Simulation Analysis and Machine Learning Based Detection of Beam-Induced Heating in Particle Accelerator at CERN

A method for a first-order approximation estimation of the longitudinal impedance of a synchrotron component, starting from power loss measurements on the device, is proposed. This method also estimates the resonance frequency and the quality factor of the impedance after the execution of several machine runs, without disconnecting the device. After a detailed description of the method, its suitability is demonstrated through a practical case study using power loss measurements of the Large Hadron Collider (LHC) at the the European Organization for Nuclear Research (CERN). Then, electromagnetic simulations were used to benchmark recent theoretical models and assess their possibility to compute the two beam power loss. It is shown how beam-induced power loss can largely differ from the single beam case when two beams are present in the same component. Simulation studies are shown in the case of a resonant pillbox cavity. This benchmark also allowed simulating cases, for which the lumped impedance assumption of the available analytical formula may not be valid anymore. Finally, machine learning models were developed to detect heating from pressure measurements in synchrotron colliders. These results allow to analyse all the pressure measurements in the time available between two consecutive machine runs. Due to the prevalence of noise and the diversity of the behaviours, simple heuristic-based techniques do not achieve high performance. To overcome the limits of simple heuristic-based algorithms, several machine learning models have been trained, tested and compared with an heuristic-based approach which is used as base-line. In particular, it is shown for the case of the Large Hadron Collider (LHC) that machine learning models reached better performance both in precision and recall scores with respect to the baseline.