PI: Shamim Pakzad
Co-PI(s): N/A
University: Lehigh University
Industry partner: N/A
Fatigue life estimation of bridges is crucial for maintenance, prognosis, rehabilitation, and enhancing the resilience of the highway bridge network. This typically entails studying the number of loading cycles that a bridge is subjected to through the use of rain flow diagrams constructed from strain measurements. Traditionally, this required the deployment of a fixed strain gauge sensor network for continuous monitoring. However, such an approach is limited by (a) the effort and costs associated with the installation and maintenance of the sensor network, and (b) the resolution achieved by the deployed network and scalability to a bridge inventory. The first issue can be addressed by using indirect sensing alternatives such as the use of accelerometers. Recent work from the research team demonstrates the effective use of AI-based tools for predicting strain response and rain flow diagrams from field acceleration response data. However, the scalability of such a paradigm to bridge inventories is still an open problem. In this project, we propose to develop a transfer learning scheme that can extend the AI-based paradigm from individual bridges to extensive bridge inventories. This will involve the development of metrics that define the similarity between bridges and the implementation of domain adaptation for the reuse of trained AI agents for multiple similar bridges for strain response prediction for subsequent fatigue assessment. The efficacy of the proposed framework will be demonstrated through field experiments. The proposed framework will facilitate the development of a novel and holistic tool for bridge maintenance to enhance the resilience of the road transportation infrastructure.