Join us using Zoom on January 28th at 10 AM (CET).

Abstract

Linear accelerator 3 (Linac3) is the starting point for the ions used in experiments at CERN. It provides lead ions for the Large Hadron Collider (LHC) and for fixed-target experiments. It is a vital part of the whole accelerator system here at CERN.

The goal of the Linac3 is to produce a stable and sufficiently powerful beam of ions. In order to achieve that with the accelerator, the operators can in real-time adjust 14 different settings in order to stabilize the Linac3 output beam. In addition, it also has 18 sensors that are read-only. 

The ion beam current is measured at 4 different places along with the accelerator. The most critical part that affects the quality of the current are the first two parts and therefore it is sufficient to forecast the beam current measured at MTR05 (see Figure 1).

Linac3 is a very complex machine and there is no analytical solution on how to adjust the settings to produce an ion beam with optimal properties. The beam often starts to decay or shows other problems after which an action has to be taken by human operators. It is also often unclear how to adjust the settings so that the beam becomes stable again. The aim is to make a predictive model that can forecast such decays sufficiently in advance and thus helps to reduce maintenance costs and increase uptime of the accelerator. The dataset we provide consists of 470k data points with 35 columns that describe the behavior of the Linac3 in the year 2018 with a sampling rate of 1 minute.

There are two major milestones that we propose: 

1. Design a model with much better forecasting performance of the beam current than a baseline persistence model (see Figure 2). 

2. Conduct a global sensitivity analysis: analyze which features are important at different timescales predictions

For more details, please contact Filip Široký (filip.siroky@cern.ch) or Marc Bengulescu (marc.bengulescu@cern.ch).