Abstract: An identification is proposed for the torque disturbance in a rotating multi-bearing rotor system. It is based on negative selection algorithm of artificial immune system taking the rotor speed and bearing load as feature vector. First of all, self-patters of the rotor system were obtained according to the normal working status and the initial detectors were produced at random. Secondly, non-self-patterns responded to the abnormal working status after the disturbance of the rotor system were learnt and memorized, taking advantage of the evolution learning mechanism based on the artificial immune theory. At last, the mature detector was produced after evolution learning under the typical torque disturbance, and the corresponding zones of different torque types on states space were distinguished and marked using the mature detector. The experimental results show that the method is effective in detecting abnormal torque disturbance of the rotor system and identifying the torque type.