Moisture Measurement in Transformer Oil Using Sensor-Based and Computational Modelling Approaches

Abstract

Moisture contamination in transformer oil significantly reduces its dielectric strength and accelerates insulation ageing, making accurate measurement essential for transformer health monitoring. Traditional laboratory-based methods such as Karl Fischer Titration (KFT) provide high accuracy but are not suitable for continuous or on-site monitoring. This study proposes a hybrid approach that integrates a low-cost capacitive sensor, dielectric parameter measurements, and predictive modelling to estimate moisture content in transformer oil. A total of 30 oil samples with moisture levels ranging from approximately 50 to 250 ppm were prepared and tested under controlled temperature conditions. Moisture values measured by the proposed sensor showed a strong correlation with KFT results (R² = 0.94), demonstrating good sensitivity to moisture variation. Breakdown voltage (BDV) and dielectric loss factor (tan δ) measurements confirmed the expected trends, with BDV decreasing and tan δ increasing as moisture increased. A multivariable regression model improved prediction accuracy (R² = 0.96). The Artificial Neural Network (ANN) model achieved the best performance, with RMSE = 5.4 ppm and MAE = 3.9 ppm. The findings indicate that the proposed hybrid method is effective for accurate moisture estimation and has strong potential for implementation in real-time or IoT-based transformer condition-monitoring systems