Powered by Indico
v3.2.2
Plasma based particle acceleration is a hot topic of contemporary physics. The CERN - AWAKE Experiment aims to accelerate electrons in the wakefield of a train of proton microbunches in rubidium plasma. The plasma is created by an intense, infra-red laser pulse via photoionisation in a tenuous rubidium vapour. The vapor is 10 meters long and has a diameter of a few centimeters. In the experiment it is essential to determine the geometrical dimensions of the plasma channel, namely, the location of the center of the channel within the vacuum chamber, the width of the fully ionised region and the characteristic lenght of the transition region between the plasma and the vapour. In the experiment, the plasma channel is investigated in a Schlieren-imaging setup. However, determining the relevant geometrical dimensions from the Schlieren-signals is not a simple task. For this purpose, we used a machine learning approach. We constructed different neural networks that were able to predict the geometrical parameters of the plasma channel with high accuracy. We also tested the sensitivity of our networks for errors that can be present in the experiments and found that for slight errors, the predictions still remain valid, furthermore, the robustness of our models against these errors can be enhanced efficiently.
[Authors: Gábor Bíró, Mihály Pocsai, Imre Ferenc Barna, Joshua T. Moody and Gábor Demeter]
Zoom: https://cern.zoom.us/j/67590004845?pwd=TmhMaVpCOFhQZ3RCRUJwVUVMSFV0dz09
Meeting ID: 675 9000 4845
Passcode: 344007