Self-sensing magnetic bearings: From indirect inference to sensor architecture design

초록

Active magnetic bearings (AMBs) conventionally rely on dedicated displacement sensors, which increase cost, integration complexity, and reliability risk and can limit collocated bending-mode observation in ultra-high-speed rotors. Eliminating external sensors requires extracting displacement information directly from actuator electrical signals, allowing the bearing itself to function as an integrated electromagnetic displacement transducer. Self-sensing magnetic bearings address this objective by exploiting the dependence of coil inductance on rotor position through magnetic reluctance variation in the air gap. However, existing approaches-including observer-based inference, high-frequency injection, pulse-width modulation (PWM) ripple demodulation, non linear compensation, and physics-informed learning-have largely evolved without a unified framework that explains their sensing principles and performance trade-offs. This review organizes the field into a four-stage conceptual framework: indirect inference (Stage 0), induc tance probing (Stage 1), nonlinear compensation (Stage 2), and sensor architecture design (Stage 3). The central insight is that high-performance self-sensing depends not only on improved estimation algorithms, but also on the degree of physical separation between sensing and actuation. In particular, separate sensing and control windings can reduce shared-flux interference, improve signal-to-noise ratio, and provide a cleaner baseline for compensation and learning-based refinement. Recent studies combining inductance probing, nonlinear compensation, and sensor architecture design have reported performance approaching that of commercial eddy-current sensors under defined experimental condi tions, suggesting a credible route toward reduced reliance on external displacement sensors in industrial AMB systems.

키워드