Freiburg 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

Q: Fachverband Quantenoptik und Photonik

Q 67: Machine Learning

Q 67.3: Talk

Friday, March 15, 2024, 15:15–15:30, Aula

Optimizing the active isolation of an optical table with machine learning — •Jan-Niklas Feldhusen, Artem Basalaev, and Oliver Gerberding — Institut für Experimentalphysik, Universität Hamburg, 22761 Hamburg, Germany

Environmental seismic disturbances, also called seismic noise, limit the sensitivity of ground based gravitational wave detectors.

These disturbances couple via the optical components into the signal. To mitigate this noise, the optical components are passively isolated with suspensions. Parts of the suspension system include an active isolation, which suppresses the inflicted movement by knowing the transfer function of the suspension system and the motion on the ground.

We study if it is possible to improve the active isolation with an artificial neural network. In our laboratory at Universität Hamburg we have a large vacuum chamber with a seismically isolated optical table inside, intended for in-vacuum testing of interferometric inertial sensors - a task that has qualitatively similar requirements for seismic isolation as the first isolation stages of gravitational wave detectors. In this study we show that it is possible to infer averaged spectral density of motion of the table from measurements with seismometers on the floor, by utilizing artificial neural networks. We can get a better estimate of the seismic noise spectral amplitudes on the optical table than a Wiener Filter. We also investigate the ability of the neural network to predict future motion to get a real-time active isolation by feedforward of the inverted anticipated motion.

Keywords: gravitational wave physics; neural network; seismic isolation