Abstract: Vibration phenomenon, as one of the main manifestations of faults in hydro-generator units, contains a wealth of fault data information. This paper first utilizes the acoustic detection method as a means of vibration detection in hydro-generator units, outputting its monitoring process. Based on this, it designs and constructs an online abnormal vibration analysis system for hydro-generator units, which includes four core components: sensor network, data acquisition and transmission module, data processing module, and data storage and access platform. In order to make the vibration analysis of key equipment in hydro-generator units more reliable and effective, this paper proposes two processes for vibration fault diagnosis based on data-driven and knowledge base approaches. Finally, using the case study of the water turbine unit at the Shilianghe Reservoir Hydroelectric Power Station in Jiangsu Province, the practical effects of the two processes are verified, with the knowledge base-based process achieving the best results, showing a fault detection accuracy of 95% and a fault diagnosis speed of only 1.5 minutes.