Machine Learning Classification of LIGO Seismic States

The Laser Interferometer Gravitational-Wave Observatory achieves attometer sensitivity over a 4-km baseline to detect tiny distortions in space-time. While it is susceptible to many types of noise, seismic noise is not currently a limiting factor (Buikema 2020). However, a correlation is suspected between seismic activity and detector anomalies-"glitches": loud transient noise artifacts in the detector signal, and "locklosses": loss of interferometer control. Evaluation of seismic activity is difficult because of the broad range of possible motions; therefore, identification requires analysis of a large quantity of data. Machine learning has been applied to the gathered data, utilizing clustering methods to discover seismically similar data periods and then evaluate their correlation to detector state. This approach has successfully identified known seismic states. Strong correlations have been discovered between these states and detector anomalies; this is a significant motivator for future engineering studies on the effects of seismic motion on the detector.