Skip to Main Content
Blind source separation (BSS) can be used to separate mixed signals which is combined by the original data linearly, and obtain the source component which is statistically independent. The capacity of independent component analysis (ICA) is usually affected by the phase difference of the mixed signals. For this reason, an improved method called frequency domain BSS is proposed. By the properties of linear addition and phase loss of spectrum analysis, the engineering signals are transformed to frequency domain firstly, and then the spectra are processed by ICA. Simulation results and the application of eddy-current sensor fault detection both demonstrate that for ICA the correct preprocessing according to signal structure contributes to feature extraction of engineering signals effectively.