Bearing Fault Diagnosis Based On Vibration Signals Nonlinear Method

As one of the more easily damaged parts, rolling bearing is widely used in rotating machinery, which directly affects the operating condition of the mechanical system. But bearing fault signals which are generally nonlinear, non-stationary and subject to random noise interference, is difficult to be detected.According to this condition, a wavelet denoising method based on soft/hard compromising threshold, a novel method of time-frequency analysis, Hilbert-Huang Transformation (HHT) and the comparatively recent development of pattern recognition techniques, Support Vector Machines (SVMs), are combined and applied to the rolling bearing fault diagnosis in this thesis. The HHT analysis consists of two parts, the Empirical Mode Decomposition (EMD) and Hilbert spectral analysis, according to which two different methods for bearing fault diagnosis are proposed. One method is the extraction of fault feature frequency based on Hilbert margin spectrum. In this method, the collected signal de-noised by wavelet is decomposed by the self-adaptive EMD. Then the Hilbert marginal spectrum is obtained by Hilbert transformation, from which the bearing fault frequencies are extracted for the fault diagnosis. Another is a combined intelligent diagnosis based on IMF energy feature extraction and support vector machines (SVMs). In this method, the fault signal is processed by the wavelet de-noising and EMD and a number of inherent states (IMFs) components are obtained. Then the energy feature vectors are extracted from the IMFs, as inputs of the support vector machines for pattern recognition.The emulation results from bearing vibration signals with Matlab show that the combination of HHT analysis and the wavelet de-nosing can extract the wake rolling bearing fault features more effectively, compared to the simple HHT analysis for diagnosis. When SVM is combined with the EMD-based feature parameters, the rolling bearing working conditions and fault patterns can be identified intelligently and accurately in the case of small samples.