Estimation of Differential Current in Transformers Using Gaussian Process Models for Fault Diagnosis

The detection of inter-turn short circuits (ITSC) in transformer windings is critical, because these faults are among the most common failures encountered during transformer operation. Although conventional offline diagnostic methods have been widely applied for ITSC detection, recent research has increasingly focused on real-time monitoring approaches. In this context, this paper proposes a fault diagnosis algorithm based on Gaussian Process model to estimate the differential current of each phase under healthy operating conditions. The proposed model is trained using data from a fault-free transformer and subsequently employed to estimate the differential currents during operation. Fault detection is then performed by analyzing the error between the measured and estimated currents. When the estimated error exceeds a predefined threshold, the presence of an ITSC fault is detected. Experimental validation was conducted using a 5 kVA laboratory transformer. The results demonstrate the feasibility of the proposed approach for continuous transformer condition monitoring and real-time ITSC fault detection.