Impact of data dimensionality reduction on vibration-based structural health monitoring for flexible satellites

Sensing approaches based on compressed data have been studied and applied in Structural Health Monitoring (SHM) for data acquisition and analysis. However, their use in the space domain remains relatively unexplored. This study investigates the Symbolic Aggregate approXimation (SAX) technique as a data compression method in structural damage identification for space systems. The SAX-based approach transforms raw time series data into symbolic representations, significantly reducing data dimensionality. This compression is adopted to retain critical information in vibration signals while shortening the input sequence length. Applied to a spacecraft scenario featuring large solar panels monitored by distributed accelerometers, the method demonstrates enhanced classification performance compared to a prior benchmark in a more challenging multi-class damage identification problem. The data reduction is evaluated in combination with bi-LSTM, GRU and TCN network architectures to assess the robustness of the SHM approach across different models, highlighting the dual benefits of improving damage identification accuracy and computational efficiency during training.