Research On Grinding Gear Surface Roughness Measurement Methods Based On Machine Vision And Machine Learning

Gear Surface roughness is an important index of surface quality of gear,it is closely related to the gear's cooperation properties,wear resistance,fatigue strength,corrosion resistance,which has important influence on the service life and reliability of gear.With the development of precision manufacturing technology,the surface roughness of gear has been paid more and more attention.Meanwhile the precision measurement of gear surface roughness has become a research hotspot.The measurement of traditional contact roughness measurement is time-consuming and linear measurement is not enough to describe the roughness of the whole surface,as well as the problems such as small measurement area and expensive instrument equipment in traditional non-contact measurement methods.In view of the above reasons,in order to measure the roughness of grinding gear surface quickly,this paper proposes a new non-contact measuring method for grinding gear surface roughness on the basis of the combination of machine vision and machine learning.This method has the advantages of high detection efficiency,high precision,non-contact and high cost performance.The research contents of this paper mainly include the following points:(1)The definition,generation,the impact on the gear of gear surface roughness and roughness evaluation parameters are introduced.According to the method of measurement,the general situation of contact measurement and non-contact measurement are described.Then,light scattering mechanism and imaging mechanism of rough surface are analyzed.Finally,the main hardware composition and selection basis of machine vision roughness measurement system are summarized and the hardware and imaging system suitable for this thesis are selected.(2)In view of the weak sensitivity of gear surface roughness to the index S of color distribution statistics matrix(CDSM),based on the index of coincidence degree(S),the clustering index(CI)with more sensitivity to grinding gear surface roughness is proposed by using clustering method of machine learning.The experiment results show that there is a good correlation between grinding gear surface roughness and clustering index(CI),which indicates that the clustering index is more sensitive and more suitable for the evaluation of grinding gear surface roughness.(3)For a more comprehensive evaluation of grinding gear surface roughness,the clustering index(CI),the fitting ellipse eccentricity(E),the arithmetic mean value of gray scale(Ga)and the average power spectrum index(F3)of the image in frequency domain are proposed based on the grinding gear surface image.These indexes form the set of characteristic indexes of gear surface roughness,then the surface roughness is evaluated by the characteristic index set.For the random initialization of the weights and thresholds of BP neural network results in the problem that the training time is too long and the training accuracy is affected,so genetic algorithm(GA)is used to optimize the initial weights and thresholds of BP neural network,and compare it with RBF network model.Experimental results show that the prediction performance of RBF neural network is better than that of BP neural network model,and the prediction results are good.Thus the feasibility of RBF neural network model for the prediction of grinding gear surface roughness is demonstrated.(4)Based on LabVIEW and MATLAB,a prototype of grinding gear surface roughness measurement system based on machine vision and machine learning is developed.The system can complete the whole process of gear surface image acquisition,image processing,feature parameter extraction,and the gear surface roughness prediction,which can provide the necessary software support for the on-line measurement of grinding gear surface roughness.The research work of this paper is a bold attempt on the application of the roughness measurement method based on machine vision and machine learning,which has significant practical application value.