Research On Weak Fault Feature Extraction And Performance Degradation Evaluation Of Railway Axle Box Bearing

As railway transport in China is moving towards higher speeds and greater loads,while the density of the line network is also increasing.Due to the harsh working environment of the axle box bearings,they are extremely susceptible to failure.If a fault is not detected in time,it will lead to an emergency stop,thus affecting the operation order of the main line,or even cause a major accident that destroys the train,resulting in serious adverse social impact.Therefore,axle box bearing service condition monitoring and fault diagnosis is of great practical significance to ensure the safety of traffic and improve operational efficiency.In contrast to mechanical equipment installed in fixed positions,the rapid relative motion between train and track results in that the vibration signals collected on axle box bearings include not only the health information of the bearing itself,but also the vertical/transverse excitation force of the wheel and rail,the vibration caused by the longitudinal traction/braking force and adjacent components,etc.The complex signal composition,the weakness of the early failure characteristics,the non-smooth operating conditions,the signal transmission path between the failure site and sensor bring great challenges to early fault diagnosis and performance degradation assessment.In order to improve the early fault diagnosis of axle box bearing and the performance degradation evaluation effect within the repair cycle,reduce the risk and operating cost,this paper makes some exploration in the early weak fault feature extraction and the establishment of performance degradation evaluation model.The main research work and innovative achievements of this paper are as follows:(1)Aiming at the problem of difficulty in extracting weak feature information of axle box bearing faults under strong background noise,in this paper,the concept of multiresolution parameter is introduced into the Teager Energy Operator algorithm to remedy this defect.Firstly,construct the Multiresolution Teager Energy Operator(MTEO)by setting multiresolution parameter,And then MTEO is employed to calculate the energy of signal,which can suppress In-band noises and enhance cyclic impact feature hidden in vibrations of faulty bearings in the form of signal energy.Finally,the frequency spectrum of the signal energy is then given to determine the health condition of bearings.The effectiveness of the method is examined by using both synthetic and experimental data of rolling element bearing with different kind of faults.The experiment verifies that the MTEO method can improve the application effect and fault identification rate.(2)In order to solve the Influence of transfer path on extraction of weak feature information for axle box bearing faults,the multi-level feature extraction method for adaptive fault diagnosis of rolling bearings is proposed.This paper uses the maximum second-order cyclostationary blind convolution(CYCBD)to weaken the influence of disturbances,enhancing weak transient shock components.Considering the strict fault frequency component periodicity in the envelope spectrum,the envelope spectrum multi-point kurtosis(ESMK)is proposed as a metric,and used the grey wolf optimization algorithm(GWO)to obtain an adaptive sparse decomposition method(GWO-RSSD)for extracting locomotive bearing transient fault shocks to eliminate the signal effects of high amplitude disturbance shocks and background noise.Through the analysis of simulation signals,self-made rotor-bearing fault simulation test data and axle box bearing data of a certain type of railway locomotive,it is verified that the proposed method can effectively weaken the influence of signal transmission path.(3)Considering that each characteristic parameter has different ability to characterize bearing state or fault degree,a combination of multiple characteristic parameters is considered to improve the generalisation of the model,a new method to predict the performance degradation of rolling bearing based on MSET and SWM was proposed.The average deviation degree is introduced to measure the discrepancy between observation vector and estimation vector.Experimental results demonstrate the effectiveness of the approach which elaborated in this paper in improving the accuracy of health prediction.(4)Reconstruction-based performance degradation assessment models are not sufficiently significant in terms of reconstruction errors when dealing with weak signals to detect early performance degradation.Therefore,an Auto encoder-dictionary learning(AEDL)neural network is proposed for bearing degradation level assessment.The two-layer reconstruction by auto encoder and dictionary learning amplifies the reconstruction error and enhances the sensitivity of the performance degradation index.The proposed method is verified to be effective in enhancing the sensitivity of the performance degradation index and amplifying the reconstruction error by simulating the signal and XJTU-SY rolling bearing data.(5)Considering the importance of timing information in performance degradation assessment,an end-to-end model based on deep residual shrinkage network and long shorterm memory network(DRSN-LSTM)is proposed.DRSN is used to extract local abstract features from the vibration signal,while noise reduction is applied to the signal,and LSTM is used to extract the vibration signal timing information,followed by two non-linear layers to reconstruct the original signal.The analysis of simulated signals,laboratory rolling bearing fatigue test data,and experimental data of an axle box bearing verifies that the proposed method fully exploits the timing information while removing noise from the signal,maximises the restoration of the signal,especially the faint signal of the fault,and accurately assesses the degree of performance degradation.