Rotor Displacement Self-sensing Technology of Six-pole Radial Active Magnetic Bearing Based on KalmanNet

In this paper, a KalmanNet-based self-sensing method for radial displacement of a six-pole radial active magnetic bearing rotor is proposed to replace sensors and improve system reliability. First, a mathematical model of the levitation force when the rotor is eccentric is derived using the Maxwell tensor method, and the left-reversibility as well as the observability of the system are verified to validate the feasibility of using KalmanNet for displacement prediction. Then for the high nonlinearity of the system, a rotor displacement prediction model is constructed using a hybrid architecture of KalmanNet physical model constraints and data-driven compensation. Considering the oscillation problem triggered by the sudden speed change, a gated loop unit is used to memorize the state information before speed switching to achieve a smooth transition. Finally, the proposed method is validated on a six-pole radial active magnetic bearing experimental platform. The experimental results show that the proposed KalmanNet-based rotor displacement self-sensing method can realize stable levitation operation.