Abstract

Industry and science have been growing attention to developing systems that ensure the integrity of high voltage devices like power transformers. The goal is to avoid unexpected stoppages by detecting incipient failures before they become a major problem. In this context, the detection of discharge activity is an effective way to assess the condition operation of power transformers since this type of flaw can lead the transformer to total failure. The effectiveness of the fault diagnosis systems is related to their capability to distinguish the types of discharges since different flaws require different maintenance planning. This article proposes a new data analysis, which combined the frequency spectrum of the signals with the Karhunen-Loève Transform to perform self-organization maps. The effectiveness of this analysis was validated by comparing it with the Fundamental Signals Properties Classification Technique, which is widely applied for pattern recognition. Two types of sensing techniques were assessed in order to enhance the capability of the new approach. Results indicated that the new methodology presented lower standard deviation for data classification, being a promising tool to monitoring systems.