The main driving forces are to reduce maintenance costs, prevent forced outages, with related consequential costs and to work existing equipment harder and longer
[4].
In the past, winding faults in transformers were judged based on the expertise of the inspectors by simply com-paring the voltage or current waveforms. Since fault de-tection is essential, new signature methods were being continually discovered. With availability of fast digital recorders and personal computers, these waveforms were stored digitally, leading to their analysis rather than just visual observation. Primarily, differences in waveforms were amplified and compared [5].
Subsequently, Transfer Function (TF) was developed and applied successfully by the industry utilities. Instead of time domain analysis, which shows weak deviation from normal and fault conditions, Frequency Response Analysis (FRA) has been used widely to obtain transfer functions. Fast Fourier transform (FFT) has been the stan-dard technique used in the frequency response analysis [1]; hence, TF can be computed as follows:
)(
)(
in
out
VFFT
VFFT
TF = (1)
where, Vout and Vin are the output and input voltage.
TF can be obtained from the transformer under different conditions. Any significant deviations from the "fingerprint" (normal conditions) in the transfer functions may indicate the incidence of faults on the windings, e.g., short circuits. The difference will indicate possible winding abnormalities, and may suggest specific maintenance work.
FRA has shown that it is possible to detect a variety of different internal conditions, such as changes in the condition of the core, changes in the condition of the winding insulation and winding structure integrity, and internal mechanical deformation of transformers [6]. At lower frequencies, the FRA (fingerprints) shows that the transformer is characterized