Predicting the ion beam current instability of LINAC3

Abstract

The GTS-LHC ion source provides heavy ions to the Large Hadron Collider (LHC) ion injector chain situated at the European Organisation for Nuclear Research (CERN) to conduct numerous physics experiments. The execution of such experiments relies on the stable operation of the ion source, which depends on frequent changes of the source’s settings by a specialist. The objective of this research is to identify patterns in time series data from 2021 and forecast when the machine is going to fail, allowing the source specialist to take preventive action. In this study, a classical machine learning algorithm and five different neural network architectures were analysed and implemented to predict a beam decay. The results of the forecasting methods were compared to a baseline model to decide whether patterns of a beam decay exist. The implemented models were able to reach the performance of the baseline model but not surpass it. Given the current measurements, it is not possible to predict a beam decay in a short-term or long-term forecast. Additionally, two change point detection algorithms were provided to recognise abrupt changes in the status of the ion source on streaming data. After the implementation of a high voltage breakdown filter, it is possible to identify quickly and efficiently a beam decay in the present by reducing the number of false alarms.