Abstract:Due to the limited amount of gas turbine fault samples in power plants in active service and previous fault diagnosis methods relying on a significant number of data with fault labels, which can’t achieve the expected diagnostic effect in practical production, this paper focuses on using Bayesian network for counterfactual reasoning to analyze the causes of gas turbine faults. This paper first introduces the basic principles of Bayesian networks, then applies fault mode and effects analysis as well as fault tree analysis to construct Bayesian networks, addressing the lack of professional knowledge support in data-driven fault diagnosis methods. Finally, through practical case analysis, it is demonstrated that this method can analyze possible faults and their causes based on abnormal phenomena during gas turbine operation, helping operators and maintenance personnel to timely identify and resolve faults. This flexible, efficient, and reliable method provides a new approach for gas turbine fault diagnosis in practical production.