International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

www.elsevier.com/locate/ijhmt

A numerical model for segmented flow in a microreactor

N. Harries a, J.R. Burns b, D.A. Barrow a,*
, C. Ramshaw b

a
Laboratory for Applied Microsystems, Cardiff School of Engineering, Cardiff University, QueenÕs Buildings, P.O. Box 689,
Cardiff CF2 3TF, UK
b
Department of Chemical & Process Engineering, University of Newcastle, Newcastle upon Tyne NE1 7RU, UK
Received 6 August 2002

Abstract
A numerical model for segmented flow in a microreactor has been developed. The model is based on computational
fluid dynamics (CFD) which means that the flow field and mass transfer are described by a set of partial differential
equations. A general purpose CFD code was extended in order to predict the internal flow patterns of fluid segments
and the transfer of dissolved chemical species within segments and across fluid segment interfaces. The model has been
validated by comparing predicted data with experimental microreactor titrations.
Ó 2003 Elsevier Science Ltd. All rights reserved.

Keywords: Numerical model; Segmented flow; Microreactor; Computational fluid dynamics

1. Introduction by Burns and Ramshaw [1,6]. Vortex flow patterns

generated by the shearing motion within fluid segments
Many industrially important chemical processes in- enhances mixing within segments and improves mass
volve multiphase flow where reactions are led by diffu- transfer across the interface.
sion across the contacting fluid phases, for example, Efficiency of segmented flow systems in terms of re-
gas–liquid hydrogenation and immiscible liquid–liquid action time is influenced by a number of factors in-
nitration [1]. Developments in fabrication methods for cluding flow velocity, slug length, channel size, and
microfluidic devices are increasingly providing this area liquid viscosity ratio. Slug length depends upon flow
of chemistry with opportunities to take advantage of velocity, channel geometry, liquid viscosity and surface
the benefits of downscaling [2]. Microreactors based on properties [1]. The complex relationship between design
multiphase flow have emerged in the last few years. parameters for the segmented flow microreactor means
These are mainly gas–liquid systems employing bubbly that optimisation of the system can involve extensive
or segmented flow, an example of which is described by experimental work. A predictive model for the flow field
Hessel et al. [3]. More recently, attention has been paid and mass transfer in a liquid–liquid segmented flow
to immiscible liquid–liquid systems at microscale with a system would provide a design tool offering a faster and
focus on intensification of reactions in these systems by cheaper design process. Such a model would predict the
segmentation of the fluids. Liquid–liquid contacting mass transfer performance for a given segmented flow
devices most commonly use diffusion between parallel system. The model could be used to produce design
streams of the two phases as described by Harper [4] and guidelines, indicating the significance of individual de-
Harston et al. [5]. The use of segmented flow to enhance sign parameters on system efficiency, for microreactors
diffusion in a liquid–liquid microreactor is demonstrated with segmented flow.
Computational fluid dynamics (CFD) can provide
detailed flow field information for fluid systems [7]. The
*
Corresponding author. Tel.: +44-29-2087-5921; fax: +44- method employed by CFD is to describe the flow with
29-2087-4716. the governing conservation equations, which always
E-mail address: barrow@cf.ac.uk (D.A. Barrow). include equations for the conservation of momentum
0017-9310/03/$ - see front matter Ó 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0017-9310(03)00120-0
3314 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

Nomenclature

s seconds l laminar viscosity

t time q fluid density
u momentum in the x-coordinate direction
Other symbols
u velocity field
L slug length (lm)
v momentum in the y-coordinate direction
W channel width (lm)
w momentum in the z-coordinate direction
Sy source term for y-momentum equation
Greek symbols
a1 total mass of acid in water
a2 total mass of acid in kerosene

