Dissolved gas analysis using evidential reasoning

Abstract

A novel approach to the analysis and handling of dissolved gas analysis (DGA) data from several traditional methods, namely Roger's Ratio Method, Dornenburg's Ratio Method and the Key Gas Method, is presented. Ideas taken from fuzzy set theory are applied to ‘soften’ the fault decision boundaries employed by each of the three methods. This has the effect of replacing traditional Fault or No Fault crisp reasoning diagnoses, with a set of posssible fault types (i.e., those fault types distinguishable by one particular method) and an associated probability of fault for each. These diagnoses are then considered as pieces of evidence ascertaining to the condition of the transformer and are aggregated using an evidential reasoning (ER) algorithm. The results are presented as probabilities of four possible general fault types: overheating of cellulose (cellulose degradation), thermal faults, partial discharge and arcing (corona). Finally the remaining belief is assigned to the possiblity that no fault exists. The results show that the pseudo fuzzy representations of the traditional methods, perform adequately over a wide range of test values taken from actual failed transformers, and that the overall system can effectively combine the evidence to produce a more meaningful and accurate diagnosis.