Commissioning of the New CERN PS Booster Charge Exchange Injection System: Optimising and Automating Transverse Phase Space Painting

The CERN injector complex was upgraded within the framework of the LHC Injectors Upgrade project to produce high-brightness beams for the High Luminosity era of the LHC (HL-LHC). A key aspect of this upgrade is the connection of the newly built Linac4 to the Proton Synchrotron Booster (PSB) and the related installation of a new 160 MeV H$^-$ charge exchange injection system. Increasing the PSB injection energy from 50 to 160 MeV increases the relativistic factor $\beta_\text{r} \gamma_\text{r}^2$ by a factor of two, which allows doubling the beam brightness while keeping space charge forces as before the upgrade. The new PSB charge exchange injection system was commissioned in winter 2020/2021. It comprises a graphite foil for electron stripping and a horizontal injection bump to move the circulating beam to the stripping foil. Customising the field decay of this injection bump during beam accumulation, also called phase space painting, enables tailoring the transverse beam characteristics for the different experiments at CERN. This thesis investigates solutions for commissioning the new system and pushing its performance by efficiently relating the requested transverse beam distributions to optimised operational paintings and required hardware parameterisations. In the first part, painting schemes for the different users are developed using self-consistent multi-particle simulations and benchmarked experimentally during beam commissioning. These studies mainly focus on optimising phase space painting to reduce space charge effects and minimise losses during the production of high-intensity fixed target beams. The second part of the thesis explores novel applications of machine learning and numerical optimisation techniques to enhance the efficiency and performance of the new system. The feasibility of applying numerical optimisation algorithms for configuring the painting functions is demonstrated on the machine. The performance of different optimisation algorithms is assessed and compared offline on a data-driven surrogate model. The beam dynamics studies are complemented by first investigations towards using deep learning concepts to define the electronic circuit parameterisation of the phase space kicker magnet pulse generators.