Early Anomaly Warning of Environment-Induced Bridge Modal Variability through Localized Principal Component Differences

Abstract

Accurate elimination of environmental variability on bridge modal frequency is a prerequisite for high-quality structural performance evaluation. However, the non-Gaussian and nonlinear characteristics of data distribution associated with variable environments restrict the application of anomaly warning methods with inaccurate or unreliable detection results. Consequently, an early warning method in abnormal modal frequency based on the localized principal component differences model through integrating the slow feature analysis (SFA) and k-nearest neighbor rule is proposed in this paper. SFA is first used to extract the measured slowly features of modal frequency for dimensionality reduction and redundant information elimination. Second, the localized modal set of each sample can be automatically searched from the training database based on the Euclidean distance metrics. Third, the estimated slowly features of modal frequency can be calculated using the mean vector of this set. Finally, the environmental variability can be suppressed through the principal component differences between measured and estimated slowly features. After this analysis, an early warning index of modal abnormality (i.e., Mahalanobis distance) is defined for enlarging slight changes in abnormal frequency. The warning results of Z24 bridge indicate that the proposed method discards the environment-induced modal variability without environmental measurements by fully considering both the nonlinearity between modal variables and the non-Gaussianity of data distribution, and the detectability of frequency anomalies outperforms conventional methods under various modal order combinations.

Eliminating environmental and operational effects on structural modal frequency: A comprehensive review

“…Note that environmental feature matching method is mainly used for anomaly detection without the removal of external interferences. Among them, outlier analysis can be considered a special cluster, but the data should obey a multivariate Gaussian distribution 136 …”

“…Among them, outlier analysis can be considered a special cluster, but the data should obey a multivariate Gaussian distribution. 136 Under nonlinear conditions, clustering and classification algorithms are more common; the purpose of data classification is to assign new data objects to a correct category according to their attributes, while cluster analysis is to divides data into multiple clusters so that the data in each cluster are similar to each other and far away from other clusters. However, the limited training datasets, outliers, and reasonability of generated clusters under complex environments may be unfavorable to discard environmental variations.…”

Bridge Damage Detection Using Complexity Pursuit and Extreme Value Theory

“…These methods treat EOV as a latent variable, which can be extracted using feature extraction methods such as principal component analysis (PCA), factor analysis [17], and blind source separation (BSS). Among these, PCA is the most widely applied [18]. The PCA-based damage detection method assumes that the modes of dynamic characteristic changes caused by structural damage and EOVs are different; hence, they can be separated [19].…”