Blind Source Separation Of Convolution Mixed Rotating Machinery Fault Signal

With the development of science technology and modern industry, all kinds of industrialequipments increasingly integrated, speeding and intelligent. Equipment vibration monitoringand fault diagnosis plays an increasingly important role, the vibration signal acquisition,analysis and processing is the basis of fault diagnosis, the traditional vibration signal analysisrelatively speaking maturity, but they have their own limitations. Blind Source Separation is anew research focus in modern vibration signal processing, because it can recover the sourcesignals with the observed signals under the condition of unknown prior knowledge such as thesource signals and transmission channels. At present, it has been wildly applied in variousfields such as speech signal processing, array signal processing, data mining, imagerecognition and biomedical signal processing. This paper will focus on blind sourceseparation of the mixture model, theory algorithm and its application in rotating machineryfault diagnosis, some meaningful conclusions were made.The author starts from the instantaneous linear mixing model of Blind Source Separation,introduces several independence criterion of Blind Source Separation based on informationtheory, and simulates three kinds of algorithm(FastICA, EASI, SOBI) of instantaneous linearmixed, the test proves that the separation of FastICA algorithm is superior to EASI algorithmand SOBI algorithm.Taken the actual environment into consideration, the signals received by the sensor areusually the convolutions of the vibration source signals and the transmission channel impulseresponses, this article focuses on the issue of separation of convolution mixed signals,simulation results in time domain RLS blind deconvolution algorithm and frequency domainplural FastICA blind deconvolution algorithm shows that the time domain blinddeconvolution algorithm is more complicated than frequency domain blind deconvolutionalgorithm, the solution speed difference is about several dozen times.The vibration signals of bearing outer ring fault and gearing tooth broken fault werecollected on the experiment platform, time-domain RLS blind deconvolution and frequencydomain plural FastICA blind deconvolution for the measured signal, further, more satisfactoryanalysis results were obtained through wavelet decomposition for the deconvolution results.Blind deconvolution and wavelet decomposition combination of signal processing method canobtain clearer and more abundant fault feature information in comparison with simple blinddeconvolution method and Blind Source Separation of instantaneous mixed method.