Observing the radio sky through CINTEL

Observing the radio sky in partnership with Artificial Intelligence. Developing new tools and collaborative intelligence workflows to optimise radio telescope operations while allowing humans to focus their energy where it is needed most.

ASKAP antennas under the Milky Way. Single ASKAP antenna in the foreground with its PAF receiver lit by Moonlight

CSIRO’s ASKAP radio telescope under the Milky Way ©  CSIRO/A. Cherney

The Challenge

The latest generation of radio telescopes are highly complex multi-element instruments, with many points of potential failure that can reduce the quality and utility of their science data. CSIRO’s ASKAP telescope combines signals from 36 antennas, each with a novel CSIRO-designed phased array feed comprising 188 receiver elements. The whole system has many 1000s of monitoring points, with performance data feeds coming from each sensing point.  Currently, there is more performance data than can readily be interpreted by the small operations team, and we rely on relatively high-level metrics of system health. Consequently, precious telescope time can be lost when recorded data are identified as unsuitable for use during downstream processing.  

Our Response

CSIRO’s Collaborative Intelligence Future Science Platform (CINTEL FSP) and Astronomy and Space Science division are working to design a collaborative human-machine approach to respond to this problem. This research focuses on machine-based systems cleaning the large amount of monitoring data under human guidance and presenting anomalous events for human analysis and interpretation.

Impact

This collaborative intelligence system would enable early warning of configuration errors or fault conditions in complex control systems. Further, it may be possible to teach the system about its own interconnections; mapping the cause and effect of anomalies and collaborating with the human operations team to determine the best course of action and the level of trust in AI systems from astronomers around their early-stage effectiveness is low. 

The Team

Aidan Hotan, Vanessa Moss, Matt Austin, Zhuowei Wang, Ben Harwood, Marcus Hudson, Minh Huynh