b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/issn/15375110

Research Paper

Reliability of turbulence models and mesh types for

CFD simulations of a mechanically ventilated pig
house containing animals

Hao Li, Li Rong, Guoqiang Zhang*

Department of Engineering, Aarhus University, Blichers All
e 20, P. O. Box 50, 8830 Tjele, Denmark

article info
The use of computational fluid dynamics (CFD) to study the airflow within farm animal
Article history: buildings is increasing. The choice of turbulence model within CFD is generally considered
Received 21 March 2017 to be important due to the approximations of the turbulence in varied scales. Although
Received in revised form some studies have been conducted on evaluations of turbulence models in simulation of
5 June 2017 airflow in ventilated rooms, knowledge of airflow in the animal occupied zone (AOZ) with
Accepted 16 June 2017 airflow blockage by the animals and knowledge of thermal convection effects are still
limited. In this study, five commonly used two-equation turbulence models (standard ke3 ,
realisable ke3 , RNG ke3 , standard keu, and SST keu) were applied to a CFD model of the
Keywords: airflow in a mechanically ventilated pig room with animals housed in pens. In addition to
CFD modelling turbulence models, the effects of non-conformal meshing which combines several
Livestock housing ventilation computational sub-domains and connects the boundary of domains using interfaces, were
Airflow in AOZ tested. The effect of mesh ratio on the interface (i.e. ratio of the grid number of the up and
Mesh configuration down interface) was studied based on fully structural hexahedral mesh (SH). The investi-
gation of mesh type effect was conducted by application of an unstructured tetrahedral
mesh (UT) in the AOZ and SH in the rest of the domain. The results showed that the choice
of turbulence model did not have a strong effect on the main airflow pattern except for the
RNG ke3 model. The tested ratios of resolution at interfaces were also found not to strongly
impact on the predicted airflow distributions. The use of UT in the AOZ sub domain also
provided acceptable results. It was concluded that non-conformal meshes are a feasible
alternative for animal buildings with complex geometries to maintain affordable grid
numbers and also reduce the difficulties in mesh generation.
© 2017 IAgrE. Published by Elsevier Ltd. All rights reserved.

animals. Therefore, to improve the indoor environment of

1. Introduction animal house, the airflow distribution should be carefully
studied.
It has been found that flow fields in pig houses are crucial for In studies of ventilation systems, computational fluid dy-
creating a thermally comfortable and healthy environment for namics (CFD) has been shown to be superior to conventional

* Corresponding author.
E-mail address: guoqiang.zhang@agrsci.dk (G. Zhang).
http://dx.doi.org/10.1016/j.biosystemseng.2017.06.012
1537-5110/© 2017 IAgrE. Published by Elsevier Ltd. All rights reserved.
38 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

