Anomaly detection with data quality early warning systems in ATLAS

In this dissertation, the implementation of a Data-Quality Early Warning System (DQEWS) is explored. We use unsupervised Machine Learning (ML) methods to evaluate Data-Quality (DQ) in the ATLAS detector. We do so by observing and quantifying the evolution of Luminosity-Block (LB) data from Inner Detector (ID) tracking information, with a single LB towards the beginning of a run used as the reference. In this way, we obtain a trajectory that describes how the recorded LB data drift over the course of a run. Within the scope of this project thus far, the following will be shown. The version of the DQEWS algorithm defined as of the presentation of the results shown in this dissertation is shown to sufficiently flag good LBs as 'good', and bad LBs as 'bad' under the condition that the flagging criteria are evaluated on LB datasets that lie within a similar range of instantaneous luminosity as the LB datasets used to construct the criteria

Grantor

Department of Physics

Year dc:date.issued

2023

Author dc:creator

• Msutwana, Senzo

Advisors dc:contributor.advisor

• Yacoob, Sahal
• Keaveney James

• Physics

Handle dc:identifier.uri

http://hdl.handle.net/11427/39703

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/39703