Unsupervised Damage Detection for Smart Extraterrestrial Habitats Using Autoencoders and Information Fusion

Abstract

In the context of smart and resilient extraterrestrial habitats, the structural health monitoring (SHM) of habitats is crucial and challenging due to the harsh space environments. To this end, a novel anomaly detection framework is developed based on active sensing and information theory. Active sensing involves the excitation of the structure at specific locations and collecting acceleration data using sensors. The collected data from different excitation points and sensor locations are analyzed, and the extracted information is fused to further enhance anomaly detection. More specifically, an unsupervised anomaly detection framework using autoencoders (AEs) has been developed. Continuous wavelet transforms (CWTs) of acceleration signals are utilized to train AEs. Information fusion strategies are proposed to enhance the robustness of the approach to both aleatoric and epistemic uncertainties. Two unsupervised learning approaches developed by standard AE and variational autoencoder (VAE) are systematically compared. The numerical study based on the ASCE benchmark model and the experimental study based on a geodesic dome testbed have been carried out to validate the performance of the proposed framework and study its limitations. The framework's ability to extract information from multiple sources allows it to identify anomalies that might have been missed by traditional detection methods. For instance, it is shown that the proposed approach increases anomaly detection accuracy by up to 39.8% under a relatively small damage scenario compared to state-of-the-art approaches.