Numerical study of crude oil batch mixing in a long channel
 291 Downloads
Abstract
The main objective of this work is to predict the mixing of two different miscible oils in a very long channel. The background to this problem relates to the mixing of heavy and light oil in a pipeline. As a first step, a 2D channel with an aspect ratio of 250 is considered. The batchmixing of two miscible crude oils with different viscosities and densities is modeled using an unsteady laminar model and unsteady RANS model available in the commercial CFD solver ANSYSFluent. For a comparison, a LES model was used for a 3D version of the 2D channel. The distinguishing feature of this work is the Lagrangian coordinate system utilized to set noslip wall boundary conditions. The global CFD model has been validated against classical analytical solutions. Excellent agreement has been achieved. Simulations were carried out for a Reynolds number of 6300 (calculated using light oil properties) and a Schmidt number of \(~10^4\). The results show that, in contrast to the unsteady RANS model, the LES and unsteady laminar models produce comparable mixing dynamics for two oils in the channel. Analysis of simulations also shows that, for a channel length of 100 m and a height of 0.4 m, the complete mixing of two oils across the channel has not been achieved. We showed that the mixing zone consists of the three different mixing subzones, which have been identified using the averaged mass fraction of the heavy oil along the flow direction. The first subzone corresponds to the main front propagation area with a length of several heights of the channel. The second and third subzones are characterized by socalled shearflowdriven mixing due to the Kelvin–Helmholtz vortices occurring between oils in the axial direction. It was observed that the third subzone has a steeper mass fraction gradient of the heavy oil in the axial direction in comparison with the second subzone, which corresponds to the flowaveraged mass fraction of 0.5 for the heavy oil.
Keywords
Mixing Turbulence LES RANSList of symbols
 C
Mass fraction
 D
Diffusion coefficient
 H
Channel height
 \(\vec {g}\)
Gravitational acceleration
 L
Channel length
 Re
Reynolds number
 Sc
Schmidt number
 At
Atwood number
 t
Time
 \(\vec {u}\)
Velocity vector in unsteady laminar model
 \(\vec {v}\)
Velocity vector in LES model
 \(\vec {U}\)
Velocity vector in URANS model
 \(V_{\mathrm{f}}\)
Front velocity
 \(U_0\)
Inlet velocity
 \(S_{ij}\)
Mean rateofstrain tensor
Greek symbols
 \(\kappa\)
Turbulent kinetic energy
 \(\epsilon\)
Turbulent dissipation rate
 \(\mu\)
Molecular viscosity
 \(\mu _{\mathrm{t}}\)
Turbulent viscosity
 \(\nu\)
Kinematic viscosity
 \(\rho\)
Density
Subscripts
 t
Turbulent
 HO
Heavy oil
 LO
Light oil
1 Introduction
The mixing of twoliquid miscible flows has attracted a great deal of attention due to its relevance to practical applications, e.g., mixing liquids using centerline injectors (Cao et al. 2003), improving pipe wall fouling mitigation and cleaning (Regner et al. 2007), and the batch transportation of crude oil with different properties (batching) (Ekambara and Joshi 2003). In batching, where crude oil batches with different qualities and properties are transported by the same pipeline, a blended zone is created at the interface of the oil batches. The volume of the blended zone grows with time. It is important to estimate the extent of the mixing and the size of the blended zone to predict the volume of highquality light crude that will be contaminated with the lowerquality heavy crude during transportation. Ultimately, in this process, the operational question is where to cut the batches. To answer this question, we have to understand the different scenarios which can be observed during batching. One possible scenario is a case where both heavy oil (HO) and light oil (LO) flows are turbulent. In cases where a turbulentturbulent configuration exists, a fairly well defined blended zone is created between batches for which, with a good accuracy, it is possible to say that if the mixed zone is cut in half, it contains 50% of each crude oil. Another possible scenario is a case where, due to the high viscosity of the heavy oil, the heavy oil flow is laminar, while the light oil flow is turbulent (laminarturbulent configuration). Complications arise when the turbulent flow follows the laminar flow. In other words, when the light oil batch is transported after the heavy oil batch in the same pipeline, a complicated blended zone is created between the batches in which long tails of the heavy oil stretch into the light oil. In these cases, it is very difficult to estimate the phase mass fraction in the blended zone that is created. This study focuses on such cases.
