Abstract
Electrical power transformers are the most exorbitant and tactically prominent components of the South African electrical power grid. In contrast, they are burdened by internal winding faults predominantly on account of insulation system failure. It is essential that these faults must be swiftly and precisely uncovered and suitable measures should be adopted to separate the faulty unit from the entire system. The frequency response analysis (FRA) is a technique for tracking a transformer’s mechanical integrity. Nevertheless, classifying the category of the fault and its gravity by benchmarking measured FRA responses is still backbreaking and for the most part, anchored in personnel proficiency. This work presents a quantum leap to normalize the FRA interpretation procedure by suggesting an interpretation code criteria based on an empirical survey of transformers ranging from 315 kVA to 40 MVA. The study then proposes an analysis of variance (ANOVA) based interpretation tool for diagnosing the statistical significance of FRA fingerprint and measured profiles. The latter cannot be relied upon by an expert or by the naked eye. Additionally, descriptive FRA frequency sub-region data statistics are proposed to evaluate the shift in both the magnitude and measuring frequency characteristics to formulate the recommended interpretation code criteria. To corroborate the code criteria by incorporating ANOVA and descriptive statistics, the study presents various case studies with unknown FRA profiles for fault diagnosis. The results constitute proof of the reliability of the proposed code criteria and a proposed hybrid of ANOVA and descriptive statistics.