Online optimization of beam lifetime by Bayesian Optimization in TPS

The beam lifetime in synchrotron light sources is critically influenced by nonlinear beam dynamics, particularly in low-emittance storage rings where the Touschek effect dominates. At the Taiwan Photon Source (TPS), a third-generation 3 GeV storage ring, optimization of harmonic sextupole magnet strengths has been conducted using Bayesian Optimization (BO) techniques to minimize beam loss monitor (BLM) readings and thereby enhance beam lifetime. This study demonstrates an integrated approach to nonlinear optics tuning using machine learning, with constraints on chromaticity correction. The experimental implementation involves tuning harmonic sextupole families. Results indicate significant improvements in lifetime while maintaining machine stability, demonstrating the potential of data-driven control in modern accelerator operations, establishing this technique as a powerful tool for synchrotron light source operation. This paper details the implementation, experimental validation, and performance analysis of BO in optimizing beam lifetime at TPS.