Reconstruction of Transverse Beam Distribution using Machine Learning

CERN高辐射环境中的束流横向分布是通过使用基于辐射硬管的内部生产的相机对粒子束撞击反射屏产生的光进行成像来测量的。由于全球范围内辐射硬管生产的停止，CERN正在研究使用与普通CMOS相机耦合的耐辐射光纤将光束图像传输到低辐射地区。在此框架下，使用单根大芯多模光纤重建光束横向分布的开创性工作于2020年开始。它利用了使用深度学习方法（如卷积神经网络）的生成建模的进步，并试图将其应用于光束诊断。在这个博士项目中，学生将通过对成像光纤进行更逼真的光学建模来完善模拟数据集，同时考虑到温度和振动效应等环境因素。然后，学生将优化光纤的参数，如直径和数值孔径，并对现有的抗辐射光纤进行市场调查。在下一步中，学生将筛选可用的图像翻译网络，特别是卷积U-Net，广泛用于生物医学图像分割和已经使用的生成对抗网络。然后，学生将使用模拟数据集开发基于机器学习的模型，并评估其性能。在这些结果的基础上，学生将开发一个实验装置来验证模拟结果，并在CERN的CHARM辐照设施进行测量活动，以验证和研究模型中包含的累积剂量相关降解效应。学生将有机会获得Cockcroft研究所在加速器科学方面的综合研究生培训，以及LIV.INNO的结构化培训。学生将在英国度过第1年和第4年，并在第2年和第3年期间在CERN工作。

The beam transverse distribution in CERN's high-radiation environment is measured by imaging the light generated by the particle beam hitting a scintillating screen, using cameras produced in-house based on radiation hard tubes. Due to the cessation of radiation hard tube production worldwide, CERN is investigating the transport of the beam image to low-radiation areas using radiation tolerant optical fibers coupled to normal CMOS cameras. In this framework, pioneering work to reconstruct the beam's transverse distribution using a single large-core multimode optical fiber began in 2020. It takes advantage of advances in generative modeling using deep learning methods, such as convolutional neural networks, and attempts to apply them to beam diagnostics. In this PhD project, the student will refine simulated data sets with more realistic optical modeling of the imaging fiber, taking into account environmental factors such as temperature and vibration effects. The student will then optimize the fiber's parameters, such as diameter and numerical aperture, and perform a market survey for available radiation-resistant fibers. In a next step, the student will screen available networks for image translation, in particular the convolutional U-Net, widely used for biomedical image segmentation and the already used generative adversarial networks. The student will then develop a machine learning-based model using simulated datasets and evaluate its performance. On the basis of these results, the student will develop an experimental setup to validate the simulation results and carry out a measurement campaign at CERN's CHARM irradiation facility to verify and study the accumulated dose related degradation effect to be included in the model. The student will have access to the Cockcroft Institute's comprehensive postgraduate training in accelerator science, as well as to LIV.INNO's structured training. The student will spend years 1 and 4 in the UK, and be based at CERN during years 2 and 3.