Multi-Stage Binary Classification Technique for Incipient Fault Diagnosis of Oil Immersed Power Transformers Based on ANFIS

Abstract

Conventional Dissolved Gas Analysis (DGA) methods show a poor success rate for predicting incipient faults in oil immersed power transformers. It is because their rule foundation is entirely heuristic and non-mathematical. Artificial intelligence is employed in this work to emulate human expertise and ability in fault diagnosis. Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to implement multistage binary classification scheme to diagnose incipient faults. The published TC-10 data from the faulty oil-immersed transformers is utilized to evaluate the ANFIS models’ performance. This developed multistage binary classification technique based on ANFIS gives superior results than the single stage multi class classification. It substantially increases the diagnosis accuracy when compared to the conventional counterparts. Moreover, in this work the idea of multistage classification is conceptualized by selective hybridization of various DGA methods viz., Roger’s Ratio method, IEC-60599 and Doernenberg’s method. The steps of fault classification traverse from determining whether a fault exists or not to further determining its nature and severity. The diagnosis accuracy is significantly improved by integrating AI, binary classification, the hybridization concept and hence this technique altogether is proven as a proactive tool for online fault diagnosis in power transformers.