Research On Fault Diagnosis Of EMU Axle Box Bearing Based On Wavelet Packet Decomposition And FPA-SVM

With the continuous expansion of China’s high-speed rail,people pay more and more attention to its safety.As a key component of the running gear,the axle box bearing bears complex load and poor working environment,which leads to great dispersion of its life.Once the failure is not repaired in time,it will bring great potential safety hazards and affect the driving safety.Therefore,improving the accuracy of fault diagnosis of rolling bearings is of great practical significance to ensure the safety of trains.In this paper,taking the axle box bearing of EMU running part as the research object,a fault diagnosis process based on wavelet packet decomposition(WPD)and support vector machine(SVM)optimized by flower pollination algorithm(FPA)is proposed,aiming at improving the efficiency and accuracy of fault diagnosis.The main work accomplished is as follows:(1)The fault types and vibration mechanism of rolling bearings are studied,the fault characteristic frequencies are analyzed,and the research status of related technologies for fault diagnosis of rolling bearings is summarized.On this basis,the research ideas and contents of this paper are established.(2)The wavelet packet threshold denoising algorithm is studied,and an improved scheme is proposed based on the traditional threshold function.In addition,the selection of basis function,threshold and decomposition layers in the process of noise reduction are studied,and the solutions are given respectively.Through the analysis of simulation data and measured data,the results show that the noise reduction effect is good when using Db10 wavelet basis,improved threshold function,4-layer decomposition and Heursure threshold rule.(3)To solve the problem that the traditional single-domain features are not comprehensive enough for fault description,a mixed feature domain is constructed by extracting 44-dimensional features in time domain,frequency domain and wavelet packet domain.When extracting wavelet packet domain features,the energy distribution features in different frequency bands after wavelet packet decomposition are proposed as fault features,which effectively utilizes the ability of different statistical features to discriminate faults.(4)Aiming at the problem of data redundancy and computational efficiency caused by the high dimension of mixed feature domain,kernel principal component analysis(KPCA)is introduced to carry out secondary feature selection.At the same time,the selection of kernel parameters is analyzed,and feature ratio is introduced to select kernel parameters.The results show that the best kernel parameters selected by feature ratio can improve the dimension reduction effect.(5)The advantages of flower pollination algorithm over other algorithms in convergence rate and iteration rate are analyzed,and a support vector machine model based on optimization of flower pollination algorithm is proposed.By comparing the diagnostic accuracy and efficiency of PSO-SVM and GA-SVM models,the results show that the diagnostic accuracy of FPA-SVM model can reach 100%,and the diagnostic efficiency is the highest.(6)By setting up the simulation transmission test-bed of EMU running gear,the measured data are collected and analyzed.The results show that the fault diagnosis process proposed in this paper has higher fault diagnosis accuracy than other algorithms,which verifies the effectiveness and engineering practicability of the process.51 pictures,26 tables,81 references...