LINAC2026 - 33rd Linear Accelerator Conference

Machine learning-based calibration and surrogate modeling of the photoinjector at PAL-XFEL

Not scheduled

2h

Daejeon Convention Center

Poster Presentation MC1.A02 Electron and ion sources, guns, photo injectors, charge breeders Poster Session

Speaker

Won Jang (Pohang University of Science and Technology)

Description

To achieve a high-brightness X-ray free-electron laser (FEL), it is essential to perform precise tuning of electron beam parameters at the photoinjector section, where the intrinsic emittance is determined. At PAL-XFEL, injector tuning is typically performed via manual, parameter-scan-based emittance optimization, which can be time-consuming. In addition, simulation-based beam matching often shows discrepancies with measured data, making beam prediction more challenging. To address this issue, we have developed a machine learning-based surrogate model to calibrate the simulation model and represent injector beam dynamics. The model is initially trained using particle tracking simulation data that include realistic UV laser distributions. This initial training is then refined through transfer learning with measured data. The resulting surrogate model enables fast prediction of beam parameters and can be used to guide beam tuning. This approach is expected to reduce tuning time and support more efficient accelerator optimization.