Research On Gearbox Fault Diagnosis And Remaining Useful Life Prediction Of Urban Rail Vehicle

As one of the major public transportations,urban rail transit solves many urban traffic problems,such as traffic congestion,energy consumption and air pollution.It has the advantages of fast and efficient,and bringing convenience to people's lives.As urban rail transit increasingly favored by many people,safety issues can not be ignored.Gearbox is an indispensable part in power-transmission system,the running state directly affects the safety performance of vehicles.Therefore it is necessary to real-time monitoring,analysis for gearbox,timely and accurately grasp the state and predict the remaining useful life.Developing the targeted maintenance plan,decision and prevention of accidents.In this paper,the fault diagnosis and remaining useful life prediction of gearbox are studied systematically.The main contents as follows:(1)Research on feature extraction of gearbox vibration signal.Use local mean decomposition method for signal decomposition,compare with the empirical mode decomposition method,shows that LMD can compensate the shortcomings of EMD.Energy ratio and the correlation between the single component with the original signal as two measurement indexes of effective single component,then extract the features;use principal component analysis to reduce the dimension of the feature,and bring about the multi angle feature fusion.(2)Research on gearbox fault detection based on cyclostationary statistics.The multation of maximum spectral correlation density function value in different characteristic frequencies is used as early fault basis present in this paper,achieve early gearbox fault recognition based on CUSUM theory.In order to solve the problem of noise interference,the spectral correlation density slicing method is used to analyze the fault characteristics of the bearing.The fault detection of actual gear data is performed based on the autocorrelation slice and the spectral correlation density contour.Compared with previous algorithms,the demodulation performance of the cyclic autocorrelation function is verified(3)Research on fault isolation of gearbox based on variable prediction model.Laplasse score method is used as the basis for selecting the variables,and the influence of the eigenvalue on the prediction accuracy of the model is reduced.Use robust regression method to estimate the parameters of the model instead of the traditional least square method.Based on the minimum estimated error value,the optimal model type and order are obtained.The method is compared with others,and the accuracy and efficiency of the variable prediction model proved.Then the sensitivity of different eigenvalues to different faults is analyzed.(4)Research on remaining life prediction method based on co-traning semi supervised.To solve the problem of lack of fault data,considering the non-marked data contains the decline information of gearbox,and taking into account the differences between different methods.Combining with the theory of semi supervised,co-training of PSO-BP neural network and support vector regression.Taking into account the characteristics of the change in the rate of decline,two points as the input of the algorithm is used,and the BP neural network parameters and structure optimization design,and the prediction accuracy is improved.Finally,compared with the previous algorithms,it is proved that the co-traning semisupervised method has higher accuracy and better generalization ability.