HYBRID PHYSICS-INFORMED NEURAL SOLVERS FOR REAL-TIME TURBULENCE PREDICTION IN UNSTEADY MULTIPHASE FLOWS ACROSS ADAPTIVE HYDRAULIC INFRASTRUCTURES

Real-time turbulence prediction in unsteady multiphase flows remains a critical challenge for intelligent water management systems, particularly in adaptive hydraulic infrastructures such as dynamic dams, floodgates, and urban drainage systems. Recent advancements in physics-informed neural networks (PINNs) have opened new avenues for integrating physical laws into deep learning models, enhancing predictive accuracy and generalizability in fluid mechanics. This study proposes a hybrid physics-informed neural solver that synergizes data-driven learning with traditional Navier-Stokes-based solvers to enable fast and robust predictions of turbulent multiphase interactions. We test our model in complex flow scenarios involving adaptive hydraulic structures and demonstrate substantial improvements in prediction speed and physical consistency over baseline models. Results affirm the potential of hybrid solvers for deployment in real-time control environments.