深度学习在MRPC性能分析中应用研究

Time of flight system (TOF) based on MRPC technology is widely used in modern physics experiments, and it also plays an important role. With the increase of accelerator energy and measurement precision, it is very urgent to develop high time resolution TOF system. In order to improve the time resolution of TOF system, in the hardware aspect, we have to development narrower gap MRPC and we also have to develop high bandwidth FEE. In the same time, we will use high speed pulse shape sampling technology to get the output pulse shape of MRPC. Compared traditional TOT method, this technology is undoubtedly more advanced. So we can not use traditional time calibration method. Deep learning is a new field of computer science and artificial intelligence. It has been widely used in the field of high energy physics in recent years. In order to make full use of the abundant information provided by signal waveforms, this project first uses deep learning methods to study the time performance of MRPC. First,we will construct a neural network for deep learning. Then we will use the simulated MRPC output waveform signal to train the neural network. After above two steps we can input the measured signal into the neural network to get the exact time of the particle incident on the detector. This technology greatly improve the time performance of the MRPC, and also the TOF particle identification ability. This is of great significance for promoting China's high-precision particle detection technology and cultivating innovative nuclear physics research scientists.

基于多气隙电阻板室（MRPC）技术的飞行时间谱仪（TOF）在现代物理实验的粒子鉴别中得到了广泛应用，发挥了重要作用。随着加速器能量及测量精度的提高，研制更高时间精度的飞行时间探测器势在必行。为提高TOF系统的时间分辨，硬件方面是要研制更窄气隙的MRPC，研制高带宽前置放大器，采用高速波形采样技术得到其输出波形。这种技术路线较传统过阈定时方法无疑具有先进性，这样我们就不能沿用传统的时间刻度方法。深度学习是计算机科学、人工智能领域的新兴领域，近年来在高能物理领域得到广泛应用。为了充分利用信号波形所提供的丰富信息，本课题首次采用深度学习的方法来研究MRPC的时间性能。构建深度学习的神经网络，用模拟得到的MRPC输出波形信号对神经网络进行学习，将实测信号输入神经网络，得到粒子入射到探测器的准确时间，大大提高MRPC的时间性能，提高TOF的粒子鉴别能力。

基于多气隙电阻板室（MRPC）技术的飞行时间谱仪（TOF）在现代物理实验的粒子鉴别中得到了广泛应用，发挥了重要作用。随着加速器能量及测量精度的提高，研制更高时间精度的飞行时间探测器势在必行。为提高TOF系统的时间分辨，硬件方面是要研制更窄气隙的MRPC，研制高带宽前置放大器，采用高速波形采样技术得到其输出波形。这种技术路线较传统过阈定时方法无疑具有先进性，这样我们就不能沿用传统的时间刻度方法。深度学习是计算机科学、人工智能领域的新兴领域，近年来在高能物理领域得到广泛应用。课题紧密围绕应用深度学习技术分析MRPC探测器的性能这个中心点，在探测器模拟，神经网络时间重建，窄气隙MRPC设计及测试，CEE实验端部MRPC研制，气体污染及解决方案研究等方面取得了很好的进展。课题基于Geant4软件开发出一套独立于物理实验框架的MRPC信号模拟程序，模拟结果与实验符合度较高，对MRPC的研发、设计与优化有重要作用。首次深入地对MRPC的本征时间分辨能力进行了定量地理论和模拟研究，总结出决定MRPC本征时间分辨率的主要因素，为超高时间分辨MRPC的研制提供了重要指导。结合高速波形采样技术在飞行时间谱仪系统中的应用，首次提出了一种基于深度学习和神经网络的MRPC探测器时间重建算法，并根据MRPC信号的特点，首次提出了一种ComLSTM神经网络，改善了时间分辨率。实验表明，对于四室8气隙、单气隙厚0.104 mm的MRPC，该算法重建出的最佳时间分辨率为16.8 ps，较传统方法提升了约20%，处于世界领先水平。