FPGA-Based LQG Control of a Low-Speed Active Magnetic Bearing Joint: Implementation and Validation

Abstract Active magnetic bearings (AMBs) enable contactless support and are attractive for precision mechanisms in particle-sensitive environments. This work presents a complete discrete-time linear-quadratic-Gaussian (LQG) control pipeline on a field-programmable gate array (FPGA) for a low-speed robotic joint using a three-degree-of-freedom active magnetic bearing (two radial axes and one axial levitator). A translational physics-based model explicitly captures bias currents and nominal air gaps, followed by equilibrium linearization and zero-order-hold discretization at 2 kHz. The estimator–controller ( linear–quadratic regulator , LQR, plus Kalman filter) is implemented in fixed-point logic with sub-200 \((\mu)\)s end-to-end latency. Experiments on a dedicated testbed compare the proposed LQG against a tuned proportional–integral–derivative (PID) controller and an observer-based pole-placement baseline under identical sensing and actuation. In the radial loop, LQG reduces startup settling time by approximately 44% and integrated absolute error by approximately 36.5%, with overshoot below 0.5% and actuator limits respected; recovery to a \((\pm 5%)\) band under external disturbances is about 13% faster. A nonlinear-plant robustness map indicates stable, non-overshooting behavior under \((\pm 20%)\) variations of magnetic and electrical parameters. Particle measurements during continuous operation remain below the International Organization for Standardization (ISO) Class 5 thresholds, supporting the suitability of the proposed joint for cleanroom robotics.