A neural network-based closed loop identification of a magnetic bearings system

This paper deals with the identification of a radial-type active magnetic bearing (AMB) system using Artificial Neural Network (ANN). Identification and validation experiments are performed on a laboratory magnetic bearing system. Since the electromechanical configuration is inherently unstable, the identification data is gathered while the AMB is operating in closed loop with a controller in the loop. From this data, the identification procedure generates an open-loop plant model. A NNARX (Neural network autoregressive external input model) structure is proposed and evaluated for emulating the system's dynamic. The model is implemented by a Neural network, constructed using a multilayer perceptron (MLP) topology, and trained using as inputs the rotor's displacements and excitation currents. Validation tests are performed under perturbation conditions (impact applied on the rotor). Results show that the neural network based model presented here is a powerful tool for dynamic plant's identification, and that it could be also suitable for robust control application.