Source Separation Methods Based On Time-frequency Analysis Of Mechanical Failure

Lot of blind source separation of machine faults limited to non-Gaussian, stationary and mutually independent source signals, and requires more the number of observation signals than the number of source signals ,it will often have a lot of questions in machinery equipment fault diagnosis, because machinery source signal do not often meet these assumptions. For this shortage, at the National Natural Science Foundation of China (No.: 50775208), Henan Province Natural Science Foundation Office of Education (No.: 2006460005, 2008C460003) funding, to Cohen Class time-frequency distribution, Fractional Fourier transform and empirical mode decomposition as an example, a blind source separation method based on time-frequency analysis of the mechanical equipment of non-stationary signals is carried out in-depth research, make some innovative results, and its main contents are as follows:Chapter one recites briefly the subject and its significance. Then research of blind source separation at home and abroad and blind source separation(BSS) in fault diagnosis research are summarized. In the analysis of the lack of existing sources separation of machine faults ,the centre and novelty of this dissertation is brought forth..Chapter two discusses the basic theory of the blind source separation and its two inherent uncertainties: that is the uncertainty of the amplitude and phase and the uncertainty of sort, The uncertainties do not affect the outcome of blind source separation. Then discusses some important concepts of the blind source separation , such as principal component analysis, singular value decomposition, independent component analysis. And discusses distinction and contacts. Between the independent component analysis and principal component analysis, blind source separation.At last,we discusses three typical algorithms which are used in the paper: JADE algorithm,Infomax algorithm and FastICA algorithm, give out the main calculation formula and steps. Finally, give performance indicators of the blind source separation used in the paper.Chapter three: for the existing mechanical failure based on blind source separation methods ignoring non-stationary signals, Combined the advantage of Cohen class time-frequency analysis and blind source separation, time-frequency analysis is a powerful tool to deal with non-stationary signals,it can describe the characteristics of its spectrum changes with time, blind source separation of multiple signals in the separation of aliasing is very purposeful. A blind separation method of non-stationary signals in the mechanical equipment based on Cohen class time-frequency distributions is proposed . This method makes the existing blind source separation method of mechanical fault diagnosis change to Cohen class time-frequency distribution, with the time-frequency distribution of signal to achieve the purpose of machinery equipment many faults seperation. At the same time, the method proposed is compared with the traditional mechanical source separation method for non-stationary signals. The simulation show that the proposed method is superior to traditional mechanical failure blind source separation method. Blind source separation for machinery equipment non-stationary signals must take full advantage of non-stationary signals in order to achieve good separation. The experimental results further verify the effectiveness of the method. The characteristics of the method are as long as the source signals with different time-frequency distributions, we can achieve effective separation. Finally, based on a variety of quadratic time-frequency distribution in blind source separation mechanical failure compared and analyzed, and through root-mean-square error to reflect the effect of source separation of signals.Chapter four: existing mechanical methods of blind source separation based on time-frequency analysis are only confined to Cohen Class time-frequency distribution, and not extended to other time-frequency distribution. Fractional Fourier transform as a new time-frequency analysis method, is the generalization of classical Fourier transform, has a close relation with the time-frequency analysis and also provides a certain characteristics which Fourier transform do not have. Fractional Fourier transform is a powerful tool for treating non-stationary signals, In this, combined with fractional Fourier transform and blind source separation, a blind source separation method for the non-stationary signal of the mechanical equipment based on fractional Fourier transform is proposed, the method first whitening observation signals to get a new observation signal, and then calculating FRFTof the new observation signal, then estimates generalized correlation matrix, Thus the joint approximate diagonalization for estimated generalized correlation matrix in order to receive estimated source signal.The proposed method has two major advantages: it does not requirethe assumption that the signals' powers vary over time, and it does not require a pre-processing stage for selecting the points in the time-frequency plane to be considered. Finally, the method is applied to blind source separation of both inside and outside the ring bearing failure, experimental results show that the method is effective.Chapter five: Against the lack of the existing the mechanical failure source separation based on time-frequency analysis method requires more the number of observation signals than source signals, Combined the advantage of empirical mode decomposition (EMD) and BSS ,principle component analysis (PCA), two determining blind separation methods of machine faults which are called EMD-BSS method and EMD-PCA method , are proposed. In EMD-BSS method, using EMD method decomposing mixed observation signals, will break down of all the weight and the IMF and the original observation of mixed-signal re-formation of the new observation signal structure, all the IMF components which are decomposed and original mixed observation signal re-form a new structure of the observation signal, put determined BSS problem into overdetermined BSS problem. And, whitening treatment and United diagonalization for new observation signal to get the estimated source signal. One distinctive feature of this method is can not only deal with the separation of stationary signal, but also deal with the separation of non-stationary signals, another distinctive feature of this method is that it is suitable to the source of a few more than the number of observation signal separation,also applies to the source of a few less than the number of observation signal separation. The simulation results show the effectiveness of the method, and is superior to traditional mechanical failure blind source separation method. Finally, the proposed method applied to the motor -reducer coupling experiments further showed the effectiveness of the method. In EMD-PCA method, the basic idea is similar to EMD-BSS methods. The difference is, EMD-PCA method,the use of PCA on the new observation signal common analysis to be the major components of the source signal. It has the same excellent characteristics with EMD-BSS method. The simulation and experiment results show that the proposed method is very effective.Chapter six: summarize the full text content and noted that the content is worthy of further study.