Speaker
Description
Transverse beam profile monitoring in high-radiation areas is challenging due to camera degradation. A proposed solution employs a multimode fiber (MMF) to relay optical signals from the radiation zone to a shielded area, where a standard CMOS camera can operate safely. However, MMF transmission introduces significant distortions, producing complex speckle patterns at the output. This contribution presents a machine learning–based method to reconstruct the transverse beam distributions from these patterns. It will also discuss fibre choice and coupling, performance under different radiation levels, and the impact of fibre length.
The model was trained solely on synthetic data generated via Stochastic Gaussian Mixture simulations, in which the samples were displayed on a laser-illuminated digital micromirror device (DMD) and relayed through a 5-meter MMF. The same setup was used for testing, with beam images on a scintillating screen from CERN’s CLEAR facility replayed on the DMD. The
model achieved 1.49% mean relative RMSE across four key beam parameters—77% lower than the 6.59% baseline, demonstrating the potential for diagnostics in radiation-constrained environments.
Funding Agency
Work supported by the Science and Technology Facilities Council through the LIV.INNO Centre for Doctoral Training under Grant No. ST/W006766/1, and by CERN.
| I have read and accept the Privacy Policy Statement | Yes |
|---|---|
| Supervisor's name | Carsten P. Welsch |
| Supervisor's email | welsch@liverpool.ac.uk |
| Paper submission software | LATeX |