(Navier–Stokes equations) and mass and may also in- of Coventor, Inc. [11]. Most commercial CFD codes are
clude equations for the conservation of energy, con- user friendly but vary in the add-on models they provide
centration of chemical species and other scalar variables which may include models for particle tracking or
depending on the flow under investigation. This set of chemical reaction. Codes also vary in the flexibility they
partial differential equations provides continuous data offer for user development.
for the flow domain, however, direct solution of this A numerical model for segmented flow in a micro-
equation set is not possible. The flow domain is divided reactor was developed and validated in this study.
into a mesh of volumes and the partial differential Model development involved extending a CFD code to
equations are discretised over the computational mesh predict the flow field and mass transfer for immiscible
yielding a set of algebraic equations which are then liquid–liquid segmented flow. The numerical predictions
solved by an iterative calculation procedure. Solution of for this study were performed with the CFD code CFX,
the algebraic equation set provides velocity, pressure which is a general purpose commercial code offering a
and scalar (such as, temperature and concentration of range of add-on models appropriate for microfluidic
chemical species) data for each mesh volume in the fluid applications and sufficient access to the source code to
domain. This vast array of solution data is processed to enable user software developments.
provide graphical representations of the solution in the Microreactor experiments carried out at Newcastle
form of speed maps, velocity vectors, streamlines and University provided data which was used to validate the
scalar concentration maps [8]. model [1]. Titration reactions were carried out to dem-
Availability of fast processors, computer memory onstrate the mass transfer performance of segmented
and advanced graphics capabilities at reasonable cost flow. Predicted data was compared with experimental
together with commercially available user friendly CFD titration times and with photographic images of seg-
codes has meant that most areas of fluid engineering mented flow titration.
now use CFD as a design or research tool [8]. A number Development and validation of the model are des-
of commercially available CFD codes exist which are cribed in this paper.
appropriate for microfluidic applications. The majority
of these are general purpose codes, such as CFX,
property of AEA Technology plc. [9], and FLUENT, 2. Segmented flow
property of Fluent Inc. [10]. A small number of avail-
able codes are designed specifically for microfluidic ap- Segmented flow is comprised of alternating aqueous
plications with preset options for microfluidic geometries and organic segments (see Fig. 1). The aqueous phases
and flow regimes, for example, CoventorWare, property do not mix but transfer of momentum (effects of vis-

Fig. 1. Photographic image of liquid/liquid segmented flow in a microchannel (kerosene/water).

 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322 3315

Fig. 2. Diagram illustrating the internal vortex flow patterns

within fluid segments (flow direction is left to right).
Fig. 3. Photographic image of fluid segments generated with a
‘‘T’’ piece.
cosity) and mass (dissolved chemicals) occurs across the
fluid interface. As the fluid segments move along the
channel internal vortex flow patterns are generated (see in this study represents established segmented flow.
Fig. 2). These circulation patterns are a result of shear- Modelling the generation of segmented flow is not
ing motion within the low Reynolds number flow com- within the scope of this report.
bined with fluidic interfaces. This fluid motion enhances Glass microchannels of square cross-section and 380
mixing within the fluid segments and also improves the lm wide were employed for the experimental titrations
rate of diffusion of solute across the liquid/liquid inter- and fluid segments were generated using a ‘‘T’’ piece [1].
face by the shearing disruption of concentration gradi-
ents within the segments.
The alternating segments of two liquid phases pro- 3. Computational fluid dynamics model
vide an environment for mass transfer and chemical
reaction. Typically, in segmented flow microreactors, Segmented flow was represented by two adjacent
solute in one phase diffuses into the adjacent segments rectangular units in a microchannel with moving walls
where it reacts with dissolved chemicals in that phase. (see Fig. 4). The fluid segments under observation were
This study is concerned with modelling titration reac- held in the calculation domain while the moving walls
tions in segmented flow, where acetic acid dissolved in created the flow field. The adjacent fluid segments were
the organic phase (kerosene) partitions into the aqueous linked at both ends, much like a strip of paper wrapped
phase and neutralises the dissolved base (KOH/NaOH). around into a continuous loop. In this way, the fluid
Segmented flow may be generated in microchannels segments behaved as though they were in a channel of
by various methods [1]. Commonly, a ‘‘T’’ shaped in- segmented flow experiencing the effects of momentum
tersection is employed where the two liquid phases meet and mass transfer from both ends.
head-on [6]. Segments are generated by each phase cut- Two fluid phases were defined, identified by density
ting the flow of the other as both phases vie to travel and viscosity. Both phases were incompressible and
down the exit channel (see Fig. 3). The model developed Newtonian in their viscous behaviour.

Fig. 4. Diagrammatic representation of modelled fluid segments.

