Research On Methods For Mechanical Failture Diagnosis Based On Blind Source Separation

Mechanical failture diagnosis is essentially a problem of pattern recognition,the process includes signal detection,feature extraction,state recognition and diagnostic decision-making.The signal detection is the prerequisite for the fault diagnosis,the feature extraction is the key part of the fault diagnosis,and the state recognition is the core of the fault diagnosis.The inadequacies of the existing methods of failture diagnosis theory are mainly in the process of signal detection and feature extraction.It make the process of feature extraction became more difficult that the mixed of the external interference or noise and the redundancy of the information which came form multi-channel sensor.In view of the above problems,the blind source separation method is introduced in this paper,in the sensor observation,if the signal of the run of the machine and the external interference can be seemed as a separate external interference source,or the different background source of vibration of the internal machines can be seemed as statistical independence,then we can remove the external interference and redundancy of the information by using the source separation.In this paper,a new framework for failture diagnosis is proposed which is based on a source separation,and the overall solution about blind source separation method is given.Independent component analysis is a specialized method which solves the problem of blind source separation methods.Now,the method of solving the problem of blind source separation is based on this.The difference is the basis of determine and the optimization algorithm.There are two methods often used in practice which are information greatly act(Informax) which based on gradient calculated and largest law negentropy(Fast ICA) which based on fixed-point iteration calculation.In this paper, these two methods are summarized,and compared by the experiments.At the same time, the feasibility of the blind source separation methods is verified,then the effectiveness and importance of the blind source separation methods using in the mechanical failture diagnosis is discussed.Blind deconvolution is a more practical method of blind source separation.The mixed model of source signal is deconvolution mixed,which is more close to the actual. In this paper the method of solving intersymbol crosstalk is given which can be used in evaluating multi-channel deconvolution.It also point out the direction for the study of multi-channel deconvolution.Besides,we successfully separated the miscellaneous department of mixed-signal after solving the problem of blind deconvolution using the nuclear function method-nonparametric estimation.