|
The Very Large Telescope Interferometer (VLTI) must control its Optical Path Differences (OPD) to extremely high precision in order to achieve its characteristic and desired high performance. This proves a challenge when using Very Large Telescope’s (VLT) 8 meter Unit Telescopes (UT) given they are not fully dedicated to interferometry and can be equipped with up to three different instruments each. Among the several important control systems that allow the VLTI to achieve the necessary precision for this task is Manhattan II (MNII), which measures vibrations along the Optical Path (mirrors M1 to M7) and sends Optical Path Length (OPL) corrections to the Delay Lines (DL). In the context of GRAVITY+ upgrade, MNII is being extended to cover a larger portion of the light path (previously M1 to M3) and expanded with Phase-locked Loop (PLL) to improve OPD control by targeting specific frequencies. Alongside, several options are being explored to further improve the capabilities of the system. Active compensation is improved by the upgrade of MNII’s PLL. In addition, better troubleshooting tools and automatic Anomaly Detection (AD) systems are needed to constantly monitor and react to the changing vibration signature of the UTs. Furthermore, similar AD systems will be fundamental in the future for the operation of the upcoming Extremely Large Telescope (ELT). This work is about the ongoing efforts to develop an automatic AD system using Machine Learning on MNII’s vibration data. We focus on the different methods and models used in the proof of concept which include Auto-encoders, clustering and classical statistical methods as well, the infrastructure required to have a working end-to-end prototype, the data pipeline, preprocessing and the future envisioned production system.
|
