NEURO-FUZZY SYSTEM AND GENETIC ALGORITHM FOR CUBESAT LQR BASED ATTITUDE CONTROL SYSTEM

CubeSat is next generation, promising, and small-sized satellites that can be easily assembled using commercially available components with a low investment cost. Attitude control is one of the core subsystems in the CubeSat system that deals with how to orient the CubeSat in any desired direction. Proper attitude controller of satellite design is necessary since devices that need pointing direction like antennae, camera, and some measurement devices are always mounted on the satellite. Uncontrolled CubeSat may even cause the entire mission loss.
Fuzzy controllers and linear quadratic controller are among the commonly employed controllers in CubeSat attitude control. Although manually tuned linear quadratic controller design shows good performance without actuator saturation, it is not optimal in handling the tradeoff between the desired performance and actuator saturation. A genetic algorithm is employed to handle this optimization problem. Besides, developing fuzzy controllers is challenging for multiple inputs and multiple outputs systems using expert knowledge and intuitive rational guess. An adaptive neuro-fuzzy inference system is proposed to develop a fuzzy system using training data sampled from the simulation of a genetically tuned linear quadratic regulator. These fuzzy systems mimic the linear quadratic regulator that is tuned by the genetic algorithm. MATLAB is used for the genetic algorithm optimization of linear quadratic regulator and learning based fuzzy controller design. The attitude kinematics is modeled using quaternion while the dynamics of the CubeSat's attitude considers reaction wheel actuation and gravity gradient torque