Integrating AI based DGA fault diagnosis using Dempster–Shafer Theory

Conventional dissolved gas analysis (DGA) methods and artificial intelligence (AI) techniques based on DGA data have been used for long to diagnose incipient faults in transformers. The Dempster–Shafer Evidential Theory (DST) has been applied to various AI oriented applications where there is uncertainty and conflict. This paper uses DST to integrate the results of incipient fault diagnosis of back propagation neural networks (BP-NN) and fuzzy logic, so as to overcome any conflicts in the type of fault diagnosed. The proposed approach is applied to independent data of different transformers and case studies of historic data of transformer units. This method has been successfully used to identify the type of fault developing within a transformer even if there is conflict in the results of AI techniques applied to DGA data.

► We have integrated two evidences using Dempster–Shafer Theory for transformer fault diagnosis. ► The dissolved gases in oil are taken as input for both the evidences. ► First evidence is normalized output of BP neural networks and second evidence is normalized results of fuzzy logic. ► Results are compare with IEC/IEEE method and proposed method is applied to historic data/case studies. ► The method can detect faults when conventional methods fail and there is conflict in the diagnosis by AI methods.