Reconstruction of missing data of the PandaX-IIIexperiment TPC using neural network

Abstract

The PandaX-III is a Neutrinoless Double-Beta Decay (NLDBD) experiment which uses a Time Projection Chamber (TPC) detector with a readout plane formed of Micromegas modules, which allows reconstruction of track topology for the background discrimination as well as reconstruction of the energy of the events. In NLDBD experiments, in order to achieve the highest sensitivity to the decay, it is necessary for the detector to have a high energy resolution, the background level should be low, and techniques for background discrimination must be applied as well. In reality, inhomogeneous signal gain at each module and the presence of missing channels lead to an incorrect energy reconstruction of the events. In this work, a method based on a Convolutional Neural Networks (CNN) aiming to reconstruct the kinematics of the event from imperfect data with missing channels is presented. Preliminary results of the reconstruction of the missing data using CNN are showing an increase in detection efficiency. The detection efficiency was evaluated on the simulated data with three channels randomly chosen per Micromegas module and artificially set as missing. Direct reconstruction of the energy gives the efficiency of 78%, while after applying CNN it increases to 86%, providing a promising application of this technique.