Aiming at the shortcomings of the existing time-frequency analysis methods in the diagnosis of rotor oil film instability faults， synchro-squeezed wave packet transform （SSWPT） is proposed to analyze the non-smooth multi-component signals during the operation of the steam turbine. By selecting different main frequency wavelets for different kinds of signals， the signal time-frequency graph is obtained， and the original signal can be reconstructed from the time-frequency graph by the algorithm， and the accuracy of the existing time-frequency method is compared. Taking a 1 000 MW unit of a power plant as the research object， the rotor oil film oscillation fault diagnosis is carried out by using SSWPT method in view of the major vibration problem of shafting during commissioning. Using the on-site turbine diagnosis managment （TDM） system to collect data and carry out wavelet packet transformation to obtain the wavelet transform coefficient and the instantaneous frequency of non-stationary fault signals. Finally， the wavelet packet transform coefficient is compressed under the instantaneous frequency scale. More accurate frequency components are obtained. The results show that the method is superior to the feature extraction of non-stationary signals， and can accurately judge the location and type of faults， which provides a reliable basis for the unit's later fault processing.