A novel approach to power transformer fault diagnosis based on ontology and Bayesian network

This paper proposes an application framework to reinforce the existing process for ontology-based transformer fault diagnosis with formal probabilistic semantics using the Bayesian Network. This framework allows users to quantify a certain fault with Bayesian Network, based on the knowledge embedded in a transformer ontology regarding relationships of faults and their features such as causes, symptoms and related diagnostic methods. Firstly, the essential principle of Ontology and Bayesian Network are introduced. Then a transformer ontology, which can capture the knowledge of fault diagnosis for transformers, is described as the basis of the developed framework. This framework has three functionalities, i.e. ontology-based Bayesian network generation, uncertainty assignment, evidence assignment and beliefs update. All of these functionalities are discussed in detail. Finally the proposed framework is exemplified by a nine-node Bayesian Network model for transformer fault diagnosis, which demonstrates the potential of the developed framework for transformer fault diagnosis.