Anomaly Identification of Structural Health Monitoring Data Using Dynamic Independent Component Analysis

Abstract

Independent component analysis (ICA) has the potential to identify anomalies in structural health monitoring (SHM) data due to its non-Gaussian data-processing ability. In order to additionally take into account the dynamic property between current and past measurements, this paper proposes to employ the concept of dynamic ICA (DICA) for anomaly identification. However, no standard criterion is available for dimensionality reduction, i.e., to extract the systematic and noisy parts. Canonical correlation analysis (CCA) is therefore used to preprocess the time-delayed SHM data where the dynamic behavior is included (CCA is introduced here to serve as a dynamic whitening tool). A direct criterion (i.e., whether the canonical correlation coefficient equals zero) is then presented for extracting systematic and noisy parts, followed by the formulation of a modified DICA method. After that, two statistics are defined to detect potential anomalies, and two corresponding indices are deduced to locate anomaly sources. Case studies using SHM data from a numerical benchmark structure and an actual cable-stayed bridge are finally considered to verify the availability and effectiveness of the proposed method.