Rolling Bearing Fault Diagnosis Based On Iterative Resonance Sparse Signal Decomposition And Hidden Markov Model

Aiming at diagnosing the early typical faults of rolling bearing, a method based on the combination of the iterative resonance-based sparse signal decomposition (IRBSSD) and the hidden Markov model is analyzed. First, IRBSSD is used on four bearing state (health, outer ring, inner ring and ball failure) of vibration signal for noise reduction. Then, the characteristics in the time domain and frequency domain parameters under different states are calculated separately, using principal component analysis to reduce dimension of feature vectors, fewer features that can retain the original information characteristics of composite index are obtained. And these features make up the characteristic vector. Finally, the HMM can be used to recognize the fault pattern.The main contents discussed are as follow:1. A novel de-nosing method is proposed. The method bases on resonance sparse decomposition iterative to diagnosis the rolling bearing fault early.This method firstly set the initial value of high Q factor according to the component spectrum distribution of the default. Decompose the bearing vibration signal by resonance sparse. Then determine whether a low resonance component needs to be decomposeda second time according to the spectral kurtosis of decomposed high resonance component size. Finally, calculate the envelope spectrum of the low cycle after the termination of resonance components, judge fault type according to the frequency of extreme value point. We use the simulation signal and the measured signal to validate the proposed method respectively. The diagnosis results are compared with the wavelet packet de-noising method. The results show that the method can effectively filter out the noise and preserve signal in the transient impact component of bearing fault signal.2. Calculating feature parameters of the denoised rolling bearing under different status of time domain and frequency domain vibration signal, using the principle component analysis (PCA), converts multiple characteristic parameters for a small amount of comprehensive index, construct new compound instead of the original feature parameters on the basis of keeping the original information as much as possible to achieve the purpose of reducing the number of parameters and calculation.3. The bearing fault pattern recognition based on HMM is studied. Bring the feature vector formed by PCA dimensionality reduced compound feature parameters dimension reduction into the HMM model for training and recognition. The recognition accuracy reached 97.5%. Compare the extracted feature vector of bearing signal without noise reduction processing with the feature vector of denoised signal, the IRBSSD and HMM are more effective bearing the early failure mode. The IRBSSD and HMM are of high applicability, they can be used in the actual bearing system fault diagnosis.