Combination of Measurement Data and Domain Knowledge for Simulation of Halbach Arrays with Bayesian Inference

TL;DR

This paper presents a Bayesian inference method that combines measurement data and domain knowledge to improve the simulation accuracy of Halbach array magnets, validated on both simulated and real measurements.

Contribution

It introduces a novel Bayesian approach to fuse measurement data and domain knowledge for more accurate magnet modeling of Halbach arrays.

Findings

Posterior magnet model describes magnetic flux density ten times better than prior.

Method validated on simulated data and applied successfully to FASER detector measurements.

Significant improvement in magnet model accuracy achieved.

Abstract

Accelerator magnets made from blocks of permanent magnets in a zero-clearance configuration are known as Halbach arrays. The objective of this work is the fusion of knowledge from different measurement sources (material and field) and domain knowledge (magnetostatics) to obtain an updated magnet model of a Halbach array. From Helmholtz-coil measurements of the magnetized blocks, a prior distribution of the magnetization is estimated. Measurements of the magnetic flux density are used to derive, by means of Bayesian inference, a posterior distribution. The method is validated on simulated data and applied to measurements of a dipole of the FASER detector. The updated magnet model of the FASER dipole describes the magnetic flux density one order of magnitude better than the prior magnet model.