Hassan Reza Mirzaei

Transient overvoltages within a power grid may lead to electrical faults in power transformers. Moreover, short circuit currents passing through the transformer may cause mechanical deformations in its windings. Engineers typically use Frequency Response Analysis (FRA) to identify defect characteristics in transformers. The severity of defects varies based on the extent of the short circuit current and power flowing through the fault. Among electrical faults, disk-to-disk faults in the disk windings are particularly significant, as they can reduce transformer efficiency and increase repair costs. In this study, a nonlinear Kernel Principal Component Analysis Algorithm (KPCAA) is employed to estimate the location and severity of these faults. In the present work, a 1.2 MVA, 10 KV non-interleaved double-disk winding is used and subjected to various artificial faults. Firstly, the high-dimensional Transfer Functions (TF) of this winding are obtained using the Low Voltage Impulse (LVI) method under both healthy and faulty conditions. Next, the KPCAA is applied to the obtained TFs to extract significant features in a lower-dimensional space. Afterwards, the parameters of the selected kernel functions associated with KPCAA are adjusted to achieve optimal results. Finally, the proposed technique’s advantages in determining the faults’ characteristics are presented.

Keywords: Transformer fault detection, disk to disk fault, Kernel PCA, Low Voltage Impulse (LVI) method, Dimension Reduction