Detection of Failures Within Transformers by FRA Using Multiresolution Decomposition

Abstract

The detection of failures within power transformers is considered an important issue since these components are of critical importance for power system reliability; moreover, their replacement cost is extremely high. In monitoring the transformer condition along its useful life, frequency-response analysis (FRA) has gained great interest due to its sensitivity to failures in the windings and the iron core. These failures can be detected by evaluating transfer function changes by means of statistical and mathematical indices and classified according the frequency band in which these changes take place. However, this procedure involves evaluation inaccuracies due to disturbances or minor changes during FRA measurements. The new methodology is based on the decomposition of the original responses in several levels of decomposition (filtering) using the discrete wavelet transform, and the subsequent comparison using smooth versions of the responses. Fault detection is further supported with statistical indices calculated using the frequency band where abnormal differences appear. This procedure gives more robustness to the method and reduces the possible influence of disturbances during measurement in the diagnosis result. The methodology has been tested using different failure cases and two of them are used for validation purposes in this paper.