A Physics Informed Neural Network Integrated Digital Twin for Monitoring of the Bridges

Abstract

In recent years the Digital Twin (DT) paradigm has been studied as a futuristic tool for the next generation of infrastructures. Due to the interdisciplinary nature of the design, construction, monitoring, and maintenance of the infrastructures and the cooperation of several stakeholders throughout their lifetime, it is indispensable to introduce a comprehensive platform for the digital representation of infrastructures. Although the DT emphasizes the role of digital modeling and data analysis, there is a gap between physical modeling and data-driven tools. The newly introduced Physics Informed Neural Networks (PINNs) are capable of not only filling this gap but also representing a unified real-time platform for different users from various fields. These algorithms suggest an agile environment for users to introduce different criteria from the design stage to the health monitoring period. The PINN integrates both physical modeling and data analysis in a unique algorithm, helping them interact simultaneously and providing real-time, reliable responses. By means of the PINN, the DT can learn and update the model from various data sources with a unique platform, which plays an essential role in the rapid flow of information and transparency of data-based calculations. The dynamic ambiance of the PINN enables the users to interact with the modeling procedure and track the analysis. In this study, the details of the proposed platform for the integration of the PINNs in the DT are addressed for monitoring the bridges. Extensive numerical studies are provided for various scenarios of sensor equipment, including sensor type, data accuracy, and installation pattern. The performance of the proposed platform is evaluated for predicting subsequent responses to ensure the reliability of the responses in future decision makings.