Disclosed is a water turbine fault prognosis method based on on-line monitoring. On-line monitoring of the pendulum state of a water turbine set and real-time fault prognosis are realized through collection of a pendulum data sample of the water turbine set, real-time library reading and writing, data preprocessing and making of inference rules. The water turbine fault prognosis method based on on-line monitoring comprises the steps of collecting and displaying real-time data, realizing fault early warning and vibration tread displaying, recording and displaying the diagnosis result, fault analysis and processing suggestions, and storing diagnostic reports.