Subdomain principal component analysis for damage detection of structures subjected to changing environments

Abstract

Modal frequencies that capture structural dynamics are the most commonly used damage-sensitive features in vibration-based structural health monitoring. However, environmental changes usually lead to nonlinear frequency variabilities, which further mask the damage effects on modal frequencies. In this paper, a subdomain principal component analysis (Sub-PCA) method is proposed for damage detection of structures with robustness/immunity to environmental interference. Considering in mind the linear nature of PCA, it is performed in each of the disjoint linear subdomains of nonlinear frequency data. Two statistical analysis techniques including the Gaussian mixture model and the likelihood ratio test are adopted for subdomain division. Then damage indicators, i.e., the Mahalanobis and Euclidean distances, and their thresholds are calculated according to each Sub-PCA model. After that, weighted averages of the threshold-normalized indicators are defined to detect damages. Two case studies involving an experimental wooden bridge and an actual concrete bridge are employed to verify the effectiveness of the Sub-PCA method. The analysis results demonstrate that the Sub-PCA models, especially those constructed from the likelihood ratio test, are particularly robust/immune to changing environments in terms of damage detection.