Artificial intelligence (AI) is everywhere, to win the game of go, drive cars, chat with humans, create illustrations… And now to help astronomers. For years, scientists have been using these techniques to automatically clify the different types of galaxies, or to identify exoplanets disrupting the light flux of their star. Now these methods are creeping into the very heart of telescopes.
“Deep learning has become clic in our discipline for data exploitation. But tomorrow it will be used to operate the telescopes, warns Damien Gratadour, CNRS researcher at the Laboratory for Space Studies and Instrumentation in Astrophysics (Lesia, Paris Observatory and Paris-Sciences et lettres) in Meudon, in the Haus-de-Seine. Future instruments will even be so complex that AI will undoubtedly be necessary to use them. »
For four years, this researcher has been immersed in algorithms to prevent him and his colleagues from finding themselves helpless when the future Extremely Large Telescope (ELT) in Chile and its 40-meter mirror or the SKA radio telescope and its 130,000 antennas in South Africa and Australia will be at full capacity for the 2030s.
“Our first results obtained since the spring on the Subaru telescope in Hawaii are encouraging”, indicates the researcher, who has started to deploy his programs on an essential technique for terrestrial observations, adaptive optics. The latter uses a mirror whose surface is deformable thanks to thousands of actuators, in order to compensate for the turbulence in the atmosphere which blurs the images. The principle is “simple”: an analyzer measures the small differences between the light beam received from the sky and a reference, then a computer calculates the commands in order to cancel these differences, by creating hollows and bumps by displacements of a few micrometers. All in less than a millisecond, which is the characteristic time of variation of the disturbances.
But the more actuators there are, the more calculations are required, while remaining within the very short time window. And the more precise we want to be, the more we must compensate for very rapid or very weak disturbances. This is where current techniques find their limit. “For the telescope in Chile, Very Large Telescope (VLT), we need to control between 200 and 1,600 actuators. For the next one, the ELT, with a main mirror five times larger, the complexity will be twenty-five times greater, with up to 40,000 actuators for the most complex instruments.recalls Damien Gratadour.
You have 65% of this article left to read. The rest is reserved for subscribers.