Research On Weak Fault Feature Extraction And Intelligent Diagnosis Method Of Axle Box Bearing Based On Sparse Representation

The rolling bearing is affected by noise interference and random sliding of rollers,which makes it difficult to diagnose its faults.Axle box bearing is one of the core rotating parts of high-speed train running gear,which plays a vital role in the operation safety of high-speed trains.The axle box bearing not only has the difficulty of fault diagnosis of ordinary rolling bearings,but also has a new level of difficulty in fault monitoring and diagnosis due to its unique engineering background,such as complex operating conditions and the influence of external geographical environment,line conditions,variable load,wheel rail interaction and other factors.Based on the theory of sparse representation,this paper deeply studies the strong background noise interference of axle box bearing during high-speed operation,the high amplitude impact interference caused by rail shortwave disease,and the periodic interference caused by wheel tread failure.The main contents include:(1)Under certain conditions,the weak faults of axle box bearings are classified.The fault experiments are carried out using the railway bearing comprehensive test bench and the high-speed train single axle rolling vibration test bench.The collected experimental signals are preliminarily analyzed.The effects of speed,load,transmission path,and wheel rail excitation on the fault diagnosis of axle box bearings are studied.It is found that the increase of speed will lead to the increase of the vibration amplitude of the faulty bearing,and the vibration noise will also increase.The increase of load has a greater impact on the amplitude of time domain signal and a smaller impact on the energy distribution in frequency domain.The wheel rail excitation will increase the difficulty of fault feature extraction of axle box bearing.(2)In order to solve the problem of strong noise interference in axle box bearing vibration signal and high amplitude impact interference caused by rail shortwave disease,a method based on pre-identification singular value decomposition is proposed to filter the signal and reduce the noise.Through the analysis of simulation signals and experimental signals,it is found that compared with traditional methods,this method can effectively identify the high amplitude impact caused by rail shortwave diseases,and has a good noise reduction effect.(3)An index guided sparse representation diagnosis method is proposed to solve the problem of adaptive fault feature extraction of axle box bearings under variable load conditions.The frequency domain correlation kurtosis index is introduced into the K-SVD dictionary learning process,which can enhance the repeated transient impact of the reconstructed dictionary,make it more consistent with the fault characteristics of the bearing,and strengthen the sparsity of the reconstructed signal.Through the analysis of simulation signals and experimental signals under variable load conditions,it is found that this method has obvious advantages over other methods under variable load conditions.(4)An auto correlation sparse representation method is proposed to extract the fault features of axle box bearing with tread fault,which solves the problem that the frequency source cannot be effectively identified when the frequency domain of wheel tread fault and the harmonic generated by wheel rail interaction are consistent.Through the analysis of simulation signals and bench test signals,the method can effectively solve the problem of multiple source faults.(5)An intelligent identification method of axle box bearing fault based on sparse subspace clustering of reduced dimension graph is proposed,which makes up the deficiency that bearing parameter information and speed information need to be known in advance for fault diagnosis based on frequency domain information.By analyzing the simulation data and the axle box bearing fault data collected from the single shaft rolling vibration test bed,the superiority of the method in identification accuracy and clustering compactness is verified.By studying the noise reduction method of axle box bearing vibration signal filtering,the adaptive fault feature extraction method and the intelligent diagnosis method without prior knowledge are developed to solve the problem of multi-source fault diagnosis under variable load conditions.The research results will provide a solid theoretical basis for condition monitoring and precise diagnosis of train axle box bearings,and have certain theoretical significance and practical application value.