Analysis of Artificial Intelligence based diagnostic methods for satellites

The growing utilization of small satellites in various applications has emphasized the need for reliable diagnostic methods to ensure their optimal performance and longevity. This master thesis focuses on the analysis of artificial intelligence-based diagnostic methods for these particular space assets. This work firstly explores the main characteristics and applications of small satellites, highlighting the critical subsystems and components that play a vital role in their proper functioning. The key components of this study revolve around Diagnosis, Prognosis, and Health Monitoring (DPHM) systems and techniques for small satellites. The DPHM systems aim at monitoring the health status of the satellite, detecting anomalies and predicting future system behavior. The reason why advanced DPHM systems are of interest for the space operators is the fact that they mitigate the risk of satellites catastrophic failures that may lead to service interruptions or mission abort. To achieve these objectives, a hybrid architecture combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks is proposed. This architecture leverages the strengths of CNNs in feature extraction and LSTM networks in capturing temporal dependencies. The integration of these two neural network architectures enhances the diagnostic capabilities and enables accurate predictions for small satellite systems. Real data collected from an operational satellite is utilized to validate and test the proposed CNN-LSTM hybrid architecture. Based on the experimental results obtained, advantages and drawbacks of the exploitation of this architecture are discussed.