Enhanced condition monitoring of power transformers through improvement in accuracy of DGA interpretation

Transformer failure in electricity supply grids has a high financial impact, due to failure to meet commercial contract and possibility of transformer replacement cost. Therefore, detecting fault inception is very important in order to keep the transformer operating with recommended efficiency, hence ensuring stability of the electric network. Dissolved gas analysis (DGA) of a transformer can provide clear indication of thermal and electrical stresses on power transformer insulation and is considered as one of the most effective tools for oil-filled power transformer diagnostics. DGA is used to detect incipient faults in order to manage the fault severity. Both on-line and off-line condition monitoring methods can be applied to obtain gas content, thereafter there are many interpretation techniques for DGA results. The accuracy of these techniques is dependent on the operator’s experience and knowledge of the materials and equipment involved. In this work, a combined fuzzy logic analysis technique for monitoring of power transformers based on DGA analysis is proposed, the system uses the 7 key gases to diagnose the health of the transformer and, where applicable, fault type. Initially, gas levels are considered using the IEEE standard as a basis to indicate the health of the transformer. A combined fuzzified analytical tool, based on Duval Triangle, Doernenburg ratio and Key gas method, are analysed to identify the fault type, improving on the accuracy of the individual interpretation techniques. The analytical tool has been applied to 444 sample faults reported in the literature to assess the accuracy of the proposed system. Results presented show the system’s overall decision has improved capability of identifying the transformer condition over individual methods. The proposed system is proved to have 99 % accuracy in identifying the transformer normality. For cases where transformers were faulty, the approach has 98.76% accuracy in recognising the actual fault, superior to individual approaches.