www.cet-journal.

com Page 1 Chemical Engineering & Technology

Comparative evaluation of OpenFOAM® and ANSYS® Fluent for the modeling

of annular reactors
1 1,2 2,3 2 1,2
Cristian Ariza , Cintia Casado , Ruo-Qian Wang , E. Eric Adams , Javier Marugán *
1
Department of Chemical and Environmental Technology, ESCET, Universidad Rey Juan Carlos,C/ Tulipán s/n,
28933, Móstoles, Madrid, Spain
2
Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, MA 02139, USA
3
Department of Civil and Environmental Engineering, University of California, Berkeley, 760 CA 94720, USA
*Correspondence: Javier Marugán (E-mail: javier.marugan@urjc.es), Department of Chemical and
Environmental Technology, ESCET, Universidad Rey Juan Carlos,C/ Tulipán s/n, 28933, Móstoles, Madrid, Spain
Abstract
A three dimensional numerical model was solved for a turbulent incompressible steady flow and the
® ®
discrepancies and similarities between the software OpenFOAM and ANSYS Fluent on the flow modelling are
analyzed. The evaluation was done for the meshing, setup and simulation step of the modelling process.
Outcomes have been contrasted with experimental data to validate the fluid flow obtained with both software.
With this aim, an experimental pulse-response technique was used to obtain the residence time distribution
(RTD) curve to be compared with the RTD curves obtained from the modelling. Additionally, images of the
tracer transport in the experimental reactor were taken, showing good agreement with the CFD predictions.
® ®
Keywords: CFD modeling, Fluid Dynamics, OpenFOAM , ANSYS Fluent, k-ε model
1. Introduction

Governing equations of flow processes in chemical reactor modelling are often complicated, non-
linear and coupled. For complex geometries, analytical solutions are not possible and numerical
modeling is required to reproduce and analyze the flow. Moreover, the deviations from ideal flow
configuration in chemical reactors may lead to inefficiency of the system and prevent the reactors
from scaling up due to the lack of understanding in the nature of the mixing and flow field.
Computational Fluid Dynamics (CFD), traditionally a design tool in aerospace engineering [1], has
been exploited in the environmental [2] and chemical [3] engineering fields. Considering the central
role played by reactors in chemical process industries, there is a tremendous potential on applying
computational flow-modeling tools to improve reactor engineering.
In this work, the system of partial nonlinear differential equations that describes momentum and
mass transport was solved numerically for an annular reactor using a finite volume method. Aside
from the ANSYS® Inc. package [4], one of the most widely used software in CFD modelling [5], the
open source software OpenFOAM® [6] was evaluated. One example of the scaling up effort using

Received: August 03, 2017; revised: February 06, 2018; accepted: May 16, 2018

This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which
may lead to differences between this version and the final Version of Record (VOR). This
work is currently citable by using the Digital Object Identifier (DOI) given below. The final
VoR will be published online in Early View as soon as possible and may be different to this
Accepted Article as a result of editing. Readers should obtain the final VoR from the
journal website shown below when it is published to ensure accuracy of information. The
authors are responsible for the content of this Accepted Article.

To be cited as: Chem. Eng. Technol. 10.1002/ceat.201700455

Link to final VoR: https://doi.org/10.1002/ceat.201700455

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 2 Chemical Engineering & Technology

OpenFOAM® can be found in [7]. Under the license of open source, OpenFOAM® is free for the
general use to study, distribute and modify. It provides an alternative to closed source CFD packages.
Since the source code of OpenFOAM® is open to the public, it is an advantage to not only use the
built-in codes at their full potential, but also to create and modify solvers for customized
applications. On the other hand, ANSYS® Fluent includes a fully developed GUI's (Graphical User
Interface) and support systems to make easier the setting-up of the simulations, especially for first
users. Basic process such as the 3-D CAD geometries design and meshing are more automated in
ANSYS® Fluent than in the basic interface of OpenFOAM®, which decreases the number of
uncertainties for first users who can focus more in learning. OpenFOAM®, can be also installed with
third party GUI such as HELIX [8], and the number of GUI interfaces developed for OpenFOAM® is
increasing quickly [9]. Some other software as SALOME [10] could be also used as a more automated
alternative mesher for OpenFOAM®.
The selected geometry of the annular reactor was typical in many applications of chemical
engineering processes and has been extensively studied especially for photoreactors [11]. Despite its
success and wide applications, mass transport could be limited due to suboptimal kinetics, operating
conditions, and geometrical design.
OpenFOAM® has been used to resolve the fluid dynamics for different applications in chemical
engineering, including the modeling of bubbling fluidized bed [12], continuous gravity settler [13],
electrochemical cells [14] and microreactors [15]. ANSYS® Fluent has also been applied to various
chemical reactors including a series of studies on annular reactors [11,16,17] in which the inlet and
outlet port configurations, and the internal accessories were found important in improving the mass
transport performance of the reactor. Attempts to compare the two CFD codes have been limited,
but include work related to incompressible turbomachinery [18], and industrial aerodynamics
[19,20]. In the chemical engineering field, both software have been evaluated in the heat transfer
modelling of packed-bed reactors [21],for bench-scale anaerobic gas-lift digesters [22] and gas-solid
fluidized bed hydrodynamics [23]. In this work, a comparison of both programs for the modeling of
hydrodynamics in a tubular reactor and in an annular reactor with high turbulent mixing was carried
out. The results were validated against experimental observations, including the residence time
distribution (RTD) curve derived from a pulse-response experiment and a tracer transport
experiment using methylene blue. The discrepancy of the comparison is further analyzed, confirming
the relevance of the CFD modelling and the potential of open source software package for the
chemical reactor engineering field.

