Journal of Bridge Engineering

Active-learning Kriging (AK) was developed as a surrogate-aided reliability technique to address the need for efficient reliability estimation when assessing complex limit states. The results of AK analyses are sensitive to the choice of the regression function, correlation function, learning function and associated stopping criteria, and reliability estimation technique, with unique sets of these input parameters referred to as AK configurations. For the reliable use of AK analysis in bridge reliability assessment, recommendations regarding the best-performing AK configurations are needed to balance the desired accuracy-to-efficiency of the simulation. The objective of this study was to recommend sets of AK configurations for the reliability analysis of reinforced-concrete bridge girders and piers that can be readily used by engineers to perform AK analysis for bridge design optimization and assessment. An extensive parametric analysis, using 432 unique AK configurations and over 3,000 AK analyses, was performed, combined with the application of a comprehensive metric system to recommend the top five best-performing AK configurations for bridge analysis based on the root mean square error, the absolute average error, the degree of consistency, and total number of training points.

Establishing an accurate finite-element (FE) model that faithfully replicates the nonlinear behavior of steel–concrete composite girder bridges is crucially important for effective model-based structural health monitoring (SHM) for this specific bridge type. While Bayesian model updating methodology has gained widespread acclaim for its application in data-driven FE models, apprehensions have emerged regarding the rationality of the updated model parameter values. These concerns predominantly stem from the use of oversimplified FE beam models in model updating studies. This study addressed these concerns by introducing nonlinear constitutive models for the steel–concrete composite girder bridge, as well as comprehensive and detailed three-dimensional (3D) FE model updates. Key parameters for nonlinear constitutive models were estimated based on structural responses from a large-scale static loading test. Bayes’ theorem was employed to infer posterior probability density functions (PDFs) for the model parameters. A transitional Markov chain Monte Carlo sampler was used as a computational tool to generate samples for representing the posterior PDFs. A two-step model updating approach designed to achieve a balance between computational efficiency and simulation accuracy was proposed in the context of nonlinear model updating. Initially, deflection and neutral axis height data were used to update the linear segment of the constitutive model. Load–deflection curves were then used to update the nonlinear segment. Following the model updating with deflection and strain data, the nonlinear simulation results showed improved comparability to the measured data, indicating a significant improvement in the model accuracy. Furthermore, the model update effectiveness was cross verified successfully by comparing load–strain curves of concrete and reinforcing bars obtained during the experiment. The updated model showcases its capability for structural performance evaluation and its potential application in the domain of SHM.

This study introduces an innovative approach for identifying surface defects in underwater bridge structures through the fusion of deep learning and three-dimensional point cloud. The method employs a U²-Net neural network enhanced with residual U-blocks to effectively capture defect features across scales and merge multiscale underwater image attributes to produce significant probability images for defect detection. By leveraging three-dimensional digital image correlation techniques, the method reconstructs the bridge pier surfaces’ physical dimensions from point cloud, enabling precise defect contour and size recognition. The fusion of deep learning’s semantic segmentation with the accurate dimensions from point cloud significantly improves defect detection accuracy, achieving pixel accuracies of 0.943 and 0.811 for foreign objects and spalling and exposed rebars, respectively, and an Intersection over Union of 0.733 and 0.411. The method’s millimeter-level precision in point cloud reconstruction further allows for detailed defect dimensioning, enhancing both the accuracy and the quantitative measurement capabilities of underwater bridge inspections, and shows promise for future advanced applications in this field.