Digital twins-boosted identification of bridge vehicle loads integrating video and physics

Abstract

Traffic loads are very critical in bridge digital twins for assessing the deterioration state and structural integrity of road bridges. The existing load rating methods are complicated and time-consuming, necessitating more efficient and intelligent approaches to identify and evaluate safe load capacities. This paper presents a digital twins-boosted approach to identify vehicle loads on road bridges by integrating video records and related physic information. The convolutional neural network (CNN) is adapted with a proposed pixel scale factor (PSF) method to track the motion and dimension of vehicles crossing the bridge. Based on the tracked vehicle data, the time-dependent traffic flow is regenerated via traffic simulation models. Due to the correlation in vehicle loads within a road network, the detailed weight of each vehicle in the traffic flow is inferred using related vehicle load models, e.g., the model established from nearby tollgate data in the case study. After a preliminary verification in the laboratory, a field trial test is carried out to validate the proposed approach in identifying the traffic flow. Then, finite element (FE) simulations are integrated into the approach to predict the vehicle-inducted structural response of an urban arch bridge. The prediction shows a satisfying agreement with the measurement by sensors, which validates the proposed approach in identifying traffic loads. Moreover, compared with purely data-driven methods, the proposed approach demands less training effort and provides more details due to the integration of physics. In general, the output not only offers a promising solution for the digital twins of traffic loads at low costs, but also highlights the integration of visual data and physics in solving engineering issues.