NEURAL NETWORKS TO ANALYZE SURFACE TRACKING ON SOLID INSULATORS

Surface tracking on solid insulators is one of the most severe breakdo wn mechanisms associated with polymeric materials under long term serv ice conditions. A wide range of relays can detect failure in a transmi ssion line and prevent a total breakdown in the systems, but due to th e non-healing characteristics of solid insulators, in most cases it mi ght be too late to save the insulator after tracking initiation and gr owth. The method described here is employed mainly in detecting severa l conditions, such as discharges, leakage current, dry conditions, sev ere damage and tracking initiation. Initially a BPN (back propagation network) type NN (neural network) is trained with different signal typ es. Due to the nature of NN, which always require similar values of in put nodes, the system uses the FFT (fast Fourier transform) of the inp ut signal, which might have high amplitude frequency components other than the fundamental frequency depending on the condition of the surfa ce. The system works on a real time basis and warns the user with the first indication of severe damage on the surface and can protect the i nsulator from excessive damage.