Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Implementation and Deployment of an Injection Tuning Tool Using Bayesian Optimization at the SuperKEKB Accelerator

Shinnosuke Kato, Gaku Mitsuka

TL;DR

The paper tackles the challenge of sustaining high luminosity at SuperKEKB by automating the injection-tuning process, which was previously manual and time-consuming. It applies Bayesian optimization enhanced with segmented region optimization, a step-by-step update schedule, and proximal biasing to safely maximize injection efficiency in real time. The approach yielded up to 32% improvement in injection efficiency during Nov–Dec 2024 tests, with no beam aborts, and SHAP analyses showing day-to-day variability in which tuning parameters mattered most. This work demonstrates a practical, safe framework for automated injector tuning that can be adopted by collider facilities to improve operational efficiency and stability.

Abstract

As of July 2025, the SuperKEKB accelerator, which collides 7 GeV electrons with 4 GeV positrons to abundantly produce particles such as B mesons and tau leptons, holds the world record for the highest instantaneous luminosity. Continuous operation and upgrades are underway to achieve even higher luminosities. Maintaining a high instantaneous luminosity requires sustaining high beam currents in the storage rings, which in turn demands efficient beam injection from the injector. In particular, a high injection efficiency, defined as the ratio of the beam current successfully accumulated in the ring to the current delivered from the beam transport line, must be ensured. In the present study, we developed a tool to automate the injection tuning process using Bayesian optimization, a machine-learning-based technique, in order to improve the injection efficiency. During test operations conducted in November-December 2024, this tool successfully enhanced the injection efficiency by up to 32%.

Implementation and Deployment of an Injection Tuning Tool Using Bayesian Optimization at the SuperKEKB Accelerator

TL;DR

The paper tackles the challenge of sustaining high luminosity at SuperKEKB by automating the injection-tuning process, which was previously manual and time-consuming. It applies Bayesian optimization enhanced with segmented region optimization, a step-by-step update schedule, and proximal biasing to safely maximize injection efficiency in real time. The approach yielded up to 32% improvement in injection efficiency during Nov–Dec 2024 tests, with no beam aborts, and SHAP analyses showing day-to-day variability in which tuning parameters mattered most. This work demonstrates a practical, safe framework for automated injector tuning that can be adopted by collider facilities to improve operational efficiency and stability.

Abstract

As of July 2025, the SuperKEKB accelerator, which collides 7 GeV electrons with 4 GeV positrons to abundantly produce particles such as B mesons and tau leptons, holds the world record for the highest instantaneous luminosity. Continuous operation and upgrades are underway to achieve even higher luminosities. Maintaining a high instantaneous luminosity requires sustaining high beam currents in the storage rings, which in turn demands efficient beam injection from the injector. In particular, a high injection efficiency, defined as the ratio of the beam current successfully accumulated in the ring to the current delivered from the beam transport line, must be ensured. In the present study, we developed a tool to automate the injection tuning process using Bayesian optimization, a machine-learning-based technique, in order to improve the injection efficiency. During test operations conducted in November-December 2024, this tool successfully enhanced the injection efficiency by up to 32%.

Paper Structure

This paper contains 12 sections, 2 equations, 15 figures.

Table of Contents

  1. Introduction
  2. Injection Scheme into the Main Ring
  3. Bayesian Optimization
  4. Segmented Region Optimization
  5. Step-by-Step Function
  6. Proximal Biasing Method
  7. Optimization Results
  8. Uncertainty in Injection Efficiency
  9. Verification of Reproducibility
  10. Effect of Proximal Biasing Level
  11. Test Deployment of the Injection-Tuning Tool
  12. Conclusion

Figures (15)

  • Figure 1: Overview of the SuperKEKB accelerator SuperKEKB:TDR_overview
  • Figure 2: Schematic of the injection system in the vertical plane.
  • Figure 3: Schematic of the injection system in the horizontal plane SuperKEKB:TDR_BT.
  • Figure 4: Photograph of the magnets used for injection tuning in this study.
  • Figure 5: Conceptual illustration of the segmented region optimization. The blue regions represent the parameter domains for each run, which are redefined around the parameter set that achieved the highest injection efficiency in the previous run.
  • ...and 10 more figures