Speaker
Description
The new accelerator complex, FAIR (the International Facility for Antiproton and Ion Research), will soon be commissioned to deliver ion beams using its injector and the GSI accelerator complex. In order to extend the operation of the GSI to include the FAIR and due to some ageing components, a new control system has been implemented.
The time- and resource-efficient setup of complex ion beams, a long-standing challenge, has been addressed with a Java-based application called DeviceAutomator. This application handles different optimization routines based on modern machine learning technology. Apart from selecting the most suitable algorithm, challenges arise from data quality and the number of independent parameters. However, since the application is fully integrated into the FAIR control system, all operators in the control room can access and freely configure it without the need for coding.
This contribution will describe the implementation and real-life testing of different algorithms: Bayesian, Genetic, and Random Walk. Using a long section of a low-energy ion transport beam line, it has been demonstrated that larger parameter spaces extending well beyond ten parameters present significant challenges for the Bayesian algorithm. However, the genetic optimization routine remains capable of identifying optimal values. It was also evident that a well-chosen optimization routine has the potential to make ion beam setup faster and less labor-intensive.
| Paper status | Proceeding files received |
|---|