Novelty Detection Using Sparse Auto-Encoders to Characterize Structural Vibration Responses.

Deep learning techniques have been increasingly popular for detecting structural novelties in recent years. The deep learning notion originates from the theory of neural networks, and it comprises several machine learning approaches that were primarily created to solve high-dimensional and nonlinear problems due to their great data mapping capabilities. Although the basic ideas of such algorithms were established in the 1960s, their use in damage detection situations is still relatively new. In so doing, the current study assesses the Sparse Auto-Encoder (SAE) deep learning method when applied to the characterization of structural anomalies. The fundamental concept is to employ the SAE to extract significant features from monitored signals and the well-known Support Vector Machine (SVM) to classify those features within the framework of a Structural Health Monitoring (SHM) program. The proposed method is evaluated using vibration data from a numerical beam model and a highway viaduct in Brazil. The results demonstrate that the SAE can extract relevant properties from dynamic data, making it valuable for SHM applications. [ABSTRACT FROM AUTHOR]