Structural Health Monitoring and Uncertainty Quantification in Large Steel Frame Structures

Abstract

Large structural systems, under extreme natural hazard loadings, exhibit strongly nonlinear behavior that challenges commonly used diagnosis and modeling techniques. In such situations, the initial model may be quite inadequate for structural health monitoring. This paper presents the first step toward a digital twin approach that continuously updates the model with health monitoring data, focusing on extreme wind events, and includes uncertainty quantification in both sensor data and model prediction. A two-step modeling methodology is developed for diagnostic and prognostic analysis and uncertainty quantification. The first step consists of Principal Components Analysis (PCA) on strain data from the undamaged structure to compute the Q_Statistic, a damage index vector based on the PCA reconstruction error. The obtained damage index vector is used to compute anomaly/damage thresholds. The strain values, obtained from the finite element analysis simulation of are projected onto the constructed PCA baseline model for damage/anomaly detection. The damage index vector obtained from the PCA projection is used as the output in constructing a Bayesian predictive surrogate model of the structure along with environmental variables of temperature, wind speed, and wind direction as model inputs. The surrogate model is verified, calibrated, and validated with real sensor data from the field, and the uncertainty in the surrogate model prediction is quantified. The validated model is then used for probabilistic prognosis to inform proactive decision-making ahead of critical events. The proposed methodology is illustrated for a large steel frame and rolling door used in aircraft hangars in Florida. The proposed methodology results show that both the PCA and its projection-based predictive model successfully detect damage on the door structure and predict damage indicator for the door structure.