Measurement of Moisture in Transformer Insulation Using the Intelligent High-Frequency Sensor System

A high-frequency sensor-based system is proposed for evaluating the degradation of the transformer insulation through the measurement of moisture concentration. The proposed system employs an intelligent open-ended coaxial probe microwave technique. An optimized artificial neural network (ANN) model is designed by selecting appropriate features to obtain accurate dielectric characterization. Redundant feature elimination reduces the computational burden and also bypasses the repeated calibration refresh required to complete RF measurement in the conventional technique. The ANN model is trained in such a way that measured impedance mismatch due to the termination by the sample at the probe end is related to its permittivity. The introduction of an intelligent standardization procedure led to a low error in the measurement of moisture (<+/- 3%) with a sensitivity of 0.027/ppm. This method is found to be simple and faster compared with other existing techniques. Due to the advantage of high-speed measurement and low error, this sensing technique is very suitable for transformer condition monitoring.