Improving the Hydra Classifier for Data Quality Monitoring at GlueX

This thesis investigates the potential expansion of Hydra, an AI-based system for data quality monitoring at the Thomas Jefferson National Accelerator Facility. Hydra is designed to assist in monitoring graphical data (histograms) generated by existing systems and improving efficiency and accuracy in identifying data anomalies. Currently, Hydra can classify these histograms into multiple categories. We are mainly concerned with the 'Good' and 'Bad' categories. During this thesis, we wrote a plugin that emulates malfunctioning electronics in a detector. We trained a model to identify malfunctioning electronics with 99% accuracy by adding 1% of the dataset containing only those malfunctioning electronics to our original training dataset. Additionally, we developed a non-machine-learning algorithm that can pinpoint which electronic device has malfunctioned in ~ 6 seconds. This algorithm, if optimized, could be integrated into Hydra's workflow in the future.