| dc.description.abstract | Sophisticated guidance, navigation and control (GNC) algorithms play a vital role in the success of space missions. The complex GNC schemes must meet the mission’s robust stability and performance requirements despite uncertainties and disturbances. Identifying any lack of robustness of the GNC scheme at the earlier stages of the design is economical and reaffirms the success of missions. Given the stability and performance requirements, determining the worst case scenarios for the designed GNC in the presence of various uncertainties is critical. Further, given a multi-dimensional bounded uncertain parameter space, it is essential to demarcate the region of the uncertain parameter space where multiple performance requirements satisfy, or not, with a certain confidence level. Widely used Monte Carlo approaches are computationally restrictive to determine these precisely. Classical μ-analysis methods to modern integral quadratic constraints analysis cannot entirely address these analysis problems for a general class of complex systems. In contrast, optimisation based methods are adaptable to a broader class, and the computational complexity depends on underlying optimisation algorithm complexity, as expected.
Motivated by these, this thesis develops efficient algorithms for spacecraft control system’s Verification and Validation (V&V ), based on Bayesian approaches. The proposed methods use a Gaussian process (GP) based algorithm to 1) find the worst case given the uncertainty, 2) divide the uncertain parameter space into safe and unsafe regions, and 3) predict the system’s response given a set of inputs. The Bayesian approaches developed in this thesis are applied to validate guidance and control systems of Euclid and Proba-3 missions in industrial projects funded by Sener Engineeria and the European Space Agency (ESA).
In conclusion, this thesis proposes Bayesian approaches to validate complex control systems. Following GP regression and classification introduction, the Bayesian optimisation and sensitivity analysis are utilised to determine the worst case uncertain parameter combination. The estimation of safe and unsafe regions, possibly disjointed ones, over the uncertain parameter space using the GP classification idea is developed and applied to examples to demonstrate the efficacy of the proposed approach. The concept is developed further to determine the domain of attraction of the low-order dynamical systems, a classical, still active research problem. Last, a novel streaming spare Gaussian process algorithm is proposed to build a surrogate model that can replace the original time-consuming spacecraft model. | en_GB |
