1. Shenhua Guohua(Beijing)Electric Power Research Institute Limited Company, Beijing 100025, China; 2. China Real-time Technology Co. Ltd., Beijing 100192, China

As the future development trend of power equipment maintenance mode will be condition based maintenance, making judgment of early fault with weak feature-information or in the bud is the key to condition based maintenance. Therefore, the real-time information data(PI data)of large units was developed to build dynamic early warning models for several turbine generator units through the grey theory and similarity principle, realizing the early warning function of equipment vibration fault. A new method of intelligent monitoring and evaluation on the health state of equipment was provided by the hypersphere similarity analysis technology. And an abnormal state of the unit was discovered by this warning system as proof of the early warning function of vibration fault in large turbine generator sets in actual application.