Physics-Informed Neural Networks (PINNs) are an emerging approach in structural health monitoring (SHM), combining machine learning with the principles of physics to provide accurate and efficient models for assessing structural integrity. Unlike traditional data-driven neural networks, PINNs integrate governing physical laws, such as partial differential equations (PDEs), directly into the training process. This allows the network to make predictions that are not only guided by the available data but also constrained by the underlying physics of the system. For SHM, PINNs can be used to model complex structural behaviors, detect anomalies, and predict damage with reduced reliance on extensive sensor data. By embedding physics into the neural network, they provide a robust framework for real-time monitoring of infrastructure, improving prediction accuracy and generalization, while ensuring that the solutions remain physically meaningful. The advantages of using Physics-Informed Neural Networks (PINNs) for Structural Health Monitoring (SHM) are significant. First, PINNs incorporate the fundamental physical laws governing structural behavior, directly into the learning process. This reduces the need for large amounts of training data, as the model is guided by known physics rather than solely relying on empirical data. Second, PINNs provide more accurate and reliable predictions, as they ensure that the solutions adhere to real-world physical constraints, minimizing the risk of unrealistic or non-physical results. Additionally, PINNs can generalize well to different structural conditions and scenarios, making them more versatile across various monitoring applications. Another key advantage is that they can efficiently handle noisy or incomplete data, making them ideal for real-time SHM, where sensor data may be limited or noisy. Overall, PINNs improve the robustness, efficiency, and accuracy of SHM systems, enabling more proactive and precise maintenance of critical infrastructure. This project aims to employ PIINS for drive by bridge inspection.
Civil and Environmental Engineering
Structural Health Monitoring
- Research environment
- Expected outcomes
- Supervisory team
- Reference material/links
This research will be conducted in the School of Civil and Environmental Engineering at UNSW. The student will begin with an extensive literature review, followed by extensive numerical investigations.
The expected outcomes of using Physics-Informed Neural Networks (PINNs) in Structural Health Monitoring (SHM) are substantial improvements in the accuracy, efficiency, and robustness of structural assessments. By embedding physical laws directly into the neural network, PINNs can provide more reliable predictions even with limited or noisy sensor data. This leads to more precise identification of structural anomalies in real-time, enhancing the safety and longevity of infrastructure such as bridges, buildings, and aircraft.