i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5

available at www.sciencedirect.com

w w w . i i fi i r . o r g

journal homepage: www.elsevier.com/locate/ijrefrig

Numerical simulation of temperature and velocity in

a refrigerated warehouse

Son H. Ho, Luis Rosario, Muhammad M. Rahman*

Department of Mechanical Engineering, University of South Florida, 4202 E. Fowler Avenue, ENB 118, Tampa, FL 33620, USA

article info abstract

Article history: This paper presents air velocity and temperature distribution in a refrigerated warehouse.
Received 9 August 2008 The refrigerated space under study has a set of ceiling-type cooling units installed in front
Received in revised form of the arrays of stacks of palletized product packages. Numerical solutions of the steady-
20 February 2010 state airflow and heat transfer were done using a complete three-dimensional model
Accepted 27 February 2010 and an equivalent two-dimensional model. The results obtained from both models were
Available online 11 March 2010 found to be in good agreement with each other. A parametric analysis was performed using
the equivalent model with various values of blowing air velocity and locations of the
Keywords: cooling units. It was found that a better cooling effectiveness and uniformity of tempera-
Refrigerated warehouse ture in the refrigerated space could be achieved by using higher blowing air velocity and/or
Modelling locating the cooling units lower and closer toward the arrays of product packages.
Numerical simulation ª 2010 Elsevier Ltd and IIR. All rights reserved.
Distribution
Velocity
Airetemperature

Simulation numérique de la température et la vitesse de l’air

dans un entrepôt frigorifique
Mots clés : Entrepôt frigorifique ; Modélisation ; Simulation numérique ; Distribution ; Vitesse ; Airetempérature

1. Introduction
Refrigeration for the cold storage of perishable foods has been
utilized for more than a century. The need for refrigerated
storage grows with hot weather. The frozen food industry
expanded many times in freezer storage in a few decades after
World War II. Recently, the gross capacity of refrigerated
warehouses in the United States has increased constantly
every year (USDA, 2008). Cold storage facilities are also signif-
icant energy consumers that call for attention to thermal
behavior as it greatly influences the cost. Proper design to
improve thermal behavior of a refrigerated space requires the
knowledge of air distribution and thermal conditions within
the space. The frozen food quality is sensitive to both storage
temperature and fluctuation in temperature. Inadequate
operation of a refrigeration system may result in negative

