Multi-span rotor system in air separation plant shows more and more coupled and serious faults as the equipment becomes larger and more complicated, and the online diagnosis and decision making for the fault is significantly important to the stable operation of the whole system. So a model based method for online quantitative multi-fault diagnosis of multi-span rotor system is presented. Based on finite element and harmonic balance theory, dynamic characteristics of rotor system are analyzed, and the relationship between harmonic components of vibration responses and frequency response matrix of residual equation in fault rotor system is used to derive theory and obtain diagnosis equation. The finite element models of a multi-span rotor system with rub-impact and coupling misalignment faults is established and the effectiveness and robustness of the proposed method is verified by numerical experiment and then by test rig experiment. The advantage of the proposed method is that n fault locations can be detected on line by using at least vibration response data on n 1 nodes of rotor system before and after faults occur at the same rotating speed, and also the method has a good applicability for detecting the main vibration faults in rotor system such as rub-impact, misalignment and crack.