Neural Network Identification of an Axial Zero-Bias Magnetic Bearing

Abstract

System identification for magnetic bearings can be used for various purposes, such as aggressive control design, on-line monitoring and early diagnosis of upcoming faults in rotating machinery, etc. Especially for the zero-bias active magnetic bearing (AMB) system, which omits the bias current/flux to reduce power loss and rotor heating, whereas brings out the problem of highly nonlinearity. Here, we applied the NARX neural network to the identification of an axial zero-bias magnetic bearing system. The I/O data are acquired through a stabilized magnetic bearing test rig, which are more accurate than data that generated by simulation. Two kinds of training methods are compared in different working conditions. The results show a good ability of generalization, which prove the feasibility of the NN identification.