The Research On Fault Identification Based On Gaussian Mixture Model And Subspce Methods

The process of mechanical fault diagnosis is the process of fault mode identification. The application of pattern recognition methods in mechanical fault diagnosis must combine with the specific characteristics of fault signals. The fault signals are nonlinear and non-stationary and these characteristics more and more depend on the fault level. So the research on the pattern recognition algorithm suitable for fault signals is important to improve the fault discrimination and recognition efficiency.The vibration and acoustic signals were research objects in this paper. With the gaussian mixture model (GMM) and the sub space method, the correlation algorithms of fault mode identification were researched and the feasibility and availability of GMM and sub space method used in fault mode identification were analyzed. Then, all the proposed identification algorithms were compared. The main works can be summed up as following:1) The noise reduction algorithm and its application based on phase space reconstruction and local independent component analysis.According to the principle of local projective noise reduction algorithm, the phase space reconstruction and independent component analysis (ICA) were combined to the noise reduction. According to the experiment of simulation signals, the noise reduction effect of proposed algorithm as well as the influence of three kinds of phase space reconstruction methods were analyzed and compared. The noise reduction effect of proposed method is superior to globe projective.Moreover, according to the fault characteristic of low-speed rolling bearing, a method that integrating proposed algorithm and resonance demodulation technique was introduced. By the proposed method, the rolling bearing fault of a converter trunnion was detected. The diagnosed result is consistent with the fact.2) Fault mode identification algorithm based on phase space reconstruction and GMMA new method to pattern description of fault signals that integrated phase space reconstructed and GMM was introduced. Then, the undetermined signals were classified by Bayes classifier. The method was used to the classification of vibration and acoustic signals. The results show that the gear's fault modes can be identified exactly and the noise disturbance can be overcome by proposed algorithm. Moreover, the parameters selection method of GMM was introduced after the influence of parameters to the discrimination.3) Fault recognition algorithm based on quartile deviation fractal dimension and GMMAgainst the characteristics of nonlinear and non-stationary of fault signals, the feasibility of fault recognition by detrend fluctuation analysis (DFA) was analyzed. The quartile deviation fractal dimension (QDFD) and the intercept that was got when the QDFD was calculated were composed to the eigenvector to the fault mode recognition.According to the characteristics of vibration signals, an improved method that a pre-processing to the fault signals by range standardization before the QDFD was calculated was introduced. The method decreased the sensitivity to signals'amplitude and distribute. The proposed method was combined with GMM to the recognition of gear's faults. The results show that the stronger robustness, better accuracy and faster operating speed and could be got by proposed method.4) Dimensionality reduction and fault recognition algorithm based on local kernel principal component analysis and GMMThe dimensionality reduction and fault recognition algorithm that integrating kernel principal component analysis (KPCA) and GMM was introduced. The proposed method was used to the identification of gear's fault signals. The result shows that the dimensionality reduction effect of proposed method is better than the fuzzy principal component analysis (FPCA) and the discrimination is better than the FPCA and neural network.