The discovery of the Higgs boson in 2012 at CERN has opened the door for several new studies in particle physics. The Higgs boson plays a crucial role in the biggest mysteries of modern particle physics and the precise measurement of its properties, together with the properties of the W and Z bosons, will provide deep information about the Standard Model (SM) and the physics Beyond it (BSM). At present, several projects are ongoing to design the next-generation collider and its detectors, with the primary task of measuring the Higgs properties and the other most relevant electroweak parameters. The thesis starts with a description of the proposed future colliders and the IDEA multi-purpose detector concept. Particular attention is spent on the calorimetry part, introducing the main factors that currently limit the energy resolution in hadron and jet detection. The dual-readout calorimetry method, a key element of IDEA, is then explained, being one of the most promising candidates to design high-resolution calorimeters at future high-energy e+e- colliders. In the current status, the IDEA calorimeter adopts Silicon PhotoMultipliers (SiPMs) as light readout elements. Therefore, the SiPM technology is presented together with a description of the most characteristic features such as gain, photon detection efficiency, linearity and noise sources. The second part of the thesis describes my contribution to the IDEA dual-readout calorimeter software. It starts with the study of the simulation chain (modelling both the calorimeter and the SiPM-based readout) to evaluate the performance and validate the full simulation. The detector linearity has been studied, showing that, in the measurements of electromagnetic showers and considering both the calorimeter and the readout system, a linearity within 1% can be achieved, at least in the energy range considered (20−80 GeV). Additional studies verified the correctness of the SiPM simulated-signals timing properties. Eventually, the development of deep learning algorithms is presented, together with the achievement of excellent results in performing particle identification distinguishing between neutral pions and photons on the basis of their spatial shower distribution. Two neural network (NN) structures have been considered, a VGG-like NN and a ResNet-like NN. They both can achieve excellent performance in identifying neutral pions with accuracy values of 98.9% (VGGNet) and 97.6% (ResNet) while rejecting photons with probabilities of 99.4% (VGGNet) and 98.3% (ResNet). These results demonstrate the great potential of deep neural network methods applied on data from the IDEA dual-readout calorimeter.
La scoperta del bosone di Higgs nel 2012 al CERN ha aperto le porte ad una serie di nuovi studi nell’ambito della fisica delle particelle. Il bosone, infatti, ricopre un ruolo cruciale nei più grandi misteri della fisica moderna delle particelle e la misurazione precisa delle sue proprietà, insieme a quelle dei bosoni W e Z, porterà nuove informazioni riguardo il modello standard (SM) e le fisica oltre ad esso (BSM). Con quest’obiettivo, diversi collider leptonici ad alta energia sono stati progettati. Al momento il panorama include quattro differenti proposte. Il rivelatore IDEA è uno dei rivelatori in via di sviluppo per i futuri collider. In questa tesi, dopo una descrizione dei collider proposti e del rivelatore IDEA, fornisce una base teorica. Partendo dall’interazione tra particelle e materia e la produzione di sciami di particelle, viene descritta la teoria della calorimetria, dividendo tra calorimetri elettromagnetici e adronici. Viene introdotta anche la tecnica di compensazione dual-readout, caratteristica fondamentale del calorimetro IDEA. Allo stato attuale, il calorimetro idea sfrutta fotomoltiplicatori al silicio (SiPM) come sensori di luce. Di conseguenza la tecnologia dei SiPM è descritta insieme ai loro tratti caratteristici come guadagno, efficienza in rivelazione di fotoni, linearità e sorgenti di rumore. La seconda parte della tesi mostra il mio contributo al software associato al calorimetro dual-readout IDEA. Inizia con lo studio della catena di simulazione (comprensiva del calorimetro e dei sensori SiPM) per valutare per prestazioni e validare il processo di simulazione. La linearità del rivelatore completo è stata analizzata e mostra che per la rivelazione di sciami elettromagnetici la non-linearità rimane sotto l’1% per il range di energia studiato (20 – 80 GeV). Altri studi riguardo le proprietà temporali hanno verificato la correttezza dei segnali simulati dei SiPM. In fine viene presentato lo sviluppo di algoritmi di deep learning insieme al raggiungimento di ottimi risultati nel distinguere tra pioni neutri e fotoni, basando l’analisi sulla distribuzione spaziale degli sciami prodotti. Sono state considerate due reti neurali, “VGG Network” e “Residual Network”. Entrambe hanno raggiunto eccellenti prestazioni nell’identificare i pioni neutri con valori di accuratezza di 98.9% (VGGNet) e 97.6% (ResNet) mantenendo allo stesso tempo un rifiuto di riconosce fotoni come pioni neutri pari al 99.4% (VGGNet) e 98.3% (ResNet). Questi risultati dimostrano il grande potenziale dell’applicare metodi di reti neurali profonde su dati ottenuti dal calorimetro dual-readout IDEA.
Calorimetria Dual-Readout: studi sulla simulazione del sistema readout basato su SiPM e applicazioni di deep learning.
VILLA, ALESSANDRO
2019/2020
Abstract
The discovery of the Higgs boson in 2012 at CERN has opened the door for several new studies in particle physics. The Higgs boson plays a crucial role in the biggest mysteries of modern particle physics and the precise measurement of its properties, together with the properties of the W and Z bosons, will provide deep information about the Standard Model (SM) and the physics Beyond it (BSM). At present, several projects are ongoing to design the next-generation collider and its detectors, with the primary task of measuring the Higgs properties and the other most relevant electroweak parameters. The thesis starts with a description of the proposed future colliders and the IDEA multi-purpose detector concept. Particular attention is spent on the calorimetry part, introducing the main factors that currently limit the energy resolution in hadron and jet detection. The dual-readout calorimetry method, a key element of IDEA, is then explained, being one of the most promising candidates to design high-resolution calorimeters at future high-energy e+e- colliders. In the current status, the IDEA calorimeter adopts Silicon PhotoMultipliers (SiPMs) as light readout elements. Therefore, the SiPM technology is presented together with a description of the most characteristic features such as gain, photon detection efficiency, linearity and noise sources. The second part of the thesis describes my contribution to the IDEA dual-readout calorimeter software. It starts with the study of the simulation chain (modelling both the calorimeter and the SiPM-based readout) to evaluate the performance and validate the full simulation. The detector linearity has been studied, showing that, in the measurements of electromagnetic showers and considering both the calorimeter and the readout system, a linearity within 1% can be achieved, at least in the energy range considered (20−80 GeV). Additional studies verified the correctness of the SiPM simulated-signals timing properties. Eventually, the development of deep learning algorithms is presented, together with the achievement of excellent results in performing particle identification distinguishing between neutral pions and photons on the basis of their spatial shower distribution. Two neural network (NN) structures have been considered, a VGG-like NN and a ResNet-like NN. They both can achieve excellent performance in identifying neutral pions with accuracy values of 98.9% (VGGNet) and 97.6% (ResNet) while rejecting photons with probabilities of 99.4% (VGGNet) and 98.3% (ResNet). These results demonstrate the great potential of deep neural network methods applied on data from the IDEA dual-readout calorimeter.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/12403