For the acoustic emission signals of journal bearings are with of the characteristics of random

transient

timevarying

polymorphism and susceptible to interference

the correlation dimension has been taken as characteristic value to diagnose the lubricating fault in order to extract early failure characteristics effectively and improve the efficiency of online monitoring. Firstly

the acoustic emission signals of 310 MW turbo-generator set by mean of field test is analyzed to obtain its fractal characteristics

and then compared with traditional acoustic emission characteristic value. Then the lubricating condition of the journal bearing is quantitatively analyzed by means of correlation dimension. Finally

the algorithm of classic G-P correlation dimension is improved due to larger computational complexity. The proved algorithm is of more efficient than the classic by comparing the calculation time between them. The results show that correlation dimension not only have positive correlation and correspondence with the lubricating condition of the journal bearing

but also is more sensitive to characterization of journal bearing early faults. The improved correlation dimension algorithm can effectively improve the monitoring efficiency

more suitable for the real-time monitoring field to ensure the safe and stable operation of the unit.