Abstract: Aiming at the problems that the vibration signals of turbo-generator rotor are of the multi-component and non-stationary kind, a new fault diagnosis method was proposed based on ALIF-HT by combining the novel signal decomposition method-adaptive local iterative filtering (ALIF) with the Hilbert transform (HT). The specific way is to decompose the original vibration signals of turbo-generator rotor into several signal components by ALIF, then to compute the instantaneous frequency of every component by HT to get the complete time-frequency representation of all signal components, and finally to identify the fault types according to the time-frequency characteristics. The ability of ALIF in decomposing multi-component signals was demonstrated through analysis on simulated signals and the engineering practicability of the method was verified via fault analysis of rotor oil film instability. Results show that the ALIF overcomes the mode-mixing problem existing in EMD and the ALIF-HT method has higher time-frequency accuracy than the HHT method.