Stochastic model predictive control and its application for small satellite attitude control

Thesis

Model predictive control is an attractive control method applicable to a wide range of real-world processes with the ability to handle systems subject to probabilistic constraints and stochastic disturbances. While many methods exist that handle disturbance effects, there are fewer methods that are equipped to handle systems with measurement errors. Additionally, whereas model predictive control is used in a wide range of engineering processes, its use is limited in space applications. This thesis will investigate the use of model predictive control, in a manner that handles stochastic measurement errors, to control the small satellite FalconSAT-5.

This study focused on model predictive control methods that construct a region of tubes in which the state is required to fall in order to meet the probabilistic constraints. Three such methods were studied. The first method used the probabilistic distributions of the disturbances to tighten the constraints so that, even in the presence of disturbances, the constraints would still be met. This method was then expanded to handle both stochastic disturbances and stochastic measurement errors by also including the distribution of the measurement errors as an additional constraint tightening factor. This method was superior to the first method when the system exhibited both disturbances and measurement errors. The third method used state estimation to avoid the errored measurements. When the disturbances have a greater effect on the system response, the method that considers the measurement errors was more effective than the estimation method. However, if the measurement errors have a greater effect, the estimation method was more effective. The model predictive control method was then applied to the FalconSAT-5 dynamic model for all the control modes necessary for satellite operation and the results of these control simulations were compared to the current satellite control scheme of proportional-derivative control.

The simulations showed that, for FalconSAT-5, the model predictive control method that includes the measurement error effects in the constraint definition was superior to the estimation method. This control method was used to simulate control of the satellite in the initial deployment phase, also called the detumble phase, for pointing control, and for control during a slew manoeuvre. This method was found to be superior to the proportional-derivative control in all three control modes in convergence time,control input required, and cost comparison. This study showed that it is possible to develop model predictive control algorithms that handle measurement errors and are applicable to small satellite control. Model predictive control is a viable alternative to classical control for space applications, providing superior performance and better control.

Authors: Yadlin, R

• Publication date: 2011
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: Oxford University, UK
• Language: English
• Keywords: model predictive control; astronautics; stochastic; probabilistic constraints; stochastic tubes; small satellite control
• Subjects: Aeronautical research; Control engineering; Engineering & allied sciences