An instrumentation package capable of locating partial discharge sites in cables has been developed. The digitized partial discharge (PD) signals recorded from one cable end consist of a sequence of pulses whose separations contain information on the relative location of the PD site. The signals are often contaminated by noise and undergo substantial attenuation and phase change as they travel though the cable and the detection system. Moreover, overlap of two successive pulses is possible if the PD site is close to a cable end. The authors describe and illustrate two techniques-maximum likelihood (ML) estimation and deconvolution-for extracting pulse separation from such a time series of noisy and ambiguous signals. Both real and simulated measurements are used to demonstrate the potential of these methods. A procedure whereby knowledge of the combined cable-instrumentation transfer function can be incorporated into the maximum likelihood technique is also discussed. The ML method appears to be much more effective in the presence of cable noise. The main disadvantage of the ML method is that the approximate width of the wavelet or basic PD pulse should be known to give the best compromise between noise smoothing and peak resolution. This width can be determined by an impulse response test or by knowledge of cable length and parameters.
展开▼
