Control valves are critical in process industries, yet accurately predicting multiphase flow behaviour is challenging due to nonlinear interactions between flow dynamics, valve geometry, and operating conditions. This study develops a physics-informed and data-driven framework to predict pressure gradients and energy losses in a multiphase air–water control valve system. Computational Fluid Dynamics (CFD) simulations of single-phase and two-phase flows were performed across a range of valve-opening positions (VOPs), flow velocities, and air volume fractions. Results were validated against experimental measurements, showing strong agreement. A physics-based loss coefficient correlation was derived to represent pressure gradient behaviour, achieving high fidelity with CFD data. Simultaneously, a Deep Neural Network (DNN) model was trained using CFD-generated datasets to capture nonlinear dependencies between VOP, Reynolds number, L/D ratio, and air fraction, attaining R² = 0.997. Additionally, a Response Surface Methodology (RSM) quadratic model quantified interaction effects among six key parameters, achieving R² = 0.94 and identifying optimal operating conditions. Comparative analysis demonstrates that the Buckingham Pi-based correlation provides physical interpretability, while the DNN model offers superior flexibility for nonlinear behaviours. The integrated framework reduces computational cost relative to full CFD simulations while maintaining high predictive accuracy. The results highlight the feasibility of combining physics-based correlations with machine learning to develop scalable digital twin models for real-time monitoring and optimisation of multiphase flow systems in industrial applications.
| Date of Award | 27 Aug 2026 |
|---|
| Original language | English |
|---|
| Sponsors | Petroleum Technology Development Fund |
|---|
| Supervisor | Rakesh Mishra (Main Supervisor) & Shamaila Iram (Co-Supervisor) |
|---|