2. Computational Modeling

2.1. Geometry
2.1.1. Ideal tubular reactor
For the comparison of both programs, a test was performed for the simplified geometry of a tubular
reactor. The dimensions of the pipe are 0.05 m in diameter and 4.5 m in length. The modelling for
this reactor was performed in both laminar and turbulent regimes. The model is divided in three
zones: inlet, outlet and the wall.

2.1.2. Real annular reactor

The reactor studied is an annular geometry with 0.05 m of outer tube diameter, 0.03 m of inner tube
diameter, 0.15 m of total length and 0.006 m of inlet and outlet diameter tubes. The geometry is

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 3 Chemical Engineering & Technology

angled at approximately 45 degrees to the main reactor body. The reacting volume is 1.91·10-4 m3.
The model includes a detailed description of the reactor inlet and outlet tubes, as they have a great
impact on the flow pattern. The design of the geometry is generated using an open source software,
FreeCAD [24] along with SALOME [10] in order to refine the definition of surfaces. The model was
imported into OpenFOAM® and ANSYS® meshing.
The model has several parts: fluidCylinder, the annular region and the core of the reactor; fluidInlet,
fluidOutlet: inlet and outlet surfaces and fluidRefinement, inlet and outlet cylinders. The
fluidRefinement patch is used to smoothen the transition from the inlet and outlet cylinders mesh
since the diameter is much smaller than the annulus and therefore different mesh resolutions are
needed. As inlet and outlet surfaces are not part of the wall, those did not require mesh refinement.
A scheme of the model is shown in Figure 1.

2.2. Meshing
2.2.1. Ideal tubular reactor
The tubular body was simulated as an axis-symmetric 2D model, large enough to capture the
developed velocity profile. Since the geometry is simple, blockMesh, a mesh generator of
OpenFOAM® was employed, which allows the creation of structured mesh. Also the patches were
created by blockMesh itself. For OpenFOAM®, a slice of the tubular reactor was generated, with 5000
cells, and a cyclic condition was set for the revolution planes.
In comparison, ANSYS® Fluent only requires a completely flat body and the indication of the axis of
revolution for the 2D simulation. The same number of cells, 5000 was used for ANSYS® Fluent
simulations. Figure 2 shows the generated meshes.

2.2.2. Real annular reactor

The evaluation of the mesh quality according to ANSYS® Fluent manual [25], and also used by
OpenFOAM® is done by analyzing two properties of the mesh, the skewness and the aspect ratio.
Skewness is defined as the difference between the shape of the cell and the shape of an equilateral
cell of equivalent volume, whilst the aspect ratio is a measure of the stretching of the cell. While
OpenFOAM® does not limit the aspect ratio or the skewness of cells, ANSYS® Fluent limits the
skewness to 0.98 to ensure convergence. These differences is software quality standards cause
difficulties when using a mesh from the opposite software. For that reason, the geometry was
meshed twice, following the requirements of each software.
The mesh used in OpenFOAM® was generated by snappyHexMesh, which uses unstructured cells
along with an added boundary layer close to the surfaces and important edges. The mesh has around
5.15·105 cells, including layers of refinement inside the fluidRefinement and fluidCylinder. The
boundary layer thickness is considered to be the distance at which the velocity has achieved 99% of
the developed flow velocity in the free stream, and can be calculated from the Blasius equation [Eq.
1] [26]:
5.0 5.0𝑥
𝛿≈ =
𝑈
√ √𝑅𝑒𝑥 [Eq. 1]
𝜈𝑥
where 𝑈 is the free stream velocity, ν is the kinematic viscosity, 𝑥 is the distance to the free stream
profile and Rex is the Reynolds calculated with 𝑥.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 4 Chemical Engineering & Technology

In the central section of the cylinder, x = 39·10-3 m and Rex = 1250 (U = 3.34·10-2 m s-1), the calculated
boundary layer has a thickness of 𝛿 ≈ 5.3 · 10−3 m, which is well captured by our refined mesh. The
range of Y+ recommended for using standard wall functions with 𝑘 − 𝜖 methods is between 30 and
60, leading to an estimation of the boundary layer thickness of 3.3·10-3 to 7·10-3 m, in agreement
with previous calculation.
ANSYS® Fluent uses tetrahedral cells. The surfaces of the model were refined using the Face Sizing
tool, available in ANSYS® Meshing, with the addition of inflation layers for the boundary layer near
the walls.
The final mesh has a size of around 5.1·105 cells, to allow comparison with equivalent mesh in
OpenFOAM.

Both meshes are accepted by the mesh quality checkers on each respective software. In spite of
differences (Figure 3), they constitute the optimal approach for each respective software in order to
provide high quality and mesh-independent results.

2.3. Laboratory Experiments

The experiments were carried out in an annular reactor fed by a centrifugal pump with a flow rate of
2.62 L min-1, determined experimentally, from a well-mixed reservoir tank. Due to the small inlet
section and the sudden expansion of the flow inside the annular region of the reactor, a significant
change in the fluid dynamical regime is expected. The values of the Reynolds numbers are 665 and
8870 for the annulus and the inlet section, respectively. In fully developed pipe flow, the transition
between laminar and turbulent flow occurs for Re values in the range 2300-4000 [27]; therefore we
assumed that flow is turbulent in the inlet and laminar in the annulus. A transition region exists at
the union, which is a challenge of both software.

