Daniel G. Linzell
Date of this Version
Visual inspection is often used to assess the condition of railway bridges at discrete points in time, an approach that can be subjective and possibly unsafe. Alternatively, certain bridges have their condition assessed via the installation of a large number of sensors. These sensors can be costly to place, power and maintain. Therefore, reducing their numbers and maximizing the extracted information is of utmost importance. In addition, evaluating bridge condition from measured response can be quite challenging due to loading and environmental variations, especially when a limited number of sensors are used.
The focus of this research is to develop an automated hybrid experimental-numerical framework to detect and locate damage and estimate its intensity. The framework was developed analytically, based on Proper Orthogonal Modes (POMs) and Artificial Neural Networks (ANNs), and validated experimentally using 1 and 8 weeks of measured strains collected from a monitoring system placed onto an in-service, multi-span, railway bridge. The analytical work involved using three sensor instrumentation sets and investigated structural response for two bridge spans of different type and size. To generate training data for the ANNs, Modeling uncertainties that could lead to erroneous indication or omission of damage are incorporated into framework development via a systematic analyses. The procedure was based on synergizing POMs extracted from measured structural response and POMs calculated from the numerical model with a robust damage feature independent of level and location of modeling uncertainty. A hybrid experimental-numerical approach was developed and implemented to estimate damage scenario POMs from field measured strains. ANNs were trained and tested using these POMs with DL and DI being detected. These results show the promise of the POD-ANN method as a robust, real-time fatigue damage detection tool for steel railway bridges.
Advisor: Daniel G. Linzell