Research Of Blind Source Separation Algorithm And Its Application In Fault Detection

Blind source separation (BSS), using the statistical independence between sourcesignals, is a process to only use observed signals to renew source signal, under thecondition that the source signals and the mixed channels are unknown, also known asindependent component analysis (ICA). Although classic independent componentanalysis only uses the data statistical properties can separate blind signal, the separationresults is uncertain. However, mechanical fault has some priori knowledge, such asfrequency characteristics. In this paper, these prior information is mainly used to makethe separation results of algorithm certain and this algorithm is applied to the bearingfault diagnosis.Dimension reduction and decorrelation theory based on principal component analysis(PCA) is studied. PCA processing for signals could reduce the relevance between them.Selecting criteria for different objective functions of ICA algorithm are discussed,proving that under certain condition, these criteria are equivalent. Comparing thetheoretical effect of kurtosis and negentropy as non-Gaussian measure, results show thatthe robustness of fast ICA algorithm based on negentropy is better and subsequentalgorithms are improved based on negentropy.The defects of the ICA are briefly overviewed, and a kind of ICA algorithm thatcontains reference signal constraints is presented to overcome them. Basic mathematicalmodel and principle of constraint independent component analysis (CICA) algorithmare studied and the construction principles of the reference signal and the measuremethod of similarity are discussed. The matlab program of CICA algorithm is written.Pulse signal, with the same frequency as the signal to be extracted, is selected as thereference signal, and mean square error is selected as the method of similarity measureto perform experimental simulation. Results show that CICA algorithm can separate thesignal to be extracted. Meanwhile, using bearing fault data of Case Western ReserveUniversity can effectively extract the fault signal in this experiment..To further prove the effectiveness of CICA, a bearing fault detection platform is built.Using wire cutting method to process defects on the bearing’s inner ring, outer ring and ball, and using data acquisition card supplied by National Instruments Corporation (NI)to obtain fault data and processing the data with the CICA, results showed that CICAcould extract the fault signals of the corresponding parts of bearing.In conclusion, constrained independent component analysis algorithm in blind sourceseparation is effective, and can be well used in bearing fault detection.