Research On Fault Diagnosis Method Of Parallel-Axis Vehicle Electric Drive Axle Based On Local Mean Decomposition

With the increasing number of vehicles,various types of vehicle faults are becoming more and more prominent and need to be paid attention to.The load on drive axle is complicated,and parts will inevitably deteriorate after long-term work,which will easily lead to faults and affect the safety and use of vehicles.Therefore,it is necessary to study the fault diagnosis method of drive axle.Modern signal processing methods can measure and analyze various types of signals,and judge fault phenomenon by extracting signal characteristics,which is very suitable for the field of vehicle fault diagnosis.In this paper,a parallel-axis vheicle electric drive axle is taken as an object,and improved methods of Local mean decomposition（LMD）is studied.It is combined with Support Vector Machine,SVM)and applied to the fault diagnosis of electric drive axle to realize the intelligent identification of five states: tooth surface peeling,gear tooth breaking,bearing inner ring peeling,bearing outer ring peeling and normal.Firstly,describe the common fault types of electric drive axle,and analyze the modulation characteristics of the vibration signals of gear failure and bearing failure.Then according to the existing parameters,a complete three-dimensional model of the electric drive axle is established using Pro/E,and the accuracy of the model was verified by ADAMS dynamic simulation.Then use Pro/E to simulate four types of failure: gear spalling,teeth broken,bearing inner ring spalling and bearing outer ring spalling.The corresponding vibration signals are obtained through ADAMS dynamic simulation for subsequent experiments.The signal processing effects of STFT,WT,WVD,HHT,LMD are compared,and the LMD with the best effect is selected for fault diagnosis of automobile electric drive axle.Analyze the signal decomposition principle of LMD.For the shortcomings of LMD,corresponding improvement methods is proposed: a four-point waveform continuation method based on matching error is proposed to improve the endpoint effect.A piecewise rational spline interpolation method is proposed to improve the envelope accuracy.An adaptive filtering termination condition is used to determine the screening time of LMD decomposition.Combine the comprehensive feature index with the Kmeans clustering algorithm to screen "sensitive components".Study the noise reduction effect of the singular value difference spectrum as a pre-processing method before LMD decomposition of signals.The effectiveness of the improved method is verified through MATLAB simulation.The classification principle of SVM is analyzed.Multi-scale permutation entropy is selected as the feature vector of the input to SVM,and the influence of different parameters on multi-scale permutation entropy is studied.The improved LMD was used to decompose the fault signal of the electric drive axle and 200 sets of feature vectors are calculated.The particle swarm optimization（PSO）algorithm is used to optimize SVM,and the classification results of SVM,PSO＿SVM and artificial neural network are compared,all of which can realize intelligent identification of electric drive fault types,while PSO＿SVM has the highest classification accuracy.In addition,the actual bearing fault data was selected for repeated experiments,which also realizes intelligent identification of fault types.The results show that combining improved LMD with PSO＿SVM is an effective fault diagnosis method which can be applied to the fault diagnosis of automobile electric drive axle.