Study On Separation And Identification Of The Vibration Sources Of Mechanical Svstem

The vibration signals of mechanical system is a mixture of all vibration sources and noise by different ways. With the help of modern signal processing theory and method, separation of the source of vibration signal form the mixtures can effectively extract the feature and improve the accuracy of fault diagnosis. Therefore, the separation and identification of the source of vibration in mechanical systems for the mechanical system condition monitoring and fault diagnosis has important theoretical significance and practical value.Based on the High Technology Research and Development Program of Chi-na(No:2008AA04Z410)-" Research on the condition monitoring and fault diagnosis techniques and its application system of supercritical and ultra-supercritical steam turbines and pressure generating units " and the National Nature Science Fund Project(No:50675194)-" Research on method for mechanical vibration source semi-blind source separation and reconstruction, re-search on separation and identification of the mechanical system was carried out. The thesis pri-marily studies the blind separation and identification method of mechanical vibration source.The thesis primarily studies the blind separation and identification method of mechanical vi-bration source. From the perspective of blind source separation, the characteristics of vibration source and propagation in the mechanical system was analyzed and some problems that urgent need to solve in the blind separation and identification of mechanical vibration source were de-scribed. Such as, estimating the number of mechanical vibration source, blind separation and identification of the dependent vibration source, blind separation and identification of the vibra-tion source when the number of observed signal is less than the number of the vibration source, and how a priori knowledge application in blind source separation and identification of vibration sources. Corresponding research was carried out to address the problem, a method of estimation of the number of vibration source with one observation, a method of blind separation and identi-fication of the dependent vibration source, a method of blind separation and identification of the vibration source with one mixture and a method of semi-blind separation and identification of the vibration source were proposed. Simulation and experimental verified the new methods. The de-tails were as follow:In Chapter one:An overview of the importance of the separation and identification of the vi-bration source in the mechanical system condition monitoring and fault diagnosis. The research status of the techniques to separate and identify the vibration sources were reviewed and the problems existing in the method of separation and identification of vibration source were ana-lyzed. The essentiality of introducing blind source separation to separate and identify mechanical vibration source was addressed. Blind source separation and its application in mechanical vibra-tion source separate and identify were reviewed. A frame of the BSS based mechanical system fault diagnosis was presented. Finally, according to estimating the number of source, separation and identification dependent source problems and using prior information, the research back-ground, main contents, the overall architecture and technology roadmap and innovation were presented.In Chapter two:From the perspective of statistical independence in the theory of blind source separation, discussed the concept of "source" in the mechanical system and defined an underlying source of vibration for blind separation and identification of the mechanical system vibration source. The gear box, for example, the mechanism of vibration source and its transmission and mixing characteristics were analyzed, and the characteristics of the underlying source of vibration and vibration model were studied. After reviewing the techniques to separate and identify the vi-bration sources, a frame of mechanical vibration sources separation and identification with BSS was introduced.In Chapter there:The relationship of the number of vibration source and the number of ob-servations was analyzed, the problem of different methods to estimate in vibration signal were discussed. Aiming to estimate the vibration source number for under-determined mixtures, a method based on Empirical Mode Decomposition (EMD) to estimate the number of vibration sources by single observation was introduced. Single observation was decomposed by EMD to construct virtual multi-observations. The eigenvalues of sources, which were used to determine the number of sources with the method of Bayesian Information Criterion (BIC), were obtained through the relation matrix of the constructed virtual multi-observations. Simulation and experi-ment verify this approach is practicable and effective.In Chapter four:The statistical independent of the vibration source was discussed; the influ-ence of correlation of the source in the traditional BSS was researched. The traditional BSS can-not separate and identify vibration source accurately due to the existing of the dependent vibra-tion sources. Aiming to separate and identify dependent vibration sources, a method based on wavelet packet decomposition to separate and identify dependent vibration sources was present. Assumed some sub-component of correlated machine vibration sources are independent, sub-band observed signals was obtained by decomposing original observed signals with wavelet packet, the new observed signals was reconstructed by some sub-band observed signals with less dependent according to the mutual information criterion, separation matrix was estimated through the new observed signals with independent component analysis, the correlated machine vibration sources was separated by estimated separation matrix. Simulations testified the validity of the proposed method.In Chapter five:The reason why the number of vibration sources is more than the number of mixtures was analyzed. The single mixture BSS methods was researched, especially, the limita-tion of single mixture BSS based on wavelet decomposition was discussed. The principle and al-gorithm of EMD method were studied; the possibility and advantage EMD method in construct virtual mixtures was discussed. Aiming to separate and identify vibration sources by using un-der-determined mixtures, a method based on EMD to separate and identify vibration sources through single mixture was introduced. Single mixture was decomposed by EMD to obtain in-trinsic mode function (IMF), using the method proposed in Chapter three to determine the num-ber of the vibration sources, selecting several IMF and the original observation to construct virtu-al multi-mixtures. By changing to standard BSS, the vibration sources can be separated and iden-tified form single mixture. Simulation and experiment verify this approach is practicable and ef-fective.In Chapter six:The constraint independent component analysis (CICA) and independent component analysis with reference (ICA-R) were researched. The properties of the ICA-R and its influencing factors were analyzed. The limitation of general BSS method to separate and identify vibration sources wad discussed. In order to separate and identify vibration sources effectively and quickly, a method based reference source to separate and identify vibration sources was pro-posed. The reference signals that carry some information of sources were constructed. The close-ness measure between reference signals and separated sources was incorporated into contrast function as the constraints. The interested mechanical vibration source was obtained by solving the constrained optimization problem. The proposed method separated and identified vibration source simultaneously. Aiming to build reference signal with optimal phase and set threshold adaptive, an improved ICA-R method was introduced. By alternately optimizing the contrast function for ICA and the closeness measure for ICA-R, the proposed method can construct a ref- erence signals with optimum phase and determine a feasible threshold in an automatic manner. Simulations on synthetic data and real-world data have demonstrated its validity.In Chapter seven:The convolutions and innovations in the thesis were summarized, and an outlook on future research work are addressed and discussed.