Speaker
Description
Reliable hands-off injector operation calls for fast, sample-efficient tuning under drift and competing goals (e.g., capture efficiency, energy spread, transmitted charge). We present an autotuning framework for the ALS injector combining complementary online optimizers for robust performance under strict machine-protection/operability constraints. The controller alternates methods based on objective structure and information gain, fusing diagnostics across longitudinal and transverse systems. Building on prior Bayesian and multi-objective optimization, we add extensions for tracking a moving Pareto front during drifts, time-decayed learning for stability, and global-exploration bursts to escape trade-off plateaus. On the ALS linac, we target figures of merit tied to bottlenecks (e.g., controlling beam loading-driven energy spread challenging the booster acceptance) and enforce safety via bounded steps and surrogate constraints. Initial studies show shorter tuning time and improved repeatability vs. single-method baselines while preserving capture within the booster ring’s tight longitudinal window; we summarize architecture, decision logic, and portability.
Funding Agency
Funded by the US Department of Energy (BES Accelerator and Detector Research Program), and supported by the US Department of Energy, Director of the Office of Science under Contract DEAC02-05CH11231.
| In which format do you inted to submit your paper? | LaTeX |
|---|