Understanding interferometer lock losses with machine learning

Other Versions:

Abstract:: The Laser Interferometer Gravitational-Wave Observatory (LIGO) detectors are complex systems that must be extremely stable to detect gravitational-wave signals. Numerous control loops are used to maintain detector stability, or "lock," but a detector can lose lock. A time-consuming lock acquisition process must be undertaken to regain it, reducing the amount of time during which the interferometer is recording data. The causes of some lock losses are unknown. In this project, we use machine learning to analyze time series data from the auxiliary channels in the LIGO Hanford detector, which can indicate changes in the states of various detector components as well as environmental factors such as seismic noise. We first determine features that characterize the data and then perform a regression to identify which of these features can distinguish between data preceding lock losses and data from stable times. We run a clustering algorithm on the predictive features to identify groups of similar lock loss events. The ultimate goal is to minimize the number of lock losses in the future.
All code can be found in our git repository: https://git.ligo.org/jameson.rollins/locklasso

Files in Document:

Other Files:

Final presentation (LIGO_SURF_Final_Presentation.pptx, file is not accessible)
Final presentation (PDF) (LIGO_SURF_Final_Presentation.pdf, file is not accessible)
Follow-Up Presentation (Followup_Presentation.pdf, file is not accessible)
Interim Report #1 (LIGO_SURF_Interim_Report_1.pdf, file is not accessible)
Interim Report #2 (LIGO_SURF_Interim_Report_2.pdf, file is not accessible)
Project Proposal (LIGO_SURF_Project_Proposal.pdf, file is not accessible)

LIGO Document T1900370-v7