2.4. Governing equations and physical models

The governing equations are the continuity equation [Eq. 2] and the steady-state Navier-Stokes
equation [Eq 3].
∇(ρ𝑈)=0 [Eq. 2]
∇·(ρ𝑈𝑈)= -∇P+∇·τ+ρg
⃗ [Eq. 3]

where 𝜌, 𝑈, P, 𝜏 and 𝑔 are fluid density, velocity vector, pressure, viscous stress tensor and
gravitational acceleration, respectively. These equations, combined with Newton’s law of viscosity as
a constitutive equation to relate the stress tensor to the motion of the continuous fluid, allow
computing the velocity field within the reactor. The solver simpleFoam, from the OpenFOAM® package
was employed. It is a steady-state solver for incompressible, turbulent flow, but the turbulence type
can be set to off for a laminar modelling. simpleFoam uses the SIMPLE (Semi-Implicit Linked Equations)
algorithm, which is also used by default in ANSYS® Fluent. The algorithm is essentially a guess and
correct procedure for the coupled calculation of pressure and velocity. More details about the
SIMPLE algorithm can be found elsewhere [28,29]..

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 5 Chemical Engineering & Technology

2.4.1. Turbulent models

Various turbulence models are available in both software, the most common ones being 𝑘 − 𝜖,
𝑘 − 𝜔 and 𝑘 − 𝜔 SST models. Most industrial applications use the standard 𝑘 − 𝜖 model, which is a
two-transport-equation model solving for the kinetic energy 𝑘 and turbulent dissipation ϵ. This
model is usually stable and cheap to implement with a sacrifice of the accuracy, especially for
complex flows involving severe pressure gradients, flow separation, and strong streamline curvature.
So it is suitable for initial iterations, initial screening of alternative designs, and first-cut examination
of CFD configuration.
Within the 𝑘 − 𝜖 category, ANSYS® Fluent includes the realizable 𝑘 − 𝜖 approach, an improvement
over the standard 𝑘 − 𝜖 model. It contains a new formulation for the turbulent viscosity and a new
transport equation for the dissipation rate, 𝜖, that is derived from an exact equation for the transport
of the mean-square vorticity fluctuation. An immediate benefit of the realizable 𝑘 − 𝜖 model is that it
exhibits superior performance for flows involving rotation, boundary layers under strong adverse
pressure gradients, flow separation, and recirculation. OpenFOAM® has a similar realizable 𝑘 − 𝜖
model, which has been shown comparable in performance [30].
For the convenience of comparison, the simplest realizable 𝑘 − 𝜖 turbulence model is selected.
Usually 𝑘 − 𝜖 models are accurate enough to represent turbulent behavior in chemical reactors.
Additionally, an initial simulation using a 𝑘 − 𝜖 model is recommended before using a more complex
turbulent model [31]. For this reason, this model is selected as representative of the code
performance.
The 𝑘 − 𝜖 model assumes isotropic scalar turbulent viscosity, treats the flow as completely turbulent,
and ignores the molecular viscosity. [Eq. 4] describes the energy transport and [Eq. 5] describes the
dissipation rate, at steady-state, used by both ANSYS® Fluent and OpenFOAM®. Turbulent viscosity is
modeled by [Eq. 6] [28]. .
𝜕𝑘 𝜕 𝜇𝑡 𝜕𝑘 𝜕𝑈𝑖
𝜌𝑈𝑖 = [(𝜇 + ) ] + 𝜏𝑖𝑗 − 𝜌𝜖 [Eq. 4]
𝜕𝑥𝑖 𝜕𝑥𝑗 𝜎𝐾 𝜕𝑥𝑗 𝜕𝑥𝑗
𝜕𝜖 𝜕 𝜇𝑡 𝜕𝜖 𝜖2
𝜌𝑈𝑖 = [(𝜇 + ) ] + 𝜌𝐶1 𝑆𝜖 − 𝜌𝐶2 [Eq. 5]
𝜕𝑥𝑖 𝜕𝑥𝑗 𝜎𝜖 𝜕𝑥𝑗 𝑘 + √ν𝜖
𝐶𝜇 𝑘 2
𝜇𝑡 = 𝜌 [Eq. 6]
𝜖
𝜂 𝑘 1 𝑘𝑈∗
where 𝐶1 = max [0.43, ] , 𝜂 = 𝑆 𝜀 , 𝑆 = √2𝑆𝑖𝑗 𝑆𝑖𝑗 , 𝑎𝑛𝑑 𝐶μ = 𝐴 + 𝐴 𝜀
𝜂+5 0 𝑠

The model constants specified both in OpenFOAM and ANSYS Fluent are the following: 𝐶μ =
0.09, 𝐴0 = 4, C2 = 1.9, σk = 1 𝑎𝑛𝑑 σε = 1.2. Further details of the realizable 𝑘 − 𝜖 model can be
found elsewhere [32].
There are also several parameters needed in both OpenFOAM® and ANSYS® Fluent. The turbulent
intensity, 𝐼 can be estimated by:
1
𝑈′ −
𝐼= = 0.16Re𝑑 3 [Eq. 7]
𝑈𝑎𝑣𝑔 ℎ
2 2
The turbulent energy k [𝑚 𝑠 ] is defined in [Eq. 8].
3
𝑘 = (𝑈𝐼)2 [Eq. 8]
2
where 𝑈 is the free stream velocity and I is the turbulence intensity. The turbulence dissipation,
𝜖 [𝐽/(𝑘𝑔 𝑠)], is the rate at which turbulence kinetic energy is converted into thermal internal energy.
The equation used in OpenFOAM® is [Eq 9].

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 6 Chemical Engineering & Technology

3
𝑘2 [Eq. 9]
𝜖 = 𝐶𝜇
𝑙
where 𝑘 is the turbulent energy, 𝑙 is the turbulent length scale and 𝐶𝜇 is a turbulence model
constant, which usually has a value of 0.09. Some other CFD codes, like ANSYS Fluent, Phoenics and
CFD-ACE for example, use a different length-scale definition based on the mixing-length, and the
equation, [Eq. 10], is therefore different
3
3 2
𝑘 [Eq. 10]
𝜖= 𝐶𝜇4
𝑙
where 𝐶𝜇 is a turbulence model constant, which has a value of 0.09. The turbulent length scale can
be estimated for fully developed flows as 𝑙 = 0.038 𝐿 (OpenFOAM®), and when the definition is
based in the mixing length (ANSYS® Fluent) as 𝑙 = 0.07 𝐿, 𝐿 [m], being the relevant dimension, which
in this case is equal to the pipe hydraulic diameter 𝑑ℎ .
Additional parameters set in both software are summarized in Table 1.