Over the last few decades, many studies have been published on gravity and pressuredriven twofluid flows. Some works focused on the experimental and/or numerical investigation of miscible gravitydriven flows (Hallez and Magnaudet 2008; Séon et al. 2004, 2007a, b). For instance, Séon et al. (2004) studied the buoyant mixing of two fluids in tubes experimentally, evaluating both the tube’s angle of inclination and the contrast in density between the fluids. The fluids had identical viscosities. It was shown that buoyancydriven mixing in tilted tubes differs significantly from that in the vertical tubes, which was investigated by Debacq et al. (2003). In particular, Séon et al. (2004) demonstrated that Kelvin–Helmoltz instabilities play a crutial role in flows mixing in tilted tubes. Later, Séon et al. (2007a, b) published more detailed studies on buoyancydriven flows in nearly horizontal tubes, in which the dynamics of the mixing front (boundary separating two fluids) were predicted using a CFDbased model. The CFD simulations provided a criterion to distinguish between inertiacontrolled or viscositycontrolled gravitydriven interpenetration flows in tilted tubes with low angles. Moreover, the authors established basic scaling laws determining the characteristic timescales and velocities of the mixing processes.
Hallez and Magnaudet (2008) carried out direct numerical simulations (DNS) to investigate the effect of the channel geometry on the evolution of the concentration and flow fields in the gravitydriven mixing of two miscible fluids in tilted tubes. They observed striking differences between the mixing dynamics in 2D and 3D geometries during the longtime evolution of the flow. In addition, they found three different regimes for the front velocity (depending on the tilt angle), which was in agreement with the results of experimental investigations (Hallez and Magnaudet 2008). However, it should be mentioned that the authors did not adequately take into account how the mixture viscosity depended on the local concentration of each fluid. In their momentum conservation equations, the kinematic viscosity was taken as a constant. Some other researchers have focused on the displacement of miscible fluids due to an imposed flow. Taghavi et al. (2010) experimentally studied how the flow rate affected the stability of a buoyant exchange flow between two miscible fluids of equal viscosity in a long tube. They measured the evolution of the front velocity (\(V_{\mathrm{f}}\)) as a function of the imposed velocity (\(U_0\)). It was found that at low values of inflow velocity, \(U_0\), there was an exchangeflowdominated regime characterized by Kelvin–Helmholtzlike instabilities. When \(U_0\) increased, it was observed that the flow became stable and \(V_{\mathrm{f}}\) increased linearly with \(U_0\). However, at a large \(U_0\), it was concluded that \(V_{\mathrm{f}}\) is almost the same as \(U_0\). In another study, using DNS, Sahu et al. (2009) investigated the effect of buoyancy on the dynamics of pressuredriven flow of two miscible fluids in inclined channels. They examined the effect of the density ratio, Froude number, and channel inclination on the flow dynamics for different Reynolds numbers and viscosity ratios. They showed that the rates of mixing and displacement of the more viscous fluid are enhanced as the density ratio and Froude number increase. Furthermore, these rates are shown to increase when the inclination angle increased and the displaced fluid is the denser one. The most recent research, which is closely related to the case of interest in our study, has been conducted by Taghavi and Frigaard (2013) and Taghavi et al. (2012). They presented a numerical framework for estimating the degree of mixing between successive miscible fluids pumped along a nearhorizontal pipe. However, the aspect ratio of their geometry was 100 and, to reduce computational cost, they used a tube with a diameter of 19 mm. It should also be noted that in their study, to estimate the length of the blended zone, only turbulentturbulent and laminarlaminar configurations were examined.
2 Model formulation
Next, we describe the problem setup and CFDbased models used in our calculations. To better understand the mixing of two oils with different densities and viscosities, we consider a very long 2D channel. The length of the channel is 100 m and its height is 0.4 m, which provides an aspect ratio of 250. An illustration of the computational domain is shown in Fig. 1. The heavy oil occupies the first 40 m of the channel and light oil fills the rest (60 m). The heavy oil (HO) has a density of \(924 \, {\mathrm{kg}}/{\mathrm{m}}^3\) and viscosity of 0.29 \({\mathrm{Pa\,s}}\), while the light oil (LO) has a density of 881 \({\mathrm{kg}}/{\mathrm{m}}^3\) and viscosity of 0.028 \({\mathrm{Pa\,s}}\). According to Fadaei et al. (2013), a mixture mass diffusivity of \(10^{9}\) \({\mathrm{m}}^2/{\mathrm{s}}\) is used, corresponding to the Schmidt number of \(~10^4\) (calculated using the light oil properties). In this work, we consider an imposed flow with a flowaveraged velocity of 0.5 m/s corresponding to Reynolds number values of \(Re=6300\) and \(Re=637\), which are calculated using the light oil and heavy oil properties, respectively. Thus, it can be seen that in this case we may have a combination of turbulent and laminar flows depending on the mixing rate.

