Diesel Engine Bearing Fault Diagnosis Based on Underdetermined Blind Source Separation

Abstract

The method of diesel engine bearing fault diagnosis based on underdetermined blind source separation was proposed. The phase space reconstruction matrix of the vibration signal was decomposed with singular value decomposition, and reconstruction order number was determined by the method of dynamic clustering in order to construct virtual observed signal, so the blind source separation was changed from underdetermined to determined or over-determined. Then, the virtual observed signal and the original signal were composed into propagation paths, and the signal containing bearing fault characteristics was separated from mixed signal by the method of blind source separation which was based on the algorithm of adaptive PARAFAC. The simulation and test results show that this method effectively isolates the source signal.