Deep Learning-Driven Structural Health Monitoring Using High-Volume Operational Data

The increasing complexity of modern spacecraft calls for advanced methods to assess structural integrity during in-orbit operation. Structural Health Monitoring (SHM) offers a pathway to continuous assessment of critical components, yet conventional approaches struggle with the vast data volumes produced by onboard sensors. This study investigates the integration of deep learning techniques to enable in-orbit SHM for large satellite structures. A distributed network of accelerometers is considered to capture multi-channel time series data, while a high-fidelity finite element spacecraft simulator provides realistic training conditions. Emerging learning models, including recurrent neural networks (RNNs) and wavelet-based convolutional neural networks (CNNs), are evaluated for damage detection. Results highlight the effectiveness of the proposed framework in identifying structural failures with improved computational efficiency. This approach enhances defect detection and supports autonomous self-monitoring, offering a foundation for next-generation space systems and on-orbit servicing.