Taking AI to New Heights

Artificial intelligence (AI) has made significant progress in a variety of areas, with recent successes in anomaly detection standing out as a notable breakthrough. Anomaly detection systems can now efficiently identify deviations from normal patterns in data using cutting-edge machine learning algorithms, allowing for the early detection of potential threats or irregularities in complex systems. This capability has been widely used in sectors such as cybersecurity, healthcare, and manufacturing, allowing for rapid responses to emerging issues and the prevention of potential disasters.

Although these successes have primarily been observed on Earth, there is a growing interest in using such AI capabilities for spacecraft and satellites. There are many potential applications for anomaly detection in space missions, ranging from identifying equipment malfunctions to detecting unusual celestial events or cosmic phenomena. By implementing AI-based anomaly detection systems on satellites and spacecraft, space agencies can improve the reliability and safety of their missions, while also helping scientists to unlock more of the secrets of the universe.

The challenge, however, is dealing with unknown environments encountered during space exploration, which necessitates the development of new AI models on the spacecraft itself. The current practice of sending data back to Earth for training purposes is not feasible for missions exploring the distant reaches of the solar system due to the prolonged data transfer time and limited bandwidth. As a result, the training process must be carried out onboard, using the limited energy and computational resources available. This is a significant obstacle, as traditional AI training methods require significant computing power, which is often impractical in the context of space missions.

A team of engineers at the Julius-Maximilians-Universität Würzburg in Germany has spent the past two years working to send AI to the stars. They have developed a nanosatellite called SONATE-2 that is scheduled to launch via a SpaceX rocket in March of next year. The initial mission will remain in an Earth orbit, allowing the researchers to test novel hardware and software technologies under the harsh conditions found in space. Critically, SONATE-2 will be capable of capturing sensor data and using it to train AI models onboard, with no reliance on Earth-based resources.

The 6U+ cubesat model is about the size of a shoebox and is packed with cameras and processing power to test out newly developed procedures and methods. It is hoped that these techniques will enable SONATE-2 to learn about the normal geometric patterns found on the Earth’s surface, then use that information to detect anomalies. If all goes according to plan, such a system could one day be used to rapidly identify unusual objects and phenomena on a distant asteroid, moon, or planet, for example.

The satellite has already been demonstrated to be capable of enduring the stresses of spaceflight during a launch simulation. This confirmed that all the solder joints, screws, and glue can stand up to the extreme forces that will be encountered before SONATE-2 reaches orbit. The researchers expect that the satellite will remain operational for at least one year, but hope for a much longer useful life to give them plenty of time to test and refine their methods for future missions.