Application of artificial neural network (ANN) in SF6 breakdown studies in nonuniform field gaps

In SF₆-filled electrical equipment, the electric field distribution is kept rather uniform. However in practice, the electric field in the gas gap is distorted by nonuniformities. For this reason, the inhomogeneous field breakdown in SF₆ has been extensively studied by various researchers and the breakdown characteristics of compressed SF₆ have been reported. Obtaining experimental data under all conditions is not possible. Therefore, an attempt has been made in the present work to apply an artificial neural network (ANN) to obtain such data. The projection pursuit learning network (PPLN) has been used as the ANN model. Breakdown data for four different voltage waveforms were used to train the network for SF₆ pressures of 1-5 bar and rod diameters of 1-12 mm in a rod-plane geometry. The ANN was first trained with these data so as to obtain a smooth regression surface interpolating the training data. The regression surface thus obtained, was thereafter used to generate the breakdown and corona inception voltages with in the range of gas pressures and nonuniformities studied, where no data is available