CFD and machine learning for scour prediction at square bridge piers:South Platte,Mississippi,Choptank,and Pearl riversOA

Local scour around bridge piers remains a leading cause of failure in hydraulic structures,especially for shapes like square piers that promote flow separation and turbulence.This study introduces a new hybrid framework that combines high-quality Computational Fluid Dynamics(CFD)with understandable Machine Learning(ML)methods to ensure accuracy and maintain physical insight.A comprehensive dataset was assembled from 116 field measurements collected by the U.S.Geological Survey(USGS)and 90,000 synthetic samples generated through FLOW-3D simulations.The eXtreme Gradient Boosting(XGBoot)algorithm predicted scour depth,while SHapley Additive exPlanations(SHAP)analysis clarified feature importance and model behavior.The integrated CFD-ML model outperformed both empirical methods like HEC-18 and standalone ML models,achieving an R²of 0.897 and reducing root-mean-square error(RMSE)by approximately 45%.It identified three main physical factors influencing scour development:approach depth,sediment size,and pier width.Validation with field data and numerical simulations demonstrated its strong potential for real-world design and risk assessment.Overall,this accessible,data-driven hybrid approach provides a robust and versatile tool to improve scour-resistant infrastructure,especially for bridge piers with complex shapes.