* Corresponding author. Tel.: þ1 813 974 5625; fax: þ1 813 974 3539.
E-mail address: mmrahman@usf.edu (M.M. Rahman).
0140-7007/$ e see front matter ª 2010 Elsevier Ltd and IIR. All rights reserved.
doi:10.1016/j.ijrefrig.2010.02.010
1016 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5
Nomenclature
1 1
cp specific heat, J kg K
g gravitational acceleration vector, m s2
h heat transfer coefficient, W m2 K1
k thermal conductivity, W m1 K1
lc characteristic length scale of the flow, m
lm mixing length, m
ln distance from the nearest wall, m
n coordinate in normal direction, m
p pressure, Pa
Pr Prandtl number
Q heat rate, W or W m1 (for 3-D or 2-D, respectively)
q heat flux, W m2
T temperature,  C
u velocity vector, m s1
u velocity component, m s1
V blowing air velocity at outlet of cooling unit(s),
m s1
X distance from front wall to outlet face of cooling
unit(s), m
x x coordinate, m
y y coordinate, m
Z height from floor to fan centerline of cooling unit
(s), m
impacts on product quality such as accelerated deterioration
reaction at an elevated temperature and the growth in ice
crystal size during temperature fluctuation. Even in a well
insulated refrigerated warehouse with no loading or unloading
activity, there is still heat transfer through the floor, ceiling,
and walls, as well as heat generation at the lights. Cooling units
(CU) such as those equipped with cooling coils and blowing
fans have to be employed to keep the food products under
proper temperature. The ceiling-type cooling units are
frequently used in large cold storage rooms, both freezers and
coolers, because they occupy no valuable floor space, and are
out of the way and high enough off the floor thus they are not
subjected to damage caused by material handling equipment
(Woolrich and Hallowell, 1970). In a ceiling-type CU, direct
mounted propeller fans either blow the air through the coil
bank when placed on the inlet side or pull the air through the
coil bank when placed on the outlet side.
Guidelines of settings in a refrigerated warehouse can be
found in Woolrich and Hallowell (1970), Tressler et al. (1968),
and the International Institute of Refrigeration (IIR, 1966). The
manual by Woolrich and Hallowell (1970) provides a compre-
hensible background and common practice on refrigerated
warehouse construction, equipment, and management. More
specific guidelines for the preservation of foods can be found
in Tressler et al. (1968). Theoretically, frozen food products
should be stacked in solid piles in such a way that reduces air
circulation around the products to the minimum to lessen the
desiccation and oxidation except for the case in which the
products are canned or packed in sealed containers. Frozen
foods are warehoused in a number of different types of
containers including wooden boxes, fiber board containers,
tin cans, etc. The guidelines given in Tressler et al. (1968)
suggest that for any type of storage container used, the
z z coordinate, m
Greek symbols
b thermal expansion coefficient, K1
k von Karman constant
m dynamic viscosity of air, kg m1 s1
r density of air, kg m3
Subscripts
amb ambient
cool cooling air
CF concrete floor
gnd ground
i index
j index
light lightings
PU polyurethane
ref reference
s solid (product packages)
t turbulent, or eddy
x x-direction
y y-direction
z z-direction
packages must be so placed in storage as to allow air circu-
lation around them to improve cooling effectiveness. For
maintaining a vertical circulation, the packages should not be
stacked too close to the walls. A clearance of 6 in. (0.15 m) or
more should be provided on all four sides of the storage room.
Aisles between arrays of stacks are needed, mainly for the
material handling operations while providing horizontal
circulation, and should be planned reasonably so as not to
waste too much storage space. Packages should not be piled
higher than 12e18 in. (0.30e0.46 m) below the ceiling or 6 in.
(0.15 m) below the bottom of the CU and should not be stacked
within 5 feet (1.52 m) from any non-refrigerated space, such as
openings to stairs or elevator wells. Packages of frozen food
should never be stocked directly on the floor but may be piled
on fork-type pallets or on floor racks at preferably 4 in. (0.10 m)
above the floor. The practice guide by IIR (1966) provides
specific guidelines on the use of pallets in refrigerated ware-
houses. A clearance of 2e6 in. (0.05e0.15 m) width on each
side of the pallet is necessary to provide a free space of 4e12
in. (0.10e0.30 m) between two pallets to reduce the difficulty
of placement and removal of pallets as well as to induce
vertical circulation.
Although there are general guidelines on the operations of
refrigerated warehouses, they do not show how relevant
parameters affect the cooling effectiveness and temperature
uniformity in refrigerated warehouses. Therefore, further
studies are needed to evaluate the effectiveness of the
combinations of the parameters so as to efficiently utilize
them in various situations, in both design and operation of
refrigerated warehouses. Baird and Gaffney (1976) developed
a numerical model for predicting transient heat transfer in the
pre-cooling operation. The numerical model and procedures
allowed calculations of cooling rates in beds of fruits and
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5
vegetables as a function of product size, air velocity, air
temperature, and the depth of the bulk load. Later, Baird et al.
(1988) proposed the design criteria for efficient and cost
effective forced air cooling systems for fruits and vegetables.
The criteria cover the effects of many parameters including
initial product temperature, desired final product tempera-
ture, flow rate, temperature and relative humidity of the
cooling air, ambient temperature, etc., among other factors
related to characteristics of product and equipment, and cost.
An engineering-economic model was used to study the
influence of each of these variables as well as some of their
interactions as they affect cooling time and/or cooling cost.
Nicoulin et al. (1997) studied the use of general-purpose
transient computer models for simulating the energy perfor-
mance of large commercial refrigeration systems typically
