Discriminação de sinais de detectores de partículas utilizando técnicas de aprendizado de máquina

Abstract

One of the fundamental problems in particle detector instrumentation is optimizing the signal-to-noise ratio in the early stage of electrical pulse collection. Besides adopting filters and formatting, signal discrimination is usually achieved in electronic signal proces-sing by imposing an amplitude threshold value. This solution is generally enough since the noise has a lower amplitude than the signals in most detectors. However, depending on the application and the type of detector used, this method filters only a part of the noise,retaining those with high amplitude, and it also has the undesirable effect of eliminating low-amplitude signals. In cases like these, methods of discrimination independent of the size of the electrical signals should be considered. The use of machine learning techniques is an interesting alternative, given the increasing availability of high-speed programmable devices like Field Programmable Gate Arrays (FPGAs) for artificial intelligence applications. The method considers the differences in waveforms between signals and noise, which can be used in discriminating the electrical pulses generated by the detectors, regardless of their size. In this dissertation, we present a study of signal discrimination using different machine learning algorithms applied to a scintillator detector. Simulated data of signals and noise and experimental data were used to study signal discrimination using an unsupervised algorithm called Hierarchical Clustering Analysis (HCA) and supervised algorithms such as Random Forest, Deep Neural Networks, and Convolutional Neural Networks. A comparison of the performance of the different supervised algorithms with the pulse amplitude discrimination method is presented.