Audio Hmm-based Bearing Fault Diagnosis Research

Rolling bearing is the most widespread application of rotating machinery, by the status of which, the normal performance of the whole machine is influenced directly. On the basis of combining theory with practice, bearing fault diagnosis by acoustic signals based on Hidden Markov Model (HMM) is researched systematically in this paper. The details are studied as follows:First, plentiful significant information about the operation status of bearings, which is potential for the fault diagnose after processed properly, is contained in their acoustic signals. And acoustic signals can be collected with non-contact sensors, so as to be convenient and cheap. Therefore, taking acoustic signals as object, and according the characteristics of MFCC with human auditory, the Mel-Frequency Cepstrum Coefficient (MFCC) is applied to the bearing acoustic signals for the first time.Second, because all parameters are discrete values, Discrete Hidden Markov Model (DHMM), simple model, fast speed, and high diagnosis accuracy is applied to the acoustic signals emitted by bearing in this paper. Experiments results prove that, with an average training time of 55.8s, diagnosis time of 0.031s, and diagnosis rate of 90%, the presented method is effective.Third, because output probability is described logically by Continuous Gaussian Mixture density function, Continuous Gaussian Mixture Hidden Markov Model (CGHMM) is applied to the acoustic signals emitted by bearing in this paper. At the same time, the algorithms of training and diagnosis are improved by model parameters initialization based on clustering algorithm and forward-backward algorithm based on scaling coefficients. Experiments results prove that, the improved algorithm in this paper, with an average training time of 10.941s, diagnosis time of 0.028s, and diagnosis accuracy of 97.5%, is superior to the normal algorithm, with an average training time of 110s, diagnosis time of 0.810s, and diagnosis accuracy of 96.3%.Forth, the diagnosis experimental results of DHMM and CGHMM are compared. DHMM is faster in speed than the normal CGHMM, but lower in diagnosis accuracy. Using this improved algorithm of CGHMM, the time of operation is reduced greatly, and the average training time is less than a quarter that of DHMM, the diagnosis time is also less, and the diagnosis accuracy is improved obviously.Five, Developed in Visual C++ 7.0, diagnosing platform for acoustic fault HMM-based is designed, on which experiment is carried out, so as to test the effectiveness of the method in this paper which has good application value.