Performance optimisation of accelerators using machine learning techniques


Reliable and realistic accelerator models are essential for the efficient control and performance optimisation of any accelerator. They need to be coupled with a robust beam control system to guarantee high beam quality and stability and reduce losses.

To meet these requirements, this project will investigate the use of Machine Learning (ML) techniques to build predictive models for a multivariate optimisation of an accelerator/beam line. The goal is to create algorithms that find and correct anomalies in different kinds of data, e.g. orbit data, power supplies, beam losses, etc. These ML tools will then provide excellent and efficient tools to tune beam optics, improve operation and provide the required beam parameters for specific experiments.

This first part of the project will combine input from the established codes MAD-X, G4beamline and CST studio. This work will familiarize the student with geometry and boundary condition definition, as well as the structure and limitations of all individual codes. The combined approach will allow performing high precision simulations taking into account e.g. fringe field effects and inhomogeneities due to geometrical factors. Models will then be established of the different CERN rings and enhanced by implementing genetic algorithms which the group is already applying for the optimization of synchrotrons (SOLEIL) and PERLE. These will then be utilised for machine optimization studies.

The project is offered within the Liverpool Centre for Doctoral Training in Big Data Science (LIV.DAT). You will benefit from a comprehensive postgraduate training in data science provided within the centre whilst at the same time also having the opportunity to follow the Cockcroft Institute lecture program on Accelerator Science. 

 To apply for this opportunity, please click here.  


Open to EU/UK applicants

Funding information

Funded studentship

The project is funded for full 4 years and builds on an existing collaboration with CERN. In Liverpool, you will be expected to do some undergraduate demonstrations as part of the studentship. The possibility to spend years 1 and 4 at UoL/CI and years 2+3 at CERN can be discussed as part of the project setup. Project offered, subject to funding.