Operational Health Monitoring and Digital Twinning of Bridges Using Multimodal Data Fusion and Physics-Based Bayesian Model Updating
Loading...
Authors
Malekghaini, Niloofar
Issue Date
2024
Type
Dissertation
Language
en_US
Keywords
Bayesian Inference , Bridges , Health Monitoring
Alternative Title
Abstract
National surveys show that large number of obsolete and structurally deficient bridges are currently in service across United States that need to be replaced or rehabilitated. However, replacement or rehabilitation of such a large number of bridges is a monumental task, possibly requiring decades to complete. To prioritize public safety, it is imperative to assess the structural health of these bridges and offer actionable insights to guide decision-making process for their rehabilitation. For this purpose, there is a need for an operational health monitoring solution for bridges, one that is non-interruptive to traffic, cost-effective to implement, and requires minimal installation workload. This research is focused on developing a novel framework for operational health monitoring of bridges based on digital twinning process. Digital twin is a finite element (FE) model of the bridge that is trained/updated using the measured dynamic responses of the bridge. In the proposed framework, bridge digital twin is developed based on a stochastic filtering approach by combining traditional vibrational measurements of the bridge along with location of the vehicles traveling on the bridge, obtained using computer vision and artificial intelligence methods. Using the digital twin of the bridge, structural damage can be inferred quantitatively from the updated mechanics-based model parameters. Moreover, the digital twin includes information regarding the pattern and dynamic load of traffic, which can be used for load rating, as well as traffic monitoring and management. This research is carried out in three main phases: theoretical development and verification, proof of concept studies, and real-world validation. In the first phase, the theoretical foundations of the Bayesian time-domain FE model updating are developed. The proposed method and the developed formulations are verified using series of case studies based on numerically simulated data from a prestressed reinforced concrete box-girder bridge model. The acceleration responses along with the location of the vehicles on the bridge are used for joint estimation of the model parameters and vehicular loads. The estimated model parameters are then used to infer the location and extent of damage along the bridge. In the second phase, the developed framework is evaluated in an experimental setting with the purpose of understanding and addressing its limitations. The novelty of this phase includes addressing the modeling uncertainties related to boundary conditions and computational cost for estimation of input time histories. In this phase, a pair of full-scale precast prestressed bridge I-girders are used as a testbed structure. A series of forced vibration experiments are performed, and the acceleration responses of girders are utilized to jointly estimate the uncertain model parameters of the testbed structure and the input force excitation. In the last phase, the proposed framework for operational health monitoring of bridges is validated using a decommissioned steel bridge. The bridge is instrumented with networks of traffic cameras and accelerometers to collect the video recordings of traffic and dynamic response of the bridge. Using computer vision techniques, time histories of the footprint of tires are extracted from the video recordings. This data is fused with the collected acceleration responses, and together serve as the input to the time-domain Bayesian model updating process. The estimates of model parameters are used to infer the location and extent of damage in the bridge. The damage localization result shows reasonable agreement with those inferred from Ground Penetrating Radar (GPR) test on the bridge. In Summary, the work presents a systematic technology development effort, from theoretical development to experimental testing, and real-world validation. The novelties of the research include integration of stochastic filtering approaches and computer vison techniques to develop a low-cost and non-interruptive method for health monitoring of bridge structures. Contributing to the advancement of technology solutions for operational monitoring of bridges, this research enhances the management efficiency of the degrading civil infrastructure assets, the security of transportation infrastructure, and the public safety.