Source identification of turbine vibration based on Independent Component Analysis with virtual signal channels

The basic independent component analysis (ICA) has its own limitation, if N components are expected to be separated, the number of observed composite signals M must at least be equal to or more than N, that is M ges N, that is difficulty for implementation. This paper proposes a novel generalized ICA model, with which more independent components can be separated than the number of the input signals by means of the additional virtual channels, that is, it is available for ICA separation by M < N. The virtual channels are build up under the condition of the prior information of possible sources. Tests of this ICA model with virtual channels were carried out by using the real measured multi-channel vibration signals of a steam turbine to demonstrate its effectiveness.