Aman, Asma Mohammad Nusrat
(2023)
The development and performance evaluation of Particle Filter–H∞ Filter for attitude estimation of a small satellite in low earth orbit.
PhD thesis, University of Nottingham.
Abstract
The accurate determination of the attitude of small satellites is essential for missions with stringent pointing requirements. Small satellites such as CubeSats are subject to mass and volume constraints, which directly affects their ability to carry more extensive and demanding payloads and limits their power and computational capabilities. There is, therefore, a need to minimise the computational cost of attitude determination while maximising accuracy and robustness to make small satellites useful for a wide range of applications. This research investigates the use of low-cost commercial sensors for attitude determination, namely the three-axis magnetometer and sun sensors, and proposes a novel hybrid Particle Filter–H∞ Filter to facilitate higher performance in the attitude determination of CubeSats. Existing attitude determination methods based on Kalman Filters are often not robust enough, which reduces their estimation accuracy, and they are limited in their inability to handle non-Gaussian noise. While Particle Filters offer higher estimation accuracy and the ability to handle non-Gaussian noise, their high computational cost is a distinct disadvantage. This research focuses on the development of a novel attitude estimator, the Particle Filter–H∞ Filter, which aims to achieve high estimation accuracy, robustness and lower computational cost by incorporating an H∞ Filter into the resampling step of a Particle Filter.
The novel Particle Filter–H∞ Filter was developed and executed in a MATLAB simulation environment, in which an attitude determination system combining the relevant satellite environment models (the Simplified General Perturbations 4 orbit propagation model, the International Geomagnetic Reference Field model, the Sun position model), as well as three-axis magnetometer and sun sensor models and Euler’s equations for rigid body dynamics, including different simulation scenarios for a 3U CubeSat in low Earth orbit, was examined to verify the novel estimator. The simulation scenarios involve evaluating the estimator’s performance under different noise assumptions: Gaussian, non-Gaussian, increased process noise and coloured noise. Also, other simulation scenarios considered are magnetometer-only attitude estimation, large initial errors (observed during satellite separation and detumbling phase), the presence of measurement outliers and attitude estimation in a 6U CubeSat platform. The performance of the novel Particle Filter–H∞ Filter was evaluated against widely used nonlinear attitude estimators, namely the Extended Kalman Filter, the Unscented Kalman Filter, the H∞ Filter, and the standard Particle Filter, to determine its superiority with respect to the selected performance criteria: accuracy, robustness, computation time, and the ability to converge quickly and maintain convergence under different operating conditions. Compared to Particle Filters, the Kalman Filter-based attitude estimators are less computationally intensive, but their estimation accuracy is relatively poor. The results of various simulated scenarios have shown that the novel Particle Filter–H∞ Filter significantly improves the attitude estimation accuracy and has higher robustness compared to existing techniques. Furthermore, it reduces the computation time of the standard Particle Filter by 59.2%. The novel Particle Filter–H∞ Filter met the desired attitude estimation accuracy requirements of ±0.1° for an EO mission and showed the best performance in the simulation cases of non-Gaussian noise, magnetometer-only attitude estimation, measurement outliers, coloured noise and 6U CubeSat, demonstrating the robustness of the novel method in cases where the other attitude estimators do not provide the desired attitude estimation performance. For the case of Gaussian noise, the novel Particle Filter–H∞ Filter achieved a steady-state attitude accuracy of ±0.0526° for the roll axis, ±0.0360° for the pitch axis, and ±0.1005° for the yaw axis. Thus, this novel solution presents an advanced estimation algorithm that addresses the need for higher accuracy in small satellite attitude determination and robust performance that can be used in various scenarios of satellite operation and space environment noise. Some recommendations for future research include expanding the implementation of the attitude estimator to include other attitude sensors and further evaluating the estimator performance with an additional attitude control loop.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)