Preliminary results of crystal channelling optimisation in the LHC using reinforcement learning

The Large Hadron Collider (LHC) can operate with high intensity proton and heavy ion beams, both of which require a collimation system to ensure an efficient operation and to protect against damage to sensitive equipment along the ring. The crystal collimation scheme using bent silicon crystals as primary collimators was therefore introduced to improve the collimation efficiency for heavy ion-beams. The first operational deployment of crystal-assisted collimation was achieved in the 2023 Pb run. This demonstrated the required performance gain to safely handle high intensity ion beams, but undesired crystal rotation led to the loss of optimal performance during physics fills. The cause of this is thought to be mechanical deformation of the goniometer due to heating related to beam impedance effects. Hence, a conventional numerical optimiser was deployed to monitor and compensate for crystal angular errors based on a set of beam-loss monitors. The problem at hand, allows for the use of machine learning techniques to ensure continuous optimal channelling, minimising convergence time and eventually the optimization of crystals in multiple planes in parallel.