Abstract: In order to improve the identification accuracy of vibration faults of hydropower units,a vibration fault diagnosis method based on CEEMDAN-ELM-Adaboost is proposed in this paper. Firstly,the original vibration signal of the unit is denoised by fully adaptive noise ensemble empirical mode decomposition（CEEMDAN）,and the sample entropy of the main IMF component is extracted. Then,the hybrid feature vector is constructed by combining the conventional time-domain and frequency-domain features. Finally,the extracted mixed feature vectors are input into ELM-Adaboost to build an intelligent fault diagnosis model for hydropower units,so as to realize high-precision classification diagnosis of vibration faults of hydropower units. Taking the rub-impact fault of the runner room of a hydropower station as an example,this paper proves that the proposed fault diagnosis model based on CEEMDAN-ELM-Adaboost has advantages over the traditional model.