Abstract. Since nanosatellites operate with limited human intervention, onboard fault diagnosis is a critical factor in ensuring reliable mission execution. This paper addresses the development of a 1U–3U CubeSat nanosatellite equipped with an autonomous fault detection and isolation system based on artificial intelligence (AI). The study focuses on the design of a nanosatellite health management system employing machine learning techniques. The work compares classical threshold-based monitoring with modern AI approaches and proposes an onboard diagnostic workflow capable of detecting anomalies in key subsystems with high accuracy and low latency. The hardware architecture, a telemetry-based fault simulation environment, and an AI algorithm trained on simulated telemetry data under nominal and faulty operating conditions are described. The proposed system achieves approximately 98% fault detection accuracy, significantly outperforming classical methods, while maintaining detection latencies of only a few seconds. The results demonstrate that machine learning techniques effectively complement model-based diagnostic approaches. The developed AI-based diagnostic system enhances nanosatellite mission resilience by enabling early anomaly detection and autonomous execution of recovery actions.

Keywords: nanosatellite, CubeSat, telemetry, fault detection and isolation, artificial intelligence, machine learning, hybrid diagnostics, embedded system