Nonstationary System Identification Techniques

System identification (SI) is a critical component of structural health monitoring (SHM) practice, as it closely relates to the structural performance evaluation, damage detection and helps to gain insights to future design practice. Most of the existing SI techniques are based on the assumption of stationarity of data; however, full-scale data measured during strong winds (e.g. typhoon, downburst, and thunder storm etc.), traffic loadings and earthquakes usually exhibit nonstationarity. This feature introduces several challenges in the use of existing SI techniques, as most of the nonstationary data is transient. Accordingly, due to a lack of long segments of stationary data needed for a reliable spectral estimate, classical frequency-domain SI methods (e.g. half-power band width, frequency domain decomposition, etc.) are not applicable. Furthermore, nonstationary colored excitation could cause spectral overlapping between different modes, rendering most of the existing modal identification methods based on the assumption of stationary white noise excitation (e.g. second order blind identification) problematic. Additionally, the high amplitude nonstationary response of large-scale structures under strong earthquakes or windstorms usually exhibit nonlinear behavior, indicated by time-varying frequencies and damping ratios. The traditional SI methods based on the averaged information in the entire time duration (e.g. random decrement technique, etc.), are not able to capture these time-varying system properties. These challenges have underscored the strong need for developing nonstationary SI techniques. In light of this critical need, this research is dedicated to addressing the challenges posted by nonstationarity. Three new nonstationary SI techniques are developed to enable reliable SI from nonstationary response under both in-service and extreme conditions. These techniques are based on the continuous Morlet wavelet transform, transformed singular value decomposition and Laplace wavelet filtering (WT-TSVD-Laplace), time-frequency blind source separation, and time-varying spectra derived from time-frequency representations, respectively. In addition, a web-enabled real-time hybrid SI framework is proposed, in order to track changes in structural characteristics to provide quick decision-making support regarding post-event rescue and structural retrofitting. This web-enabled framework based on WT-TSVD-Laplace and a traditional stationary structure identification scheme, covariance-driven stochastic subspace identification, can account for the dynamic response under both normal conditions (e.g., synoptic winds and/or ambient excitations) and transient loadings (e.g. earthquakes, windstorms, traffic loadings). To demonstrate the effectiveness of the scheme, it has been implemented in the “SmartSync” system of the world’s tallest building-Burj Khalifa. The SI techniques developed in this research add significantly to the applicability of SHM systems towards assessing the structural performance under nonstationary/transient loadings and can ultimately support quick decision-making regarding structural operation.