The Interdisciplinary Centre for Security, Reliability and Trust (SnT) has launched AI4Space, its first in-space machine learning experiment. The experiment aims to fine-tune AI algorithms to detect temperature anomalies in satellite systems in a more efficient manner than currently possible. Developed by research teams in the Space Systems Engineering (SpaSys) and the Computer Vision, Imaging and Machine Intelligence (CVI2) research groups, the system will be tested on a Skykraft satellite, which will be deployed in low earth orbit, before the SnT experiment kicks off.
The AI4Space project is highly significant for the space industry and has far-reaching potential applications. The technical design work for the system was carried out according to lean methodology principles, which involves working on incremental improvements in efficiency to ensure quick iteration aimed at producing an efficient and reliable solution. The machine learning algorithms used in the system are designed to be low computation hardware efficient, enabling it to perform real-time onboard anomaly detection.
This experiment is a significant improvement in the traditional out-of-range method of measuring satellite temperature, whereby an electronic device is only considered healthy as long as its temperature stays below a certain threshold. The AI4Space experiment uses complex patterns in temperature changes to alert operators of potential problems, in real time, before they lead to major failures. The SpaSys team prepared the hardware design, the development, and integration, while the CVI2 team contributed its expertise in machine learning to train the anomaly detection algorithm and deploy it on an edge device.
The development methodology is significant for future uses of this technology, as it allows a satellite to autonomously detect and solve a problem before it becomes a major issue. This reduces the amount of supervision necessary and can also extend the satellite’s lifetime. The execution of the AI4Space project envisages updating the anomaly detection algorithm based on the results of on-board anomaly detection. This would further improve the machine learning capabilities of satellite temperature detection. The research team is anticipating producing and disseminating the results of the experiment to expand its utility to all satellite platforms, ranging from small CubeSats to large communication satellites.
