In the attitude control of CubeSats, a hybrid actuator configuration that combines reaction wheels and thrusters is an effective approach that mutually compensates for the shortcomings of each actuator. However, in over-actuated systems, where the num...
In the attitude control of CubeSats, a hybrid actuator configuration that combines reaction wheels and thrusters is an effective approach that mutually compensates for the shortcomings of each actuator. However, in over-actuated systems, where the number of actuators exceeds the degrees of freedom of the control objective, infinitely many thrust-combination solutions satisfy the upper-level controller command, and when the required torque lies outside the Attainable Control Set (ACS) of the actuators, searching for a feasible optimal solution in real time becomes difficult. Conventional pseudo-inverse-based methods have a low computational burden but incur tracking errors due to clipping correction when the actuator constraints are exceeded, whereas optimization-based methods offer superior control performance but are constrained in real-time implementation owing to their high computational burden.
This thesis proposes a two-stage thrust allocation algorithm that simultaneously achieves computational efficiency and performance optimization. In the first stage, an ACS-geometry-based intersection-facet Look-Up Table (LUT) is employed to perform LUT-based intersection computation only when an infeasible command is received. In the second stage, the null-space parameters are adjusted using an Artificial Neural Network (ANN) trained on null-space parameter datasets generated through quadratic-programming-based optimization, in which fuel efficiency and thrust equalization are defined as the cost function. Unlike conventional approaches that directly infer the final thrust command, the proposed ANN infers only the null-space parameters, which do not affect the total torque. When a command lying outside the ACS is received, the proposed algorithm computes the intersection point with the ACS boundary through the LUT and then predicts and applies the optimized null-space parameters using the ANN, thereby maintaining fuel efficiency while keeping the computational burden low.
To verify the effectiveness of the proposed thrust allocation algorithm, it was applied to the Attitude Determination and Control System (ADCS) of a 6U CubeSat operating in an Elliptical Lunar Frozen Orbit (ELFO) for a Lunar Navigation Satellite System (LNSS) mission. The overall attitude system performs LQR (Linear Quadratic Regulator)-based attitude control and EKF (Extended Kalman Filter)-based attitude determination, and employs thrusters for attitude control and for momentum dumping upon reaction wheel saturation. Through Software-In-the-Loop Simulation (SILS) conducted under lunar environment conditions, the tracking-error reduction and fuel-efficiency improvement of the proposed algorithm were confirmed, and the satisfaction of the mission requirements was verified.