2.5. Species transport modeling

The species transport equations are solved to reproduce the behavior of the tracer transport. Both
software were compared with the lab experiment with an injection of a tracer at the inlet patch. The
flow field derived through the flow simulation is used as the initial condition of the species transport
problem.
For each individual chemical 𝑖 in the computational domain, the mass conservation equation can be
expressed as [Eq. 11].
𝜕
(𝜌𝑌𝑖 ) + ∇(𝜌𝑈 ̅𝑌𝑖 ) = ∇𝐽𝑖 + 𝑅𝑖 [Eq. 11]
𝜕𝑡
where 𝑌𝑖 is the fraction of i in the mixture, 𝐽𝑖 is its diffusive flux and 𝑅𝑖 its rate of production. The
velocity vector, 𝑈, couples the mass balances to the hydrodynamic calculations. The diffusive flux can
be estimated using Fick’s first law of diffusion [Eq. 12].
𝐽𝑖 = −𝜌𝐷𝑖,𝑚 ∇𝑌𝑖 [Eq. 12]
®
The solver needed to simulate species transport in OpenFOAM is scalarTransportFoam. The resulting
flow field from the flow simulation is set as the initial field and a T file is created which specifies the
concentration of the species. To solve the species transport in ANSYS Fluent, the reproduced flow
field was used in the same way.

2.5.1. Experimental Residence Time Distribution (RTD)

The RTD, 𝐸(𝑡), describes the distribution of times for which a conservative tracer, injected
instantaneously at the reactor inlet, resides in the reactor. It reflects the flow pattern inside the
vessel volume [33,34]. For CFD hydrodynamic validation, the integral of the RTD on the annular
reactor was determined experimentally by using a tracer-response technique. Starting from the
system working under stationary circulation of water, methylene blue (Scharlab, extra pure) was
injected into the reservoir with vigorous stirring to produce a perfect mixture in the vessel at time t =
0 s. The absorbance at 660 nm in the reactor outlet was monitored in a continuous mode with an
illuminated flow cell coupled to the spectroradiometer probe. Normalized concentrations were
calculated between the baseline at 𝑡 = 0 𝑠. and the final absorbance. More information about the
experimental setup can be found elsewhere [35].

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 7 Chemical Engineering & Technology

The residence time distribution has been calculated from the experimental curve and from the
𝑑 𝐶(𝑡)/𝐶
0
outcomes of the modelling, as 𝐸(𝑡) = , where 𝐶(𝑡) and 𝐶0 are the outlet and inlet
𝑑𝑡
concentrations respectively. The mean residence time can be obtained as the first moment of the
distribution:
∞
𝑡𝑟 = ∫0 𝑡 𝐸(𝑡)𝑑𝑡 [Eq. 13]
The variance is the second central moment:
∞
𝜎 2 = ∫0 (𝑡 − 𝑡𝑟 )2 𝐸(𝑡)𝑑𝑡 [Eq. 14]
Finally, the number of equivalent tanks in series can be calculated as [34]:
𝑡𝑟 2
𝑁= [Eq. 15]
𝜎2

2.5.2. Theoretical velocity profiles and entrance length.

For the ideal reactor, a tubular pipe with smooth walls, the velocity profile obtained can be
compared with theoretical expressions for fully developed flows. For the laminar flow, the following
equation was used [36]:
r2
̅ (1 − 2 )
Uz = 2 U [Ec. 16]
R
̅ is the averaged velocity and R the tube radius.
Where U
The theoretical turbulent profile is calculated according to the following equations:
1 1
𝑈 𝑦 𝑛 𝑟 𝑛
𝑈 = = ( ) = (1 − ) [Eq. 17]
⋆
𝑉 𝑅 𝑅
8
𝑛 = 𝑘√ [Eq. 18]
𝑓
where 𝑉 is the initial velocity, 𝑦 the distance from the wall, 𝑟 the distance from the central axis, 𝑅 the
maximum radius and a calculated value of 𝑛 = 5.
The entrance length is defined as the longitudinal position at which the developed profile
appears. In laminar regime has been calculated as:
LE
D
= 0,065 · Re [Eq. 19]
For turbulent flow, the entrance length was calculated using the the empirical equation of Nikuradse
[37].
𝐿ℎ,𝑡𝑢𝑟𝑏 ≈ 40𝐷 [Eq. 20]

2.6. Physical properties, numerical schemes and boundary conditions

Prior to the simulation, all the setup conditions must be defined, in the case of ANSYS® Fluent
through the visual interface, and in the case of OpenFOAM® through the setup of the system folder
files and constant files.
In all simulations, the fluids are assumed to be Newtonian, incompressible, and isothermal, with
constant physical properties. The physical properties of water at 25 ºC were considered (ρ = 998.2 kg
m-3 ; μ = 1.003·10-3 kg m-1 s-1). A no-slip boundary condition was imposed at all reactor walls.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 8 Chemical Engineering & Technology

2.6.1. Idealized tubular reactor

For the laminar case, an inlet velocity of 0.027 m s-1 is set (5.3·10-5 m3·s-1), corresponding to a
Reynolds number of 1365. For the turbulent case, the velocity was increased to a value of 0.107 m s-1
(2.1·10-4 m3·s-1, Re = 5314). In both cases, direction is normal to the boundary. The outlet boundary
condition assumes atmospheric pressure. For both flow regimes, the transport of a tracer was
simulated (Methylene blue, 0.5 mM, diffusion coefficient 𝐷𝑖,𝑚 = 6·10-10 m2·s-1 [38]) over the steady-
state velocity field. The prediction of the non-steady state modelling was performed for 200 s with a
time step of 1 s in both software. Since the outcomes of tracer concentration at outlet are plotted
with normalized concentration, i.e. 𝐹(𝑡) = 𝐶(𝑡)/𝐶0 , the concentration at the inlet is set to be 1.

