A Framework for Vehicle Bridge Strike Detection, Strike Characterization, and Damage Estimation for Railroad Bridges

Rail bridges provide vital crossings for freight and passenger trains over natural and man made obstructions in terrain. Over time these structures develop damage due to aging or unexpected external loading events. Specifically, low clearance rail bridges are susceptible to frequent strikes from overheight vehicles or equipment. It is critical to detect those strikes once they occur to ensure the bridge and the public safety and to also meet FRA regulations of mandatory post-strike bridge inspection. Early bridge damage detection also reduces bridge closure times and prevents further deterioration. Not every bridge strike represents an immediate risk to safe bridge operation, thus this dissertation presents a comprehensive methodology that detects vehicle bridge strikes in real-time, characterizes strike severity post detection, and detects and quantifies damage if present in the bridge. First, the system leverages the improved accessibility and scalability of bridge instrumentation technology and interrogates bridge data using mechanics and Machine Learning (ML) algorithms to rapidly detect strikes and determine whether an immediate inspection is necessary or can be safely deferred. Specifically, this dissertation develops parallel heterogeneous data-fusion convolutional neural networks (PHD-CNN) operating on data collected from in service rail bridges to improve detection and classification of vehicle-bridge strikes. The method provides a mechanism to homogenize and fuse disparate data streams for use as inputs to a classifier that distinguishes bridge strikes from passing trains. Optimum PHD-CNN networks detect, on average, 95% of bridge strikes with false positive rates less than 2%. Next, operating on identified strike data the framework utilizes principal components analysis, an unsupervised machine learning technique, to characterize strike severity. The system analyzes extracted v severity-related features to group strikes with similar characteristics together and then compares them to user defined thresholds to determine strike severity. Finally, for an observed change in the system’s fundamental frequency this dissertation presents an energy-based mechanics relationship to provide a feasible domain of potential damage scenarios to detect, localize, and characterize damage. The final output of the system comprises practical guidance to inspectors by (1) indicating the presence of damage, (2) locating the damage, and (3) quantitatively estimating the severity of the damage; thus, the method attains a Rytter level 3. Rytter levels comprise four stages of damage evaluation: detection, localization, quantification, and prediction of remaining structure life. They are a widely used framework in structural health monitoring to rate the capabilities of damage assessment systems.