Vibration-based anomaly detection using sparse auto-encoder and control charts

Several approaches can be found in the scientific literature when the subject is damage detection based on vibration signals. In the last few years, increasing attention has been given to the application of Computational Intelligence algorithms in structural novelty identification. In more details, the powerful data mapping capability of computational deep learning methods has been recently exploited to develop strategies of structural health monitoring through appropriate characterization of dynamic responses. Therefore, the present work is aimed at investigating the capability of a deep learning algorithm called Sparse Auto-Encoder (SAE) to identify structural alterations of the Z24 bridge, a classical benchmark for integrity assessment studies. The main idea is to characterize the Z24 dynamic responses via SAE models and, subsequently, to detect the onset of abnormal behavior through the well-known Shewhart T control chart (T2-statistic), calculated with SAE extracted features. An advantage of the proposed methodology is that data are processed directly in the time domain, avoiding modal parameters estimation and tracking analysis. Moreover, control charts are considered suitable tools for continuous monitoring due to their relatively simple implementation. The obtained results demonstrate that the proposed strategy based on SAE and Shewhart T control chart has potential to be explored in structural damage detection problems, since it is able to distinguish between the two investigated scenarios (i.e., undamaged and damaged) of Z24 bridge.