In order to improve the diagnosic accuracy of the charging generator rotor vibration fault, the work presented in this paper focuses on using frequency spectrum analysis and Bayesian networks to diagnose the typical vibration fault of generator rotor vibration. The vibration fault has properties of randomnessand layers, and its information has the features of uncertainty and non-integrality. The method of typical rotor vibration fault diagnosis is studied based on Bayesian networks, which uses expert knowledge to determine conditional probability, turning fault information into numeric vectors. Then, the Bayesian network model of vibration fault diagnosis is established using spectrum data. The clustering algorithm is improved to minimite calculation and enhance diagnostic accuracy by optimizing connections between clustering nodes. Based on the condition that the known information is fuzzy and incomplete, the results of the experiment indicate that the Bayesian network-based method can improve the accuracy of diagnosing typical generator rotor vibration faults.