Soft-ANN based correlation for air-water two-phase flow pressure drop estimation in a vertical mini-channel

• Published in: Proceedings of the Institution of Mechanical Engineers. Part C, Journal of mechanical engineering science Journal of mechanical engineering science, 2022-02, Vol.236 (3), p.1430-1442
• Main Authors: Manuel, Barroso-Maldonado Juan, Manuel, Riesco-Ávila José, Martín, Picón-Núñez, Manuel, Belman-Flores Juan
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Correlation; Density; Diameters; Dimensionless numbers; Fluid flow; Laminar flow; Pressure drop; Reynolds number; Two phase flow; Viscosity; Water drops
• ISSN: 0954-4062; 2041-2983
• DOI: 10.1177/09544062211020329

In this paper, an Artificial Neural Network soft matrix correlation to estimate the pressure drop of air-water two-phase flow is developed. The applicability of the model is extended by using dimensionless physical numbers as inputs (Air-Reynolds number, Water-Reynolds number, and the ratio of Air Inertial Forces to Water Inertial Forces), so the model can be implemented for vertical pipes with the proper combination of diameter-velocity-density-viscosity allowing estimations of dimensional numbers within the range of: Air-Reynolds numbers (430–6100), Water-Reynolds number (2400–7200), and Air-Water-Inertial forces ratio (1.6–1834), including the diameter range from 3 to 28 mm. Experimental measurements of frictional pressure drop of water-air mixtures are determined at different conditions. A search of the most suitable density, viscosity, and friction models was conducted and used in the model. The performance of the proposed ANN correlation is compared against published expressions showing good approximation to experimental data; results indicate that the most used correlations are within a mean relative error (mre) of 23.9–30.7%, while the proposed ANN has a mre = 0.9%. Two additional features are discussed: i) the applicability and generality of the ANN using untrained data, ii) the applicability in laminar, transitional, and turbulent flow regimen. To take the approach beyond a robust performance mapping, the methodology to translate the ANN into a programmable equation is presented.