Deep Learning for Space

PROJECT SUMMARY

DL4SPACE (Deep Learning for Space) is an ESA-funded activity that investigates the development of innovative technology to support operators and improve spacecraft operations and AIT activities. The activity was carried out together with GSOC (German Space Operations Centre), GOMSpace (Luxembourg), and UniBwM (Germany).

PROBLEM WE SOLVED

AIKO has worked together with GSOC and GOMSpace to identify a set of possible interesting use cases in OPS (Operations) and AIT (Assembly, Integration, and Testing). OPS objectives were twofold: on one side, to enhance current monitoring algorithms based on out-of-limit approaches and to identify possible anomalies and root causes; on the other side, to research the possibility of predicting telemetries and analyze their trends. AIT's objective was to design a data-driven algorithm to estimate the performance of a reaction control thruster.

WHY IT IS IMPORTANT

DL4SPACE demonstrated the capabilities of data-driven approaches applied to the OPS and AIT domains. Concerning OPS, the developed algorithm is capable of detecting anomalies within the nominal operational ranges (those that a traditional out-of-limit approach cannot identify) and output a list of potential root cause candidates. The algorithm has been assessed together with GSOC, showing promising performances. On the other hand, the telemetry prediction model and trend analyzer can be used to improve the planning of mission timelines and support operators in identifying components’ aging and degradation. Concerning AIT, the developed data-driven model can predict the thrust, given as input the input command and the test pressure. In the future, it could become a useful support during the AIT campaign to save time and decrease costs.

Publications

17th INTERNATIONAL CONFERENCE ON SPACE OPERATIONS "An innovative AI-based framework for on-ground anomaly detection and root cause analysis" | To appear