Indirect Structural Health Monitoring of Bridges: Mathematical Models and Laboratory Experiments

An indirect approach is presented for determining the condition of model bridges based on data collected from instrumented vehicles that traverse the structure. The underlying idea behind this study is that vehicles traveling across bridges pick up dynamic characteristics of the structure which allow one to identify changes in the bridge condition. By using a fleet of vehicles and using a classification algorithm to discriminate between environmental factors and structural health, it would then become possible to examine efficiently the condition of a large stock of bridges using a smaller stock of vehicles. This research studies the indirect approach by means of mathematical and experimental models.;The feasibility of the indirect structural health monitoring (SHM) approach was first assessed by means of 1D and 3D mathematical models. A multiresolution classification algorithm was employed to detect the existence, severity, and location of changes in the bridge structure. Several classification experiments were run. The detection experiments, that is, classifying among "Damaged" of "Undamaged", obtained on average 95% accuracy for the 1D models and 90% for the 3D models. It was observed a little dependence of the parameters of the moving vehicle. The damaged labels were defined as stiffness reduction of structural members.;The detection capability of the approach was then tested in a laboratory setup for three different types of damage scenarios: changes in the support conditions (rotational restraint), additional damping, and an added mass at the midspan. A set of features based on discriminant frequencies was used in conjunction with a support vector machine classifier on data measured from a passing vehicle at the wheel and suspension levels, and directly from the bridge structure for comparison.;New sets of experiments were performed on the laboratory setup to explore the effects of variations of the vehicle-bridge interaction, the effect of roadway roughness and temperature gradients. The experiments were also augmented by experimenting with a second model bridge, and including vehicle variations and damage at different locations.;Across all the different variations of vehicle mass, vehicle speed, roadway roughness and ambient temperature, it could be concluded that the indirect data obtained from sensors at the wheel level or at the suspension level, in general, performed just as well or better than the data from sensors on the bridge in terms of the classification accuracy obtained with the signal processing and classification methods used.