Evaluation of drift-velocity closure relationships for highly viscous liquid-air slug flow in horizontal pipes through 3D CFD modelling

Abstract In the past 50 years, several correlations have been developed in order to predict the drift velocity of a liquid-air slug-flow system inside horizontal pipelines. However, a great portion of these correlations is based on either very limited experimental data in terms of fluid properties and operating conditions or on often invalid assumptions. Considering this, the present study assesses the accuracy and validity of seven of the most common non-complex drift velocity correlations through the development of a 3D-Computational Fluid Dynamics (CFD) model using the commercial software Star-CCM+. For this assessment, 13 experimental measurements from the literature were used to validate the CFD model, and 11 case-studies were proposed in order to expand the range of analysis to high-viscosity fluids (μ>0.3 Pa·s) often found in industrial settings. The two-phase system was modelled with an Eulerian-Eulerian approach, coupled with the Volume of Fluid (VOF) method. Three VOF parameters were calibrated in order to obtain the best configurations for different viscosity ranges. The results indicated that the CFD model offers an excellent prediction of the drift velocity given than the average absolute relative error obtained was 6%, even considering highly viscous fluids (μ ∼6.86 Pa·s). The model allowed to confirm that the slug units and gas bubbles do not behave symmetrically in the axial direction and therefore, a 3-D approach substantially improves the accuracy of the model when compared with some 2-D models developed previously. The correlations that showed the highest predicting capabilities were Jeyachandra’s 2012 correlation and Livinus’ 2018 correlation. These two studies considered the combined effect of the Viscosity and Eotvos Numbers, indicating that both parameters significantly influence the drift velocity. In addition, a new correlation was developed by the fitting of the CFD results for all the case-studies proposed. This new correlation slightly improves the prediction achieved by other authors as the overall average error obtained (18%) is 4% lower than the one found for the correlations of Livinus (2018) and Jeyachandra (2012) and around 11% and 6% lower for the high-viscosity range (Nvis > 0.1) when compared to those two correlations, respectively.