Speaker
Description
Photocathodes are critical components in advanced electron sources, and accurate characterization of their performance is essential. Direct measurements of mean transverse energy (MTE) using the Transverse Energy Spread Spectrometer (TESS) at Daresbury Laboratory often yield modest datasets. Physics-informed machine learning, particularly generative models like GANs and diffusion models, offers a way to expand these into larger sets of physically consistent synthetic images. Preliminary results show structural similarity indices (SSIM) $>0.95$ and predictive accuracies with $R^2 \approx 0.98$. A predictive model for estimating MTE at a given wavelength is also introduced. This enables rapid optimization of materials and parameters, cutting experimental overhead. These data-driven methods improve electron beam diagnostics, aid photocathode development, and may enhance accelerator performance. The approach also shows promise for optimizing free-electron laser (FEL) systems, advancing ML in accelerator science.
Funding Agency
This work is supported by STFC through the LIV.INNO Center for Doctoral Training under grant agreement ST/W006766/1.
| I have read and accept the Conference Policies | Yes |
|---|