2.6.2. Real annular reactor

Annular reactor flow field was calculated using the realizable k-ε turbulent model. The inlet velocity
is set to 1.54 m s-1 and normal to the boundary. At the outlet, atmospheric pressure was also applied.
For this case, the transport of the tracer is simulated over the steady state velocity field, for 50 s with
a time step of 0.1 s in both ANSYS® Fluent and OpenFOAM®. Results are compared with the
experimental residence time distribution curve to validate the velocity field.

Table 2 lists the numerical schemes used on both software [18]. The procedure to establish the
interpolation methods in both software is different and this step could introduce a significant
uncertainty, especially for novice users. While ANSYS® Fluent has an interface in which first order or
second order interpolation schemes can be easily assigned to the turbulence and momentum
equation, in OpenFOAM® interpolation methods can be assigned to pressure and momentum
gradients, Laplacians, interpolations schemes and divergence operators, procedures which are
accessible to more experienced users.

2.7. Convergence criteria and solution strategy

All simulations carried out in this work were done with OpenFOAM® v.3.0.1 and ANSYS® Fluent v.17.0.
Convergence of the numerical solution was ensured in both software by monitoring the scaled
residuals to a standard of at least 10-6 for the continuity, momentum variables and concentrations,
meaning that the relative change of the magnitudes among successive iterations is below this value.
Additionally, the variables of interest have been monitored at different surfaces of the computational
domain and a convergence is considered to be achieved when 50 iterations were performed without
parameter changes.

The computer used for the simulations has two Intel Xeon processors with 2.40 GHz (16 cores each)
and 64 GB of RAM. Preliminary simulations showed that 5 processors are the optimal to run
OpenFOAM®, while in ANSYS® Fluent, 6 processors were found to be optimal. A number of 5
processors is used in both software for comparison.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 9 Chemical Engineering & Technology

3. Results

3.1. Flow field

3.1.1. Idealized tubular reactor
Before simulating flow inside the real (annular) reactor, a test was performed in an ideal tubular
reactor. The flow was simulated in both laminar and turbulent conditions.
Figure 4 shows the results of the velocity profiles along the Z direction obtained from the laminar
modelling. The velocity is zero at the outer wall and free of stress in the center, which is consistent
with the boundary conditions.
The results obtained from the laminar flow case are almost the same with both software., The
highest error is found for a distance of z = 0.9 m, where a 2.7% difference between software is
obtained in the pipe center (Umax). This error decreases with distance (being the difference in Umax
lower than a 0.7 % between ANSYS® Fluent and OpenFOAM® at the pipe outlet). In the outlet section,
the developed profile is reached, and results are almost the same as the theoretical velocity profile
calculated for developed laminar flow (Eq.18). The entrance length calculated with Eq.19 is 4.4 m,
slightly lower than pipe length (4.5 m).
The difference is increased when we applied the realizable 𝑘 − 𝜖 turbulent model.Figure 5 shows
the comparison of OpenFOAM®, ANSYS® Fluent and the theoretical velocity profile at the outlet for
the turbulent flow. Differences obtained in the shape of the theoretical profile can be attributed to
the fact that although the power law is commonly used to predict turbulent flows in pipes, it does
not include the effect of wall shear stress and is inaccurate near the wall. Figure 6 shows
the maximum turbulent velocity obtained from both software along the pipe center (z axis). The
entrance length calculated with Eq. 20 is also included in the graph. ANSYS® Fluent has a stronger
overshoot than OpenFOAM®, which results in a higher velocity after the entrance length.

In Table 3 and Table 4, the maximum velocities at outlets are compared, for both laminar and
turbulent flow. The absolute values obtained for the turbulent flow are similar, with less than 3.6%
difference between the two software (instead of the 0.7% obtained with the laminar flow), so both
the simulation results are satisfactory. Compared with theoretical values, for developed flow, all
simulations have a difference lower than a 4.5% (Table 3). Table 4 shows the relative error of the
averaged values in the outlet surface, instead of maximum values, using the theoretical value as a
reference, demonstrating that the discrepancy of the simulation is very small, with a relative error
between 10−4 and 10−7.

3.1.2. Real annular reactor

For the purposes of predicting reactor behaviour and system scale-up, it is necessary to determine
quantitatively the flow field difference between the real reactor and the idealized models. If the non-
ideality is not well handled in scale changes, serious errors will happen in the final design. Figure 7
shows the velocity obtained for the annular reactor at different cross-sections using both programs.
Due to the fluctuations of the flow along the X axis some differences were found in results with the
two software represented on the XY plane, where the velocity in the z direction is plotted.. On the

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 10 Chemical Engineering & Technology

other two planes, the flow shows similar behavior of the velocity magnitude with minimal differences
between both software.
To quantify the difference, the velocity profiles along the x axis are plotted in Figure 8 at different
distances from the inlet. The reactor is not symmetrical, as the inlet is located in the left side, and
outlet in the right side (see Figure 1). Therefore, different velocity values are obtained in the z
direction for both sides of the annular geometry as it can also be observed in Figure 7. Due to the
short length of the reactor, fully developed flow is not achieved.

