Research On Fault Diagnosis Method Of Rolling Bearing Based On Hilbert-Huang Transform

Rolling bearing is widely used in rotary machine and liable to be damaged. A lot of faults of rotary machine are related to rolling bearings. Generally, rolling bearing faults lead to abnormal vibration and noise of machines, even cause the damages of machine and casualties. Therefore, it is very important and significant to research fault diagnosis methods of rolling bearing.Vibration signals of faulty rolling bearing have non-stationary characteristics, and there usually are few fault samples in practice. In rolling bearing fault diagnosis, the diagnosis results are not ideal by using traditional feature extraction methods and fault pattern recognition methods based on neural network theory. Hence, in this paper, a new non-stationary signal analysis method, Hilbert-Huang Transform (HHT) is applied in rolling bearing fault diagnosis. The methods of feature extraction and fault diagnosis were researched, based on morphological filtering, two-dimension spectrum entropy and Support Vector Machine (SVM). The present study based on HHT is involved in feature extraction methods, pattern recognition methods and fault feature extraction and diagnosis system. This research focuses on theoretical investigation, simulation, experiments and prototype system implementation. The main investigation contents are given as below:(1) The HHT theory was in-depth researched. A comparative analysis of HHT and other representative time-frequency analysis methods, such as short time Fourier transform, Wigner-Ville distribution and wavelet transform was done. The conclusion is deduced that HHT is valid and superior in the field of non-stationary signal analysis. An improved method based on endpoint extreme point estimate extending was proposed for the end issue of EMD. An improved method based on generalized morphological filtering and correlation coefficients was proposed for the mode fission and pseudo modes of EMD because of noise. Simulation and experiments were carried out to demonstrate the effectiveness of the two proposed methods. These two improved methods enhance the application of HHT in rolling bearing fault diagnosis.(2) Vibration signals of faulty rolling bearing are non-stationary, high frequency modulation and susceptible to background noise and low frequency harmonic interferences. Therefore, a fault feature extraction method was proposed, based on EMD and adaptive morphological filtering demodulation. The high frequency modulation signal was separated by using EMD, and it was demodulated by kurtosis-based morphological filtering. Finally, fault features of non-stationary signals were extracted. The results of simulation and experiments of inner ring and outer ring defects show that the fault features are effectively extracted by using this proposed approach. The proposed approach is superior to Hilbert-based envelope method. It will have a good application prospect for fault feature extraction of rolling bearing.(3) Through researching on the two-dimension spectrum entropy of signals, Genetic Algorithm (GA) and SVM, a fault feature extraction method based on EMD and two-dimension spectrum entropy was proposed. A classification method based on GA and SVM (GA-SVM) was proposed. Fault diagnosis experiments of rolling bearings were carried out. The results show that it has a higher diagnostic precision by using these two methods. It provides a new train of thought for rolling bearing fault diagnosis.(4) Based on theoretical investigation of the fault feature extraction and diagnosis methods, a fault feature extraction and diagnosis prototype system of rolling bearings was developed by Matlab. The results of simulation and experiments show that this system is effective and practical. It gives a paradigm for developing the rolling bearing fault diagnosis system.