Finite Element Model Updating for Identifying Structural Stiffness in Bridge Structures via Computer Vision and Deep Learning

Conventional structural maintenance approa- ches, such as visual inspections and vibration-based system identification, have limitations in consistency, reliability, and sensitivity to environmental factors. Simulation-based damage detection and dynamic response-based finite element (FE) model updating also have inherent constraints in practical applications. To address these challenges, this study proposes a novel methodology for identifying structural stiffness in bridge structures by integrating computer vision and deep learning techniques. The proposed method consists of two phases: 1) extracting displacement influence lines under a moving load using computer vision and 2) identifying stiffness distribution using deep learning. In the first phase, the computer vision technique accurately tracks vehicle positions and structural displacements, ensuring precise measurement of displacement influence lines. In the second phase, a deep neural network (DNN) updates the stiffness distribution of the FE model to reflect the as-is condition. The network is trained using simulation data with various stiffness distributions. A laboratory-scale test was conducted to validate the proposed method, demonstrating that the updated FE model closely matches the actual structure in terms of dynamic characteristics. The proposed method leverages influence lines to highlight the load-displacement relationship, which is directly linked to structural stiffness. By employing a well-trained DNN capable of handling various stiffness conditions, this approach enables the accurate estimation of key structural properties, facilitating effective data-driven structural health monitoring.