Application of Blind-Source-Separation algorithm for investigating transformer vibration patterns

Continues monitoring of a power transformer's winding clamping pressure could be very useful for early identification of any defect caused by depleted short-circuit strength. Analyzing transformer vibration patterns may provide a cost effective way to online monitor the clamping pressure. However, the vibration signals measured from the transformer is a mixture of signals from the winding, core and noise generated by various other devices. This problem becomes more complex in in-service transformers because the noise produced by other components around the transformers also affects the vibration signal. As a result, the sensors collects the vibration signals, which are not uniquely generated by the targeted part of the transformer, but also comprise the contribution from other components of transformer including cooling fan & tap changer. To tackle these issues and extracting the effective information from the measured vibration signal for predicting the actual condition of the transformer winding is a challenging task. This paper proposes the blind source separation (BSS) technique. BSS is a well-known technique in many areas, which allows to extract a set of signal from mixed signal without the prior knowledge of signal source and mixing process. In this paper BSS technique suitable for vibration monitoring of transformers is developed and applied for vibration data gathered from a lab based transformer. All lab measurements were conducted under controlled environment to test the effectiveness of algorithm to identify specific vibration source. The results of the presented analysis demonstrate that BSS is a possible technique to identify winding clamping conditions through vibration signal measurements.