A new method for online detection of abrupt faults based on sigularity value decomposition

A new online detection method for detecting abrupt information embedded in noise and stain signals is presented based on the research results of the singularity values' distributing by using the singularity value decomposition (SVD) about track matrix of attractor reconstructed by the time series of stain signals, abrupt information and noise respectively: The singularity values' distributing of the track matrix of attractor reconstructed by the survey signal is gained by SVD; Updating the survey signal, the new singularity values' distributing of the track matrix of attractor is gained again. Then the remnant of the singularity values can be gained by using the second group singularity values minus the first group singularity values correspondingly. The reverse transforms of SVD and the reconstructed track matrix of attractor are implemented by selecting some singular values in the second group according to the remnant values, and the outcome of transforms is reconstructed signal. The change of equipment's state can be reflected by the reconstructed signal. The method is performed according to several numerical signals, and the results display that the abrupt information reflecting the change of equipment's state is easier to be sized because satin signal and noise can be reduced at the same time by the method.