Speaker
Description
TRIUMF's Isotope Separator & ACcelerator (ISAC) facility provides beam to many pivotal nuclear astrophysics experiments. Among these is the Detector of Recoils And Gammas Of Nuclear reactions (DRAGON), which aims to explore the reaction rates of nuclear astrophysical processes by measuring resonances through radiative capture. For this, rare isotope beams delivered to DRAGON are manually tuned by operators in what is a highly variable and inefficient process. This leads to issues as there is high demand for beam time, limiting availability for DRAGON. In particular, the required beam steering cannot be directly modeled as potential sources like alignment errors or magnetic fringe fields are not accurately known. This is a black-box problem where the true functional value can only be accessed by evaluation. Our method uses machine learning to find the optimal steerer values for a given objective. This method has been primarily tested for the single-objective case where our objective is the beam transmission. The multi-objective case considers both the beam transmission and deviation from the beam axis, where initial testing has been done using simulations to prepare for online testing.
