Monitoring Polymeric Insulators with an Ultrasonic Noise Based Technique
This paper presents a monitoring and predicting technique of polymeric insulators operational condition, including pollution, based on the detection of ultrasonic noise signals. The acoustic inspection techniques are based in the detection of the mechanical waves created by instantaneous energy release of electrical discharges in regions of intense electric field. This electric field condition is very typical in cases of intense pollution or compromised physical integrity of the polymeric insulator. The noise arising from the surface discharges releases a greater amount of energy in the ultrasound range, especially when surrounded by air. The ultrasound monitoring is an alternative technique with advantage of immunity to electromagnetic interference. However, its worst disadvantage is the high occurrence of false positives and false negatives when humans do the judgment of the audio. The presented method employs the Spectral Sub-band Centroid Energy Vectors algorithm to extract attributes and process the ultrasonic noise signals, performing an estimation of contamination levels with the aid of an artificial neural network. One of the main advantages of using artificial neural networks is that they have high generalization capacity. Test patterns can be successfully classified when compared to similar, but not identical, training patterns. The Spectral Sub-band Centroid Energy Vectors algorithm splits the signal spectrum into a number of overlapping frequency sub-bands. Then, it locate the centroids of each sub band and calculates the energy in the proximity of the centroid, within a determined range. Therefore, tests were carried out in the laboratory with 230 kV polymeric insulators collected from a transmission line with of different degradation levels. The estimation performed using the artificial neural network made possible to identify and distinguish accurately the energy vectors from reference insulators (considered perfect samples), polluted insulators, and insulators with superficial damages. Furthermore, the technique evaluated the ultrasonic noise signals behavior for three different axes, arranged with a special shift of 120º, and considered the distance from the insulator and ultrasound microphone, since measurements were made from three different distances: 5, 7.5 and 10 meters. In all cases, the artificial neural network success rates were above 80% the. The results form the basis for application of the method in field measurements, which will be possible to evaluate this application on system operation.