found in food processing facilities to predict facility perfor-
mance and estimate savings with inclusion of modeling
issues specific to refrigerated warehouse systems including
warehouse loading door infiltration calculations, evaporator
model, single-stage and multi-stage compressor models,
evaporative condenser models, and defrost energy
requirements.
The increasing development of computers and the field of
computational fluid dynamics (CFD) in recent years have
opened up the possibilities of a low-cost yet effective method
for modeling and simulation of airflow and heat transfer in
refrigerated warehouses with fewer experiments required.
Smale et al. (2006) reviewed the application of various
numerical modeling techniques, focusing on CFD and briefly
on others including the Lattice Boltzmann method (LBM) and
network models, to the prediction of airflow in refrigerated
food storage applications including cool stores, transport
equipment, and retail display cabinets. Hoang et al. (2000)
presented an analysis of airflow in a cold store using CFD
approach employing the Reynolds-averaged NaviereStokes
equations with the ke3 turbulence model. A comparison with
experimental measurements showed an average difference of
26% between calculated and measured air velocities. Nahor
et al. (2005) developed a transient three-dimensional CFD
model with the use of standard ke3 turbulence model to
calculate the velocity, temperature and moisture distribution
in an existing empty and loaded cool store. The results
showed that an average accuracy of 20e22% on the velocity
magnitudes was achieved and that the model was capable of
predicting both air and product temperature with reasonable
accuracy. Chourasia and Goswami (2007) simulated the effects
of stack dimensions and stacking arrangements on heat
transfer characteristics in a stack of bagged potatoes during
cooling by using a CFD model and found a satisfactory
agreement between the experimental transient temperature
data from a commercial potato cold store and simulated
results with an average temperature difference of
1.4  0.98  C. Foster et al. (2002, 2003, 2006, 2007) reported
several studies concerning reducing air infiltration in cold
stores by means of CFD modeling and simulation for various
cases. A CFD model for air movement through a doorway was
developed and verified against conventional and laser Doppler
anemometry (LDA) measurements (Foster et al., 2002).
Measurements of infiltration through different size entrances
of a cold store at two different cold store temperatures were
1017
compared against established analytical models and CFD
models (Foster et al., 2003). More recently, they presented 2-D
and 3-D analyses of the effectiveness and optimization of air
curtain devices with CFD models and compared the results to
the measured data (Foster et al., 2006, 2007). For each study,
the CFD model was generally found to be in agreement with
experimental data to some degree. All the previous studies on
CFD modeling and simulation of refrigerated spaces consid-
ered small cool stores with a couple of product stacks.
A thorough study on a larger refrigerated warehouse with
many parallel arrays of product stacks and cooling units has
never been reported. This study aims to explore that problem
for a somewhat larger warehouse and the results are expected
to be very useful for various refrigerated warehouse design.
The present study considers a typical refrigerated ware-
house as shown in Fig. 1. A set of many cooling units is
installed along the front wall, in front of the arrays of product
packages, to provide cold airflow to maintain low tempera-
ture in the space. The products are stacked on pallets into
many parallel arrays with aisles between one another and
clearances from the walls and from the floor. Each array
consists of two back-to-back rows of four piles of three stacks
of packages with both horizontal and vertical clearances
between each other, as suggested by applicable guidelines
(IIR, 1966). This arrangement in the warehouse possesses two
parallel planes of symmetry as shown in Fig. 1. The first one
cuts through the middle of an array and separates its two
rows of stacking packages. The second one cuts through the
middle of an aisle between two arrays. The space contained
in between the two planes of symmetry represents almost
the whole space inside the refrigerated warehouse (without
taking into account the end-wall effects) by mirroring itself
through the planes of symmetry to recreate the whole space.
A CFD modeling approach using the finite element method is
used to simulate the flow and heat transfer of the air in the
refrigerated space. Incompressible fluid, steady state and
constant properties are assumed. Thermal loads as heat
transfer through walls, floor and ceiling are included into the
numerical model by using the linear heat transfer coefficient
model. The distributions of velocity and temperature and
the uniformity of temperature distribution are presented as
the most important parameters of thermal behavior of the
refrigerated warehouse. For assessing the cooling effective-
ness and the temperature uniformity, the maximum and
mean temperatures and the standard deviation of tempera-
ture distribution are calculated. Several factors may affect
these thermal behavior parameters such as cooling air
velocity and temperature; location of the CU, both horizontal
and vertical; and the product distribution pattern, i.e. the
clearance between racks, aisles, and from the walls. This
study investigates the effects of the blowing air velocity from
the CU and the location of the CU. The problem was first
solved using a three-dimensional (3-D) model; then using an
equivalent two-dimensional (2-D) model for a comparison
between the results to determine the accuracy of the 2-D
model. Both the 3-D and 2-D simulations are for the base case
(the CU is located at a central position and produces
a medium blowing air velocity). Additional 2-D simulations
including five cases for different blowing air velocities and
eight cases for different CU locations were then considered.
1018 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5