models with realistic geometry have been used in some CFD

Nomenclature studies of indoor environment inside buildings (Bjerg, Zhang,
& Kai, 2008; Gebremedhin & Wu, 2005; Seo et al., 2012).
Abbreviations Turbulence models are generally considered to be impor-
AOZ animal occupied zone tant on the results due to the approximations of the turbu-
CFD computational fluid dynamics lence in varied scales (Rong et al., 2016; Sorensen & Nielsen,
LES large eddy simulation methods 2003). Although some studies have been conducted on the
RANS Reynolds-averaged NaviereStokes method evaluation of turbulence model in simulation of airflow in
RNG re-normalisation group ventilated rooms (Norton, Grant, Fallon, & Sun, 2010; Rong
SH structural hexahedral mesh et al., 2016; Shen, Zhang, & Bjerg, 2012; Zhang, Zhai, Zhang,
SST shear stress transport & Chen, 2007), knowledge of the airflow affected by animals
UT unstructured tetrahedral mesh blocking the airflow and thermal convection is still limited.
Among turbulence models, generally there are Reynolds-
Symbols
averaged NaviereStokes (RANS) method and large eddy
r fluid density
simulation (LES) methods. RANS method is widely used in
ui velocity component
building ventilation research due to the sufficient high quality
f flow variables
and low requirements for computational power (Zhang et al.,
Gf,eff effective diffusion coefficient
2007). Therefore, five two equation turbulence models (stan-
Sf source term
dard ke3 , realisable ke3 , Re-Normalisation Group (RNG) ke3 ,
p fluid pressure
standard keu, and Shear Stress Transport (SST) keu) which
H fluid enthalpy
were commonly used and embedded in most CFD codes were
k turbulent kinetic energy
evaluated on a ventilated pig room containing animal models
3 turbulent dissipation rate
in this study.
u turbulent specific dissipation rate
Gridding is another important factor for simulation results.
mt eddy viscosity
In general, it is the hexahedral, tetrahedral, and hybrid
Gk generation of turbulence kinetic energy due to
meshes are the commonly applied in animal building simu-
the mean velocity gradients
lation. The hexahedral mesh was the first type of mesh
Gb generation of turbulence kinetic energy due to
adopted in the CFD studies and it is commonly considered to
buoyancy
have more accurate results compared to tetrahedral meshes
R3 addition term in RNG ke3 models concerning
(Duan et al., 2015; Hefny & Ooka, 2009). Hexahedral meshes
the strain rate
are mainly adopted for relatively simple geometries. For a
S rate of the strain
complex geometry, such as an animal, it is almost impossible
C3 1, C3 2, C3 3, Cm coefficients of the turbulence models
to obtain suitable meshes using hexahedral grids. Tetrahedral
Gu generation of u
meshes, however, are useful for boundaries with complex
Yk dissipation of k due to turbulence
surfaces. In addition, using most modern grid generation
Yu dissipation of keu due to turbulence
codes, they can be generated automatically. However it is
ak inverse effective Prandtl number for k
known that the prediction accuracy of the tetrahedral meshes
a3 inverse effective Prandtl numbers for 3
may be lower than hexahedral using the same grid number.
sk turbulent Prandtl numbers for k
To maintain the same accuracy, the number of cells of a
s3 turbulent Prandtl numbers for k and 3
tetrahedral mesh need to be larger than for a hexahedral
n measurement number in the evaluation
mesh with the same size cells, which means that the
y experimental results
computation time can be longer (Duan et al., 2015; Juretic &
b
y simulated values from CFD
Gosman, 2010; Yu, Yu, Sun, & Tao, 2012). Conformal hybrid
y average value of experimental data on
meshes, as a combination of hexahedral and tetrahedral
measurement line
meshes, exhibit the advantages of both these kinds of mesh,
and they have been applied to study air distribution in one
animal house (Seo et al., 2012). However, this kind of hybrid
mesh is difficult to generate, particularly with grids that have
experimental methods. The main advantages of CFD are that high aspect ratios.
it allows full control on the influencing factors and provides Non-conformal meshes, which connect adjacent domain
universal data in the computational domain with relatively boundaries using an interface, are used when complex ge-
low costs of time and expense. There is therefore an ometry is involved in the room space. For instance, a tetra-
increasing tendency to use CFD methods to study of flow hedral mesh can be used in the animal occupied zone (AOZ)
distributions in agricultural buildings (Bartzanas, Boulard, & where the geometry is complex and a hexahedral mesh can be
Kittas, 2002; Benni, Tassinari, Sonora, Barbaresi, & used in the rest of the room domain. When hexahedral and
Torreggiani, 2016; Bjerg, Svidt, Zhang, Morsing, & Johnsen, tetrahedral meshes are adopted in different domains, they
2002; De Rosis, Barbaresi, Torreggiani, Benni, & Tassinari, can be regarded as form of hybrid mesh but without requiring
2014; Lee et al., 2013; Norton, Sun, Grant, Fallon, & Dodd, the full agreement of nodes at the interface boundary.
2007; Rong, Nielsen, Bjerg, & Zhang, 2016; Wu, Zhang, Bjerg, Therefore, they are relatively easy to generate compared to
& Nielsen, 2012). To better model livestock buildings, animal the aforementioned hybrid mesh. However, since the nodes
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2
do not fully agree at the boundary, iteration data needs real-
location at the boundary and the ratio of resolution at the
interface boundary can impact on the data transfer process.
On the basis of the state-of-the-art of CFD simulations of
air distribution in pig houses, this investigation conducted a
systematic evaluation of turbulence model and mesh types.
The objectives were: (1) to identify a suitable turbulence model
for air velocity in a mechanical ventilated pig house contain-
ing heated model pigs; (2) to test the impact of ratio of reso-
lution on the interface boundary on the air speed distribution;
(3) to compare the effects of hexahedral and tetrahedral
meshes modelled in the AOZ on the prediction accuracy of air
velocity when using a non-conformal mesh. Following these
evaluations, a case containing complex-shaped pig models as
an example of using this modelling method to simulate a
realistic problem was investigated.
2. Material and method
2.1. Experiment setup
A room in the experimental pig house in AU Foulum, Denmark
with dimensions of 5.70
4.88
2.67 m (L
W
H), was used
for this experiment. The dimensions and layout of the room
followed a section of typical commercial Danish pig produc-
tion unit. The room was equally divided into two pens
(4.8
2.44 m, L
W) by a partition wall (heights 0.78 m and
1.03 m in the two parts).
Each pen had a two thirds fully slatted floor and a one third
drain floor. There was a drinking water trough on each side
wall, and one shared feeder on the partition wall in each pen.
In front of the pens, there was a 0.9 m wide inspection alley.
Underneath the floor, a 0.69 m deep manure pit was built for
each pen. There were four artificial animals placed inside the
room, in each pen two the artificial animals were placed in the
middle of the slatted floor and drained floor. The artificial
animals were made of iron in a hollow cylindrical shape. Four
100 W incandescent bulbs were installed inside each of the
artificial animals to generate the heat. The experimental
measurements were conducted under non-isothermal con-
ditions with ventilation rate of 2533 m3 h1 operated in
negative pressure and two bottom hinged flap wall inlets.
Before each measurement, the ventilation system was oper-
ated for around 3 h to reach steady state condition in the
room. During the experiments, the ventilation rate was
measured by a free running propeller and recorded by a
climate control system (Vengsys, Roslev, Denmark) each
minute. Using 8 omnidirectional air velocity transducers
(model 8475, TSI, Minnesota, USA) simultaneously together
using a CR 1000 data logger (Campbell Scientific Ltd, Lough-
borough, UK), the air speeds were measured every 0.2 s and
averaged every 1.0 s for a measurement period of 60 min at
each measuring points. The measurement was conducted on
four vertical lines in the 99 centre of the pig pen 1 (Fig. 1) at 8
heights (0.35 m, 0.7 m, 1.1 m, 1.5 m, 1.8 m, 2.0 m, 2.2 m, and
2.4 m).
Air velocity above the floor was measured but air velocity
below floor was not considered as it was assumed to be
negligible and also not the target area for this research. Room
39
temperature and humidity were continuously measured by
the sensors fitted in the room and recorded each minute by
the climate control. The surface temperature of wall and
artificial pigs was measured using T type thermocouples and a
data logger (Eltek Ltd, Cambridge, UK) every 5 min during the
experiment. Temperature profile was not studied as a
parameter in this study because the focus was on the config-
uration of mechanical ventilation, and the ventilation rate
was relatively high. In these conditions the indoor air can be
assumed to be fully mixed and with low temperature
gradients.
2.2. Computational fluid dynamics modelling
2.2.1. Calculation domain and grids
All the studies reported here were based on the same geom-
etry of the pig house as the field measurements. In order to
reduce the complexity of modelling in CFD, proper approxi-
mations were made: the light and drink system were not
included in the CFD modelling, and the feeders were modelled
as cuboid. Since the study has two main objectives, to test the
turbulence models and mesh resolution and type, the treat-
ments of the calculation domain and mesh generation were
different in these two parts. In the study of the impact of
turbulence models, the calculation domain was regarded as a
whole, and a structured mesh was adopted for the grids. To
reduce the impact from the mesh and ensure the quality of
simulation, a grid independence test was conducted first. The
slots in the slatted floor increased the complexity of the mesh.
There were 168 slots of 10 mm in width in the floor and
therefore the mesh size of the slots cannot be larger than
10 mm. The cylindrical animal models also increased the
complexity of mesh, and the skewness could be higher when
the cell size around the animal models was large. Therefore,
low Reynold number modelling approach was used to mesh
the grids around the pig models to provide a better mesh
quality around the animal models and consequently a better
simulation result on the airflow around animals. With all
these considerations, the coarsest mesh in the grids inde-
pendent test had 3.25 million cells. For the mesh in medium
and dense levels, the cell numbers were 4.25 million and 6.10
million, respectively. Although more dense meshes have been
used in CFD studies, for this application this density was
considered adequate.
Figure 2 shows the velocity profiles in four vertical
measured lines in different mesh resolutions using standard
ke3 model. No significant difference was found among the
three resolutions on the testing heights above floor. The mesh
at the medium was used for evaluating the turbulence
models. The coarse mesh was not selected since the slots only
contained one cell on the width dimension of the slots, which
was consider too coarse to generate accurate data beneath
floor, although, as mentioned earlier the velocity under floor
was not the focus of this study.
In the study of the impact of mesh interface resolution and
type, the entire computational domain was divided into three
parts (Fig. 3), namely, above the AOZ domain, in the AOZ
domain, and below the AOZ domain. It should be noted that
the reason for separating the domains in this way was to
generate meshes in a convenient way. The zone where AOZ
40 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

