Development of a portable PDC diagnostic system for discriminating transformer insulation and winding faults using Homoscedastic Probabilistic Neural Network

Assessment of the healthiness of insulation in transformers is a vital aspect for power system utilities since most in-service transformers which form the crux of the power system have reached its lifetime duration. Among many factors, moisture and ageing significantly influence the dielectric properties of a transformer. Recently, several researchers have proposed new diagnostic methods that are complementary to the classical measurement systems such as insulation resistance, power frequency dissipation factor, polarization index measurements etc. Dielectric diagnosis with polarization and depolarization currents (PDC) uses the dielectric system response in time domain. In this research a laboratory-built, compact and easy to construct PDC measurement system is implemented without compromising on the reliability of the measurement system. The capability of the measuring system is ascertained by carrying out detailed studies on scaled down laboratory models and a 315 kVA, 11kV/433V, Dyn 11, ONAN distribution transformer with artificially simulated insulation flaws namely moisture content in oil, inter-disc/ inter-turn shorts, winding asymmetry etc. Simulation studies on the transformer R-C equivalent representation are also carried out to validate the response of the PDC system. In addition, the Homoscedastic Probabilistic Neural Network (HOPNN) which utilized the Expectation Maximization (EM) with the Maximum Likelihood (ML) algorithm is implemented for obtaining a parsimonious training for subsequent identification of faults.