Model Based Automatic Tuning and Control of a Three Axis Camera Gimbal

Sammanfattning: A gimbal is a pivoted device that decouples movements of a platform from its payload. The payload is a camera which must be stabilized to capture video without motion disturbances. A challenge with this type of gimbal is that a wide span of cameras with different sizes and weights can be used. The change of camera has significant effect on the dynamics of the gimbal and therefore the control system must be retuned. This tuning is inconvenient, especially for someone without knowledge of control engineering. This thesis reviews suitable methods to perform an automatic controller tuning directly on the gimbal's hardware.This tuning starts by exciting the system and then using data to estimate a model. This model is then used to control the gimbal, thus removing the need for manual tuning of the system. The foundation of this thesis is a physical model of the gimbal, derived through the Lagrange equation. The physical model has undetermined parameters such as inertias, centre of gravity and friction constants. System identification is used to determine these parameters. A problem discussed is how the system should be excited in order to achieve data with as much information as possible about the dynamics. This problem is approached by formulating an optimization problem that can be used find suitable trajectories. The identified model is then used to control the gimbal. Different methods for model based-control are discussed. By using a method called feedback linearisation all of the parameter-dependant dynamics of the gimbal can be compensated for. Apart from being independent of model parameters the new outer system is also decoupled and linear. A PID controller is used for feedback control of the outer system. The uncertainty of the feedback linearisation is analysed to find the effects of model errors.To assure robustness of the closed loop system a Lyapunov redesign controller is used to compensate for these model errors. Some experimental results are also presented. The quality of the estimated model is evaluated. Additionally, the reference tracking performance of the control system is tested and results reveal issues with the estimated model's performance.