State- and Parameter Estimation for Spacecraft with Flexible Appendages using Unscented Kalman Filters

The problem of system identification for dynamic effects on spacecraft has become increasingly relevant with the surge of agile spacecraft, which must perform large amplitude maneuvers at high rates. Precise knowledge of the state of the spacecraft, as well as of the parameters characterizing its motion, is vital for the design of control algorithms enabling stabilization and pointing accuracy. Traditional rigid body models and estimation methods are no longer sufficient to provide this knowledge. This thesis focuses on estimation of flexibility effects and spacecraft parameters through methods based on the unscented Kalman filter, an estimator for nonlinear dynamic systems. A spacecraft model consisting of a rigid central body and a flexible appendage described as an Euler-Bernoulli beam in pure bending is built, and equations for its translational and rotational motion, as well as the deflection of the beam, are derived in the Newton-Euler framework considering the first bending mode of the flexible deformation. Observability tests are successfully conducted to ensure that estimation of the relevant states and parameters can be performed exclusively from linear and angular velocity measurements. A total of eight filters, estimating the spacecraft’s state along with different combinations of parameters, are developed, implemented, and tested on simulated data. Grouped under the common denomination “UFFE” (Unscented Filter for Flexibility Effects), they are made available as Simulink library blocks. State estimation is performed for the linear and angular velocities of the spacecraft and the modal coordinate and velocity of the appendage, with estimates following closely the truth model of the state variables and estimation errors at least an order of magnitude lower than true state values. Simultaneous state and parameter estimation is implemented from two approaches, joint estimation and dual estimation, whose performance and applications are compared. Estimated parameters include the moments of inertia of the system and natural frequency, damping ratio, and modal participation factors of the flexible appendage. Convergence to true parameter values is reached in the first 100s of the estimation for inertia terms and natural frequency, while the estimation for modal participation factors is conditioned to precise tuning of the filter. Estimates of the damping ratio are biased, most likely due to the control input not being optimal for observation of this parameter. The dual approach to parameter estimation is found to be advantageous when proper filter tuning is possible, as it enables the continuous operation of a state filter combined with short runs of the parameter filter activated at will; this configuration could be employed to track the variation of spacecraft parameters along space missions. The causes of estimation error are identified and methods for automatic tuning of the process noise and process noise covariance are researched. Five such tuning techniques are implemented and tested, with promising results found for online sampling of the process noise covariance through Monte Carlo methods. A discussion on the limitations of the chosen dynamic model and estimator, along with recommendations for extensions and future applications, concludes this work.