Advancing LHC Superconducting Circuit Models through Closed-Loop Multi-Objective Optimization

The Large Hadron Collider (LHC) comprises eight individual main dipole circuits (RB), each containing 154 superconducting magnets. Simulations of the electrothermal behavior of these cir- cuits during transients are an essential tool to verify the circuit performance and for the analysis of unexpected events occurring in these circuits. For example they are used, to analyze Fast Power Abort (FPA) events, which are triggered in case a disturbance (such as a power-supply trip, equipment mal- functioning, or a magnet quench) in the main dipole circuit is detected. Since the physical mechanisms determining the behavior of superconducting magnet circuits span different scales, it is not practical to exclusively simulate them using conventional approaches like finite element methods, as the simulation would become computationally too expensive. However, these challenging simulations can be carried out with the Simulation of Transient Effects in Accelerator Magnets (STEAM) framework, developed at CERN. The accuracy of these simulations is highly influenced by the models of two different types of circuit subcomponents: the energy extraction resistors and the magnets' by-pass diodes. The elec- trical behavior of these two components is temperature-dependent which was not taken into account in their previous STEAM models. In some cases, this led to significant deviations between simulated and measured circuit signals. The aim of this thesis was therefore to model the electrothermal behavior of these components and to implement and use a new method to automatically fit the models' free parameters to measurement data. For this purpose an already existing software interface between STEAM and the parameter optimization toolkit Dakota (developed at Sandia Labs) was extended for conducting closed-loop multi-objective optimizations. In this approach, simulation results are iteratively compared with measurement data in a closed-loop fashion. Since several signals are optimized simultaneously, the method applied in this thesis is referred to as closed-loop multi-objective optimization. The application of this method made it possible to fit the free parameters of the component models, whose influence on simulation accuracy is complex and non-linear. By following this approach, an improvement in simulation accuracy of the energy extraction resis- tance of up to 14 % could be achieved (errors are normalized to the maximum absolute value of the measured quantity). Thus, the accuracy of the simulated circuit current, a crucial quantity of the su- perconducting magnet circuit, improved by up to 2 %. The newly implemented diode model allowed to qualitatively explain the occurrence of post-quench recooling-related peaks in the forward voltage of the diodes which was not possible before. These were observed to be accompanied by a sharp drop in the diodes' temperature and a highly increased resistance of the diodes. The models are based on a more complete set of physical effects, and are thus expected to be more robust for simulating unexpected events for which experimental data are currently not available. The good transferability of the applied method to other models ensures that closed-loop optimization will be a promising option for further improving circuit simulation accuracy in future works. In the future, the STEAM framework will thus be more accurate and reliable for research and development related tasks like hardware commissioning, the analysis of FPA events and the further development of the LHC superconducting magnet circuits.