COMET Phase-I Track Reconstruction using Machine Learning and Computer Vision

Abstract

The COMET experiment will measure Charged Lepton Flavour Violation by searching for the neutrinoless decay of a muon into an electron while the muon is electromagnetically bound to an atomic nucleus. This μ-e conversion process is not allowed in the Standard Model of particle physics which makes it an excellent probe for Beyond the Standard Model physics. The first phase of the experiment will improve the current sensitivity limit on μ-e conversion from the 7.0 × 10−13 at 90% C.L. to 3.0 × 10−15 at 90% C.L. To achieve this, COMET will utilize several novel design elements to produce the world’s high-intensity muon beam to maximise the number of observed muonic atoms. The high-intensity design of this experiment poses significant challenges to both the tracking and triggering systems. A novel algorithm called the CDC Hit Filter is designed to alleviate these challenges. The algorithm utilizes machine learning classification and a circular Hough Transform to identify and remove background hits. When applied offline, it can remove 98% of background hits while retaining 98% of signal hits. When adapted to the online environment, it can remove 89% of the background hits considered by the trigger while retaining 89% of signal hits. Both of these mark a significant improvement over the more traditional cut-based approach, which can remove 75% of background hits while retaining 75% of signal hits.