A new blind-source-separation method and its application to fault diagnosis of rolling bearing

In this paper some existing methods of Blind Source Separation (BSS) are analyzed and the general framework of BSS based on Joint Diagonalization (JD) is presented. Fractional Fourier Transform (FrFT) is reviewed, and a new property of FrFT is established and proved, namely the mutually uncorrelated signals would still be uncorrelated after FrFT. So a new method of BSS based on this property is put forward. And this new method has some other strong merits compared with the existing methods, such as fast computation speed and facility to deal with time-varying or non-stationary signals. The comparison of this method with the existing ones shows that this method is more practicable when used for simulation of signal selected randomly. At last this method is used in fault diagnosis for the rolling bearing of a freight train, and the results illustrate the feasibility and potential ability of this method in fault diagnosis.