IEEE Transactions on Dielectrics and Electrical Insulation Vol. zyxwv 9, No. 4; August 200.2 555 zy Pattern Classification of Impulse Faults in Transformers by Wavelet Analysis zyx P. Purkait Instrumentation Engincering Dcpartment Haldia Institute zyxwvutsr of Technology Haldia, Midnapore 721 zyxwvut 657, India and s. Chakravorti ~ Electrical Enginecring Department Jadavpur University Kolkata 700 032, India ABSTRACT One of the tests carried out on a transformer after assembly is the lightning im- pulse test, for assessment of the integrity of its winding insulation. In the case of a fault, it has been well established that the pattern of the fault currents contains a typical signature of the nature and location of the insulation failure involved. This paper describes a new approach using the 'wavelet transforms' to classify the pat- terns inherent in different fault currents. Whereas conventional frequency- response analysis based techniques fail to identify the time-localization of a partic- ular frequency component in a time-dependent signal, the wavelets are not only lo- calized in frequency, hut also in time. The 'time-frequency localization' feature of wavelet transform is employed for pattern classification of impulse fault currents of transformers. Results for simulated models of 3,5 and zyxwv 7 MVA transformers are presented to illustrate the' ability of this approach to classify insulation failures. 1 INTRODUCTION NSUGTION failure within transformers is considered I to be one of the most important causes of failure of power transformers. Impulse testing of transformers after assembly is an accepted procedure for the assessment of their winding insulation strength to surge ovelvoltages. In such tests, impulse voltage sequences are generated in the laboratory and applied to the transformers as per stan- dards [l]. Manufacturing defects or inadequacy of insula- tion may lead to failure against impulse voltage stresses. The detection of the fault and its location of occurrence, which in many cases may take a long time, have to be determined for taking proper remedial measures. For many years, the applied voltage waveforms and the result- ing current waveforms were analyzed manually by study- ing oscillographic records [Z-61. Such manual interpreta- tion of the waveform patterns for fault identification and classification was strongly dependent on the knowledge and experience of the experts performing the analysis. With the advent of digital recorders and analyzers, there has been an increasing trend to use the frequency domain analysis, particularly the transfer function approach [7-91 for fault classification. In recent years, the use of com- puter aided techniques like expert systems [lo-121 and ar- tificial neural networks (ANN) 1131 have been employed for impulse fault classification in transformers. Wavelets are widely used in areas [14-151 such as singu- larity detection, data compression, detecting features in images, noise elimination and harmonic distortion in sig- nals. The wavelet technique is an advancement over Fourier transform, which allows each frequency compo- nent to be studied with appropriate time resolution. Re- cently, the wavelet technique has been. applied in trans- former condition assessment and fault diagnosis [16-201. The inherent noo-stationary pattern of transformer cur- rent waveforms during different fault conditions can be effectively classified using this frequency-selective feature of wavelet transform. Determination of incipient faults in- volved careful and precise recording of the current wave data [171, whereas location identification of permanent kind of impulse faults required the exact knowledge of the travelling wave velocity in the transformer winding [18]. However, due to the complex nature of the composite in- sulation system of transformers, determination of exact lo- cation of faults within transformer windings is a difficult task in many practical cases. Thus, the present paper aims at classifying different impulse faults in several sections 1070-9878/1/$17.00 zyxwvut 0 2002 IEEE