Fig. 1 e Basic arrangements in a refrigerated warehouse.

2. Mathematical model

For 2-D simulations, the computational domain is modeled as

a rectangular region shown in Fig. 2a with 12 product packages
stacked with vertical and horizontal gaps between them to
separate them from one other and from the floor and the back
wall. A CU is located in front of the array, which blows the
cooling air through its outlet side on the right face. The rele-
vant lengths L1 through L11 are given the following values:
L1 ¼ 7.0 m, L2 ¼ 4.0 m, L3 ¼ 1.2 m, L4 ¼ 0.8 m, L5 ¼ 0.1 m,
L6 ¼ 0.1 m, L7 ¼ 0.2 m, L8 ¼ 0.1 m, L9 ¼ 0.4 m, L10 ¼ 0.6 m,
L11 ¼ 0.4 m. The values of L5eL8 were chosen in accordance
with the guidelines given by Tressler et al. (1968) and IIR (1966).
The CU location is defined by the distance X from the front
wall and the height Z from the floor. This 2-D model repre-
sents the space in the warehouse on a vertical plane cutting
through the CU and the arrays of packages. For 3-D simula-
tions, the space between the two planes of symmetry is
modeled as a box region as shown in Fig. 2b including a half of
an aisle, one row of four piles of three stacks of product
packages, and the set of a large number of cooling units in
front of the array of packages modeled as one long CU. The
relevant dimensions for the 3-D model are the same as the
respective ones for the 2-D model. In the y-direction that is
absent from the 2-D model, the widths of a package and of the
computational domain are 1.0 m and 2.0 m, respectively.
The air inside the refrigerated space is considered to be an
incompressible fluid with constant properties. The equation
for the conservation of mass can be written as

V$u ¼ 0 (1)

The buoyancy force term arising from density variation is Fig. 2 e Two- and three-dimensional models of
included by means of the Boussinesq approximation based on a refrigerated warehouse.
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5
the assumptions that the variation in fluid density affects only
the buoyancy term and the fluid density is a function of
temperature only. The equation for conservation of
momentum can be written as


ru$Vu ¼ Vp þ ðm þ mt ÞV2 u þ rgb T  Tref (2)
For modeling the turbulent flow, the effective viscosity is
defined as the sum of the dynamic viscosity (physical property)
and eddy viscosity (representing effects of turbulent flow).
A similar treatment is also applied for the effective thermal
conductivity. To estimate the eddy viscosity, the mixing length
model is used. The mixing length model is a simple yet effec-
tive turbulence model involving a single unknown parameter
called the mixing length or the mean free path for the mixing of
turbulent fluid flow. The mixing length model works well for
relatively simple flows such as wall boundary-layer flows, and
jet and wake flows without requiring any additional governing
equation. For the problem at hand, this turbulence model can
be employed on considering that the rather complex flow in the
computational domain is composed by several regions of
simple turbulent flows (wall boundary-layer and jet) separated
far enough that the transport and history effects of turbulence
are insignificant. The eddy viscosity can be written in general
form (Rodi, 1984), applicable to both 2-D and 3-D computational
models, as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  ffi
vui vuj vui
mt ¼ rl2m þ (3)
vxj vxi vxj
where
lm ¼ minðkln ; 0:09lc Þ
Assuming that there is no heat generation, the equation for
conservation of energy for the fluid (air) can be written as
rcp u$VT ¼ ðk þ kt ÞV2 T
where
cp mt
kt ¼
Prt
Inside the product packages, it is assumed that heat conduc-
tion is the only mode of heat transfer. The equation for the
conservation of energy for the solids (product packages) with
constant properties at steady state can be written as
V2 T ¼ 0
The applicable thermal properties of air are taken from Kays
et al. (2005) at a reference temperature of 0  C and listed as
follows: r ¼ 1.293 kg m3, m ¼ 17.20  106 kg m1 s1, k ¼ 24.07 
103 W m1 K1, cp ¼ 1004 J kg1 K1, b ¼ 3.663  103 K1. Gravi-
tational acceleration is taken as g ¼ 9.8 m s2. The values of the
von Karman constant k in Equation (4) and the turbulent Prandtl
number Prt in Equation (6) can also be found in Kays et al. (2005) as
follows: k z 0.41, Prt z 0.85. The thermal properties of the prod-
ucts are adapted from ASHRAE (2002) as rs ¼ 840 kg m3,
ks ¼ 0.52 W m1 K1, cp,s ¼ 3.79  103 J kg1 K1.
To completely define the problem, appropriate boundary
conditions are needed at every boundary segment of the
computational domain. For Equations (1) and (2), the
boundary conditions for velocity were applied as follows:
prescribed constant velocity at the outlet side of the cooling
1019
unit, zero y-component of velocity on planes of symmetry (for
3-D case only), and zero velocity on solid surfaces (walls, floor,
ceiling, packages’ surfaces, etc.). For Equation (5), the
temperature at the outlet side of the CU was fixed at
a constant value of Tcool ¼ 0  C, so chosen for the refrigerated
space to achieve the range of temperature for the storage of
foods such as fresh fruits and vegetables, poultry and dairy
products, etc. as recommended in ASHRAE (2002) and
Hardenburg et al. (1986). Linear heat transfer models were
applied to the floor, the ceiling, and the walls of the compu-
tational domain with relevant data taken from ASHRAE (2002).
The floor was assumed to be made of 6 in. (0.152 m) un-
insulated concrete slab with a heat transfer coefficient of
hCF ¼ 1.18 W m2 K1 and a ground temperature of
Tgnd ¼ 15  C. Both the walls and the ceiling were assumed to be
made of 4 in. (0.102 m) polyurethane insulation with a heat
transfer coefficient of hPU ¼ 0.23 W m2 K1 exposed to an
outside (ambient) temperature of Tamb ¼ 35  C. There was also
lightings load of qlight ¼ 10 W m2, as recommended by IIR
(1966) considered as a uniform heat flux added to the heat
flux through the ceiling.
The boundary conditions on velocity can be written as
On CU outlet : ux ¼ V; uy ¼ uz ¼ 0 (8)
On planes of symmetry : uy ¼ 0 (9)
On all solid surfaces : ux ¼ uy ¼ uz ¼ 0 (10)
The boundary conditions on temperature can be written as
On CU outlet : T ¼ Tcool (11)
(4)
vT

