A recurrent neural network-based rotor displacement estimation method for eight-pole active magnetic bearing

Abstract

Active magnetic bearing (AMB) is a key technology in high-speed rotating machines for rotor suspension, where the displacement sensors play a crucial role in detecting and controlling the rotor position. However, the traditional displacement sensors have the problems of high cost, large volume and poor reliability. To solve these problems, this paper proposes an innovative solution that utilises a recurrent neural network (RNN) to estimate the rotor displacement from the current in the AMB controller. The proposed method offers high-quality prediction performance for the rotor displacement which is close to the high precision eddy current displacement sensors. The mathematical model of AMB is analysed to provide guidance in historical current data acquisition and design of RNN. The input dimensions and the architecture of the neural network are optimised to improve both prediction accuracy and computational complexity. Experimental results validate the effectiveness of the algorithm and demonstrate that the proposed method has high accuracy, robustness and generalisation ability.