Research On Identification Of Wear State Of Reciprocating Sliding Friction Pairs Based On Friction Vibration Signals

As one of the main failure forms of equipment,the wear state of friction pair has an important impact on the operation and management of the whole mechanical system.Abnormal wear of friction pair will seriously shorten the service life of equipment,and even cause machine destruction and death.As one of the main output information of tribology system,the friction vibration signal can be obtained in real time during the normal operation of equipment compared with the friction torque,friction coefficient,surface topography of friction pair and wear particle size.Through the real-time acquisition of friction vibration signals,noise reduction and feature extraction,the wear state of friction pair can be diagnosed,so as to provide a basis for the management personnel to take appropriate maintenance or replacement measures,and thus reduce the occurrence of production safety accidents and unnecessary economic losses caused by friction pair wear.In order to measure the friction vibration signals of different wear states,an online and real-time test device was set up to measure the vibration signals of tangential and normal directions during the process of friction and wear of reciprocating sliding friction pairs,and the friction and wear tests under different loads and reciprocating frequencies were designed.According to the variation trend of friction coefficient,the running-in wear of reciprocating sliding friction pair was separated from the normal wear,and the severe wear state was designed by changing the oil injection amount.By analyzing the variation of friction force,surface morphology and statistical parameters of reciprocating sliding friction pair in relative motion,the rationality of the division of three wear states is illustrated.In order to eliminate the influence of noise on feature extraction and recognition,a method combining EEMD decomposition and cross-correlation analysis was proposed to reduce the noise of friction vibration signals.By analyzing the cross-correlation function and the cross-correlation number of homologous signals,non-homologous signal and mixed signal,it is found that the signal homology and correlation contain inherent consistency,so the signal homology can be characterized by cross-correlation analysis method.The experimental results show that,through the method of noise reduction processing,the correlation coefficient of tangent and normal direction vibration signal has significant increased compare to the previous number of mutual relations between the two directions.Compared with the results of denoising using wavelet packet,the correlation number of frictional vibration signals in two directions is increased by about 8%after denoising by the proposed method,which indicates the superiority of the denoising effect of the proposed method.Therefore,the proposed method can be used to de-noise friction vibration signals,and lay a foundation for the subsequent research on feature extraction of friction vibration signals.In order to characterize the nonlinear characteristics of friction vibration signals under different wear states,the multifractal characteristics of friction vibration signals under different wear states were extracted by multifractal downtrend wave analysis algorithm based on fractal theory.The experimental results show that the shape and position of multifractal spectrum of friction vibration signals are different under different wear conditions.Therefore,the nonlinear characteristics of friction vibration signals under different wear states can be characterized by multifractal spectrum,and the sample feature vectors of different wear states can be constructed by the parameters of multifractal spectrum,which lays a foundation for the subsequent wear state identification.In order to recognize different wear states from friction vibration signals,SVM method is used to obtain the optimal classification model.In order to avoid the problem of sample asymmetry and indivisibility,one-to-one method is selected as the concrete solution of the three classification problems in this paper.Under the condition that the kernel parameters are all default values,the recognition accuracy of SVM with different kernel functions in the same sample set is calculated,and the radial basis kernel function is determined to be the optimal kernel function.On the basis of determining the optimal kernel function,this paper discusses the optimization of penalty coefficient and kernel function parameters by using grid optimization method,particle swarm optimization algorithm and genetic algorithm.Experimental results show that the grid optimization method takes the least time and has the highest sample recognition accuracy,which increases from the original 81.1%to 93.3%.Therefore,the grid optimization method is determined to be the optimal penalty coefficient and kernel parameter optimization method.For the generalization of the identification method,the experimental analysis is carried out by using different load test sample sets and different reciprocating frequency test sample sets.The results show that the change of load and reciprocating frequency of test sample sets has little influence on the recognition accuracy,and the recognition accuracy is more than 90%.Therefore,the proposed method can be used to identify different wear states.