On floor surface : k ¼ hCF Tgnd  T (12)
vn
vT
On wall surfaces : k ¼ hPU ðTamb  TÞ (13)
(5) vn
vT
On ceiling surface : k ¼ hPU ðTamb  TÞ þ qlight (14)
vn
(6)
vT
On other boundary surfaces : ¼0 (15)
vn
It may be noted that the moisture content (humidity ratio) of
the air in a refrigerated space is expected to be fairly low and
therefore, the water vapor in the air has not been included as
a part of the simulation model. In addition, the model
(7)
considered only the steady-state operation of the warehouse
and therefore did not include periodic maintenance opera-
tions such as defrosting of the cooling coil.
3. Numerical solution
For each simulation, the governing equations, Equations (1)
through (7), along with their associated boundary conditions,
Equations (8) through (15), were solved using the finite
element method. The computational domain was first
discretized (subdivided) into a mesh of many small elements.
Meshes of four-node quadrilateral elements and eight-node
hexahedral elements were employed for the 2-D model and
3-D model, respectively. In each element, two or three (for 2-D
or 3-D model, respectively) velocity components, pressure,
1020 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5

and temperature were approximated by using the Galerkin 10

procedure (Fletcher, 1984), which led to a system of nonlinear 9
algebraic equations defining the fluid flow and heat transfer in
8
the discretized continuum. The velocity components and the
7

