Blind source separation of vibration signals for fault diagnosis of power transformers

This paper aims to separate the sources vibration from the transformer tank vibration, which is normally collected by sensors. Blind source separation consists in the recovering of different physical sources of a system given only a set of external measurements, and makes successes in the field of machine vibration analysis and health monitoring. However it has made few contributions to the vibration analysis for a power transformer, since vibration sources in a vibrating transformer are highly correlated with each other. As a result, the assumptions of statistical independence between sources normally held in conventional BSS (Blind Signal Separation) methods cannot be satisfied for transformer vibration separation. In this paper, a TIFROM (Time-Frequency Ratio of Mixtures) -BSS method is adopted to separate the sources vibration from the transformer tank vibration, on the basis of the diversity of STFT (Short Time Fourier Transform) ratios in time-frequency (TF) domain. The experimental results indicate the effectiveness of the application of TIFROM-BSS for transformer vibration separation.