Optimization of singly-charged particles identification with the AMS02 RICH detector by a machine learning method

AMS-02 is a detector currently in operation onboard the International Space Station (ISS). One of the main scientific goals of the spectrometer is the measurement of charged particle fluxes. The detector design makes possible the identification of particles and antiparticles by precise measurement of particle momentum in the AMS-02 Silicon Tracker, and velocity in the Cherenkov (RICH) detector. The RICH is able to measure the isotopic composition of the light elements (up to charge Z = 5) in the kinetic energy range from a few GeV/n to about 10 GeV/n. However, the velocity reconstruction for charge 1 particles is particularly challenging due to the low number of photons they produce in the RICH detector which can lead to wrong event reconstruction. In this paper, we show the high potential of the Multilayer Perceptron deep learning model (MLP-BFGS) for identification of signal and the background due to interactions inside the AMS-02 detector, and to significantly improve particle identification by its mass.