Data-driven control of Gravitational Wave detectors

Control is one of the fundamental pillars in the operation of gravitational wave detectors and played a vital role in enabling one of the most remarkable scientific achievements of the 21st century: the first detection of gravitational waves in 2015. Almost exactly a hundred years after Einstein's prediction of the existence of gravitational waves, the LIGO and Virgo detectors have started a new era in astronomy by enabling the measurement of these faint signals resulting from the acceleration of massive celestial objects. Using large-scale interferometry, these terrestrial detectors have been able to measure the resulting ripples in the fabric of spacetime, which induce length changes in the detector on the order of 1 attometer (1e-18 m). This success has inspired the development of next-generation detectors such as the Einstein Telescope, aiming to measure signals of even lower amplitude, which may even contain remnants of the Big Bang. Two key metrics for the success of a detector are its sensitivity and duty cycle, and next-generation detectors require large improvements in both aspects. Control plays a fundamental role in achieving these objectives and advances in this area are therefore crucial to enable the next-generation of gravitational wave detectors. Gravitational wave detectors are large-scale, complex optomechatronic systems, which present significant challenges within the control domain. First, several subsystems integrate technology from many physical domains, making it difficult to systematically identify performance limitations and optimize overall control performance. Second, the multi-physical character of the system gives rise to complex dynamics that are often strongly coupled, time-varying and potentially non-linear. Traditional model-based methods are typically limited in their abilities to address these control-related challenges due to their lack of flexibility and robustness. Third, since experimentation and downtime are expensive, there is an increased need for control design tools and methods that make efficient use of data and minimize operational downtime. Common state-of-the-art control techniques are often not well suited to effectively address the above-mentioned challenges, due to the unique nature of these systems. The aim of this thesis is therefore to develop control approaches specifically tailored to the demands of gravitational wave detectors. This thesis presents five contributions that facilitate improved duty cycles and increased sensitivities for current and next-generation detectors. First, a method is developed that iteratively estimates a decoupling matrix to handle the time-varying coupling between several degrees of freedom in Virgo. Implementation of this method on Virgo has demonstrated that the method reduces some of the couplings by an order of magnitude and can adapt to the time-varying coupling in the system, thereby improving system robustness during operation and enabling higher-performance controllers that reduce noise coupling to the detector output. Second, a novel approach to locking an optical cavity is explored that uses a non-linear state estimator to estimate the length of the cavity. The method has been experimentally verified on a simulation and experimental data of a single cavity from Virgo, illustrating its potential to significantly reduce the time needed to lock the cavities, thereby improving the duty cycle of the detector. Third, the GraFIT toolbox is developed for measuring Frequency Response Functions, which uses a novel estimation method from the control community. Application of the toolbox on two experimental use cases in gravitational wave detectors has demonstrated that the required amount of data and therefore measurement time can be reduced by a factor of ten, while achieving similar error bounds to those of the baseline method. Fourth, a systematic approach is presented for the derivation of a model and noise budget of a complex nested feedback loop in the Virgo detector, which are subsequently integrated in an optimal control design method. Experimental validation of the new controller on the Virgo detector has demonstrated roughly a factor of 3 reduction in RMS error of the loop with similar levels of sensor noise amplification. Fifth, an adaptive online subtraction method is developed that uses observational data to estimate the subtraction filter online, aiming to minimize the coupling of noise from auxiliary degrees of freedom to the output of the detector. This approach removes the need for commissioning activities to update the filter due to the time-varying nature of the system, thereby reducing the downtime and potentially improving the sensitivity. The majority of the above-mentioned research results have been successfully implemented on the Virgo detector in Pisa, Italy. The deployment of these methods during the O4 science run demonstrates the practical impact of advanced control techniques across a range of applications, from enhancing control performance to reducing downtime and commissioning effort. Novel control techniques specifically tailored to the requirements of gravitational wave detectors will therefore play an important role in enabling the next-generation detectors.