Speaker
Description
Physics-Informed Neural Networks (PINNs) are a powerful machine learning framework that integrates physical laws, typically represented by partial differential equations (PDEs), into neural networks to solve complex forward and inverse problems. Unlike traditional neural networks, which rely solely on data, PINNs incorporate governing equations directly into the network's architecture through the loss function. This enables the model to adhere to the underlying physics of the system while learning from data, reducing the need for large datasets and improving generalization. PINNs are particularly effective in cases where experimental data is sparse or noisy, as they exploit known physical principles to constrain predictions. They have broad applications in fields like fluid dynamics, structural mechanics, electromagnetics, and other areas where modelling complex physical phenomena is essential. By fusing data-driven learning with physics-based constraints, PINNs offer a more robust and interpretable approach to solving real-world problems. This paper utilizes the concept of physics-informed neural networks (PIINs) for the structural integrity assessment of bridge structures. Structural Health Monitoring (SHM) is crucial for ensuring the safety, reliability, and longevity of infrastructure. Bridges are subjected to constant wear from traffic loads, environmental factors, and material degradation, which can lead to structural weaknesses over time. Early detection of issues such as cracks, corrosion, or excessive vibrations through health monitoring systems allows for timely maintenance and repair, preventing catastrophic failures. In particular, in this study, the concept of indirect SHM (ISHM) or drive-by inspection is adopted where structural health of bridges is assessed using sensors and data collection systems mounted on vehicles that pass over the bridge. ISHM addresses many of the limitations inherent in conventional SHM systems, such as access challenges, power supply, and cabling requirements, among others. To this aim, we develop a novel approach to simulating forward and inverse problems involving vehicle-bridge interaction (VBI) using PIINs. Physics-informed neural networks leverage the governing physics of VBI. Numerical results demonstrate that PIINs is an effective method for solving both forward and inverse problems in VBI, as well as for characterizing bridge parameters and detecting localized changes.
