Speaker
Description
The Advanced LIGO detectors are highly sensitive instruments designed for detecting gravitational waves. A critical challenge in maximizing detector uptime is the occurrence of "locklosses," which happen when the interferometer loses light resonance in one or more of its optical cavities, resulting in significant downtime required for re-acquisition. While some locklosses are attributed to known causes such as seismic disturbances, a substantial portion, 26% of total locklosses during the ongoing LIGO's O4 observing run started in 2023, remains classified as "unknown." This investigation aims to enhance the understanding of these "unknown" lockloss events by analyzing the morphological patterns in auxiliary channel data leading up to them.
To characterize the data morphology, we apply t-distributed Stochastic Neighbor Embedding (t-SNE), a non-linear dimensionality reduction algorithm. t-SNE organizes the events by morphological similarity, placing those with more similar features closer together, revealing underlying patterns in the data. We correlate the observed data morphologies with known causes of locklosses and, consequently, assign more specific reasons to the previously categorized "unknown" locklosses.