Fig. 1 e The sketch of (a) the experimental pig house and (b) the measurement points for the air speed in horizontal plane, all
units in metre.

domain located may not exactly be the AOZ that occurs in equation RANS turbulence models were compared to the
reality. The grids in these three subdomains were generated experimental measurements. The turbulence models used in
separately. The two adjacent domains were connected by in- the evaluation included the standard ke3 model, the realisable
terfaces named as “interface up” and “interface down”. These ke3 model, the RNG ke3 model, the standard keu model, and
three calculation domains were meshed with different grids the SST keu model. These two-equation RANS turbulence
resolutions and different mesh types were compared on the models were chosen because they were the models used most
AOZ domain. Detailed information of cells numbers and mesh often in both research and design applications for room
types is shown in Table 1 where they are noted as cases 1e4. ventilation (Shen et al., 2012).
Grids plot for the interface boundary is also provided in Fig. 4. The governing equations for all these five turbulence
models in a steady state can be written in a general form with
2.2.2. Turbulence models a convection term on the left and a diffusion term and a
Turbulence models are generally considered to have impor- source term on the right:
tant influence on the results of CFD due to the approximations  
of the turbulence in varied scales. To determine the most vf v vf
ruj ¼ Gf;eff þ Sf (1)
suitable turbulence model for use in the study, five two- vxj vxj vxj

Fig. 2 e Grid independence test for calculation velocity profiles on the four vertical lines.
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2 41

Fig. 3 e The divisions of the computational domain. The two adjacent domains are connected by interface boundaries, all
units in metre.

Table 1 e Combination of the three parts of computational domain in different cases.

The test of interface resolution The test of mesh type
Case 1 Case 2 Case 3 Case 4
Up AOZ domain Hexahedral Hexahedral Hexahedral Hexahedral
Mesh type/cells number 1850k 1233k 377k 377k
AOZ domain Hexahedral Hexahedral Hexahedral Tetrahedral
Mesh type/cells number 1534k 1534k 1534k 1397k
Down AOZ domain Hexahedral Hexahedral Hexahedral Hexahedral
Mesh type/cells number 868k 868k 868k 868k
Ratio of grids number on interface up 1:1 1:1.5 1:5 1:2.3
Ratio of grids number on interface down 1:1 1:1 1:1 1:1.4

Note: The cell numbers in the computational domain are the cell numbers of volume domain. And the grids ratio is defined as the ratio of grids
number of up interface to that of down interface for two adjacent faces on the interface boundary.

Fig. 4 e Grids plot on the interface boundaries of the four cases with different mesh resolutions and types. The red line in the
plot is the interface boundary. The cells above the upper red line belong to the above AOZ domain. The cells between the
two red lines belong to the AOZ domain. The cells below the lower red line belong to the below AOZ domain.
42 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

Table 2 e Key mathematical expressions of the five turbulence models selected in this study (Ansys, 2013).
f Gf;eff Sf Key constants and coefficients
Sharing variables ui m þ mt vp=vxi  rbgi ðH  H0 Þ=Cp
m mt
for (1)e(5) T sT þ sT;t SH
pffiffiffiffiffiffiffiffiffiffiffiffiffi
m þ smkt mt vT
2
(1) Standard ke3 k Gk þ Gb  r3 mt ¼ rCm k3 ; Gk ¼ mt S2 ; S ¼ 2Sij Sij ; Gb ¼ bgi Pr t vxi
; C3 1 ¼ 1:44; C3 2 ¼ 1:92;
m þ smt
2
3
3
C3 1 k3 ðGk þ C3 3 Gb Þ  C3 2 r 3k Cm ¼ 0:09; sk ¼ 1:0; s3 ¼ 1:3
mt ¼ rCm k3 ; R3 ¼ Cm h1þbh
ð1h=h0 Þ 3 2
2 3
(2) RNG ke3 k ak ðm þ mt Þ Gk þ Gb  r3 3 k ; C3 1 ¼ 1:42; C3 2 ¼ 1:68; Cm ¼ 0:0845;
2
3 a3 ðm þ mt Þ C3 1 k3 ðGk þ C3 3 Gb Þ  C3 2 r 3k  R3 ak ¼ a3 z1:393  
pffiffiffiffiffiffiffiffiffiffiffiffiffi
m þ smkt h
2
(3) Realisable ke3 k Gk þ Gb  r3 mt ¼ rCm k3 ; C1 ¼ max 0:43; hþ5 ; h ¼ S k3 ; S ¼ 2Sij Sij ; Cm ¼ 1
* ; C3 1 ¼ 1:44;
A0 þAs kU
m þ smt rC1 S3  rC3 2 kþ3pffiffiffi ffi þ C3 1 k3 C3 3 Gb
2 3
3
3 n3
C3 2 ¼ 1:9; sk ¼ 1:0; s3 ¼ 1:2
(4) Standard keu k m þ smkt G k  Yk mt ¼ a* rk u *
u ; Gu ¼ a k Gk ; Yk ¼ rb fb* ku; Yu ¼ rbfb u ; sk ¼ 2:0; su ¼ 2:0
2

u m þ smut G u  Yu
(5) SST keu k m þ smkt ~ k  Yk
G mt ¼ rk h1 ~ k ¼ minðGk ; 10rb* kuÞ; Gu ¼ a Gk ; sk ¼
i; G 1
u 1 SF2
nt F1 =sk;1 þð1F1 Þ=sk;2 ;
;
u m þ smut G u  Yu þ D u max
a * a1 u

su ¼ F1 =su;1 þð1F
1
1 Þ=su;2
.

