A Full-Plant, Physics-Aware Approach to the Planning and Control of Multi-Robot Nuclear Fusion Maintenance Systems

Nuclear fusion holds great potential as a clean, high-tech energy source. However, fusion is impeded by myriad challenges across multiple domains: plasma physics, plant engineering, commercial viability, et cetera. Of these, the maintenance of fusion power plants presents a complex challenge which is deeply coupled to both the physics and the economy of the plant: how do we safely maintain such a large and complex nuclear environment in a manner that allows fusion to move beyond a scientific experiment to become a economically-viable source of power? In this thesis, we explore the feasibility of applying autonomous multi-robot systems to the fusion maintenance problem. In so doing we conduct extensive literature reviews of the fusion state-of-the-art, as well as that of fusion maintenance specifically, and discover significant historical trends and underlooked challenges which will define the priorities of the upcoming 'demonstrator generation' of fusion power plants (i.e. those which aim to demonstrate commercial viability). We then propose and implement a method of scaling the task scheduling of safety-verified, multi-robot systems to the sizes required by a future tokamak such as EU-DEMO. Finally, we introduce two complementary methods to analyse plasma disruptions under the presence of magnetic coil failures using the JOREK-STARWALL plasma simulation code, and subsequently synthesise a control policy for a 'plasma-gnostic' fusion maintenance system to accelerate maintenance in critical scenarios.