The differences between the plots obtained with both software are attributed to various reasons.
First, the transition area may dominate the computational domain, so the results are highly sensitive
to numerical errors and significant difference can emerge. Second, the addition of the induced
rotation on the inlet and outlet can produce slightly different results due to the error accumulation
through the flow entrance to the main cylinder and internal differences in the solvers
The value of some representative magnitudes obtained from both software were averaged in the
reactor volume and summarized in Table 5.
With the exception of pressure, the differences of the parameters are less than 5%. The larger
differences with pressure may be related to the differences in the mesh resolution and mesh shape
near the wall and inlet. Morover, although the algorithm implemented is the same in both programs,
ANSYS® Fluent calculates explicitly the velocity flux in the cell faces, and the pressure implicitly in the
cell center, while OpenFOAM® solves first explicitly for the velocity in the cell center and the pressure
explicitly in the cell faces. This difference in the calculation procedure explains some differences
found between solvers. As ANSYS® Fluent calculates the velocity fluxes in the cell faces, the
adaptation is better with non-orthogonal meshes, while OpenFOAM® has more problems with non-
orthogonality, as the velocity flux has to be interpolated from cell center to faces. To deal with that,
OpenFOAM® includes a non-orthogonal corrector feature, inside the SIMPLE algorithm. Additionally,
this difference explains the higher memory consumption and also the longer calculation time of
simulation with ANSYS® Fluent, as the 3 velocity equations are solved in cell faces. The OpenFOAM®
numerical scheme reduces calculation time and memory consumption but could experience difficult
with convergence.
The volume is also compared since the meshing uses polyhedral cells, and these are different in the
two meshes.

3.2. Species Transport

3.2.1. Idealized tubular reactor
For the species transport, the concentration of the injected tracer is plotted versus time in Figure 9,
for laminar and turbulent flow.
Even though the outcomes are similar to each other, for both flow regimes, there are some
discrepancies due to non-linear amplification along the distance from the inlet. In addition, the
differences in the transport solver can also introduce a certain amount of error. The calculation of a
theoretical value is not obvious since, among other reasons, 𝐶(𝑡) varies with 𝑥: the further from the
entrance, the “wider” is the curve. Another aspect to have in mind is that, at the inlet profile, the
velocity profile goes from flat to parabolic. Therefore, roughly, the concentration curve depends
globally on the mass flow rate, the geometry and the Reynolds numbers, and locally on the distance
from the inlet.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 11 Chemical Engineering & Technology

3.2.2. Real annular reactor

To compare species transport, spectrophotometric measurements were recorded of the tracer
concentration leaving the reactor. Additionally, a fixed camera was used to record the flow pattern of
the fluid through the reactor volume. The experimental images obtained (Figure 10) for different
time increments are shown below together with the modelling displayed side by side with the
OpenFOAM® and ANSYS® Fluent.
Although it is difficult to compare pictures with computer graphics, it is easy to see tendencies in all
the pictures, such as the tracer reaching the outlet tube, the distance from the entrance of the
maximum concentration and the reactor being filled. A similar convective-diffusive behavior between
the computer generated results and the experimental images was observed.
For a quantitative analysis, the residence time curves were plotted (Figure 11), which corresponds to
the plot of the function: 𝐹(𝑡) = 𝐶(𝑡)/𝐶0 .
The results obtained from the cumulative RTD distribution are close. Given the uncertainty in the
experimental results, the current comparison is acceptable. The values of the residence time, the
variance and the equivalent number of equivalent tanks, N, were calculated for the experimental and
computational results in Table 6.
In comparison to the experimental value, both software show a similar response. The calculated
residence time obtained with OpenFOAM® is closer to the experimental residence time than the
value predicted by ANSYS® Fluent. In contrast, ANSYS® Fluent better predicts the variance of the RTD
curve and the derived number of tanks in series. Considering the short length of the reactor and
complexity of the flow, the agreement between the experimental data and the predictions of both
software is considered reasonable.

4. Conclusions

The comparison of fluid mechanics and mass transport model predictions of OpenFOAM® and ANSYS®
Fluent requires not only a deep understanding of the fundaments of the available physical and
mathematical methods, but also of the details of their implementation. A careful analysis of both
software was made in order to make the simulation models as comparable as possible. Predictions of
the flow field in a real annular reactor, with a complex laminar-turbulent regime, showed differences
of less than 12.5% when comparing simulations with experimental data. For the species transport,
both qualitative representations of the distribution of a tracer and quantitative analysis of the RTD
curves show good agreement between the experimental data and the predictions of both software.
The successful implementation of the model on the open source software OpenFOAM® proves that
OpenFOAM® can be a useful tool for simulating chemical reactors, with a predictive power
comparable to that of ANSYS® Fluent.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 12 Chemical Engineering & Technology

Acknowledgments

The authors gratefully acknowledge the financial support of the Spanish Ministry of Economy and
Competitiveness (MINECO), in the frame of the collaborative international consortium
WATERJPI2013−MOTREM of the Water Challenges for a Changing World Joint Programming Initiative
(Water JPI) Pilot Call. J. Marugán would like to acknowledge MINECO for the financial support of his
stay at Massachusetts Institute of Technology through the program Salvador de Madariaga
(PRX15/00012). C. Casado also acknowledges MINECO for the FPI PhD grant (BES-2012-056661).

