Robust Thrust Estimation Method for Electric Propulsion Under Varying Axial Inflow

Direct drive motor propeller units offer compact structure, high reliability, and fast torque response, yet thrust is still regulated indirectly through rotor speed. Under time varying axial inflow, the speed to thrust mapping degrades, and aerodynamic inversion near hover becomes ill-conditioned due to vanishing sensitivity and potential loss of bijectivity. This paper proposes a sensorless thrust estimation method for direct drive systems. Power based inversion is reformulated as a regularized global search over a discretized inflow domain. A normalized sensitivity metric drives a finite state machine that switches between dynamic inversion and a static mapping fallback upon ill-conditioning detection. An online adaptation law updates the static coefficient from trusted dynamic estimates, keeping switching transients small. The method is validated in simulation with detailed PMSM drive dynamics and nonlinear propeller aerodynamics under time varying inflow. Against static speed mapping, torque-based inversion, and Newton-style power inversion, the proposed method achieves lower estimation error and noticeably mitigates the start up dead zone across the tested conditions.