Research On Fault Diagnosis Method Of Rolling Bearing Based On Vibration Feature Extraction

Rolling bearing is one of the most commonly used parts in mechanical equipment, so the research on monitoring and fault diagnosis of rolling bearings is of great significance. When failure occurs in operation of rolling bearing, its vibrational signal contains abundant fault information, how to use modern signal processing methods to extract fault feature in fault signal has always been a research hotspot in field of fault diagnosis. Time-frequency analysis method is a kind of analysis tool, which is widely used in the mechanical fault diagnosis. With the development of modern signal processing methods, there have been some new time-frequency analysis methods, therefore it has an important value to study. Based on the research of non-stationary time-frequency analysis method, this paper introduces local mean decomposition method, and a method of improved local mean decomposition combined with neural network, support vector machine method is applied. Finally, intelligent fault diagnosis of rolling bearing was realized.The main contributions are as follows:1) Conventional time-frequency analysis methods are studied in application of analyzing non-stationary signal. Based on researches of short-time Fourier transform, Wigner-Ville distribution, wavelet transform and Hilbert-Huang transform, a method of resonance demodulation based on wavelet packet and Hilbert envelope spectrum, a method based on improved Hilbert-Huang transformation have been introduced, with which fault diagnosis of rolling bearings can be realized. The fault features used the latter method were extracted by analyzing Hilbert marginal spectrum, thus fault types of the rolling bearing were determined.2) Research is mainly focused on a new time-frequency analysis method called local mean decomposition, principle of LMD algorithm were analyzed, and made a simulation about calculating process of algorithm. Researches are focused on the method for calculating the instantaneous phase of expansion and instantaneous frequency. The differences between LMD method and EMD method, a simulation analysis of LMD method has done and this method is applied to fault feature extraction of rolling bearings. For a problem of mode mixing in LMD method, a method called ensemble local mean decomposition (ELMD) has been put forward, The simulation and comparison of ELMD and LMD method are studied.3)Ensemble local mean decomposition has combined with neural network, then a fault diagnosis method of rolling bearing based on ELMD and neural network has proposed, ELMD method is as a preprocessor to extract kurtosis coefficient and energy characteristic parameter of rolling bearing to form feature vectors, which are regarded as input of BP neural network to classify rolling bearing faults. Compared with wavelet packet decomposition as a preprocessor, through instance analysis of rolling bearing signal, the results demonstrated the effectiveness and practicability of the approach.4) Ensemble local mean decomposition has combined with support vector machine (SVM), a method of rolling bearing fault diagnosis based on ELMD and least squares support vector machine (LS-SVM) is proposed, Applying ELMD method to extract features to form fault feature vector. Working state and fault type of rolling bearing can be classified automatically, through example analysis of roller bearing, the results show the effectiveness of this method.