ANFIS-Based Health Index Estimation for Power Transformers with a Comparative Perspective on Fuzzy Logic for Predictive Maintenance

Crisanto E.; Crisanto A.; Cordero G.; Risco R.; Arrese C.C. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244689

Resumen

This paper presents the development and compara-tive evaluation of two data-driven models: Adaptive Neuro-Fuzzy Inference System (ANFIS) and Multilayer Perceptron (MLP), for estimating the Health Index (HI) of power transformers. The models are trained using key diagnostic parameters derived from dielectric oil analysis, including indicators of both oil and solid insulation degradation. The methodology includes data preprocessing, model configuration, and validation using both synthetic and real-world failure data. Results show that the ANFIS model outperforms the MLP in terms of accuracy and interpretability, particularly in scenarios with limited data availability. These findings highlight the potential of neuro-fuzzy systems as reliable tools for supporting predictive maintenance strategies in power systems.

Instituciones: Universidad de Piura