Erlangen 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 7: Data, AI, Computing, Electronics I

T 7.3: Talk

Monday, March 16, 2026, 16:45–17:00, KH 00.024

Particle Identification in OSIRIS using Deep Learning — •Martin Kandlen, Elisabeth Neuerburg, Achim Stahl und Christopher Wiebusch — RWTH Aachen

In August 2025, the filling of the 20-kton liquid scintillator based Jiangmen Underground Neutrino Observatory (JUNO) was finished. The radiopurity of the liquid scintillator (LS) was monitored by the Online Scintillator Internal Radioactivity Investigation System (OSIRIS). OSIRIS has a cylindrical acrylic vessel for 20-ton batches of LS observed by 64 20'' PMTs. The Uranium and Thorium impurities are measured via Bismuth-Polonium coincidence decay signals. I use attention-based deep learning methods for particle identification, where the input consists of the scintillation time profile. Simulation-data-mismatches are overcome by domain adaptation; the model is trained using both simulated and real detector data. In this talk, I present the methodology and performance of the classification.

Keywords: Deep Learning; Particle Identification; Transformer; Simulation-Data-Missmatch; Domain Adaptation