원문정보
초록
영어
Synchrotron radiation (SR) sources provide very high photon flux light in a very narrow opening angle with wavelength ranging from visible to hard X-rays for use in experiments related to material science, physics, chemistry and biology. In the beam lines (BL) the SR position is highly dependent on the electron beam position and angle at the source point. The tuning of accelerator for getting the desired electron beam position and angle at the source point is a time consuming and regular job done during commissioning of new BL or when accelerator is operated at new operating point. This paper presents a novel intelligent agent based operator support and beam orbit control scheme for accelerator control. The proposed multi-agent based scheme is well suited for the multilayer control system architectures of synchrotron radiation sources. The scheme successfully distributes the orbit control job to multiple low complexity reactive agents that work simultaneously and control the local orbit for individual BL and insertion devices (ID) in an optimized manner. The proposed scheme of beam orbit control in particular is very useful for machines like INDUS-2, where new BL are in the process of commissioning as this scheme reduces the operator efforts and accelerator tuning time for providing beam to new BL. It also extends the beam availability to other BL (already installed and in use) as the agent tunes the accelerator in systematic way and under constraints on local orbit bump leakage thereby enabling the use of other BL for routine experiments which otherwise was not possible. The effectiveness of the scheme is shown through simulation results obtained by applying the stated scheme on INDUS-2 storage ring model.
목차
1. Introduction
2. Subsumption Agent Architecture
3. Closed Orbit Bump and Accelerator Environment
3.1. Three Corrector Bump
3.2. Four Corrector Bump
4. Organization of Agents in Multi-agent Beam Orbit Control Environment
5. Agent Design and Implementation
6. Constraint Gradient based Reinforcement Learning
7. Early Implementation Simulation Results on INDUS-2 Model
8. Conclusion
References