Note: Gk is the generation of turbulence kinetic energy due to the mean velocity gradients, Gb represents the generation of turbulence kinetic
energy due to buoyancy, R3 is the addition term in RNG ke3 models concerning the strain rate, Gu generation of u, Yk and Yu are the dissipation of
k and u due to turbulence, ak and a3 are inverse effective Prandtl numbers, sk and s3 are the turbulent Prandtl numbers for k and 3 , respectively.

where f is the flow variables, e.g., velocity, enthalpy, and
turbulence parameters, Gf,eff is the effective diffusion coeffi-
cient, and Sf is the source term. The key mathematical ex-
pressions of the five turbulence models selected in this study
are presented in Table 2. Due to the space limitation, this
paper does not provide a detailed description of all the pa-
rameters included in these models. In this study, default pa-
rameters were used for all the simulations and the reader is
referred to the user theory guide of the commercial codes (e.g.
Fluent (Ansys Inc., Canonsburg, PA, USA)). The standard ke3
model uses k to represent the turbulence kinetic energy to
account for energy transport, and 3 as dissipation rate to
represents the turbulence diffusion (Launder & Spalding,
1974). This model is robust and widely used for industrial
application although its performance is limited in applica-
tions with swirling and rotating flows and strong separation
flows. The realisable ke3 model is a modified form of the
standard ke3 model. The dissipation of turbulent kinetic en-
ergy, in standard ke3 is described using the coefficient, Cm
which is a constant. In the realisable ke3 model it is expressed
as a function of flow properties (Shih, Liou, Shabbir, Yang, &
Zhu, 1995). The 3 term in realisable ke3 model is quite
different from those in both the standard and RNG ke3 models
and is formulated based on a dynamic equation of the mean-
square vorticity fluctuation. It can be seen in Table 2 that the
production term is not included in realisable ke3 model. It is
believed that the modified transport equation can better
describe the spectral energy transfer (Anonymous, 2013a,
2013b). Due to the revisions on calculations of turbulent vis-
cosity and dissipation rate, this model can provide appro-
priate predictions for complex shear flows involving moderate
swirls, vortices and locally transitional flows such as occur in
room ventilation. The model constants of the RNG ke3 model
are derived from RNG theory instead of empirical approxi-
mation (Yakhot & Orszag, 1986). The RNG model has been
suggested by a few researchers for room ventilation of build-
ings occupied by humans (Zhang et al., 2007). Comparing to
realisable ke3 model, this model is more difficult to converge.
The standard keu model is based on model transport equa-
tions for the turbulence kinetic energy (k) and the specific
dissipation rate (u), and the accuracy of prediction for free
shear flows is improved in this model. It improves the results
for boundary layers under adverse pressure gradients
compared to ke3 models (Wilcox, 1988). However, this model
is sensitive to inlet boundary conditions, which decrease its
robustness. The standard keu model used in Fluent codes
(Ansys Inc., Canonsburg, PA, USA) is reported to improve the
accuracy of the model for predicting free shear flows by add-
ing production terms to both the k and u equations
(Anonymous, 2013a, 2013b). The SST keu model combines the
advantages of standard keu model in simulating the near-
wall region and the robustness of ke3 model in simulating
the freestream in the far field (Menter, 1994). A blending
function is multiplied to the standard keu model and the
transformed keu model and then both equations are summed
accordingly. The standard keu model is activated by setting
the blending function to be 1.0 in the near-wall region and
zero away from the solid surface to activate ke3 model there.
The standard keu models requires a fine mesh in the near-
wall regions. However, in practice, a very fine mesh on all
the surfaces is a challenge for mesh generation and requires a
very large computational capacity. Therefore, to overcome the
difficulty in mesh generation and to reduce the grids number,
a low Reynolds modelling approach was adopted on the region
where the artificial animals located. Fine mesh was generated
on the surface of the artificial animals, giving a Yþ (non-
dimensional wall distance) on the surface around 1 and
providing better predictions of the airflow near the animals.
2.2.3. Data transfer between interfaces
Interface boundaries are generally used to calculate the flux
across the non-conformal boundary in which hanging nodes
may exist. In this study, the interface boundary connected
calculation domains were all considered as fluid zones, and an
interior zone was generated on the interface where the mesh
on the two domains overlapped and allowed the flux transfer
across the domain boundaries. Data was transferred across
the interface boundaries through a virtually generated inter-
face zone on the intersection area between the two adjacent
boundaries (Anonymous, 2013a, 2013b). Figure 5 shows an
example of data treatment at the interface. When data is
transfer from domain 1 to domain 2, data on the interface
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2 43

Domain 1 Table 3 e Boundary conditions.

I II
A B
a b c d
D E F G
III IV
Domain 2
Fig. 5 e Example on the data transfer on the interface
boundary. AB and BC represent the faces of cells I and II in
domain 1; DE, EF, and FG represent the faces of cells III, IV,
and V in domain 2. Domain 1 and Domain 2 were
connected by interface boundary and virtual interface zone
were generated. ab, bc, cd, and de represent the virtually
generated face grids in interface zone.
Inlet Pressure inlet (0 Pa), temperature of
19.8  C
C Outlet Mass flow inlet with reversed flow in
Interface boundary 1 0.862 kg s1 (equal to 2533 m3 h1)
e
Interface zone Walls and floor Non-slip wall, with the wall
boundary on an average
Interface boundary 2 temperature of 20.8  C and floor in
19.7  C
V
Pig model The two pigs closing to inlet, surface
temperature of 36.2  C;
The two pigs closing to the working
corridor, surface temperature of
33.3  C
2.2.5. Comparison between experiment and simulation
results
The coefficient of variation of root-mean-square error, CV
(RSME), was adopted to analyse the accuracy of the CFD pre-
dictions using the various turbulence models and mesh
schemes (Ntinas, Shen, Wang, & Zhang, 2017).

boundary, e.g., face AB, are distributed into faces, named ab sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
1 ðy  y Þ
b 2
and bc, in the new generated interface zone, while the flux of RSME ¼ (2)
different parameters on boundary face BC, is allocated into n
face cd and face de, in the interface zone. Then the data on the
RSME
interface zone is transferred to the corresponding cells CVðRSMEÞ ¼ (3)
y
directly, for instance, the data transfer from cells I and II to cell
IV was made by face bc and face cd, but not by face EF. Since where n is the number of measurement points in the evalua-
the data are reallocated on the interface, the accuracy of the b is the simulated values from
tion; y is the experimental data; y
simulation may depend on the mesh ratio at the interface CFD; y is the average value of experimental data of the n
between two adjacent domains. In this study, the mesh ratio measurement points on each measurement line.
is defined as the ratio of number of grids on the up interface to
that of down interface for two adjacent faces on the interface
boundary. The interface ratio is represented in a form as 1:x 3. Results and comparison with experiment
(where x is a positive rational number from calculation). data

