A new scientific review developed within the intoDBP project examines how Physics-Informed Neural Networks (PINNs) could support smarter modelling and management of water and wastewater systems.
Published in Water Research, the study, titled “Physics-informed neural networks in water and wastewater systems: a critical review,” explores how artificial intelligence can be combined with fundamental physical laws to improve the modelling of complex water systems.
Unlike purely data-driven artificial intelligence models, PINNs integrate physical equations directly into neural network training. This allows models to respect physical principles while learning from sparse, incomplete or uncertain datasets.
The review analyses applications of PINNs between 2014 and 2024 across several areas, including drinking water distribution networks, wastewater treatment plants, urban drainage systems and water treatment processes.
The authors highlight the particular potential of PINNs for solving inverse problems, such as estimating system parameters from indirect observations. The technology can also support system identification, real-time state estimation, sensor placement and model calibration.
According to the review, existing applications have shown that PINNs can reduce the amount of training data required compared with conventional neural networks while maintaining strong predictive performance under certain conditions.
However, the study also identifies important limitations. PINNs can face challenges related to training stability, scalability, complex network structures and uncertainty quantification, particularly when applied to multi-scale or highly dynamic systems.
Rather than replacing established numerical modelling approaches, the authors position PINNs as complementary tools capable of combining mechanistic understanding with data-driven learning.
Looking ahead, the review identifies promising opportunities for PINNs in digital twins, real-time monitoring and intelligent water infrastructure management.
The research was carried out by Antonino Di Bella, Maziar Raissi, Domenico Santoro and Paolo Roccaro.
By exploring the intersection between physics and artificial intelligence, the study contributes to the development of more robust and data-efficient tools for future water management.
Read the full open-access publication here.