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 13 Chemical Engineering & Technology

References

[1] H. Xia, P.G. Tucker, W.N. Dawes, Prog. Aerosp. Sci. 2010, 46, 274–283. DOI:
10.1016/j.paerosci.2010.03.001.
[2] R.Q. Wang, A.W.K. Law, E.E. Adams, Int. J. Multiph. Flow. 2014, 67, 65–75. DOI:
10.1016/j.ijmultiphaseflow.2014.08.004.
[3] S. Wang, H. Lu, F. Zhao, G. Liu, Chem. Eng. J. 2014, 236, 121–130. DOI:
10.1016/j.cej.2013.09.033.
[4] ANSYS® Academic Research, Release 14.5.
[5] W. Jeong, J. Seong, Int. J. Mech. Sci. 2014, 78, 19–26. DOI: 10.1016/j.ijmecsci.2013.10.017.
[6] H.G. Weller, G. Tabor, H. Jasak, C. Fureby, Comput. Phys. 1998, 12, 620. DOI:
10.1063/1.168744.
[7] R.-Q. Wang, E.E. Adams, A.W.-K. Law, A.C.H. Lai, J. Hydraul. Eng. 2015, 141, 6015006. DOI:
10.1061/(ASCE)HY.1943-7900.0000993.
[8] HELYX-OS Free Open-Source GUI for OpenFOAM. URL: http://engys.com/products/helyx-os
(accessed July 20, 2017).
[9] Graph. User Interfaces - OpenFOAMWiki. URL https//openfoamwiki.net/index.php/GUI
(Accessed July 20, 2017).
[10] A. Ribes and C. Caremoli. “Salomé platform component model for numerical simulation,”
COMPSAC 07: Proceeding of the 31st Annual International Computer Software and Applications
Conference, pages 553-564, Washington, DC, USA, 2007, IEEE Computer Society.
[11] J.E. Duran, F. Taghipour, M. Mohseni, Int. J. Heat Mass Transf. 2009, 52, 5390–5401.
[12] Y.F. Liu, O. Hinrichsen, Chem. Eng. Technol. 2013, 36, 635–644. DOI:
10.1002/ceat.201200625.
[13] S.K. Panda, K.K. Singh, K.T. Shenoy, V. V. Buwa, Chem. Eng. J. 2017, 310, 120–133. DOI:
10.1016/j.cej.2016.10.102.
[14] J.L.C. Santos, V. Geraldes, S. Velizarov, J.G. Crespo, Chem. Eng. J. 2010, 157, 379–392. DOI:
10.1016/j.cej.2009.11.021.
[15] T. Van Daele, D. Fernandes del Pozo, D. Van Hauwermeiren, K. V. Gernaey, R. Wohlgemuth, I.
Nopens, Chem. Eng. J. 2016, 300, 193–208. DOI: 10.1016/j.cej.2016.04.117.
[16] D.A. Sozzi, F. Taghipour, Int. J. Heat Fluid Flow. 2006, 27, 1043–1053. DOI:
http://dx.doi.org/10.1016/j.ijheatfluidflow.2006.01.006.
[17] F. Jović, V. Kosar, V. Tomašić, Z. Gomzi, Chem. Eng. Res. Des. 2012, 90, 1297–1306. DOI:
http://dx.doi.org/10.1016/j.cherd.2011.12.014.
[18] F. Bothe, C. Friebe, M. Heinrich, R. Schwarze, CFD Simulation of Incompressible
Turbomachinery — A Comparison of Results From ANSYS Fluent and OpenFOAM, in: Vol. 2B
Turbomach., ASME, 2014: p. V02BT39A025. DOI: 10.1115/GT2014-26338.
[19] A. Karvinen, H. Ahlstedt, Proc. Open Source CFD Int. Conf. 2008, 1–17.
[20] M. Balogh, A. Parente, C. Benocci, J. Wind Eng. Ind. Aerodyn. 2012, 104, 360–368. DOI:
10.1016/j.jweia.2012.02.023.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 14 Chemical Engineering & Technology

[21] A. Singhal, S. Cloete, S. Radl, R.Q. Ferreira, S. Amini, MAYFEB J. Chem. Eng. 2016, 1, 1–17.
[22] A.R. Coughtrie, D.J. Borman, P.A. Sleigh, Bioresour. Technol. 2013, 138, 297–306. DOI:
10.1016/j.biortech.2013.03.162.
[23] N. Herzog, M. Schreiber, C. Egbers, H.J. Krautz, Comput. Chem. Eng. 2012, 39, 41–46. DOI:
10.1016/j.compchemeng.2011.12.002.
[24] FreeCAD: An open-source parametric 3D CAD modeler URL: https://www.freecadweb.org/
(accessed July 20, 2017).
[25] ANSYS® Academic Research, Release 14.5, Help System, Meshing User Guide, ANSYS, Inc.
[26] I.G. Currie, Fundamental Mechanics of Fluids, Third Edition, CRC Press, 2002.
[27] J.P. Holman, Heat transfer, McGraw-Hill, 1989.
[28] ANSYS® Academic Research, Release 14.5, Help System, Theory Guide, ANSYS, Inc.
[29] J.H. Ferziger, M. Peric, Computational Methods for Fluid Dynamics, 3er Edition (2002),
Springer.
[30] M. Rakai, P. Kristóf, CFD simulation of flow over a mock urban setting, in: 5th OpenFOAM
Work. Chalmers, Gothenburg, Sweden, 2010: pp. 2–5.
[31] V. V. Ranade, Computational Flow Modeling for Chemical Reactor Engineering., Elsevier,
2001. URL: https://books.google.es/books?hl=es&lr=&id=Cv6x-
Ipik3IC&oi=fnd&pg=PP2&dq=Computational+Flow+Modeling+for+Chemical+Reactor+Engineering&ot
s=FU16dCszHP&sig=hWktYam0YJDrUiCdJhOAwKo1eJs#v=onepage&q=Computational Flow Modeling
for Chemical Reactor Engineering&f (accessed January 25, 2018).
[32] T.H. Shih, W.W. Liou, A. Shabbir, Z. Yang, J. Zhu, Comput. Fluids. 1994, 24, 227–238.
[33] E. Sahle-Demessie, S. Bekele, U.R. Pillai, Catal. Today. 2003, 88, 61–72. DOI:
http://dx.doi.org/10.1016/j.cattod.2003.08.009.
[34] O. Levenspiel, Wiley, New York . 1999,. DOI: 10.1021/ie990488g.
[35] C. Casado, J. Marugán, R. Timmers, M. Muñoz, R. van Grieken, Chem. Eng. J. 2017, 310, 368–
380. DOI: 10.1016/j.cej.2016.07.081.
[36] R.B. Bird, Appl. Mech. Rev. 2002, 55, R1–R4.
[37] J. Nikuradse, J. Appl. Phys. 1943, 14, 399. DOI: 10.1063/1.1715007.
[38] N. Miložič, M. Lubej, U. Novak, P. Žnidaršič-Plazl, I. Plazl, Chem. Biochem. Eng. Q. 2014, 28,
215–223.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 15 Chemical Engineering & Technology

