Speaker
Description
This contribution presents the results of an experimental test for the automatic tuning of a FEBIAD ion source and electrostatic beamline elements, performed at the OFFLINE 2 facility at ISOLDE. The algorithms were developed for the automatic tuning of the ISOL@MYRRHA facility at the Belgian Nuclear Research Centre and were adapted for testing at the OFFLINE 2 facility. The integration of automatic algorithms to assist in the parameter tuning process of Isotope Separator Online (ISOL) systems is essential due to the time-consuming nature of manual beam tuning and ion source optimization. Optimizing these systems requires adjusting a large number of parameters, while meeting specific performance requirements. As a result, the quality (purity) and quantity (intensity) of the Radioactive Ion Beam (RIB) delivered to the experimental end station are highly dependent on the precise tuning of these parameters and their interactions. In this work, we propose optimization techniques that simultaneously tune the beam optics and ion source parameters, with the goal of extracting a beam with the highest current and optimized beam shape while ensuring efficient transmission through the dipole magnet. Optimizing an ISOL system poses a constrained, multidimensional optimization problem, where evaluating the objective function is time-intensive due to the data acquisition duration of the diagnostic devices.
To address this challenge, we propose using Bayesian optimization (BO), known for its ability to optimize expensive objective functions in relatively few iterations. Furthermore, we employ a Gaussian Process (GP) as a surrogate model to capture parameters effect on the objective function. While the combination of Gaussian Process and Bayesian optimization has been successfully applied in accelerator tuning, in this work we extend its functionality by simultaneously tuning ion source parameters along with ion beam optics at the ISOLDE OFFLINE 2 facility (YOL2). The beam optics parameters optimized include the voltages of three electrostatic quadrupoles and four voltages from two horizontal and vertical steerers located before the mass separator dipole magnet. For the FEBIAD ion source, the anode voltage and ion source magnet coil current were tuned. The optimization objective was to maximize the beam current, align the beam, and minimize beam size at the entrance focal point of the dipole magnet. Instead of using a Faraday cup to measure the beam current, as is common practice, we computed the objective function using only the wire scanner located before the separator magnet.
Lastly, to accelerate the algorithm's convergence, we utilized archival data by formulating a data-informed Gaussian Process. This approach learns correlations between design parameters, enabling faster convergence toward optimal parameter combinations. This formulation was also tested on the beam optics of YOL2.
