A Fault Diagnosis Approach for Rolling Bearing Based on Wavelet Packet Decomposition and GMM-HMM

Considering frequency domain energy distribution differences of bearing vibration signal in the different failure modes, a rolling bearing fault pattern recognition method is proposed based on orthogonal wavelet packet decomposition and Gaussian Mixture Model-Hidden Markov Model (GMM-HMM). The orthogonal three-layer wavelet packet decomposition is used to obtain wavelet packet decomposition coefficients from low frequency to high frequency. Rolling bearing raw vibration signals are firstly decomposed into the wavelet signals of different frequency bands, then different frequency band signals are reconstructed respectively to extract energy features, which form feature vectors as the model input of GMM-HMM. A large number of samples are trained to get model parameters for different bearing faults, then several groups of test data are adopted to verify GMM-HMMs so different fault types of rolling bearings are recognized. By calculating the current state appearance probability of monitoring data in GMM-HMMs, different failure patterns are recognized and evaluated from the maximum probability. Similarly, we establish GMM-HMMs for different grade fault samples and evaluated the performance degradation state. Test results show that the proposed fault diagnosis approach can identify accurately the fault pattern of rolling bearings and evaluate performance degradation of bearings.