3316 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

A Cartesian coordinate system was employed with a The acid/base reaction in Eq. (3) was implemented
no-slip condition applied to the channel walls. The axi- into the segmented flow model:
symmetric nature of the flow allowed a two-dimensional
KOH þ CH3 COOH ¼ CH3 COO Kþ þ H2 O ð3Þ
slice of the channel to be considered. The two-dimen-
sional calculation permitted a high resolution compu- Acid was initialised in the organic segment to represent
tational mesh to be employed for the axial plane. acetic acid in kerosene. Solute is identified in the CFD
The interface between the fluid segments was as- code by its diffusivity in each phase and is initialised by
sumed to be flat. An adjustment factor was applied to concentration. Assisted by the recirculating flow pat-
titration time to compensate for the reduced surface terns within the fluid segments the acid diffuses into the
area resulting from the flat interface model. Alternative aqueous segment. Concentration maps representing the
modelling options for the fluid interface are discussed in distribution of acetic acid as it diffuses across the or-
Section 6. The fluid interface was defined by setting zero ganic/aqueous are presented in Fig. 6.
axial velocity for the mesh cells which form the interface. The recirculating flow patterns act to refresh the acid
This was achieved by setting the source terms for the concentration at the interface of the organic and inor-
v-momentum equation to zero (Eq. (1)––conservation ganic segments. In the aqueous segment, the recirculat-
equation for momentum in the y-coordinate direction ing flow draws the acid away from the interface along
[12]). the centre line. A true representation of movement of
acetic acid from kerosene to water is partitioning 85:1 in
dðqvÞ dq
þ divðquvÞ ¼ divðl grad vÞ  þ Sy ð1Þ favour of the water phase. The source terms of the scalar
dt dy equation representing dissolved acid were calculated
to represent the partitioning movement of acetic acid
Diffusion of momentum and mass remained active
across the water/kerosene interface. Fig. 7 illustrates
across these cells. For momentum this is a condition of
how the concentration of acid was recalculated for the
equal stress across the interface (Eq. (2)):
mesh cells either side of the fluid interface. In the dia-
l1 du=dx ¼ l2 du=dx ð2Þ gram a1 and a2 represent the total mass of acid in water
and kerosene respectively. The numerical expressions
In addition to the boundary conditions at the interface below the mesh cell diagram indicate how a1 and a2 were
between the fluid segments in the calculation domain, redefined in the source terms of the scalar equation
diffusion of mass and momentum were defined at the representing dissolved acid.
segment ends which formed connections with segments Partitioning now replaces diffusion of acid at both
outside the calculation domain. Wall velocities were fluid interfaces. Concentration maps representing the
graduated to zero at the segment interfaces in order to distribution of acetic acid as it partitions across the or-
maintain continuity of velocity in mesh cells near the ganic/aqueous interface are represented in Fig. 8.
boundaries. This approximate model for the fluid seg- Solute was initialised in the aqueous segment to
ment interface required a relaxation of viscosity ratio of represent KOH in water. KOH remains in the aqueous
the aqueous and organic phases in order to reach a phase. As acetic acid partitions into the aqueous phase
converged solution with practicable computation times. the acid neutralises the base (Eq. (3)). Source terms for
A viscosity ratio of 1:1 was employed for the validation the scalar equations representing the dissolved acid and
exercise. The calculated flow field is presented as a speed base were calculated to represent the titration reaction.
map with overlaid velocity vectors in Fig. 5. The vector Concentration maps representing diminishing base
map shows the vortex flow patterns within the fluid during titration are presented in Fig. 9.
segments. Also, the wall velocity distribution is shown Most of the problem definition for the segmented
by the speed map. flow model was achieved by Fortran coded subroutines.

Fig. 5. Speed map and vectors representing calculated flow field.

 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322 3317

Fig. 6. Concentration maps representing the distribution of acetic acid as it diffuses from the organic to the aqueous phase (flow speed
0.1 mm/s). (a) t ¼ 0:1 s, (b) t ¼ 0:5 s, (c) t ¼ 1:0 s, (d) t ¼ 6:0 s.

times, were employed. Solutions for direct comparison

with experimental titration employed a fine computa-
tional mesh (1–2 lm) to maximise accuracy. Fine mesh
solutions require increased computer memory and
storage, and have longer run times.
Converged flow fields were achieved with 100 time
steps of 0.001 s with 50 iterations of velocity and mass
equations at each time step. Up to this point in the so-
lution the time steps have no physical meaning. The flow
field is now established and the acid and alkali can now
Fig. 7. Diagrammatic representation of mesh cell either side of
be initialised and time reset to zero. With the flow field
fluid interface.
frozen, that is zero iterations of the velocity and mass
equations, the scalar variables were calculated with time
Two computers were used for running the CFD simu- steps of 0.001 s and 10 iterations of the scalar equations
lations in this study. A Sunblade100 Workstation (500 at each time step. Typically, computation time for a
MHz processor, 256 MB memory and 15 GB internal numerical solution to full titration is 30 h with a course
disc) was used for course mesh simulations during code mesh and 50 h with a fine mesh.
development. A Sun Enterprise 6500 (30  400 MHz
processors with 6 GB memory) was used for the fine
mesh simulations. 4. Model validation
A computational mesh size of 0.001  slug length (1–
2 lm mesh cells) achieved solutions with a high degree of Three sets of experimental titrations were modelled,
mesh independency (50% change in mesh size producing each set representing a different base/acid concentration
less than 10% variation in solution data). During code ratio. Table 1 indicates the base/acid ratios of the
development meshes of 10–20 lm, requiring short run experimental titrations. A number of flow rates were
3318 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

