Gear Fault Diagnosis Based On Recurrence Analysis Method

Based on the complexity and nonstationary characteristics of the gear vibrationsignal,those traditional methods are constrained of low SNR, non-stationarity andunder-sampling. By using Empirical mode decomposition method for signal de-noisingprocessing in this article, the vibration signal is decomposed into a finite number ofIntrinsic mode function components, screening the IMF components containing faultinformation, retain state information of the gear signal. Getting rid of uselessinformation, the EMD method has the characteristics of adaptability, being suitable fornonlinear and non-stationary signal feature.This article uses the RP analysis method toanalyze gear vibration signals after de-noising,and the recurrence nature is one of thebasic properties of dynamics system, through the recurrence nature can reflect thebehavior of the system in phase space.We found that the analysis of gear signal RP,graphical features of RP reflect the state of gear model, according to the distribution ofrecurrence points to distinguish between the gear model. This article puts forwardchoice recurrence points in the figure as a characteristic, combining with the Gaussianmixture model and the Bayesian classifier for gear fault pattern recognition,successfully distinguishing the normal, wear, circular pitch error and tooth broken fourfailure modes of gear vibration signals. Meanwhile,on the basis of the de-noising byEMD, the use of Recurrence quantification analysis method, the RQA by statistic RPfeatures a quantitative analysis method.Through the study of the experimental data,this paper select and determine the DET and the LAM as the characteristic, to gearfault pattern recognition, the classification results compared with the recurrenceanalysis. Finally, the classification results show that: the RP has a higher recognitionrate and greater reliability than RQA, it’s more suitable for gear fault diagnosis.