Research On The Mechanical Fault Diagnosis Method Based On The Infinite Factor Hidden Markov Model

A kind of method for the hidden Markov model(iFHMM) based on the infinite factor is presented in this paper,which is based on on the new method of mulitiple faults composite diagnosis variational Bayesian independent component analysis in the National Natural Science Foundation, according to the existing hidden Markov model(HMM) and factorial hidden Markov model(FHMM) problems.And the iFHMM is applied to the rotating machinery fault pattern recognition of single channel;Combined with the blind source separation(BSS) analysis of the independent component(ICA) method, the iFHMM is applied to the rotating machinery fault pattern recognition of multi channel and to predict the missing data.The above data are based on the same set of data from the original rotating machine in different fault modes. Meanwhile, Experiments and simulations show the effectiveness of iFHMM.Chapter 1 discusses the proposing and significance of this topic and the development status of pattern recognition technology and several dynamic pattern recognition tools of HMM and FHMM are introduced. On this basis, this section gives the main research contents and innovation of this article.Chapter 2, Due to the idealization of mathematical hypothesis of the HMM and FHMM, and the traditional dynamic pattern recognition tools’ deficiency, As a basis for this, This section discusses the theory and algorithm of iFHMM. Given the structure of iFHMM, This paper focuses on the theoretical algorithm of iFHMM and its derivation process by using a two element matrix,introducing the India restaurant process(IBP) and the section bar structure model(Stick-Breaking)to modeling the iFHMM.Chapter 3,The i FHMM is applied to the rotating machinery fault diagnosis by using the signal processing ability of the dynamic pattern recognition model of iFHMM,And a method of the rotating machinery fault diagnosis base on iFHMM is proposed in this section.Some experiments were carried out to study mechanical vibration signals of the rotating machinery under 5 conditions,which respectively are "normal", "balance", "loose base", "oil whirl". These 5 sets of vibration signals to train iFHMM are used to get the corresponding parametric iFHMM for each mechanical failure mode. Experiments show that iFHMM can correctly identify the failure mode of rotating machinery.Compared the recognition results of iFHMM with the results of FHMM identification, the recognition rate of iFHMM was significantly higher than that of FHMM. On this basis, this paper presents a fault diagnosis method of rotating machinery based on iFHMM.Chapter 4,This section introduced the theory and model of the ICA,including RTC and RMI.Next, it proposes the ICA-iFHMM method based on ICA. This method,a complete set of signal processing methods, can be used for the separation and identification of signals.It can be also applied to the multi channel recognition of mechanical vibration signals and speech noise reduction. The experimental results show that the proposed method is effective.Chapter 5,on the model parameter estimation for iFHMM.EM algorithm can only be used into the area of local optimization, so, it makes the algorithm in the defects of early convergence to local minima prematurely. Using particle swarm optimization(PSO) has the characteristics of global optimization, PSO- iFHMM model is put forward, in the improved iFHMM, using PSO and a rolling bearing based on PSO-iFHMM state prediction method.The state of rolling bearings, and the identification and prediction.Experiments show that the method is effect is good.Chapter 6,this thesis research content has carried on the comprehensive summary, and of the need for further research work is prospected.