AI-based system for fault detection, recovery, and resource management in satellites
The project addresses the need for greater autonomy in managing large satellite constellations, where traditional ground-based fault management is costly, slow, and increasingly impractical. The main challenge is to detect, isolate, and resolve failures on board in real time, while predicting potential degradations before they affect service.
To tackle this, the consortium combines advanced AI methods with neuromorphic hardware, enabling fast, low-power decision-making directly on spacecraft. This proof-of-concept approach will demonstrate how predictive monitoring and autonomous recovery can reduce outage and restoration times by an order of magnitude and pave the way for resilient, self-managing space systems.
fortiss will investigate CNN-based classification and detection algorithms on the BrainChip Akida platform. In addition, we will explore pattern-matching techniques, encoders, and complementary network architectures to process heterogeneous multi-channel inputs. As a first step, we will use the public ESA anomaly dataset to identify suitable network models for fault detection. Building on these results, we will validate the implemented algorithms using real-time data from Kepler and the OHB constellation simulator.
01.09.2025 - 31.08.2027
