Speaker
Description
Beam imaging cameras can be rapidly damaged in radiation environments such as CERN accelerator beamlines. To mitigate this, a remote beam imaging system is being developed, transporting scintillation light from a screen through a 15 m large core multimode fiber (MMF) to a shielded camera. A key challenge is reconstructing the transverse beam distribution from the MMF output when limited real beam data are available for training.
This contribution investigates a convolutional autoencoder (CAE) for reconstructing transverse beam distributions from MMF transmitted scintillation light under limited training data and for estimating the real beam dataset size needed. Using experimentally acquired data from the CLEAR facility, three compact training set strategies are compared: random sampling in image space, latent space density guided selection, and augmentation using an approximately orthogonal response basis derived from line scans with the beam size minimised on the screen. The study evaluates how these methods affect reconstruction accuracy, generalization, and the practical minimum dataset size required for reliable MMF based transverse beam imaging.
Funding Agency
This work was supported by the Science and Technology Facilities Council (STFC) through the LIV.INNO Centre for Doctoral Training under grant agreement ST/W006766/1 and CERN.
| In which format do you inted to submit your paper? | LaTeX |
|---|