Network Synthesis of Transformer Winding for FRA Interpretation

Condition monitoring of power transformers are becoming critical as there is a thrust to operate the grid closer to its limits. Frequency response analysis has emerged as one of the principal methods in appraising the state of the physical structure of the transformer core and winding assembly. Changes in winding signatures, however, can still only be interpretted by experts. This paper explores the use of network synthesis as a means to interpret changes in winding signatures. A large set of transformer windings is simulated using complex RLC meshes. Neural networks are then trained to match the responses to network parameters. A number of different neural networks with varying parameters were tested and results show that Radial Basis Networks perform the best in correctly matching circuit parameters.