The periodic signals acquired from a rotor by equal-time-interval sampling stop being periodic during the startup or shutdown stages, making methods based on Fourier transform no longer applicable. To estimate vibration of a startup or shutdown rotor with both non-stationary amplitude and frequency, qualitative and quantitative analysis of a startup rotor was performed with fractional Fourier transform (FRFT), and a new rotor startup running estimation method was proposed. In this method, fault feature extraction of the rotor was performed with FRFT combined with the holographic principle, and feature classification was conducted with the fractional PCA method. Experimental results showed that the proposed method effectively solved the problem of extracting transient information in the startup process of a rotor, precisely extracting the fault feature, and accurately realizing fault feature quantitative classification of several different typical faults, as well as the same rotor fault at different degrees.