Speaker
Description
We propose a machine-learning method to reconstruct longitudinal phase-space parameters of an injected beam from its transient longitudinal motion. An ELEGANT model of HEPS storage-ring injection simulated 500-turn evolution of the longitudinal distribution while scanning arrival-time offset, relative momentum offset, bunch length, and energy spread. A dataset of 72,000 simulated motion images trained a multi-output ResNet18 model, mapping each 224 x 224 image to the four injection parameters. For 1,000 simulated test samples, R-squared values were above 0.98, with mean absolute errors of 2.31 ps, 0.051 percent, 0.253 mm, and 0.026 per mille. The method was tested using streak-camera measurements of injected-beam motion at the HEPS visible-light beamline under eight RF-frequency settings. After matching time scale, turn number, intensity, and image size, the predicted energy offset showed a reasonable correlation with the expected RF-frequency dependence, and the predicted bunch length and energy spread were broadly consistent with streak-camera estimates. These preliminary results suggest that transient longitudinal motion images can support rapid diagnostics of injected beams.
| I have read and accept the Privacy Policy Statement | Yes |
|---|