Mixing due to molecular diffusion (where \(D_{\mathrm{m}}\) is the diffusion coefficient) since the oils are miscible.

Mixing due to turbulent diffusion caused by the chaotic turbulent motion.

Buoyant mixing due to the density difference between the fluids, which is induced by Rayleigh–Taylor (RT) instabilities.

Shearflowdriven mixing due to a velocity difference at the heavy oil–light oil interface, which is induced by Kelvin–Helmholtz (KH) instabilities.
2.1 Unsteady laminar model
It should be noted that in the commercial CFD solver ANSYSFluent used in this work, it is possible to select a transient solver along with the ‘laminar’ model as the viscous model. The term ‘laminar’ here implies that the Navier–Stokes equations are solved directly without any turbulence modeling. We term these selections as the ‘unsteady laminar’ model. This unsteady laminar model applied to our 2D mesh is similar to a 2D direct numerical simulation (2D DNS) in that the Navier–Stokes equations are solved along with the continuity and mass balance equations directly. However, our unsteady laminar model is different from a true 2D DNS in that it does not resolve all spatial and temporal scales. At the same time, we expect that if the mixing in the system is dominated by largerscale motions, the result from the unsteady laminar model will be substantially the same as that obtained by 2D DNS. A griddependence study will be used to determine the significance of the coarser grid on the result. We note, however, that even a 2D DNS tends to underestimate mixing because it cannot capture the vortexstretching phenomenon that contributes to the energy cascade (Ameen and Abraham 2016). In terms of the abovementioned mechanisms, we expect that the mixing dynamic predicted by the unsteady laminar model will capture molecular diffusion and K–H and R–T instabilities, but will slightly underpredict the role of mixing due to turbulent diffusion. At this stage, we intend only to compare our selected models to the behavior expected in a pipelinemixing scenario, to interpret the results in terms of realism and dominant mechanisms and to estimate the mixing zone size. Therefore, the unsteady laminar model is expected to be good candidate to satisfy these purposes. By choosing the unsteady laminar model, we benefit from not having to neglect or model numerical terms while at the same time reducing the computational cost for a given grid resolution.
2.2 URANS model
The vast majority of turbulent flow computations for industrial flows are based on the RANS method due to its simplicity and low computational cost. The RANS method functions by timeaveraging the Navier–Stokes equations and is capable of obtaining turbulent timeaverage properties (Versteeg and Malallasekera 2007). It is also possible to implement an unsteady RANS (URANS) approach that attempts to separate larger and smaller timescale motions by considering shorter timeaverages of the flow properties and marching the solution through time from previous time steps. In this work, our URANS approach uses \(\kappa \varepsilon\) model as the turbulence model. Contrary to the unsteady laminar approach, in this approach the spatial and temporal turbulence scales are modeled as functions of the smalltimeaverage mean flow properties. To accomplish this modeling, two extra equations for \(\kappa\) and \(\varepsilon\) are introduced to predict the Reynolds stresses and the scalar transport terms. Therefore, this approach has a significantly lower computational cost compared to the unsteady laminar approach. However, there are also several drawbacks. For instance, the underlying assumption in the \(\kappa \varepsilon\) model is that the turbulent viscosity is isotropic, which means that the ratio between Reynolds stress and mean rate of deformation is the same in all directions. This assumption fails in many complex flows. The \(\kappa \varepsilon\) model also assumes highReynoldsnumber flow that is well developed in time. We expect the effect of these assumptions on the current flow configuration to be an increase in smallscale mixing and an overestimated diffusion (molecular and turbulent) at the fluid interface. In other words, in URANS the interface will spread so that the sharp interface between phases will be less distinct. Though in principle, the KH and RT instabilities can be predicted by the URANS model, since these instabilities arise from sharp phase interfaces, we expect them to be damped significantly. The role of turbulent dispersion, however, is expected to dominate.