2.2.4. Boundary condition and numerical methods 3.1. Effect of turbulence model on the results
To represent the experimental pig room, proper boundary
conditions were necessary. The wall inlet in the physical The experimental results of the measured air velocity on the
model was defined as pressure inlet. And the exhaust of the four vertical lines are summarised in Table 4. Figure 6 shows
ventilated room was regarded as an outlet to ensure the distributions of the experimentally measured and numeri-
control of the ventilation rate. All the surfaces of the wall, floor cally predicted velocity magnitudes on the four vertical mea-
and pig models were set as non-slip walls with no roughness, surement lines in pen 1. Generally, good agreement between
and an enhanced wall treatment (a blended wall model in experimental and numerical values was found for most of the
which two separate wall models were blended by use of a turbulence models used in this test, except for the RNG ke3
damping function) was activated in the three ke3 models. The
temperatures of the inlet air and the wall surface were aver-
aged from the measurement. The boundary conditions were Table 4 e The air speeds at different heights on the four
summarised in Table 3. measurement lines.
Based on the commercial CFD software Fluent 15.0 (Ansys Heights (m) Air speeds, mean value ± standard deviation
Inc., Canonsburg, PA, USA), the simulation used the finite (m s1)
volume differencing scheme in Steady RANS method and Line 1 Line 2 Line 3 Line 4
SIMPLE (Semi Implicit Method for Pressure-Linked Equations)
0.35 0.38 ± 0.10 0.34 ± 0.08 0.19 ± 0.06 0.43 ± 0.12
numerical algorithm. Second-order upwind schemes were
0.7 0.33 ± 0.08 0.31 ± 0.08 0.45 ± 0.10 0.60 ± 0.13
used for both the convection and the viscous terms of the 1.1 0.35 ± 0.09 0.44 ± 0.10 0.39 ± 0.12 0.53 ± 0.13
governing equations. The criteria of convergence were set 1.5 0.36 ± 0.09 0.28 ± 0.10 0.20 ± 0.10 0.33 ± 0.11
such that the residuals for u, v, w, k, 3 less than 103, and for 1.8 0.30 ± 0.07 0.20 ± 0.07 0.20 ± 0.07 0.25 ± 0.09
the residual for energy were less than 106. The net flow rate 2 0.57 ± 0.15 0.44 ± 0.15 0.44 ± 0.16 0.38 ± 0.16
of the geometry, which was monitored as a supplementary 2.2 2.39 ± 0.17 1.38 ± 0.21 1.00 ± 0.21 0.66 ± 0.19
2.4 0.64 ± 0.11 1.46 ± 0.18 1.40 ± 0.15 1.04 ± 0.16
convergence criterion, was set to be less than 104.
44 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2
Fig. 6 e Air speed profiles of the four vertical lines by simulation using varied turbulence model and measurement in non-
isothermal condition.
model, which had higher deviation than others although its
trends were similar to the other models and the experimental
results. The CV (RSME) values were 0.29, 0.27, 0.25, and 0.28 for
standard ke3 model, realisable ke3 model, standard keu
model, and SST keu model, respectively. The CV (RSME) was
0.39 for RNG ke3 model.
The contours of velocity magnitude on the middle plane in
the tested pen are shown in Fig. 7. Similar to the velocity
profile, the RNG ke3 model generated a different airflow
pattern compared with the other turbulence models all of
which had a similar flow pattern above floor. Similar results
were also found on the horizontal contours of velocity at
height of 0.35 m (Fig. 8).
In general, it can be concluded that most of the CFD
models provided a reasonable description of the airflow
pattern in the room. Using the experimental data as evalu-
ation reference, it is evident that RNG ke3 model did not
perform well in this specific ventilation configuration. The
results are different from some research that suggests RNG
ke3 models have high performance under different ventila-
tion conditions (Huawei, Lingying, & Yuanhui, 2007; Liu et al.,
2013). However, our results agree with the results by Rong
et al. (2016) in a study conducted under similar conditions.
The discrepancies in the performance of turbulence models
may come from the influence of geometry and ventilation
configurations. There are other evaluations of the current
five turbulence models but conducted with different pur-
poses. Shen et al. (2012) concluded that the SST keu turbu-
lence model had the best performance for predicting
ventilation rate in a naturally ventilated house. For simula-
tion of the external airflow around various designs of build-
ing roof, the SST keu model was found to be the most
suitable model for the prediction of turbulent kinetic energy
and the RNG ke3 but Realisable ke3 model performed better
for the prediction of velocity (Ntinas et al., 2017). It appears
that the performance of turbulence models on the prediction
of airflows could be case dependent. Therefore, a suitability
evaluation on turbulence model test is recommended in the
given simulation if no such evaluation has previously been
done for similar case.
The geometry of the case considered here was complex,
and a robust turbulence model is required for the iterations to
reach convergence. Since the standard ke3 is the most widely
used in engineering application, and is considered to be very
robust, it was chosen for the study on the impact of interface
resolutions and mesh types.
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2 45

Fig. 7 e Contours of the velocity magnitude on the middle plane in pen 1 based on different turbulence model.

3.2. Impact from the resolution of interface
Figure 9 shows various distributions of experimental and
numerically predicted velocity magnitudes along the sym-
metrical plane in the pen 1 when the resolution of the inter-
face was varied. The data in each domain were transferred
continuously. Cases 1e3 (Table 1) tested in the study were
found sharing the same tendency with the experimental re-
sults on the four measured vertical lines. The differences be-
tween these three cases were relatively small, the CV (RSME)
of these three ratios were at similar level, i.e., 0.20, 0.27, and
0.22 for cases 1e3, respectively. The prediction of velocity
magnitude in the vertical plane in the centre of pen 1 (Fig. 10)
showed that the information was transferred through the
interface smoothly. Comparing the contours between the
cases with varying ratio of interface resolutions, the differ-
ence was not significant above the floor. Same results were
also shown on the horizontal contours of velocity at height of
0.35 m (Fig. 11). Data is reallocated on virtual interface zone
between the two adjacent interface boundaries during the
transportation through calculation domains. Therefore, the
resolution on the interface boundary should have impact on
the data transfer, since detailed information can be lost in the
process of data redistribution when the ratio of the interface
46 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2
Fig. 8 e Contours of the velocity magnitude on the horizontal plane at height of 0.35 m based on different turbulence models.
boundary is too high. Interface boundaries with non-
conformal mesh have been adopted by many researchers (Li,
Rong, & Zhang, 2016; Li, Rong, Zong, & Zhang, 2016), and
good simulation results have been achieved generally. It is
always preferred that the ratio between the two adjacent
interface boundaries is close to 1. The interface boundary ra-
tios in our simulations ranged from 1:1 to 1:5. Within this
range, no significant differences were found. The results
provided reference for the grid size ratio on the interface
boundary in this type of study. Although the interface grid size
ratio was found not to be significant in this study, it is still
notable that interface should not be used where sudden flow
transition may occur, since the data may be affected by the
interface during data transfer (Anonymous, 2013a, 2013b). The
use of the interface, on the other hand, provides the possibility
to mesh the calculation domains in a same geometry with
varied resolution. For instance, the place where detailed in-
formation is needed can be meshed in denser meshes, while
the rest of the domain could be meshed using a coarse mesh,
and connected to each other by the interface. This approach
also reduces iteration time since the total number of cells is
reduced when the interface is adopted. In this study, the it-
erations were conducted using 4 nodes and a parallel pro-
cessor on a PC with i7292 4770 CPU and 16 GB memory, the
iteration times for the three cases were 38.3 h, 30.5 h, and
22.1 h, respectively, with the ratios on the interface boundary
increased when the grids was fixed number in below AOZ
domain.
3.3. Impact from the mesh type in AOZ
Figure 12 compares the measured airflow distributions with
simulated airflow patterns at a cross section in the symmet-
rical plane in the pen 1. With the structured mesh (Case 1), all
three sub-domains were in structured meshes and the inter-
face ratio in the AOZ to above AOZ and to below AOZ were all
1:1. In unstructured mesh (Case 4), only in the AOZ domain was
unstructured gridding applied; below AOZ and above the AOZ a
structured mesh remained. The ratios of AOZ interfaces to
above AOZ and below AOZ domains were 1:2.3 and 1:1.4,
respectively. The difference between these two different mesh
types on predicting the airflow profile in the four tested vertical
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2 47

