Research On Fault Diagnosis Of Rolling Bearing Based On Hmm

Rolling bearing, a vulnerable component, is widely used in rotary machine. So it leads to a high frequency of failure of rotary machine. The failure of rolling bearing shall create abnormal vibration and noise; destroy the machine, or even worse cause personal injuries and other serious accidents. Therefore, the research on diagnosis and monitoring of fault roller bearing is significant.Therefore, the thesis is used a pattern recognition technology-Hidden Markov Model (Hidden Markov Model, HMM), which is developing rapidly in voice recognition technology field in recent years, to detect and diagnose the rolling bearing faults. The characteristics of HMM is stronger pattern classification ability, fewer training samples, quicker computing speed. It is suitable for rolling bearing fault vibration signal analysis which is non-stationary and repeated reproducible poorly. The basic method of HMM is to extract features through vibration signal, train Hidden Markov Model of corresponding status number, and then calculate the probability of similar signals that to be detected, according to the maximum in the similarity probability and signal condition of judged corresponding model, to achieve the signal pattern classification purpose.The major work of this thesis is to use HMM to diagnose the fault of rolling bearing, including four research themes:1. On the basis of HMM theory's study and research, the thesis is discussed the practical function of three classical types of HMM algorithm in fault diagnosis, and developed a MATLAB-based implementation procedures to verify the feasibility of HMM theory in rolling bearing fault diagnosis. According to the results of wavelet analysis, author uses standardization and vector measures to classify the eigenvalues, and brings into discrete HMM to train and judge. 2. For the characteristics of rolling bearing are vibration signals with non-stationary, modulation, and vulnerable to background noise, the thesis is used wavelet transform to decompose the vibration signal to extract one-dimensional signal in low frequency coefficients as the eigenvalues. The value of the extracted feature vector quantization according to VQ (Vector Quantization) theory to its transformation, or using the Gaussian probability density function correction HMM parameters, and finally to discriminate input HMM. The experiments are proved that extracted the features by wavelet analysis combined HMM is feasible in practice. According to the discrete HMM and continuous HMM, combining with experimental data, author analyzes the advantages and disadvantages of the two HMM in fault rolling bearing diagnosis, and provides a new line of further research.3. Using the new method of non-stationary signal analysis-Hilbert-Huang transform to extract fault eigenvalue, combined with discrete Hidden Markov Model to identify the status of rolling bearing. As the eigenvalue extracted by HHT is simple, the speed of HHT training and diagnosis is up, ultimately increased the diagnostic accuracy.4. On the basis of above theories, the thesis is used MATLAB GUI to design a virtual system to extract and diagnose the fault features of rolling bearing, and used simulation and experimental signal to test the validity and practicability of the system. The thesis provides an example of diagnosis system of fault rolling bearing.