2.3 LES model
As noted already, turbulence is inherently 3D in nature, which means the turbulent flows are characterized by 3D timedependent structures. Therefore, to examine whether the 3D mixing dynamics in our system lead to a significantly different result from that in the 2D system, we use the socalled dynamic Smagorinsky Large Eddy Simulation (LES) model (ANSYSFluent, Inc. 2013) in a 3D version of our 2D channel. The LES approach assumes that momentum, mass and energy are transported mostly by large eddies. Therefore, only large eddies are resolved. Small eddies, however, are not resolved; instead, these small eddies and their effects on the resolved scales are modeled. Since the KH and RT instabilities are generally large scale, it is expected that their role will appear in the mixing dynamics predicted by LES. Resolving only the large eddies allows us to use coarser mesh and larger timestep sizes in LES which reduces the computational cost compared to the case with the 3D unsteady laminar model (pseudoDNS). The dynamic Smagorinsky method differs from the standard Smagorinsky method in that the Smagorinsky constant is not set a priori but is calculated from the local dynamics of the resolved scales of motion (ANSYS, Inc. 2016).
2.4 Boundary conditions and numerics

In contrast to a classical approach, the overall flow rate in ‘movingwall’ model is zero

Noslip end walls, see Fig. 3a, start to affect the instantaneous dynamics of the mixing (caused by noslip conditions on the vertical walls) when the mixingfront approaches end walls.
Commercial software (ANSYS, Inc. 2016) was employed to solve the problem under consideration. In detail, the governing equations for each model were solved using an implicit finitevolume technique. For pressure velocity coupling, the SIMPLE algorithm was used (Patankar 1980). The convective terms in momentum conservation and species equations for Laminar and URANS models were discretized by means of the QUICK scheme. A secondorder upwind scheme was used in the \(\kappa\) and \(\varepsilon\) equations to discretize the convective terms. A firstorder implicit scheme was activated for transient terms. For equations in the LES model, the socalled bounded central differencing scheme was used for momentum conservation equations and the MUSCL scheme for the species conservation equation. A secondorder implicit scheme was chosen for the time derivative terms in the LES model. The time step in all models was set to \(\Delta t=5\times10^{3}\) s and the maximum number of iterations per time step was 40.
Mesh sizes used in grid study
Name  Mesh size (H \(\times\) L)  Number of CV 

Mesh25  \(25 \times 6250\)  \(0.156 \times 10^6\) 
Mesh50  \(50 \times 12{,}500\)  \(0.625 \times 10^6\) 
Mesh75  \(75 \times 18{,}750\)  \(1.41 \times 10^6\) 
3 Results
Before we proceed with a description of the simulation results, let us recall the main phenomenology on buoyant miscible displacement flows in nearhorizontal 2D ducts. In numerous publications, e.g., see Alba et al. (2014), Hallez and Magnaudet (2008), Taghavi et al. (2010, 2012), it was shown that after ‘an interface is opened’ separating two fluids perpendicular to the channel axis, a heavier fluid displaces a lighter fluid downwards and the front between two fluids accelerates. During the socalled adjustment period of time, the flow is controlled by inertia (Séon et al. 2007b) and later by viscous effects. This effect is observed in almost all displacement flows. As time progresses, Kelvin–Helmholtz (K–H) instabilities appear along the interface between the two fluids, enabling strong transverse mixing. This effect strongly depends on the density difference between the fluids, the fluid viscosities and the channel Reynolds number. It should be noted that gravity currents resulting from the release of a heavy fluid into a light fluid have similar flow scenarios. The main difference comes from the impact of the imposed flow, characterized by the Reynolds number.
Figure 11 compares all three models using axial profiles of the averaged mass fraction of heavy oil calculated at \(t=100\) s. It can be seen that the axial profile of \(\bar{C}_{\mathrm{HO}}\) predicted using URANS has a shortest mixing zone length, where \(0< \bar{C}_{\mathrm{HO}}<1\), in comparison with LES and laminar models. The mixing zone length calculated using the unsteady laminar model has the maximum length, which includes three different subzones with different axial gradients of \(\bar{C}_{\mathrm{HO}}\). The differences in the obtained mixing lengths can be understood by the discussions from the expectations for each model given in Sect. 2. We can clearly eliminate observe that URANS is unsuitable for this problem because the mixing is much more complete than we would expect for this problem, and indeed the high rate of mixing is anticipated based on the assumptions in the \(\kappa \varepsilon\) model. Differentiating between the LES and unsteady laminar simulations to determine the best option is more difficult. However, the LES appears to overestimate mixing on the side of the heavy oil when compared to the results of Sahu et al. (2009), which points to the unsteady laminar case being the most appropriate. Since this case also has a much smaller computational cost than the LES, it appears to be the best model for future work. On the side of the light oil, the LES and unsteady laminar concentration profiles are nearly identical, so it appears that the challenge is to determine the best way to resolve central finger proceeding into the heavy oil. The unsteady laminar approach appears to be the most effective to carry out this work.
