Monitoring the vibration state of engine rotors is one of the important methods to improve the reliability of aero-engines. However, in actual operation and vibration monitoring process, it is found that the engine speed changes rapidly, and the rotor vibration component is weak, which makes it difficult to track and extract the rotor harmonic vibration component in real time. To solve this problem, this paper proposes an extraction method for aero-engine rotor vibration components based on sparse harmonic product spectrum and adaptive Vold-Kalman filter. Firstly, the characteristics of the engine casing signal is analyzed. Through combining the harmonic relationship between blade passing frequency and rotation frequency with the idea of harmonic product spectrum, a method for calculating the engine rotation frequency rapidly is proposed, without the need for accurate phase information. Secondly, the residual error of the order spectrum is minimized by using the method of variable step iteration, and the optimal filtering parameters of the Vold-Kalman filter are determined, to realize the accurate extraction of the harmonic vibration component of the engine rotor. It is verified by simulation signals that this method can effectively extract the weak rotor vibration components from the casing signal under relatively large noise, and this method is compared with a variety of classical signal decomposition methods. Finally, actual engine signals are further analyzed, and the calculation and verification are carried out for two typical unsteady working conditions of engine start/stop and acceleration/deceleration, which proves the superiority of the proposed method.