Physics-informed neural networks for predicting sediment transport in pressurized pipe flows

Abstract

This study presents the development of a Physics-Informed Neural Network (PINN) for predicting sediment transport rates, integrating physical laws governing sediment transport dynamics to improve prediction accuracy. The model was evaluated against traditional machine learning models, including Random Forest and Support Vector Regression (SVR), as well as empirical formulas, demonstrating superior performance with an average \(R^2\) of 0.9573 and low error metrics. SHapley Additive exPlanations (SHAP) analysis revealed that dimensionless bed shear stress (\(\eta _b\)) and relative grain size (Z) were the most significant contributors to model predictions. A Graphical User Interface (GUI) was also developed to facilitate real-time interaction with the model, making advanced predictions accessible to hydrological engineers. The study underscores the potential of combining machine learning with physics-based constraints to enhance the predictive capabilities of sediment transport models, offering a practical tool for environmental management.