Fig. 8. Concentration maps representing the distribution of acetic acid as it partitions from the organic to the aqueous phase (flow
speed 0.1 mm/s). (a) t ¼ 0:1 s, (b) t ¼ 0:5 s, (c) t ¼ 2:0 s, (d) t ¼ 6:0 s.

investigated within each set, these varied from 0.55 to The predicted titration times are similar to the ex-
27.0 mm/s. Fluid segment length varied with flow rate perimental values for all three sets of data indicating
with smaller segments being generated at higher speeds. that mass transfer has been predicted well by the model.
Segment lengths were between 1 and 3.8 mm. At different The predicted titration times are all within 30% of the
acid/base concentrations the same flow rates did not experimental times. The majority of predicted titration
produce the same segment length. This can be explained times (80%) are within 20% of the experimental values.
by the dependence of segment length on fluid properties. The base concentration maps in Figs. 13 and 14 show
In order to compare experimental titration times with similar distribution patterns. Fig. 13 is experimentally
predicted values the concentration of base was averaged derived, whilst Fig. 14 is theoretically predicted. The
for the aqueous segment and the endpoint of titration similarity of the concentration maps qualitatively indi-
was taken when 95% of the base had been neutralised. cates that the general pattern of the flow is predicted
well. Fig. 13 is a photographic image indicating the base
distribution in an aqueous segment 4.0 s after slug for-
5. Results mation at 2.8 mm/s. Fig. 14 is a concentration map
representing the predicted base distribution in an aque-
The predicted and experimental titration data were ous segment 4.0 s after the start of titration at 2.8 mm/s.
compared in the form of base concentration maps and Comparison of the titration photograph in Fig. 13 with
titration times. the predicted base map in Fig. 14 shows a similar dis-
Figs. 10–12 graphically present the measured and tribution of base in the aqueous segment as it is neu-
predicted titration times for each base/acid ratio: tralised during titration.
Titration time vs flow speed graphs show similar
(i) titration time vs flow speed; power law profiles for measured and predicted data
(ii) titration time vs segment length; and (Figs. 10–12). This indicates a good prediction by the
(iii) measured vs predicted titration time. model of the influence of flow speed on titration time.
 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322 3319

Fig. 9. Concentration maps representing the distribution of base during titration (flow speed 0.1 mm/s). (a) t ¼ 0:5 s, (b) t ¼ 1:0 s,
(c) t ¼ 5:0 s, (d) t ¼ 9:0 s.

Table 1 6. Discussion
Base/acid concentration ratios employed in the experimental
titrations Although the fluid segment interface was approxi-
Experiment Base (moles/ Acid (moles/ Base/acid ratio mate in terms of both (i) the flat interface, (ii) wall ve-
litre) litre) locities near the interface and (iii) relaxed viscosity ratio
Titration (a) 0.25 NaOH 0.50 acetic 0.50 for the organic and aqueous phases, the overall perfor-
Titration (b) 0.25 KOH 0.68 acetic 0.37 mance of the model in predicting titration times and
Titration (c) 0.40 KOH 0.68 acetic 0.59 base distribution was good. A fine computational mesh
(1–2 lm) was employed for the titration simulations.
This brought the numerical solution data close to mesh
Similarly, titration time vs segment length correlation independence. Finer computational meshes give greater
shows close power law profiles for the measured and solution accuracy but demand greater computational
predicted data (Figs. 10–12) indicating that the influence resources (CPU time, RAM and disk storage). A com-
of segment length on titration time is well predicted. puter capable of running the model with a fine compu-
The influence of base/acid concentration on titration tational mesh would require at least a 400 MHz
time is accurately represented by the model. This can be processor with 1 GB internal memory and 20 GB in-
seen from the close correlation between the simulated ternal disc.
and experimental titration times for each of the base/ The model may be further refined by the implemen-
acid concentration ratios. tation of a free surface model to provide a more accurate
Measured and predicted titration times show a representation of the aqueous/organic segment interface.
high level of correlation (Figs. 10–12). R2 values were Representing the fluid segment interface in this way
0.90, 0.96 and 0.92 for titrations (a), (b) and (c) re- would allow the solution to exhibit finer details of the
spectively. flow field, such as small vortices near the parabolic
3320 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

Fig. 11. Titration (b) data (base/acid mole ratio 0.37).

