The work is carried out by Rajesh Reddy, Dr. Krupa Shah and Dr. Manjunath Kallamadi.
The main reasons for transformer winding failures are abnormal over-voltages and short-circuit forces. Therefore, the condition assessment of the transformer is essential to identify the changes in winding geometry at an early stage. Frequency response analysis (FRA) is a technique that has gained wide attention due to its sensitivity in identifying changes in winding deformations. As the winding deformations get reflected as changes in the high-frequency behaviour of the transformer, characterizing its high-frequency behaviour is essential. For this purpose, a physically realizable high-frequency ladder circuit of the transformer winding can be built from the FRA data. However, synthesizing the ladder circuit model is challenging. To achieve this, a study is performed to identify the best optimization algorithm such that estimated circuit variables are repeatable and accurate, and prior knowledge of the initial guess and search space is avoided. From the experimental study, it is identified that the circuit variables estimated by the gradient-based sequential quadratic algorithm are repeatable, accurate, and reliable. Also, the human involvement in adjusting algorithm parameters is less.
