Early Warning And Diagnosis Methods Of Train Bearing Faults Under Complex Conditions

Research on fault diagnosis and early warning technology of rotating machinery system is of great significance to ensure the safety and reliability of mechanical equipment and prevent accidents.In this paper,the vibration signal of the key components of the rotating system is taken as the research object to study the signal processing methods such as wavelet packet decomposition,empirical mode decomposition,spectral kurtosis,visibility graph,spectral coding,and their applications in feature extraction under early weak fault,strong background noise and variable conditions.On this basis,combined with incremental support vector data description,K nearest neighbor,multilabel k-nearest neighbor and semi-supervised classification are used to realize fault warning and fault diagnosis in complex scenes.The main contents of this paper are as follows:(1)Based on the characteristics of various signal processing methods,an open framework for feature extraction of vibration signal of rotating machinery is proposed,which provides a new idea for constructing the feature extraction method of vibration signal of rotating machinery.(2)In view of the difficulty in extracting the early weak fault of rolling bearing,an early fault feature extraction method based on wavelet packet filtering and spectral coding is proposed.In this method,kurtosis index is introduced into the selection of wavelet packet decomposition components to chose the component containing fault information,and then the square envelope spectrum of several signals is averaged to further suppress the influence of random components.Then,the amplitude of rotation frequency and fault characteristic frequency is selected to form the feature vector.Aiming at the problem of fault identification with only normal samples,an online state identification method based on incremental support vector data description is proposed.The experimental results of bearing life cycle show that the feature extraction method based on wavelet packet filter and spectral coding can effectively extract bearing early fault features,and the model based on incremental support vector data description can adapt to bearing early fault warning with different life length.(3)In order to solve the problem of fault feature extraction caused by strong noise,a feature extraction method combining empirical mode decomposition filtering method and spectral coding based on tile coding is proposed.The index of kurtosis and correlation coefficient is introduced into the selection of components to achieve the purpose of selecting faulty components.The square envelope spectrum of several selected signals is averaged to further suppress the random components in the spectrum.Then,the spectrum coding method based on tile coding is used to encode the spectrum,which can preserve the spectrum information as much as possible and reduce the data dimension.Experiments under different signal-to-noise ratios show that the method can effectively extract fault features under strong noise background.(4)Aiming at the problem of difficult feature extraction caused by variable operation conditions,a visibility graph feature(VGF)extraction method with natural insensitivity to variable operation conditions is proposed.The vibration signal is transformed into visibility graph,and the relevant visibility graph features are extracted as the features of vibration signal.The simulation results show that the method is robust to variable operation conditions.On basis of VGF,to deal with imbalanced unlabeled data,graphbased rebalance semi-supervised learning(GRSSL)for fault diagnosis is proposed.In GRSSL,a graph based on a weighted sparse adjacency matrix is constructed by the knearest neighbors and Gaussian Kernel weighting algorithm by means of unlabeled samples.Then,a bivariate cost function over classification and normalized label variable is build up to rebalance the importance of labels.Finally,the proposed VGF-GRSSL method was verified by data collected from Case Western Reserve University Bearing Data Center.The experiment results show that the proposed method of bearing fault diagnosis performs effectively to deal with the imbalanced unlabeled data under variable conditions.(5)Bearing faults and rotor imbalance exist at the same time frequently,leading to misdiagnosis.To deal with this challenge,this paper presents a novel signal processing scheme: spectral-kurtosis-based feature extraction in conjunction with log-tile coding(SK-SA-LTC)for concurrent fault detection.In this scheme,(a)spectral kurtosis,as a useful signal processing algorithm is used to extract components with max kurtosis,which contains faulty bearing information.(b)Spectral linear interpolation and spectrum average based on the Squared envelope spectrum(SES)is used to suppress random component in the spectrum.(c)LTC is used to reduce dimensions of features with the waveform characteristics the spectrum.(d)KNN are used to verify the availability of features extracted by the proposed method.Finally,the proposed method was applied to fault diagnosis of a concurrent fault case consist of bearing fault and rotor imbalance.The results prove that the proposed method is effective in processing concurrent fault.