| Author: | Luo, Lanxin |
| Title: | System identification of bridges integrating physics-based and data-driven models |
| Advisors: | Xia, Yong (CEE) |
| Degree: | Ph.D. |
| Year: | 2026 |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | xxxii, 248 pages : color illustrations |
| Language: | English |
| Abstract: | Bridge system identification aims to establish a mechanical model and identify characteristic parameters of the structure based on measurement data. Existing methods can be categorized as either physics-based or data-driven. Recent advancements in artificial intelligence have promoted the development of physics-data co-driven approaches, which exploit the complementary strengths of both paradigms. Meanwhile, structural health monitoring systems collect abundant data for bridge system identification, though these multi-source data introduce significant uncertainties. In this context, this thesis attempts to develop advanced techniques for bridge system identification by integrating physics-based and data-driven models. First, a physics-based finite element model updating (FEMU) method using static and dynamic measurement data is developed for structural parameter identification. The theoretical and measured static responses are efficiently obtained by fusing traffic videos, vehicle weight data, and structural responses through computer vision-based load identification and signal processing techniques. A parallel particle swarm optimization algorithm is employed to extract the stiffness and mass parameters in an optimal manner. The method is applied to a three-span bridge model. Next, the hierarchical Bayesian theorem is introduced into FEMU for uncertainty quantification. A hierarchical Bayesian FEMU method using static and dynamic data is then developed. The hyperparameters for both updating parameters and error functions are introduced to provide a comprehensive probabilistic representation. A two-stage sampling strategy, combining maximum a posteriori estimation with Markov Chain Monte Carlo sampling, is proposed to compute the posterior distribution. The method is applied to a three-span bridge model. Later, the influences of temperature and traffic loads on structural characteristics are incorporated into the hierarchical Bayesian FEMU framework. A linear relationship between temperature and stiffness parameters is established to explain the observed correlation between temperature and natural frequencies. The equivalent vehicle loads estimated from the weigh-in-motion data are introduced into the FE model to capture the traffic-induced variations in dynamic characteristics. The method is validated on a long-span arch bridge. The comparative analysis of the three proposed FEMU methods highlights the importance of considering operational loads and their associated uncertainties for FEMU and long-term condition assessment. Subsequently, a physics-based sparse Bayesian learning method for joint load-parameter identification is developed to address unknown loading conditions. The spatial sparsity of both loads and damage is embedded as prior information within the Bayesian framework. The expectation-maximization algorithm and a modified Metropolis-Hastings sampling are used to iteratively estimate damage indices, loads, and precision variables. The method is applied to both numerical and experimental bridge models. Finally, two physics-data co-driven approaches are proposed to enhance model generalizability and flexibility. A physics-encoded neural network is proposed for joint load-parameter identification. The FEMU module for parameter identification and the autoencoder for load identification are seamlessly integrated within the framework of the Newmark-beta method. The gradient information for optimization is automatically computed through backpropagation on the deep learning platform. This approach requires only the forward formulation to be explicitly defined, thereby eliminating the repeated and complex inverse derivations required by conventional methods. The applications of this method to numerical and experimental examples demonstrate that it outperforms a filter-based method in both accuracy and efficiency. Subsequently, a hybrid finite element (FE)-neural network approach is developed for nonlinear boundary condition (NBC) identification without prior knowledge of the nonlinear types. Multilayer perceptrons are used to represent the mechanical behaviors of unknown NBCs and are integrated into the FE model. The stabilized central difference algorithm is adopted as the network architecture. Numerical studies demonstrate that the proposed approach can accurately identify both elastic and hysteretic NBCs, and remains robust under varying noise levels, sensor layouts, and nonlinear characteristics. In summary, this thesis addresses three key issues in bridge system identification: multi-source data fusion, uncertainty quantification, and the integration of physics-based and data-driven models. The proposed techniques effectively enhance the accuracy, robustness, and generalizability of system identification under scenarios with varying prior knowledge and data availability. These developments provide a sound basis for long-term structural health monitoring, condition assessment, and decision-making in bridge maintenance and management. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14508