The ICARUS T600 detector is the far detector for the Short-Baseline Neutrino (SBN) program at Fermilab, aiming to identify potential neutrino oscillations using the O(1 GeV) neutrino energy Booster Neutrino Beam (BNB). ICARUS is the largest Liquid Argon Time Projection Chamber (LAr-TPC) currently used in neutrino physics, containing 500 tons of Argon in its active volume. LAr-TPC technology offers precise spatial and energy measurements based on the electron drift signal from Argon ionisation, in addition to accurate timing measurements from a prompt scintillation signal. These prompt scintillation signals are detected by 360 Photomultiplier Tubes (PMTs) and are used for triggering purposes as well as in the determination of the interaction time, which is needed in conjunction with the electron drift signal for particle tracking.
As ICARUS is a large detector operating at shallow depths, it faces the considerable challenge of identifying genuine neutrino interactions on top of a pervading background from cosmic rays. Even within the short BNB neutrino production window, cosmic rays are expected to outnumber neutrino interactions by more than three to one. In this work we investigate the possibility of using a machine learning based approach to separate neutrino interactions from such cosmic backgrounds. We train a 3D convolutional neural network using low level timing and charge readout information from the PMTs. Preliminary simulated results suggest we are able to reduce cosmic background from 77% to 34% whilst maintaining a neutrino interaction selection efficiency of 91%.
machine learning; CNN; photodetectors; background rejection; LArTPC; event filtering
|Subcategory||Experimental Methods & Instrumentation|
|Collaboration||other (fill field below)|