Abstract: A Bayesian network model is adopted in the maintenance of hydroelectric generating sets to provide maintenance personnel with a better tool in decision making of auxiliary maintenance. Based on the sensor diagnosis strategy and a Bayesian network model, we design a condition-based maintenance method that uses the strategy to handle the cases of false alarm, redundant alarm, and potential fault diagnosis on sensor monitoring signals. We apply the Bayesian network model to effective diagnosis of transmission faults, and carry out risk assessment via combining fault probability and fault risk. The results of risk assessment provide reasonable information for the maintenance personnel to set up a procedure for failure checking during maintenance. Finally, we use accuracy analysis to analyze the method and its practical operation, and verify the calibration curve by comparing the modeled results with field data, showing that the operating accuracy of the system is 80%.