Dissolved Gas Analysis (DGA) is usually performed for power transformer condition based maintenance (CBM) activities. In reality, the interpretation of DGA data is complex and done through several knowledge sources. Lack of information within these knowledge sources necessitates an information fusion methodology. Constructed upon DempsterShafer (DS) theoretic approach, this paper proposes a novel interpretation rule and algorithm. Weibull functions were utilized to quantify the degree of belief in major fault according to respective DGA parameters. Threshold belief mass (THB) that distinguishes incipient fault from major fault, which is located within the interval of degree of belief in major fault were observed. In this paper, based on minimum threshold values, a novel rule termed as Threshold Belief Mass (THB) interpretation rule is proposed and compared with Dempster's and Yager's combination rule. Furthermore, from the result of THB interpretation rule and belief or plausibility function, different level of transformer's condition is suggested. Several operational and maintenance decision-making proposal are introduced for a system operator with respect to the belief and plausibility function. In summary, the new method enables a system operator to interpret DGA data more systematically.

Diagnostic knowledge derived from dissolved gas analysis (DGA) is complex, imprecise and may lack information about the actual condition of power transformer. To deal with this uncertainty, Dempster-Shafer (DS) Theory is often utilized for the fusion of evidence obtained from various DGA diagnosis [1,2]. In the Dempster-Shafer theoretic approach, several rules for combining the evidence were proposed [3,4,5], amongst them is unified combination rule by Inagaki [5]. Unified combination rule unifies various types of combination rule from Yager's combination rule to extrapolation combination rule through the introduction of functions. The combination of belief structures derived from power transformer's DGA evidence is implemented through Inagaki's unified combination rule. Expected monetary value [6] was utilized for determining the most optimum combination rule of the unified combination rules. Furthermore, we proposed maximax and maximin criterion for selecting the best decision policy for the mapping of belief structures from transformer's condition spaces to system operator's decision making spaces.