Detection of emerging faults in power transformers using self-organising maps

Abstract

Power transformers are a crucial part of the power system, one of the largest infrastructures in industrialised countries. In particular, wind turbine transformers are subjected to frequent thermal cycling as a function of varying turbine loads. Thus transformers are prone to developing faults and defects that can involve high repair costs for instance due to the repeated thermal stress on the winding. Faults develop mainly when the insulation produces small leakage currents between turns, which if not detected early, might become short circuits that can result in interruptions in electricity supply, and difficult and costly repairs. An optimum overhaul of damaged transformers is not accomplished often because of lack of appropriate inspection tools. Detailed assessment and preventive maintenance work, which will allow the detection and repair of failures at early stages, is believed to be the only suitable way to cope with power transformer degradation at low cost. This paper presents a methodology based on the analysis of current signals converted by the S transform for the detection of incipient faults in transformers. The procedure is based on calculating the energy of the zones of the time-frequency spectrum. Its main advantage is its possible real time implementation that can be applied while the transformer is in use. Experimental results with PSCAD are presented.