IoT Framework for Condition Assessment of Railroad Bridges
Advisor: Professor Billie F. Spencer
Abstract
Railroad bridges are vital to the U.S. freight network, supporting more than 40% of the nation’s freight transportation. With over half of these structures exceeding 100 years of service, growing freight demands combined with aging infrastructure present increasing safety concerns. Despite their critical role, in-service data on key structural parameters, such as bridge deflection under revenue traffic, remains limited. The remote locations of many railroad bridges, combined with the irregular and short-duration nature of train crossings, make continuous monitoring particularly challenging. Traditional wired sensing systems are often impractical for long-term deployment under such power and access constraints. Moreover, early identification and assessment of damage in railroad bridges remain non-trivial due to limited instrumentation and the complexity introduced by the linear time-varying nature of the train-bridge structural system.
This research addresses these challenges by developing an Internet of Things (IoT)-enabled framework for monitoring and condition assessment of railroad bridges. The framework integrates low-power wireless smart sensors, edge computing, and vision-based methods to support synchronized multi-modal data collection and analysis. A hardware-software solution is developed to enable high-accuracy, synchronized data acquisition from external digital sensors using wireless smart sensing platforms. Building on this, a deployable wireless vision-based system is designed for remote displacement monitoring under in-service train loading. To address challenges in drone-assisted inspection during campaign monitoring, a strategy is introduced for obtaining multi-point displacement measurements by compensating for drone motion during flight. In addition to data collection, a deep learning-based pipeline is proposed to automate the identification and mapping of visual damage from drone survey videos to numerical bridge models. To leverage the collected multi-modal data for structural assessment, a physics-informed neural network-based model updating approach is developed to combine sensor data, visual inspection results, and prior numerical models for damage identification and condition assessment. The proposed framework is demonstrated and validated through numerical and experimental studies, including simulated damage scenarios on a full-scale railroad truss bridge in Chicago, Illinois. The results highlight the potential of the developed approach to advance data-driven, scalable, and automated monitoring and assessment of railroad bridges.