Fig. 9 e Air speed profiles of the four vertical lines by simulation using varied interface resolution ratios and measurement.
Case 1 represents the case with the up interface ratio as 1:1; Case 2 represents the case with up interface ratio as 1:1.5; Case
3 represents the case with up interface ratio as 1:5. The down interface ratios are all 1:1 in the 3 cases.

lines was not significant, and both of them showed good

agreement with the measured results, with the CV (RSME) at 4. Further investigation using pig models in
0.22 and 0.23 for Case 1 and Case 4, respectively. In Fig. 10, the similar body shapes of pigs
velocity magnitude contours of the centre plane in pen 1 are
shown. Compared with other cases using interface, the airflow Pigs in pens have more complex geometry than cylinders and
patterns were found to be similar. The results were also also have random distributions under realistic conditions. All
applicable on the horizontal contours of velocity at height of these will increase the difficulties in mesh generation if exact
0.35 m (Fig. 11). It has been reported that hexahedral meshes building geometries need to be used. On the basis of results
can provide higher accuracy in predicting the airflow distri- from the theoretical study in Section 3, an investigation using
bution (Duan et al., 2015; Hefny & Ooka, 2009; Juretic & Gosman, real pig geometries was conducted.
2010; Yu et al., 2012). However, no significant performance
increase was observed in this study for hexahedral meshes. A
4.1. Model geometry and mesh
possible reason can be the resolution of the AOZ domain was
In this part of the investigation, the room space was the same
high and that provided a greater prediction accuracy. The un-
as that earlier but the geometry was further subdivided
structured tetrahedral mesh (UT) was widely used when the
separated into three domains (Fig. 13). The difference is that
geometry of model is very complex, and normally it generated
simple cylindrical models in the AOZ domain were replaced
more cells than the hexahedral mesh based on same geometry.
with pig models with a similar shape to pig. 16 pigs in 50 kg
Using interface to combine the advantages of both hexahedral
mass were modelled in each pen. The mass was controlled by
and tetrahedral mesh can reduce the difficulty in modelling the
surface area of model based on the relationship of Brody,
complex geometry in agricultural research, e.g., the cases
Comfort, and Matthews (1928). The distribution of the pigs
which contain slatted floor or animal models.
48 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

Fig. 10 e Contours of the velocity magnitude on the middle plane in pen 1 using interfaces. Case 1 represents the case with
the up interface ratio as 1:1; Case 2 represents the case with up interface ratio as 1:1.5; Case 3 represents the case with up
interface ratio as 1:5. The down interface ratios are all 1:1 in Cases 1e3. Case 4 represents the case combines both structured
and unstructured mesh. The up interface ratio is 1:2.3 and the down interface ratio is 1:1.4.

was that referred to in literature (Bjerg, Zhang, & Kai, 2011).
A UT was used to generate the grids in the AOZ domain since it
is almost impossible to apply a structured hexahedral mesh in
such complex geometry.
4.2. Boundary condition and numerical method
The boundary conditions were similar in the aforementioned
cases with the cylindrical models. The changes mainly
happen with specific values. The ventilation rate was set as
3200 m3 h1. Since the ventilation rate was relatively high, the
heat transfer between the indoor air and wall surface and floor
surface was assumed to be very small in a steady condition,
and therefore, the surfaces of the floor and the wall were
treated as being adiabatic. For the temperature of pig surface,
a constant temperature of 39  C was selected as boundary
condition.
For the numerical method, the calculation was calculated
using the same numerical treatment as mentioned in Section
2.2.3.
4.3. Results and discussion for the case study
Although the number of pig models was larger than the cy-
lindrical models used in the validation study, the physical
models for these two cases were similar. It has been reported
that complex geometry can allow airflows to be fully devel-
oped more easily (Tian, He, & Tao, 2015). Therefore, the
example case can be validated by the case with simple cylin-
drical models.
The contour of velocity magnitude in the symmetry plan in
pen 1 is illustrated in Fig. 14. As with the cases studied above,
there was a large vortex above the AOZ because the me-
chanical ventilation from the inlet dominated the airflow. The
jet from the inlet provided the energy for the movement of the
air inside the pig room. In the AOZ, the airspeed was found to
be very low. The average velocity at a height of 0.35 m was
only 0.25 m s1. The bulk of the pig models blocked the
movement of air. It showed that the airflow passing across the
back of pig can be the main force for the convective term via
the heat loss from pig. By modelling the relationship between
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2 49

Fig. 11 e Contours of the velocity magnitude on the horizontal plane at height of 0.35 m using interfaces. Case 1 represents
the case with the up interface ratio as 1:1; Case 2 represents the case with up interface ratio as 1:1.5; Case 3 represents the
case with up interface ratio as 1:5. The down interface ratios are all 1:1 in Cases 1e3. Case 4 represents the case combines
both structured and unstructured mesh. The up interface ratio is 1:2.3 and the down interface ratio is 1:1.4.

