Speaker
Description
Accurate acquisition of the longitudinal phase space distribution is crucial for synchrotron optimization, but traditional tomography requires minutes per reconstruction, preventing real-time diagnostics. To resolve this, we propose a novel spatio-temporal neural network integrating 1D Convolutional Neural Networks (CNN) and Transformers. This hybrid model achieves end-to-end continuous reconstruction from 1D beam projections to 2D phase space dynamic evolution.
The network is trained on a high-fidelity dataset generated via the BLonD code. It incorporates nonlinear space charge effects based on the machine parameters of the Xi'an 200MeV Proton Application Facility (XiPAF). Results demonstrate the model accurately restores complex phase space topological structures. It effectively captures both high-density cores and low-density edge halos. The model achieves a longitudinal line density projection error under 1% in simulations and under 2% using real Fast Current Transformer (FCT) measurements from the XiPAF facility.
Furthermore, the framework delivers single-frame inference times of 0.109 ms on a GPU and 4.557 ms on a standard CPU. This sub-millisecond processing speed successfully crosses the engineering threshold for online real-time diagnostics. Ultimately, it establishes a reliable new continuous imaging paradigm for automated beam real-time feedback control in high-intensity accelerators.
| In which format do you inted to submit your paper? | LaTeX |
|---|---|
| Preprint marking on your proceeding paper | I wish my paper to be marked as preprint. |
| I no longer wish to present this contribution, please withdraw it. | Keep my contribution |