been changed.5 main features.4.The geometry,color and texture features of the gear image are identified by three kinds of classifiers:random forest,BP neural network and support vector machine.For geometric features,the accuracy of random forest recognition reaches 0.76,BP neural network identification.The accuracy rate only reached 0.5,and the support vector machine reached 0.92.For the color texture feature recognition wear,scratch and crack characteristics,the average recognition rate of random forest reached 0.91,the improved PCA-BP reached 0.95,and the multi-classified support vector machine reached 0.92,because the primary factor affecting the finished product of the gear is its geometric size feature,this thesis chooses the support vector machine as the geometrical feature classifier model,and at the same time,for the identification of other surface defects,PCA-BP is selected as the classifier model.