Detection of insulation faults in transformer using wavelet analysis

Abstract

Recognition of key insulation faults in power transformers through impulse testing was certainly not seen as a big problem as it had emerged today but talking of minor faults which are often neglected after the impulse testing by the naked eye had been a challenging task for a very long time in power transformers. Hence there is seen a need of such a tool which should be capable of verifying the signals/waves after these di-electric tests, as the recognition of such faults is immensely essential to overcome any disastrous situation in the longer run. This work proposes an influential approach which is proficient in detecting such minor faults. The methodology uses wavelet analysis technique, the dyadic-orthonormal wavelet transform (DOWT) in particular. The principle idea behind the working is to detect the fault (noise) at the particular time instance after decomposition of recorded faulty current responses into detailed and smoothed description of the usual signal. The results showed that for three different frequency scales i.e. 10-5MHz for scale 1, 5-2.5MHz for scale 2 and 2.5-1.25MHz for scale 3, higher localized filter coefficient i.e. L=45 is seen to be much more efficient in detecting the fault at a particular instant than the L= 8 filter coefficients under the dyadic-orthogonal wavelet transform function. Therefore the projected technique proved to be robust and way far efficient as compared to the other methods to resolve such group of faults.