Improving the modelling of the water content of power transformer insulation in the frequency domain

One technique to measure the water content of the insulation of an operating transformer is to use online water activity sensors with cellulose adsorption isotherms. In a cellulosic pressboard/paper insulation system, the water migrates between the cellulose and oil until a known equilibrium partition is reached, when the cellulose adsorption isotherms can be used to estimate the water content of pressboard from the concentration of water dissolved in the oil. However, a significant drawback is that the time constants involved in the diffusion of water through pressboard are very high, and it is very unlikely that the transformer will be at one temperature long enough for the water partition to reach equilibrium. In the past this has been resolved by using long term averages. However, an obvious question is for how long a long term average should be applied for before it filters out actual features of the measurement data, and how short it may be to suppress the short term fluctuations. We set out resolve this question by analysing the frequency spectrum of the water content of pressboard calculations for an operating power transformers. This showed a difference between the very low frequencies attributable to the migration of water, and the higher frequencies caused by the daily temperature cycling of the transformer. A low pass filter was then applied to the measurement data. This technique was found to be more effective than using a long term average, although further research is required to optimise such a filter.