4 Conclusions
In this work, we used a simple noinflow model of oil mixing with moving boundaries, to model the batchmixing of two miscible crude oils with different viscosities and densities in a long (100 m) 2D channel. The mixing behavior of the light and heavy crude oils was investigated numerically at a Reynolds number of 6300 and Schmidt number of \(10^4\). Simulations were conducted using an unsteady laminar model, URANS and LES. The results of all models showed that even a 100mlong channel with a height of 0.4 m was not sufficient to determine a steadystate mixing zone length. The URANS model predicted the shortest mixing zone at \(t=100\) s, while the laminar model showed the longest mixing zone corresponding to the channel length where \(0< \bar{C}_{\mathrm{HO}} < 1\).
Using the averaged mass fraction of the heavy oil along the channel, we identified three subzones in the mixing zone. For the first subzone, in which mixing front propagation occurs, there is a very steep mass fraction gradient. In both the second and third subzones, there is passive mixing due to the Kelvin–Helmholtz instabilities. However, in comparison with the third subzone, the second subzone has a lower mass fraction gradient and occupies a smaller fraction of the mixing zone length, which is an indication of higher mixing in that subzone. Therefore, it can be said that most of the mixing occurs in the second subzone. The third subzone is the longest of the three subzones, which is due to the presence of long tails of the heavy oil in the light oil.
It can be concluded that the best and most efficient approach to tackle the twofluid mixing problem is using the unsteady laminar model. Although it is a 2D simulation and contrary to a 2D DNS it filters some of the spatial and temporal scales, this approach gives us enough information about mixing dynamics in the system. The model captures the mixing mechanisms of molecular diffusion, KH and RT instabilities very well. These are the most important mechanisms observed in a laminarturbulent configuration of the twofluid mixing. At the same time, the unsteady laminar model takes up much less computational capacity compared to 3D LES and even 2D DNS. In addition, considering the industrial pipe flow observations explained in the introduction section, we found that the URANS is not a suitable tool to model the crude oil flow mixing as it noticeably does not capture the main mixing mechanisms affecting the mixing dynamics. On the other hand, the unsteady laminar model predictions conform with what is observed in batching.
However, we should point out that using the movingwall approach, even with a 100mlong channel, was not enough to determine the mixing length. Also, at higher Re, the spatial and temporal scales become smaller. Therefore, if we want to study the mixing dynamics at higher Re, finer mesh sizes will be needed to capture these scales. To extend this study, we can conduct simulations in longer 2D channels with finer grid size using the unsteady laminar model. In future work, it will be necessary to conduct the simulations in a 3D pipe geometry to fully quantify 3D mixing effects. For that work, LES will be required because it is able to capture the important mixing mechanisms and its computational cost is significantly less than 3D unsteady laminar or 3D DNS.
Notes
Acknowledgements
This research was conducted with the support of the NSERC Industrial Research Chair in Pipeline Transport Processes (RSS). The contributions of Canada’s Natural Sciences and Engineering Research Council (NSERC) and the Industrial Sponsors (Canadian Natural Resources Limited, CNOOCNexen Inc., Saskatchewan Research Council’s Pipe Flow Technology Centre\(^{\mathrm{TM}}\), Shell Canada Energy, Suncor Energy, Syncrude Canada Ltd., Total SA, Teck Resources Ltd., and Paterson and Cooke Consulting Engineering Ltd.) are recognized with gratitude.
References
 Alba K, Taghavi SM, Frigaard IA. Miscible heavylight displacement flows in an inclined twodimensional channel: a numerical approach. Phys. Fluids. 2014;26:122104. https://doi.org/10.1063/1.4903822.CrossRefGoogle Scholar
 Ameen MM, Abraham J. Are ‘2D DNS’ results of turbulent fuel/air mixing layers useful for assessing subgridscale models? Numer. Heat Transf. Part A. 2016;69(1):1–13. https://doi.org/10.1080/10407782.2015.1052312.CrossRefGoogle Scholar
 ANSYSFluent. Inc. ANSYSFLUENT Theory Guide, Release 14.0; 2013Google Scholar
 ANSYS, Inc. ANSYSFLUENT™ V 16.2—commercially available CFD software package based on the Finite Volume method. Southpointe, 275 Technology Drive, Canonsburg, PA 15317, USA, www.ansys.com; 2016.
