Design and performance of a large-area scintillator-based chamber for the MID subsystem of ALICE 3

TL;DR

This paper details the design, construction, and testing of a scintillator-based chamber for muon detection in the ALICE 3 upgrade, achieving high muon efficiency through machine learning.

Contribution

It introduces a novel chamber design with orthogonal scintillator layers and demonstrates effective muon identification using machine learning at CERN.

Findings

Muon efficiency above 99% with the ML algorithm.

Fake-muon efficiency decreases exponentially with absorber length.

Successful beamline testing of the chamber prototype.

Abstract

This paper reports on the design and construction of a chamber for the muon identifier detector (MID) of the ALICE 3 upgrade project. The chamber consists of two sensitive layers separated by a 1 cm air gap. Each layer holds 24 scintillator bars ($1\times4\times100$ cm$^3$) manufactured by FNAL-NICADD. The bars are equipped with Kuraray wavelength shifting fibers and the readout is provided by a silicon photomultiplier from Hamamatsu. The bars in the second layer are orthogonal to the bars in the first layer, thus providing an overlapping cell size of 4$\times$4 cm$^{2}$. The bar assembly as well as the design of the mechanical structure is described. The design of the chamber is close to that considered in the ALICE 3 letter of intent. The chamber was tested at the CERN T10 beamline using 3 GeV/$c$ pion-enriched and muon beams. The chamber was placed behind an iron absorber, with different absorber lengths considered in the test. The muon identification is performed using a Machine Learning algorithm, which was trained and tested using muon (signal) and pion (background) data (50% of the available statistics). The trained ML algorithm was applied to muon data, yielding a muon efficiency above 99% for the OR condition (detection in either layer 1 or 2). The implementation in the pion-beam data gives the fake-muon efficiency as a function of the absorber length that is well described by an exponential function with a slope parameter of 18.79 cm. The next steps towards finalizing the optimization are outlined.