LIV.INNO student wins best poster prize at Cockcroft Institute conference

Published on

A student presenting a poster.

Second year LIV.INNO student Qiyuan Xu, from the University of Liverpool’s QUASAR Group has been awarded the Best Poster Prize at the Cockcroft Institute Postgraduate Conference 2024. The event took place on Tuesday, 29th October 2024, at the Cockcroft Institute, Daresbury Laboratory.

In the opening session on Accelerator Applications several Liverpool students presented their individual research projects. They delivered excellent presentations and answered questions thoughtfully, receiving very positive feedback from the conference panel.

Reflecting on the conference Qiyuan said: “I believe it was a very enjoyable experience. Presenting my work and receiving feedback helped me identify areas to improve my presentation and poster design skills. It was also great to connect with students from other universities and learn about their research.”

Qiyuan’s award-winning poster and presentation, titled "Synthetic Data-Driven Reconstruction of Transverse Beam Distribution with Machine Learning," addressed a critical challenge at CERN. The radiation-hard tube cameras traditionally used for beam imaging in high-radiation areas are stopping production, creating a need for alternative solutions. Qiyuan’s research explores using a single large-core multimode optical fibre to transmit beam imaging signals from high-radiation areas to standard CMOS cameras located in safer zones.

The main challenge is reconstructing the original beam distribution from the distorted output caused by mode coupling and scattering within the fibre. To tackle this, Qiyuan developed a simulation algorithm that generates synthetic training data with high variance using 2D Gaussian fields and trained an Autoencoder model using this data. To test the model's performance, real beam images and simulated data from CERN's CLEAR facility were transmitted through the same multimode fibre system. These images were used to create the training and testing datasets at DITALab. The results showed high reconstruction accuracy and demonstrated the model's strong generalisation ability over unseen beam images.

The conference was chaired by Dr. Oznur Apsimon and Dr. Morgan Hibberd and featured a judging panel composed of academic staff and Cockcroft Institute partners from the University of Manchester, the University of Lancaster, and Dr. Narender Kumar from the University of Liverpool. The event concluded with a social gathering, marking the end of a productive day filled with knowledge exchange and networking opportunities.