부산대학교 도서관

This paper provides an example of how to improve DGA fault severity interpretation using screening test statistics and optimization curves with transformer failure data. Initial comparison of ‘cookbook’ fault severity methods in Draper & Dukarm 2021 [1] showed that IEEE C57.104-2019 had fallen short of its potential by not having a clearly defined status code higher than a 3. The IEEE guidelines provide for an undefined ‘Extreme DGA’ category which we attempt to define in this paper using failure data. We found that the overall performance of IEEE 2019 as a predictor of near-term transformer failure can be considerably improved by just multiplying the status code 3 limits by a factor of seven. Nevertheless, this simple modification to IEEE 2019 is not able to achieve as high of a positive predictive value or diagnostic odds ratio as PFS (IEC 60599–2015 / CIGRE TB 771) or Reliability-based DGA. Further algorithmic changes using population data with actual failure cases will be required to improve the performance of future DGA fault severity interpretation methods.