Anomaly Resilience, Detection, and Reaction in Astrodynamics Problems

Abstract

This manuscript is divided into three parts, each associated with three of the main projects I undertook throughout my doctoral journey. In particular, this dissertation focuses on the topics of anomaly resilience, detection, and reaction in astrodynamics problems, which are crucial aspects to consider for ensuring mission success, especially when dealing with the harsh space environment. In this manuscript, resilience, detection and reaction are investigated in two selected domains: trajectory design in cislunar regime and satellite constellations, both susceptible to the possibility of unexpected events. Therefore, effective identification capabilities and swift reaction to unforeseen phenomena become vital to support mission integrity. Resilience to off-nominal behaviors is investigated for the trajectory design problem in the first part of this manuscript. In particular, we explore the convergence and dynamical structure of the trajectory design space associated with lunar landing and ascent abort scenarios for a crewed module departing from and returning to the Deep Space Gateway. Numerical methods for the identification of abort trajectory solutions are employed within a two-step optimization pipeline, through which we discover the existence of regions of stiff convergence where traditional pipelines may fail. Hence, we present an extensive analysis of problem parameters, demonstrating how the presence of such regions can be traced to the formulation of employed correction algorithms, problem dynamics sensitivity, and transfers geometry. To reduce the computational cost, we introduce a three-step optimization pipeline relying on surrogate models trained via adaptive sampling for the fast generation of initial guesses, which are then corrected for the recovery of abort trajectories within the defined scenario. Results obtained from the application of the optimization pipeline on the two scenarios underscore the complexity of the solution space, while providing useful information to inform the trajectory design process. Anomaly detection and reaction in the context of Proliferated Low Earth Orbit (P-LEO) satellite constellations are then explored in the remaining two parts of this manuscript, with a focus on anomalous behaviors originating from adversarial actions against a constellation. For the detection problem, we present a transformer neural network- based pipeline for the identification of anomalous connections between satellites, modeling a P-LEO as a dynamic graph, which is capable of capturing spatial-temporal correlations characterizing a temporal network. Our analyses demonstrate how temporal and spatial signals alone are insufficient for effectively discriminating anomalous connections, requiring additional features to enrich the information extracted from the network dynamics. Notably, we discover how the introduction of edge-frequency information positively impacts our algorithm performance, reaching up to 95% in AUC score, here used as quality metric. Additionally, extensive analyses on variations of problem parameters demonstrate the robustness of the method over a wide range of scenarios, and highlight the existence of interesting couplings between satellite dynamics, spatial ground node distribution, and algorithm performance. For the reaction component, a combination of competitive coevolutionary algorithms and genetic programming is employed to evolve reactive strategies to respond to adversarial actions against a constellation system. In particular, genetic programming trees are employed for the representation of reactive policies to respond to presented and different, unseen scenarios. The analysis demonstrates how the utilized approach provides effective solutions, beating both minimally complex strategies and human-developed ones, and showing adaptability to the introduction of multiple constraints. In both the detection and reaction problems, the proposed methodologies display the potential to reduce the cognitive load on operators of large constellation systems, and to possibly enable resolution of critical situations in a timely manner.