air velocity at a certain height in the pig house and convective
heat transfer, a ventilation control algorithm can be devel-
oped based on this relationship. In the simulation, all pig
models were placed in a half standing and lying position, fully
lying or standing postures were not modelled due to the
complexity of geometry. However, under real conditions, the
pig may spend 90% time lying on the ground (Ekkel, Spoolder,
Hulsegge, & Hopster, 2003). The model in this study cannot be
regarded as a strict description for the real case. The purpose
of the example here is to demonstrate how the interface and
unstructured meshes work to make the complex geometry
meshing possible. Detailed model and meshing methods
depend on the research objectives under real conditions.
5. Limitations of current study and research
perspectives
This research contains some limitations. All simulations were
conducted under a steady-state condition, while real
conditions may not be steady; ventilation rates can fluctuate
and periodic vortices can be generated in the wake of cylin-
ders in airflows. The transient RANS or LES calculation may
provide better predictions for airflow patterns, but their re-
quirements for computational power and time are much
higher compared to the steady-state RANS method. Normally,
the thermal comfort inside the livestock building depends on
the temperature and airflow distribution over a long time span
but not in a transient period. The measurement result in the
validation experiment was averaged from the data for at least
1 h. From this point, the usage of RANS which modelled the
turbulence energy by time averaged method was therefore a
reasonable way to describe the physical model in this study.
As this study focused on a mechanically ventilated animal
house, only airflow distribution was measured and compared
with simulation results. The distributions of temperature,
humidity and ammonia were not measured or presented. This
can be a limitation of this study. However, since mechanical
ventilation dominates the airflow, the indoor air can be
considered at a fully mixed condition at high ventilation rate.
50 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2

Fig. 12 e Comparison of air speed profiles of the four vertical lines modelled by structured and unstructured mesh. Case 1
represents the case with fully structured mesh; and Case 4 represents the case with combined structured and unstructured
mesh, the above AOZ zone and below AOZ zone are structured meshes, while the AOZ zone is unstructured mesh.

