Abstract: With the increasing demand of power energy, the safe and stable operation of high parameter and large capacity turbo-generator sets is of great significance to power production. The vibration fault mechanism, signal detection, signal analysis, feature extraction and fault diagnosis methods of turbine generator set were summarized, respectively. Moreover, an advanced sensing technology and a new generation of intelligent machine learning technology represented by deep learning were introduced to solve the problems that traditional intelligent diagnosis methods are faced with, such as large amount of sampled data, difficulty in extracting signal features and shortage of fault training samples. It is summarized that the future vibration fault diagnosis technology of turbo generator sets should be based on artificial intelligence, big data, and cloud computing, supplemented by fusion virtualization and three-dimensional visualization technology, to achieve the unity of fault diagnosis speed and accuracy.