Experimental Evaluation of System Identification Techniques and Development of a Bayesian Damage Detection Strategy for Drive-By Health Monitoring

Drive-by health monitoring (DBHM) is an indirect structural health monitoring (SHM) strategy developed to reduce costs associated with traditional SHM systems on highway bridges. Before DBHM can be deployed as a reliable alternative or supplement to traditional damage detection practices, however, limitations with the methodology\u27s system identification and damage detection capabilities need to be addressed. In this dissertation, experimental DBHM system identification studies are conducted, and a novel multi-level damage classification methodology is developed to further cultivate the DBHM methodology into a reliable strategy for monitoring the nations highway bridge network. The main objectives of this work are to: 1) experimentally analyze the feasibility of employing DBHM on short span bridges; 2) characterizes the effectiveness of various experimental and OMA system identification procedures to help overcome limitations with DBHM on short bridge spans; 3) develop a novel Bayesian estimation technique for multi-level damage classification that leverages experimental DBHM data to update vehicle-bridge finite element models (FEMs); 4) present a novel strategy for relating crack damage identified in embedded FEMs to approximate levels of cracking on a physical bridge; and 5) formulate the path for future research to improve the performance of the Bayesian estimation technique when examining physical bridge data. To accomplish the aforementioned objectives, this work consists of four complementary studies. The first study focuses on addressing gaps in experimental DBHM research by investigating the feasibility of employing operational modal analysis (OMA) techniques in DBHM to identify short span bridge properties from the dynamic response of passenger vehicles. Multiple OMA techniques are employed to identify if any approach offers superior system identification capabilities under the given framework. Lastly, various testing procedures are evaluated to establish best practices for identifying short span bridge properties from passenger vehicles. Results from the study demonstrate that each of the employed OMA techniques is capable of detecting the fundamental bridge frequency in the response of a passenger vehicles. Additionally, a combination of OMA techniques is recommended for identifying bridge properties and the advantages of each approach are highlighted. The second study focuses on addressing the need for a new multi-level damage classification strategy in DBHM that doesn\u27t reference labeled data, is noise tolerant, and can detect damage across the length of a structure at moderately fast speeds. This study presents a novel Bayesian estimation technique that leverages spike and slab prior specifications on an embedded simplified vehicle-bridge FEM to perform multi-level damage classification without referencing baseline or labeled data. A novel damage-mapping methodology is also proposed to relate crack ratios identified on the embedded FEM to representative levels of cracking on a higher-fidelity bridge FEM. Through this approach, simplified and computationally efficient vehicle-bridge models can be employed for damage classification. The feasibility of the damage classification and mapping strategy is evaluated through analytical studies for a variety of damage states and operating conditions. Specifically, the classification and mapping of a 0.05 crack ratio is studied across different locations while considering varying levels of noise, vehicle velocities, number of experimental vehicle passes, and model errors. The success of the overall methodology, even in the presence of noise, indicates the DBHM approach will likely be successful handling physical data. In particular, the feasibility studies demonstrate that the DBHM methodology is capable of leveraging noisy experimental data to reliably detect, locate, and quantify small levels of crack damage across the length of a bridge while the vehicle is traveling at velocities as high as 20.11 m/s. The third study focuses on addressing limitations within the Bayesian estimation technique associated with model uncertainties and the impact they can have when updating embedded vehicle-bridge FEMs. The study addresses limitations with the damage classification methodology by: discussing factors to consider during testing and model development; demonstrating how to leverage uncertainty quantification to identify unknown model parameters and damage; performing a sensitivity analysis to reduce the unknown parameter space; and outlining concepts to consider when employing Bayesian estimation in SHM. Ultimately, the purpose of this study is to provide a generalized framework for concepts and procedures that need to be considered when employing the Bayesian estimation technique for model updating and DBHM. The fourth and final study focuses on the need to reduce uncertainties in simplified vehicle-bridge FEMs through the development of a comprehensive theory based methodology for modeling the combined effects of frozen bridge bearings and bridge thermal properties. During this study, focus is placed on modeling the coupled mechanic of frozen mechanical bearings and linear/ nonlinear temperature gradients; mechanical bearings are chosen, as they are known to be more susceptible to deterioration than other types of bridge bearings. Data from lab-scale and full-scale bridge studies are leveraged to validate the modeling methods. Lastly, a sensitivity analysis and tuning study are conducted to demonstrate how the reliability of the simplified FEM can be improved. This dissertation concludes with a summary of significant contributions to the development of the DBHM methodology and suggestions for future research