Speaker
Description
We present an EPICS-native Bayesian optimization toolkit for automated beam commissioning at CSNS. Built on the Pyapas framework, the system integrates seamlessly with existing controls via EPICS Channel Access, treating all hardware setpoints and diagnostics as PVs. It supports both single-objective BO and multi-objective BO using qEHVI to optimize beam transmission and loss simultaneously. A layered machine protection scheme—including hard parameter bounds, firmware interlocks, and automatic beam aborts—ensures safe operation. Experimental campaigns across the linac and RCS demonstrate significant improvements in transmission efficiency and beam loss reduction. The modular architecture enables rapid, non-disruptive deployment across accelerator subsystems.