merical solution. The model system employed in this

report conveniently allowed the acquisition of labora-
Fig. 10. Titration (a) data (base/acid mole ratio 0.5). tory experimental data to compare with theoretical
predictions. Quantitative comparisons of experimental
and predicted base concentration distribution are a
shaped interface. In turn, mass transfer would be more subject for future research which would support the
accurately represented. The continuum surface force qualitative comparisons demonstrated in this study.
(CSF) model developed by Brackbill et al. [13] and im-
plemented into CFX by Burt et al. [14] would be ap-
propriate for this purpose. Incorporating surface tension 7. Conclusion
calculations into the segmented flow model would sub-
stantially increase the complexity of problem definition The validation results indicate that the model accu-
and the computational effort for solution. Following the rately predicts the flow field and mass transfer for the
implementation of a free surface model into the seg- segmented flow system. This provides confidence in the
mented flow model it would be appropriate to assess the model as a design tool for segmented flow in microre-
gain in accuracy against the efforts to achieve a nu- actors. However, numerical improvements to the nu-
 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322 3321

Fig. 13. Photographic image indicating base distribution dur-

ing titration 4.0 s after slug formation (flow speed 2.8 mm/s).

merical model may be achieved by incorporating a free

surface interface between contiguous segments and vis-
cosity differentials. Performance of segmented flow may
be assessed by using the model to predict mixing time-
scales for diffusion alone (static walls) and for segmented
flow configuration at different speeds. The segmented
flow model could be advantageously extended to calcu-
late industrially important chemical reactions, such as
the nitration of benzene or toluene. These complex
chemical reactions would be implemented through the
chemical kinetics routines already in place in CFX. Once
again, the complexity of problem definition and the
computational effort for solution would be substantially
increased with the inclusion of chemical reaction calcu-
lations.

Acknowledgements

This research was supported through the UK Labo-

ratory-on-a-Chip consortium comprising 17 UK based
academic and industrial members and funded by the
Engineering and Physical Science Research Council.
Numerical predictions were performed with the CFD
software CFX, property of AEA Technology plc. A Sun
Enterprise 6500 computer was made available to this
Fig. 12. Titration (c) data (base/acid mole ratio 0.59).
project by the Cardiff Centre for Computational Science
and Engineering, Cardiff University.

Fig. 14. Base concentration map of predicted titration after 4.0 s at 2.8 mm/s.
3322 N. Harries et al. / International Journal of Heat and Mass Transfer 46 (2003) 3313–3322

References International Conference on Microreaction Technology,

2000, pp. 133–140.
[1] J.R. Burns, C. Ramshaw, The intensification of rapid [7] S.V. Patankar, Numerical Heat Transfer and Fluid Flow,
reactions in multiphase systems using slug flow in capil- Taylor & Francis, 1980.
laries, Lab Chip J. (1) (2001) 10–15. [8] J.D. Anderson, Computational Fluid Dynamics, McGraw-
[2] E. Oosterbroek, Modelling, design and realisation of mi- Hill, 1995.
crofluidic components, Ph.D. thesis, University of Twente, [9] CFX Update, Computational Fluid Dynamics Software
Enschede, The Netherlands, 1999. ISBN 90-36513464. and Services. ISSN 0965-1608, no. 21, Winter 2001.
[3] V. Hessel, W. Ehrfeld, Th. Herweck, V. Havercamp, H. [10] FLUENT UserÕs Guide, vol. 4, 1996, Release 4.4, Fluent
Lowe, J. Sechiewe, Ch. Wille, Gas/liquid microreactors, in: Europe Ltd., Sheffield.
Hydrodynamics and Mass Transfer, Proceedings of the 4th [11] MEMS and Microsystems Design, Designer Reference and
International Conference on Microreaction Technology, Tutorials, Coventor, Inc., 2001.
2000, pp. 174–186. [12] R.W. Johnson, The Handbook of Fluid Dynamics, CRC
[4] M.J. Harper, Method and apparatus for diffusive transfer Press and Springer-Verlag, 1998.
between immiscible liquids, International Patent WO 97/ [13] J.U. Brackbill, D.B. Kothe, C. Zemach, A continuum
39814, 1997. method for modelling surface tension, J. Comput. Phys.
[5] P. Harston, J.R. Burns, C. Ramshaw, Reactions of 100 (1992) 355.
aromatic compounds, International Patent WO 99/22858, [14] D.J. Burt, J.W.J. Ferguson, H. Pordal, Numerical compu-
1999. tation of surface tension effects, Proc. ASME 3 (1996)
[6] J.R. Burns, C. Ramshaw, A microreactor for the nitration 439.
of benzene and toluene, in: Proceedings of the 4th