Machine learning method for 12 C event classification and reconstruction in the active target time-projection chamber

Q: What problem does the document address for active target time projection chambers?

It addresses the challenge of analyzing the large data volume produced by AT-TPCs and replacing time-consuming, biased manual event selection with automated methods.

Q: Which machine learning models are evaluated for event classification and reconstruction?

The study evaluates Visual Geometry Group (VGG) and Residual neural Network (ResNet) models for classifying and reconstructing events from the excited 2+2 state decays in the 12 C Hoyle rotation band.

Q: What performance results are reported for the best ResNet models?

ResNet-34 achieves a precision of 0.99 on simulation data. ResNet-18 provides energy resolution σE < 77 keV and angular reconstruction deviation σθ < 0.1 rad.

Aug 5, 2026

Upload by Patrick

Active target time projection chambers are key tools for studies involving low-energy radioactive ion beams and gamma rays. This work applies machine learning to AT-TPC data, testing VGG and ResNet architectures for event classification and reconstruction of decays from the excited 2+ state in the 12 C Hoyle rotation band. ResNet-34 reaches 0.99 precision on simulation data. ResNet-18 achieves energy resolution σE < 77 keV and angular deviation σθ < 0.1 rad, indicating Monte Carlo-trained models can support future experimental event classification and prediction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Motivation and background for AT-TPC

Machine learning for detector data analysis

The Hoyle state in 12 C and the study goal

Machine learning method for 12 C event classification and reconstruction in the active target time-projection chamber Huangkai Wua,b , Youjing Wangc , Yumiao Wangc , Xiangai Dengc,d , Xiguang Caoa,b,e , Deqing Fangc , Weihu Mac , Hongwei Wanga,b,e , Wanbing Hec,∗, Changbo Fuc,∗, Yugang Mac,d,∗ a Shanghai Institute of Applied Physics, Chinese Academy of Sciences, 201800, Shanghai, China b University of Chinese Academy of Sciences, 100049, Beijing, China c Key Laboratory of Nuclear Physics and Ion-beam Application (MOE), Institute of Modern Physics, Fudan University, 200433, Shanghai, China d Shanghai Research Center for Theoretical Nuclear Physics, NSFC and Fudan University, 200438, Shanghai, China e Shanghai Advanced Research Institute, Chinese Academy of Sciences, 201210, Shanghai, China Abstract Active target time projection chambers are important tools in low energy radioactive ion beams or gamma rays related researches. In this work, we present the application of machine learning methods to the analysis of data obtained from an active target time projection chamber. Specifically, we investigate the effectiveness of Visual Geometry Group (VGG) and the Residual neural Network (ResNet) models for event classification and reconstruction in decays from the excited 2+2 state in 12 C Hoyle rotation band. The results show that machine learning methods are effective in identifying 12 C events from the background noise, with ResNet- 34 achieving an impressive precision of 0.99 on simulation data, and the best performing event reconstruction model ResNet-18 providing an energy resolution of σE < 77 keV and an angular reconstruction deviation of σθ < 0. 1 rad. The promising results suggest that the ResNet model trained on Monte Carlo samples could be used for future classifying and predicting experimental data in active target time projection chambers related experiments. Keywords: Machine learning, Convolutional neural network, Active targets, Time projection chamber, Hoyle rotation band 1. Introduction Radioactive ion beams (RIBs) and gamma beams play important roles in modern nuclear physics studies [1–4] . However, the beams have their own shortages. Such as, beam intensities are too weak, or lifetimes of RIBs are too short etc. To overcome them, the technology of active target time projection chamber (AT-TPC) has been developed in the past two decades [5, 6], which has revolutionized the fields. Typically, AT-TPCs utilize various gases serving as targets and detectors at the same time, which enables them to offer high-resolution and efficient reconstruction of charged particle trajectories and energies. This feature enables AT-TPCs as a popular choice in nuclear physics research, particularly in experiments involving RIBs and gamma beams. Several AT-TPCs have been developed in recent years, such as the MAIKo by Kyoto University for investigating shell evolution and cluster structure [7], TexAT by Texas A&M University for the study of shell evolution [8], and MATE by the Institute of Modern Physics, Chinese Academy of Sciences for investigating heavy-ion fusion reactions at stellar energies [9] . For more information on AT-TPCs, one can refer to a review article by Bazin et al. [5, 6] . ∗ Corresponding author Email addresses: [hewanbing@fudan.edu.cn](hewanbing@fudan.edu.cn) (Wanbing He), [cbfu@fudan.edu.cn](cbfu@fudan.edu.cn) (Changbo Fu), [mayugang@fudan.edu.cn](mayugang@fudan.edu.cn) (Yugang Ma) Although AT-TPCs offer many advantages, analyzing the large amount of data they generate presents a significant challenge. For instance, a typical week-long experiment produces approximately 10 terabytes of raw data, which must be processed to obtain charge deposition and spatial information. Conventional analysis methods typically rely on manual inspection and event selection, which is time-consuming and subject to human bias. However, convolutional neural networks (CNNs) have demonstrated great potential in

FAQs

What problem does the document address for active target time projection chambers?

It addresses the challenge of analyzing the large data volume produced by AT-TPCs and replacing time-consuming, biased manual event selection with automated methods.

Which machine learning models are evaluated for event classification and reconstruction?

The study evaluates Visual Geometry Group (VGG) and Residual neural Network (ResNet) models for classifying and reconstructing events from the excited 2+2 state decays in the 12 C Hoyle rotation band.

What performance results are reported for the best ResNet models?

ResNet-34 achieves a precision of 0.99 on simulation data. ResNet-18 provides energy resolution σE < 77 keV and angular reconstruction deviation σθ < 0.1 rad.

Aug 5, 2026

Upload by Patrick