 Cao Q, Ventresca AL, Sreenivas KR, Prasad AK. Instability due to viscosity stratification downstream of a centerline injector. Can. J. Chem. Eng. 2003;81(5):913–22. https://doi.org/10.1002/cjce.5450810501.CrossRefGoogle Scholar
 Debacq M, Fanguet V, Hulin JP, Salin D, Perrin B, Hinch EJ. Buoyant mixing of miscible fluids of varying viscosities in vertical tubes. Phys. Fluids. 2003;15:3846. https://doi.org/10.1063/1.1624838.CrossRefGoogle Scholar
 Ekambara K, Joshi JB. Axial mixing in pipe flows: turbulent and transition regions. Chem. Eng. Sci. 2003;58(12):2715–24. https://doi.org/10.1016/S00092509(03)001027.CrossRefGoogle Scholar
 Fadaei H, Shaw JM, Sinton D. Bitumentoluene mutual diffusion coefficients using microfluidics. Energy Fuels. 2013;27(4):2042–8. https://doi.org/10.1021/ef400027t.CrossRefGoogle Scholar
 Hallez Y, Magnaudet J. Effects of channel geometry on buoyancydriven mixing. Phys. Fluids. 2008;20(5):053306. https://doi.org/10.1063/1.2918379.CrossRefGoogle Scholar
 Patankar S. Numerical Heat Transfer Fluid Flow. Washington: Hemisphere Publishing Corp; 1980.Google Scholar
 Regner M, Henningsson M, Wiklund J, Östergren K, Trägårdh Ch. Predicting the displacement of yoghurt by water in a pipe using CFD. Chem. Eng. Technol. 2007;30(7):844–53. https://doi.org/10.1002/ceat.200600357.CrossRefGoogle Scholar
 Sahu K, Ding H, Valluri P, Matar O. Pressuredriven miscible twofluid channel flow with density gradients. Phys. Fluids. 2009;21(4):043603. https://doi.org/10.1063/1.3122779.CrossRefGoogle Scholar
 Séon T, Hulin JP, Salin D. Buoyant mixing of miscible fluids in tilted tubes. Phys. Fluids. 2004;16:L103. https://doi.org/10.1063/1.1808771.CrossRefGoogle Scholar
 Séon T, Znaien J, Perrin B, Hinch EJ, Salin D, Hulin JP. Front dynamics and macroscopic diffusion in buoyant mixing in a tilted tube. Phys. Fluids. 2007a;19:125105. https://doi.org/10.1063/1.2821733.CrossRefGoogle Scholar
 Séon T, Znaien J, Salin D, Hulin JP, Hinch EJ, Perrin B. Transient buoyancydriven front dynamics in nearly horizontal tubes. Phys. Fluids. 2007b;19(12):123603. https://doi.org/10.1063/1.2813581.CrossRefGoogle Scholar
 Taghavi SM, Frigaard IA. Estimation of mixing volumes in buoyant miscible displacement flows along nearhorizontal pipes. Can. J. Chem. Eng. 2013;91(3):399–412. https://doi.org/10.1002/cjce.21754.CrossRefGoogle Scholar
 Taghavi SM, Seon T, Martinez DM, Frigaard IA. Influence of an imposed flow on the stability of a gravity current in a near horizontal duct. Phys. Fluids. 2010;22(3):031702. https://doi.org/10.1063/1.3326074.CrossRefGoogle Scholar
 Taghavi SM, Alba K, Frigaard IA. Buoyant miscible displacement flows at moderate viscosity ratios and low atwood numbers in nearhorizontal ducts. Chem. Eng. Sci. 2012;69(1):404–18. https://doi.org/10.1016/j.ces.2011.10.065.CrossRefGoogle Scholar
 Taylor G. The dispersion of matter in turbulent flow through a pipe. Proc. R. Soc. Lond. Ser. A. Math. Phys. Sci. 1954;223(1155):446–68.Google Scholar
 Versteeg HK, Malallasekera W. An Introduction to Computational Fluid Dynamics. The Finite Volume Method. 2nd ed. London: Pearson Education Limited; 2007.Google Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.