Fig. 14 e Contours of the velocity magnitude on the vertical

Fig. 13 e Geometry and the division of the computational plane in centre of pen 1 in the case with pig model in real
domain. shape.
 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2
Also, these parameters are also highly linked with the airflow
distribution; the performance of turbulence models and mesh
scheme on airflow distribution can be sometimes represen-
tative of other parameters, such as temperature, humidity,
and ammonia.
It is also worthwhile pointing out that the streamline and
airflow pattern beneath slatted floor were different from each
other. However, due to the lack of data for airflow pattern
beneath slatted floor in this study, it is difficult to identify the
most suitable method to simulate airflow in pit. However, the
air speed below the slatted floor, was also found to be very low,
i.e. below 0.1 m s1 in the simulations. It should be noted that it
is a challenge to measure/simulate airflows with low airspeeds.
In other relevant studies, the results from the zone beneath
slatted floor produced mean relative errors, between simula-
tion and measurement, ranging from 19% to 6439% (Wu, Zong,
& Zhang, 2013; Zong & Zhang, 2014). Measurement and simu-
lation under low air speed conditions remains a challenge.
6. Conclusions
To reach the research goal for developing a strategy for
modelling indoor climate of livestock housing, this study
evaluated the performance of five commonly used two equa-
tions turbulence models, three cases using different ratios of
interfaces boundary, and two mesh types on the AOZ sub-
domain for predicting airflow distributions inside an experi-
mental pig house. An example which showed how the inter-
face and unstructured mesh work on modelling complex
geometry contain pig models was provided. The conclusions
are as follows:
The selected turbulence models had not strong effect on
the main airflow pattern except the RNG ke3 model which
deviated from other models in predicting the airflow distri-
butions. The suitability of turbulence model can be case
dependent. It is recommended to conduct turbulence model
evaluation if no test has been done on similar cases. Within
ratios of the interface resolutions from 1:1 to 1:5, no significant
difference was found on the airflow distribution. It indicated
that the interface resolution ratio can be as high as 1:5 in a
similar study. The application of interface can also be an
efficient way to reduce computational load and time. The
usage of unstructured mesh in the AOZ sub domain also
provided acceptable results like the structured mesh. It is
concluded that the non-conformal mesh can be applied in the
study of animal building with complex geometry like animals.
Using both interface boundary and unstructured mesh, the
complex geometry of animal house can be separated into
several subdomains and meshed in different grids strategies.
When suitable turbulence model is used, this approach can be
a good way to modelling indoor environment, e.g., airflow
distribution and heat transfer.
Acknowledgement
Thanks to the support from research grant of Innovation Fund
Denmark/Advanced Technology Foundation, New hot climate
ventilation system for poultry and pigs (j.nr. 17-2013-3).
51
references
Anonymous. (2013a). ANSYS FLUENT theory guide (release 15.0).
U.S.A.: ANSYS, Inc.
Anonymous. (2013b). ANSYS fluent user's guide (release 15.0). U.S.A.:
Ansys Inc.
Bartzanas, T., Boulard, T., & Kittas, C. (2002). Numerical
simulation of the airflow and temperature distribution in a
tunnel greenhouse equipped with insect-proof screen in the
openings. Computers and Electronics in Agriculture, 34, 207e221.
Benni, S., Tassinari, P., Sonora, F., Barbaresi, A., & Torreggiani, D.
(2016). Efficacy of greenhouse natural ventilation:
Environmental monitoring and CFD simulations of a study
case. Energy and Buildings, 125, 276e286.
Bjerg, B., Svidt, K., Zhang, G., Morsing, S., & Johnsen, J. O. (2002).
Modeling of air inlets in CFD prediction of airflow in ventilated
animal houses. Computers and Electronics in Agriculture, 34,
223e235.
Bjerg, B., Zhang, G., & Kai, P. (2008). Porous media as boundary
condition for air inlet, slatted floor and animal occupied zone
in numerical simulation of airflow in a pig unit. In Agricultural
and biosystems engineering for a sustainable world. International
conference on agricultural engineering, Hersonissos, Crete, Greece,
23e25 June, 2008. European Society of Agricultural Engineers
(AgEng). pp. OP-1520.
Bjerg, B., Zhang, G., & Kai, P. (2011). CFD analyses of methods to
improve air quality and efficiency of air cleaning in pig
production. In Chemistry, emission control, radioactive pollution
and indoor air quality (pp. 639e654).
Brody, S., Comfort, E., & Matthews, J. S. (1928). Further
investigations on surface area with special reference to its
significance in energy metabolism. Missouri Agricultural
Experiment Station Bulletin, 115.
De Rosis, A., Barbaresi, A., Torreggiani, D., Benni, S., &
Tassinari, P. (2014). Numerical simulations of the airflows in a
wine-aging room: A lattice Boltzmann-immersed boundary
study. Computers and Electronics in Agriculture, 109, 261e270.
Duan, R., Liu, W., Xu, L. Y., Huang, Y., Shen, X., Lin, C., et al. (2015).
Mesh type and number for the CFD simulations of air
distribution in an aircraft cabin. Numerical Heat Transfer Part B-
Fundamentals, 67, 489e506.
Ekkel, E. D., Spoolder, H. A. M., Hulsegge, I., & Hopster, H. (2003).
Lying characteristics as determinants for space requirements
in pigs. Applied Animal Behaviour Science, 80, 19e30.
Gebremedhin, K., & Wu, B. (2005). Simulation of flow field of a
ventilated and occupied animal space with different inlet and
outlet conditions. Journal of Thermal Biology, 30, 343e353.
Hefny, M. M., & Ooka, R. (2009). CFD analysis of pollutant
dispersion around buildings: Effect of cell geometry. Building
and Environment, 44, 1699e1706.
Huawei, S., Lingying, Z., & Yuanhui, Z. (2007). Evaluating RNG ke3
models using PIV data for airflow in animal buildings at
different ventilation rates. ASHRAE Transactions, 113, 358e365.
Juretic, F., & Gosman, A. D. (2010). Error analysis of the finite-
volume method with respect to mesh type. Numerical Heat
Transfer Part B-Fundamentals, 57, 414e439.
Launder, B. E., & Spalding, D. B. (1974). The numerical
computation of turbulent flows. Computer Methods in Applied
Mechanics and Engineering, 3, 269e289.
Lee, I. B., Bitog, J. P. P., Hong, S. W., Seo, I. H., Kwon, K. S.,
Bartzanas, T., et al. (2013). The past, present and future of CFD
for agro-environmental applications. Computers and Electronics
in Agriculture, 93, 168e183.
Li, H., Rong, L., & Zhang, G. (2016). Numerical study on the
convective heat transfer of fattening pigs in mechanical
ventilated pig house. In International conference on agricultural
engineering CIGR e AgEng 2016.
52 b i o s y s t e m s e n g i n e e r i n g 1 6 1 ( 2 0 1 7 ) 3 7 e5 2
Li, H., Rong, L., Zong, C., & Zhang, G. Q. (2016). A numerical
study on forced convective heat transfer of a chicken
(model) in horizontal airflow. Biosystems Engineering, 150,
151e159.
Liu, W., Wen, J. Z., Lin, C. H., Liu, J. J., Long, Z. W., & Chen, Q. Y.
(2013). Evaluation of various categories of turbulence models
for predicting air distribution in an airliner cabin. Building and
Environment, 65, 118e131.
Menter, F. R. (1994). Two-equation eddy-viscosity turbulence
models for engineering applications. AIAA Journal, 32,
1598e1605.
Norton, T., Grant, J., Fallon, R., & Sun, D. W. (2010). Optimising the
ventilation configuration of naturally ventilated livestock
buildings for improved indoor environmental homogeneity.
Building and Environment, 45, 983e995.
Norton, T., Sun, D. W., Grant, J., Fallon, R., & Dodd, V. (2007).
Applications of computational fluid dynamics (CFD) in the
modelling and design of ventilation systems in the
agricultural industry: A review. Bioresource Technology, 98,
2386e2414.
Ntinas, G. K., Shen, X., Wang, Y., & Zhang, G. (2017). Evaluation of
CFD turbulence models for simulating external airflow around
varied building roof with wind tunnel experiment. Building
Simulation, 1e9.
Rong, L., Nielsen, P. V., Bjerg, B., & Zhang, G. (2016). Summary of
best guidelines and validation of CFD modeling in livestock
buildings to ensure prediction quality. Computers and
Electronics in Agriculture, 121, 180e190.
Seo, I. H., Lee, I. B., Moon, O. K., Hong, S. W., Hwang, H. S.,
Bitog, J. P., et al. (2012). Modelling of internal environmental
conditions in a full-scale commercial pig house containing
animals. Biosystems Engineering, 111, 91e106.
Shen, X., Zhang, G. Q., & Bjerg, B. (2012). Comparison of
different methods for estimating ventilation rates through
wind driven ventilated buildings. Energy and Buildings, 54,
297e306.
Shih, T.-H., Liou, W. W., Shabbir, A., Yang, Z., & Zhu, J. (1995). A
new ke3 eddy viscosity model for high Reynolds number
turbulent flows. Computers & Fluids, 24, 227e238.
Sorensen, D. N., & Nielsen, P. V. (2003). Quality control of
computational fluid dynamics in indoor environments. Indoor
Air, 13, 2e17.
Tian, E., He, Y. L., & Tao, W. Q. (2015). Numerical simulation of
finned tube bank across a staggered circular-pin-finned tube
bundle. Numerical Heat Transfer Part AeApplication, 68, 737e760.
Wilcox, D. C. (1988). Reassessment of the scale-determining
equation for advanced turbulence models. AIAA Journal, 26,
1299e1310.
Wu, W. T., Zhang, G. Q., Bjerg, B., & Nielsen, P. V. (2012). An
assessment of a partial pit ventilation system to reduce
emission under slatted floor e Part 2: Feasibility of CFD
prediction using RANS turbulence models. Computers and
Electronics in Agriculture, 83, 134e142.
Wu, W. T., Zong, C., & Zhang, G. Q. (2013). Comparisons of two
numerical approaches to simulate slatted floor of a slurry pit
model e Large eddy simulations. Computers and Electronics in
Agriculture, 93, 78e89.
Yakhot, V., & Orszag, S. A. (1986). Renormalization-group analysis
of turbulence. Physical Review Letters, 57, 1722e1724.
Yu, G. J., Yu, B., Sun, S. Y., & Tao, W. Q. (2012). Comparative study
on triangular and quadrilateral meshes by a finite-volume
method with a central difference scheme. Numerical Heat
Transfer Part B-Fundamentals, 62, 243e263.
Zhang, Z., Zhai, Z. Q., Zhang, W., & Chen, Q. Y. (2007). Evaluation
of various turbulence models in predicting airflow and
turbulence in enclosed environments by CFD: Part 2 e
Comparison with experimental data from literature. HVAC&R
Research, 13, 871e886.
Zong, C., & Zhang, G. Q. (2014). Numerical modelling of airflow
and gas dispersion in the pit headspace via slatted floor:
Comparison of two modelling approaches. Computers and
Electronics in Agriculture, 109, 200e211.