Physics-based and Data-driven Methods with Compact Computing Emphasis for Structural Health Monitoring

This doctoral dissertation contributes to both model-based and model-free data interpretation techniques in vibration-based Structural Health Monitoring (SHM). In the model-based category, a surrogate-based finite element (FE) model updating algorithm is developed to improve the computational efficiency by replacing the FE model with Response Surface (RS) polynomial models in the optimization problem of model calibration. In addition, formulation of the problem in an iterative format in time domain is proposed to extract more information from measured signals and compensate for the error present in the regressed RS models. This methodology is applied to a numerical case study of a steel frame with global nonlinearity. Its performance in presence of measurement noise is compared with a method based on sensitivity analysis and it is observed that while having comparable accuracy, proposed method outperforms the sensitivity-based model updating procedure in terms of required time. With the assumption of Gaussian measurement noise, it is also shown that this parameter estimation technique has low sensitivity to the standard deviation of the measurement noise. This is validated through several parametric sensitivity studies performed on numerical simulations of nonlinear systems with single and multiple degrees of freedom. The results show the least sensitivity to measurement noise level, selected time window for model updating, and location of the true model parameters in RS regression domain, when vibration frequency of the system is outside the frequency bandwidth of the load. Further application of this method is also presented through a case study of a steel frame with bilinear material model under seismic loading. The results indicate the robustness of this parameter estimation technique for different cases of input excitation, measurement noise level, and true model parametersIn the model-free category, this dissertation presents data-driven damage identification and localization methods based on two-sample control statistics as well as damage-sensitive features to be extracted from single- and multivariate regression models. For this purpose, sequential normalized likelihood ratio test and two-sample t-test are adopted to detect the change in two families of damage features based on the coefficients of four different linear regression models. The performance of combinations of these damage features, regression models and control statistics are compared through a scaled two-bay steel frame instrumented with a dense sensor network and excited by impact loading. It is shown that the presented methodologies are successful in detecting the timing and location of the structural damage, while having acceptable false detection quality. In addition, it is observed that incorporating multiple mathematical models, damage-sensitive features and change detection tests improve the overall performance of these model-free vibration-based structural damage detection procedures. In order to extend the scalability of the presented data-driven damage detection methods, a compressed sensing damage localization algorithm is also proposed. The objective is accurate damage localization in a structural component instrumented with a dense sensor network, by processing data only from a subset of sensors. In this method, first a set of sensors from the network are randomly sampled. Measurements from these sampled sensors are processed to extract damage sensitive features. These features undergo statistical change point analysis to establish a new boundary for a local search of damage location. As the local search proceeds, probability of the damage location is estimated through a Bayesian procedure with a bivariate Gaussian likelihood model. The decision boundary and the posterior probability of the damage location are updated as new sensors are added to processing subset and more information about location of damage becomes available. This procedure is continued until enough evidence is collected to infer about damage location. Performance of this method is evaluated using a FE model of a cracked gusset plate connection. Pre- and post-damage strain distributions in the plate are used for damage diagnosis.Lastly, through study of potential causes of damage to the Washington Monument during the 2011 Virginia earthquake, this dissertation demonstrates the role that SHM techniques plays in improving the credibility of damage assessment and fragility analysis of the constructed structures. An FE model of the Washington Monument is developed and updated based on the dynamic characteristics of the structure identified through ambient vibration measurement. The calibrated model is used to study the behavior of the Monument during 2011 Virginia earthquake. This FE model is then modified to limit the tensile capacity of the grout material and previously cracked sections to investigate the initiation and propagation of cracking in several futuristic earthquake scenarios. The nonlinear FE model is subjected to two ensembles of site-compatible ground motions representing different seismic hazard levels for the Washington Monument, and occurrence probability of several structural and non-structural damage states is investigated