Design of high-performance digital twins to enhance magnetic resonance imaging patient positioning

Precise MRI patient positioning requires high accuracy across varying patient weights and over the system’s operational lifetime, despite strong magnetic fields. Digital twins can improve motion control by adapting to such changing conditions, but the choice between high-fidelity and lightweight models is not straightforward. This paper presents two digital twin architectures for the belt-driven positioning system of a commercial MRI scanner and evaluates them on the same hardware as two complementary case studies, each against the disturbance class it targets. The first is a physics-based twin built on a multibody model with over 50 parameters, updated monthly through offline system identification and constrained controller optimisation to compensate for mechanical ageing and wear. The second is a lightweight performance twin using a second-order two-parameter model; it identifies the dominant dynamics online within one second of each movement and provides real-time feedforward compensation for weight variations. Both augment the same cascaded PI controller. Experimental validation shows that the physics-based twin adapts to induced ageing effects, while the performance twin reduces positioning errors across the full patient weight range. As each architecture addresses a different disturbance timescale, combining periodic deep calibration with continuous online adaptation can provide robust long-term positioning performance.