Model-free fault detection framework for spacecraft health monitoring

Subsystem failures in spacecraft can lead to significant performance degradation and compromised mission safety if not addressed promptly. To mitigate such risks, this paper introduces a model-free fault detection framework aimed at enhancing onboard health monitoring without relying on detailed system models. The proposed approach integrates a support vector machine with a bio-inspired negative selection algorithm for effective self/nonself classification, while a clonal selection algorithm dynamically optimizes the model’s hyperparameters to improve adaptability under uncertain conditions. The framework is validated on a spacecraft attitude determination and control system testbed, where a series of fault scenarios are simulated. Results show that the proposed architecture outperforms a traditional standalone negative selection data-driven method by achieving higher overall efficiency in detecting anomalous behaviors. Specifically, the proposed method achieved a 10.5 % reduction in fault activation time, an 82.5 % decrease in execution time, and a 5.4 % improvement in accuracy for capturing deviations from nominal conditions. These findings highlight the value of bio-inspired machine learning in supporting the development of autonomous and fault-tolerant aerospace systems capable of operating reliably in complex environments.