Enhancing fault detection and classification in distribution transformers using non-contact magnetic measurements: A comparative study of tree models and neural networks

Ensuring the reliability of an electric power supply network (ESPN) requires accurate and rapid fault detection in distribution transformers. This paper presents a Finite Element Analysis (FEA) approach for non-contact magnetic measurements to capture magnetic flux density (MFD) values during both short-circuit (SC) and open-circuit (OC) fault conditions. These MFD measurements are then classified using various machine learning algorithms, including decision tree (DT), gradient boosting (GB), random forest (RF), and artificial neural network (ANN). The RF model achieves the highest accuracies, with 99.74 % for SC faults and 93.02 % for OC faults, outperforming all other models. The ANN model shows accuracies of 98.71 % and 92.38 %, while the DT model achieves 92.85 % for SC faults and 88.75 % for OC faults, and the GB model records 95.63 % for SC faults and 90.55 % for OC faults. Additionally, the DT model demonstrates fast prediction times of 0.0028 s for 7203 SC samples and 0.0019 s for 4802 OC samples. The novelty of this research lies in the use of FEA-based non-contact magnetic measurements to collect MFD values, which enhances safety and fault detection accuracy compared to traditional voltage and current signals. Unlike previous studies focused on overhead line protection, this method provides equipment-specific protection for transformers. Furthermore, integrating MFD data with machine learning models significantly improves fault classification speed and accuracy, providing a significant advancement in transformer fault detection.