Polynomial pattern recognition analyses for evaluation of transient signals in transformers

Abstract

The polynomial pattern recognition analyses are proposed for identification of winding insulation failure during lightning impulse testing of transformer. The polynomial pattern recognition analyses enables the computation of winding responses (transient signal) measured at neutral terminal of the transformer winding due to impulse voltage excitation. Initially, simple polynomial analysis is performed through residual graph, mean square error to visualize the correlation between the transient signals. The polynomial analysis is further extended for identification of type of relationship between the transient signals. Further, polynomial approach is utilized through Akaike's information criterion to estimate the degree of association between the responses. It is a reliable additional tool which can be used to conclude if a winding insulation of transformer has withstood the rated lightning impulse test voltage or not. To prove the proposed analyses for lighting impulse test 66.7 MVA (138/69/13.8 kV) and 250 MVA (500/275/33 kV) are considered.