Speaker
Description
High-quality beam diagnostics and control are essential for achieving stable accelerator operation and high performance. In future electron–positron colliders, precise monitoring and optimization of beam parameters will be particularly important for achieving high luminosity. Machine learning offers a promising approach to fast beam diagnostics by extracting beam parameters directly from measured beam profiles.
At the Japan Proton Accelerator Research Complex (J-PARC), negative hydrogen ions (H⁻) are accelerated in the LINAC and subsequently stripped of their electrons to produce a high-intensity proton beam. At the exit of the RFQ, the first accelerating section of the J-PARC LINAC, and in the following Medium-Energy beam transport1 (MEBT1), the beam energy is as low as 3 MeV, and strong space-charge effects significantly affect the beam distribution. Therefore, accurate estimation of the Twiss parameters and emittances, which characterize the beam phase-space distribution, is important for beam tuning.
Conventionally, these beam parameters at the RFQ exit have been estimated by iteratively comparing beam profiles measured by beam monitors in MEBT1 with Particle-in-Cell (PIC) simulations that account for space-charge effects. However, this approach requires considerable computation time, making it difficult to apply to real-time beam tuning during accelerator operation.
In this study, we develop a machine-learning model for real-time beam diagnostics at the J-PARC LINAC. The model takes beam profiles generated by PIC simulations as input and simultaneously estimates nine beam parameters: the Twiss parameters and emittance in each of the X, Y, and Z directions. Once trained, the model can estimate these parameters without iterative PIC simulations, enabling rapid beam diagnostics. In this presentation, we report the prediction performance of the developed model and evaluate its estimation errors. Although the beam conditions at J-PARC differ from those of electron–positron colliders, this study demonstrates the potential of machine-learning-based beam diagnostics for fast beam parameter estimation and optimization in future accelerators.