Reconstructing transient pressures in pipe networks from local observations by using physics-informed neural networks

Area of Science:

• Fluid Dynamics
• Computational Mechanics
• Data Science

Background:

• Reconstructing transient states in complex pipe networks is challenging due to nonlinear dynamics, system uncertainties, and limited data availability.
• Accurate monitoring of pressure and flow variations is crucial for infrastructure safety and failure analysis.

Purpose of the Study:

• To develop a novel method for reconstructing transient states in pipe networks using Physics-Informed Neural Networks (PINN) with limited sensor data.
• To integrate the PINN framework with an efficient elastic water column (EWC) model for diverse pipe network configurations.

Main Methods:

• A novel approach combining Physics-Informed Neural Networks (PINN) with an efficient elastic water column (EWC) model.
• Integration of the PINN framework with the EWC model to handle complex pipe network topologies.
• Utilizing limited and potentially noisy sensor data for state reconstruction.

Main Results:

• The proposed PINN method accurately reconstructs pressure and flow variations at unmonitored locations.
• The method demonstrates robustness even with noisy data and limited sensor availability.
• Successfully captures extreme values significant for pipe infrastructure integrity.

Conclusions:

• The PINN-EWC approach offers a promising solution for reconstructing transient states in pipe networks.
• This method enhances pipe failure analysis and safety management by accurately predicting critical parameters.
• Laboratory validation confirms the efficacy and reliability for real-world applications.