Pressure field estimation of single-phase flows across a tube bundle using physics-informed neural networks

This paper presents a methodology for estimating the pressure field from experimental velocity fields using physics-informed neural networks (PINNs). The proposed model estimates pressures directly from raw velocity data, without requiring boundary conditions or regular meshes. Manufactured solutions for a lid-driven cavity problem were employed to evaluate the model's performance, demonstrating reasonable accuracy and minimal dependence on data refinement. Time-averaged vector velocity fields of single-phase flows across a tube bundle, obtained through spatial filter velocimetry, were applied to predict the pressure field from real experimental data. The results confirmed that convective effects dominate the momentum transport equations in tube bundles, and the effects of flow separations became more pronounced as the Reynolds number increases. While inherent uncertainties in the experimental data led to higher deviations compared to the manufactured solutions, data refinement and geometry had a minimal impact on the model's convergence. The proposed method highlights the high versatility of PINNs in pressure field estimation compared to traditional numerical methods, enhancing the extraction of information from experimental data and narrowing the gap between simulations and experiments.