Table 1. Model Parameters

Annular reactor1 Ideal reactor
Hydraulic Diameter [m] 0.025 0.020
Turbulent Intensity 0.054 0.071
Reynolds Number 6000 8870
Mixing Length [m] 0.019 2.28×10-4
1
Calculated in the inlet/outlet sections

Table 2. Numerical schemes and boundary conditions used in the simulations.

OpenFOAM ® ANSYS ® Fluent
Numerical method
Solver simpleFoam Pressure based
Pressure-velocity coupling SIMPLE SIMPLE
Multigrid GAMG AMG
Spatial discretization
Convective terms 2nd order upwind 2nd order upwind
Gradient Gauss linear Green Gauss node based
Boundary conditions
Velocity fixedValue,
Inlet Velocity inlet
pressure zeroGradient
Outlet 𝑝=0 𝑝=0
Walls No-slip No-slip

Table 3. Maximum velocities at the outlet of the ideal tubular reactor

Laminar Flow Turbulent Flow
Umax, outlet (m s-1) Difference Umax, outlet (m s-1) Difference
Theoretical 0.0546 - 0.1410 -
OpenFOAM ® 0.0540 1.10% 0.1420 0.71%
ANSYS ® Fluent 0.0544 0.37% 0.1473 4.47%

Table 4. Discrepancies with the theoretical values of the outlet velocity in the ideal tubular reactor
̅ 𝒐𝒖𝒕𝒍𝒆𝒕 − 𝑼
𝑼 ̅ 𝒐𝒖𝒕𝒍𝒆𝒕,𝒕𝒉𝒆𝒐
𝑼̅ 𝒐𝒖𝒕𝒍𝒆𝒕,𝒕𝒉𝒆𝒐
OpenFOAM ® laminar -1.61×10-4
OpenFOAM ® turbulent -5.90×10-4
ANSYS ® Fluent laminar +1.10×10-6
ANSYS ® Fluent turbulent -9.36×10-8

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 16 Chemical Engineering & Technology

Table 5. Averaged magnitudes in the total volume of the annular reactor

Magnitude OpenFOAM ® ANSYS ® Fluent Difference (%)

C(t) at 𝑡 = 10 𝑠* 9.27×10-1 9.01×10-1 2.21 %

𝑃 [Pa] 1.70×103 1.52×103 11.84 %

𝑃𝑖𝑛𝑙𝑒𝑡 [Pa] 1.67×10 3
1.49×10 3
12.08 %
𝑈𝑥 [m s ]-1
1.40×10 -2
1.44×10 -2
2.86%
𝑈𝑦 [m s-1] -1.86×10 -2
-1.89×10 -2
1.61 %
𝑈𝑧 [m s-1] -8.64×10-3 -8.43×10-3 2.61 %
3 -4 -4
Volume [m ] 1.9127×10 1.9121×10 0.03 %
*Since the concentration of the tracer is set at 1 on the inlet, C(t) can be considered as
non-dimensional.

Table 6. Residence time, variance and number of equivalent tanks.

𝒕𝒓 (s) 𝝈 (s) N
Experimental 5.92 3.81 2.41
OpenFOAM ® 6.02 2.98 4.06
®
ANSYS Fluent 7.02 4.01 2.98

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 17 Chemical Engineering & Technology

FIGURE CAPTIONS

Figure 1. 3D generated computation model of the annular reactor.

Figure 2. Discretization of the ideal tubular reactor: A) OpenFOAM ® mesh. B) ANSYS ® Fluent mesh.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 18 Chemical Engineering & Technology

Figure 3. Discretization of the real annular reactor: A) OpenFOAM ® mesh. B) ANSYS ® Fluent mesh.

Figure 4. Velocity profiles in the ideal tubular at different distances from the inlet, laminar flow.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 19 Chemical Engineering & Technology

Figure 5. Dimensionless turbulent developed profiles in the ideal tubular reactor, turbulent flow.

Figure 6. Maximum velocity along the rotational axis for the ideal tubular reactor and turbulent flow.
Vertical line represents the entrance length value calculated with Eq. 20.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 20 Chemical Engineering & Technology

Figure 7. Comparison of velocity contours at different planes of the annular reactor. Left: OpenFOAM
®
; right: ANSYS ® Fluent.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 21 Chemical Engineering & Technology

Figure 8. Velocity in the z-direction for different XY planes.

Figure 9. Average concentration of the tracer at the outlet of the laminar case on the ideal tubular
reactor.

This article is protected by copyright. All rights reserved.

www.cet-journal.com Page 22 Chemical Engineering & Technology

Figure 10. Images at time intervals showing the transport of the tracer on the annular reactor. Top:
experimental; middle: ANSYS ® Fluent; bottom: OpenFOAM ®.

Figure 11. Comparison of the predicted and experimental tracer concentration at the outlet of the
real annular reactor.

This article is protected by copyright. All rights reserved.