Andrew Fieldsend-Roxborough / National Grid Company
Xiaozhou Mao / The University of Manchester
Zhongdong Wang / The University of Manchester
An increasing number of power transformers operating in the developed countries do not have OEM technical support. There is an urgent need to obtain transformer design information such as winding types through interpretation of measured FRA data. Winding types may be manifested as different features in FRA traces within specific frequency ranges. The objective of this paper is to identify the winding type via FRA measurement data in order to help manage transformer assets. Support Vector Machines (SVM) are supervised learning models for classification analysis in machine learning. Given a set of training FRA examples with known winding categories, an SVM algorithm is built to assign new FRA traces to a most possible winding category. A group of 400/275 kV transformers are tested in this study with excellent classification results of winding types, indicating a promising “digital twin” technological approach to transformer asset management.