Abstract: A fault prediction method for shafting of main engine is developed based on shaft vibration monitoring. Ensemble Empirical Mode Decomposition（EEMD） and Enhanced Intermittent Unknown Input Kalman Filter（EIIKF） are introduced into the fault prediction method. The vibration signal is mixed with a white noise before decomposition to prevent the modal mixing and improve the decomposability. The vibration signal, after filtering and reconstruction, is processed by sequential analysis to get the characteristic curve of the signal. EIIKF is used to analyze the characteristic curve and do working status prediction. In this processing, by introducing intermittent parameters, the uncertainty caused by some unknown input items is compensated. Fault diagnosis is carried out by checking the working status against a fault discrimination model. The method is verified with actual data from engine operation. The fault prediction capability of the method is seen better than that of conventional mode decomposition and Kalman filtering in terms of accuracy and timeliness.