Structural Dynamics Identification via Computer Vision and Data Analytics in Structural Health Monitoring

Among the current efforts to improve structural safety and reliability, Structural Health Monitoring (SHM) plays a very important roll, in which real-time data are acquired and analyzed for the ultimate goal of status awareness, condition-based maintenance and retrofit, operational risk minimization, and human life safety. Yet any such condition assessment is inevitably constrained by the observability of the sensing hardware and corrupted by various kinds of uncertainties, which degrades the SHM decision-making capability dramatically, leading to unidentifiable system dynamics, as well as both false damage alarms and missed detections or classifications. In this presentation, the research activities at UMass Lowell related to non-contact full-field system testing and identification will be introduced. Specifically, the enhanced modal identification and damage detection on wind turbine blades, by means of video motion magnification and other computer vision techniques, will be presented. The second half the presentation concerns the uncertainty quantification, and its applications to the SHM decisions. Probabilistic uncertainty quantification models which characterize the actual distributions of selected SHM features as random variables will be briefed. Structural damage is detected via outlier analysis based on the established probabilistic model, given the quantified significance level of decision threshold. Receiver operating characteristics will be compared among different scenarios to have an optimal trade-off between sensitivity and specificity. Bayesian decision theory is then adopted to classify the type of damages in a recursive manner, and posterior probability facilitates the selection of the correct model.