Physics-aware adaptive gain scheduling for magnetic detumbling of a PocketQube satellite with embedded hardware validation

Magnetic detumbling is a critical early-phase requirement for PocketQube satellites, and classical B-dot control remains widely used due to its simplicity and robustness. However, fixed-gain implementations are typically tuned conservatively to ensure stability across varying geomagnetic conditions, which can limit performance. This work introduces a physics-aware adaptive gain scheduling approach that augments the classical B-dot controller by modulating its scalar gain as a bounded function of geometry and energy-related state variables. A compact feedforward neural network is used to compute gain adjustments while preserving the original control structure. The method is evaluated through a 200-case Monte Carlo campaign under realistic low Earth orbit magnetic-field conditions representative of spring-ejection deployment rates. Compared to a conservatively tuned fixed-gain baseline, the adaptive approach reduces mean detumbling time by 14.2%, with improvements most evident under unfavourable magnetic geometries. Real-time feasibility is demonstrated on flight-representative embedded hardware, confirming negligible computational overhead within a 10 Hz control cycle. The results suggest that physics-aware gain adaptation can enhance detumbling performance while maintaining bounded behaviour and embedded-system compatibility, offering a practical route toward data-assisted control in resource-constrained small satellites.