Transverse phase space tomography in the CLARA accelerator test facility using image compression and machine learning

TL;DR

This paper introduces a machine learning-based method for rapid 4D transverse phase space tomography in accelerators, outperforming traditional algorithms in speed and data efficiency while maintaining comparable accuracy.

Contribution

It presents a novel approach combining image compression and machine learning for faster, data-efficient 4D phase space reconstruction in accelerator physics.

Findings

Machine learning method matches algebraic reconstruction accuracy.

Significantly reduces data set size through image compression.

Enables rapid beam behavior characterization in accelerators.

Abstract

We describe a novel technique, based on image compression and machine learning, for transverse phase space tomography in two degrees of freedom in an accelerator beamline. The technique has been used in the CLARA accelerator test facility at Daresbury Laboratory: results from the machine learning method are compared with those from a conventional tomography algorithm (algebraic reconstruction), applied to the same data. The use of machine learning allows reconstruction of the 4D phase space distribution of the beam to be carried out much more rapidly than using conventional tomography algorithms, and also enables the use of image compression to reduce significantly the size of the data sets involved in the analysis. Results from the machine learning technique are at least as good as those from the algebraic reconstruction tomography in characterising the beam behaviour, in terms of the variation of the beam size in response to variation of the quadrupole strengths.