First neural-network sensing and control in a gravitational-wave detector
GEO600 scientists demonstrate first-ever successful implementation
Modern kilometer-sized gravitational-wave detectors such as LIGO, Virgo, KAGRA, and GEO600 are very complex systems that rely on precisely aligned multi-component suspended optics. This alignement must be kept as close as possible to an optimum configuration at all times while being disturbed by environmental influences. Deviations from precise aligment lower the gravitational-wave measurement sensitivity of the detectors. To overcome limitations of current techniques for aligment, researchers at the Max Planck Institute for Gravitational Physics (Albert Einstein Institute) and at Leibniz University Hannover have successfully implemented for the very first time an aligment sensing and control based on neural networks in a gravitational-wave detector. Their demonstration is not only yet another detector technology breakthrough by the GEO600 team, but also a promising first step towards more general machine learning-based control in current and future generation gravitational-wave observatories.
Paper abstract
Suspended optics in gravitational wave (GW) observatories are susceptible to alignment perturbations and, in particular, to slow drifts over time due to variations in temperature and seismic levels. Such misalignments affect the coupling of the incident laser beam into the optical cavities, degrade both circulating power and optomechanical photon squeezing, and thus decrease the astrophysical sensitivity to merging binaries. Traditional alignment techniques involve differential wavefront sensing using multiple quadrant photodiodes, but are often restricted in bandwidth and are limited by the sensing noise. We present the first-ever successful implementation of neural network-based sensing and control at a gravitational wave observatory and demonstrate low-frequency control of the signal recycling mirror at the GEO 600 detector. Alignment information for three critical optics is simultaneously extracted from the interferometric dark port camera images via a CNN-LSTM network architecture and is then used for MIMO control using soft actor-critic-based deep reinforcement learning. Overall sensitivity improvement achieved using our scheme demonstrates deep learning’s capabilities as a viable tool for real-time sensing and control for current and next-generation GW interferometers.