Transformer Winding Condition Assessment Using Feedforward Artificial Neural Network and Frequency Response Measurements

Frequency response analysis (FRA) is a well-known method to assess the mechanical integrity of active parts power transformer. The measurement procedures FRA are standardized as described in IEEE and IEC standards. However, interpretation results far from reaching an accepted definitive methodology there no reliable code available standard. As contribution this necessity, paper presents intelligent fault detection classification algorithm using results. based on multilayer, feedforward, backpropagation artificial neural network (ANN). First, adaptive frequency division developed various numerical indicators used quantify differences between traces obtain feature sets for ANN. Finally, model ANN detect classify different transformer conditions, i.e., healthy windings, windings with saturated core, deformations, electrical faults, reproducibility issues due test conditions. database study consists measurements 80 transformers designs, ratings, manufacturers. obtained give evidence effectiveness proposed diagnosis FRA.