temperature were approximated using bilinear (for 2-D model)
or trilinear (for 3-D model) interpolation functions. The pres-
sure appearing in Equation (2), which is the vector form of the
NaviereStokes equation, was also discretized. Unlike the
velocity and the temperature, the pressure was approximated
by a constant value at the centroid of each element, which
resulted in element-to-element discontinuities in the pres-
sure approximation. Pressure unknowns were also integrated
into the global system of equations and were solved for
together with all the other unknowns to satisfy the conser-
vation of momentum in each coordinate direction and the
conservation of mass simultaneously.
For the 2-D model, the fully coupled successive substitu-
tion algorithm was used to solve the system of nonlinear
algebraic equations. Two convergence criteria were used: the
relative errors of the solutions between iterative steps and the
residuals of the nonlinear equations. The iterative procedure
for the solution was considered converged when both criteria
were satisfied. The tolerances of 1E-4 and 1E-2 were used for
these criteria, respectively. For the 3-D model, the segregated
algorithm was used to solve the system of nonlinear algebraic
equations with a relative error tolerance of 1E-6.
The distribution of the element size in the computational
domain was determined from a mesh independence study by
systematically changing the mesh density in all space direc-
tions, both globally and locally, to obtain a mesh that can give
converged solutions of acceptable accuracy. High quality
mesh elements (squares for 2-D model and cubes for 3-D
model) were used to fill most of the interior of the domain
while layers of refined mesh elements were assigned along
boundaries such as inlet, outlet, and solid surfaces to capture
the high rates of change of momentum and heat transfer that
exist there. Mesh independence study was performed by
solving the base case using a series of meshes of decreasing
element sizes. For the 2-D model, ten meshes represented by
their nominal element sizes of 0.2 m, 0.19 m, 0.175 m, 0.15 m,
0.125 m, 0.1 m, 0.075 m, 0.05 m, 0.033 m, and 0.025 m were
used. They had the numbers of elements of 1725, 2097, 2169,
2829, 3432, 5056, 7929, 15424, 31202, and 52360, respectively.
Fig. 3 presents the maximum and mean values of the
temperature solution as functions of the number of elements.
It shows that mesh independence of the numerical solution is
achieved as the number of elements increases over 5000. The
mesh with 5056 elements (nominal element size of 0.1 m) was
found quite adequate for accurate numerical prediction.
Therefore, the nominal element size of 0.1 m was used for
Temperature, ºC
6 Max. temp.
Mean temp.
5
4
3
2
1
0
0 10000 20000 30000 40000 50000
Number of elements
Fig. 3 e Mesh independence study.
generating the meshes for all the other simulation cases.
Numerical uncertainty is known to result from discretization
error and iterative convergence error (Freitas, 2002). Iterative
convergence error was bounded by the tolerances mentioned
previously. Discretization error was estimated using the
procedure recommended by Celik et al. (2008). From the three
meshes of element nominal sizes of 0.1 m, 0.15 m, 0.2 m, the
numerical uncertainty of the solution from the mesh of 0.1 m
nominal element size was found in the form of grid conver-
gence index (GCI) of 0.9% and 4.6% for maximum temperature
and average temperature, respectively. For the 3-D model, the
nominal element size of 0.1 m, which resulted in a mesh of
about 109,000 hexahedral elements, was also found to yield an
adequately accurate numerical solution.
In order to verify the numerical solution, global energy
balance at macroscopic level was performed by applying the
first law of thermodynamics. For the steady-state problem
under study, the rate of heat gained from lighting and trans-
mission through walls, floor, and roof (where the heat through
the roof and from lighting were combined into the heat
through the boundary entity “ceiling” in the models) should
equal the rate of equivalent heat extracted at the CU. This
relationship can be expressed as
Q floor þ Q ceiling þ Q front wall þ Q back wall ¼ Q extracted at CU (16)
Table 1 presents component and total heat gain and total
heat extracted at the CU for the base case of both 2-D and 3-D
models. It shows that global energy balance was satisfied with
a reasonably low error. The difference between two sides of
Equation (16) is 1.1% for the 2-D model and 2.5% for the 3-D
model.

Table 1 e Global energy balance for base case.

Model (unit) Heat gain from transmission and lighting Heat extracted at CU Difference

Floor Roof & lighting Front wall Back wall Total

2-D (W/m) 82 120 32 28 262 265 3 1.1%

3-D (W) 199 355 67 62 683 700 17 2.5%
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5 1021

All simulation cases are summarized in Table 2. For the This kind of short circuit can be happening several times to
parametric analysis, fourteen 2-D simulations with different a flow until it exits the CU at an “unfavorable” starting point.
settings were performed. For simulation cases 2e7, the CU is There, because the zone under the CU is already filled with
fixed at the base location whereas the blowing air velocity V short-circuit flows; the “unfavorable” flow is forced to go
varies. For simulation cases 8e15, the blowing air velocity V is outside that zone, all the way down to the floor. There it
fixed whereas the CU location varies. The pseudo-direction makes a turn to avoid the obstacles (the product packages)
codes NE, NW, SE, SW, N, S, E, and W (see Fig. 2) are used to and moves toward the unoccupied zone in the aisle, sweeps
indicate the relative location of the CU to the base (central) through the floor until reaching the back wall. Then it goes up
location that dependents on the values of X and Z. along the back wall, fills the space above the stacks under the
ceiling and moves toward the CU to get forced down to the
floor again and repeat the cycle until it get to a “favorable” spot
4. Results and discussion allowing it to feed to the CU or its momentum dies out at some
stagnant point. The flow along the back wall does not reach
Fig. 4 presents the solutions of air velocity, pressure and the ceiling, leaving a still air zone at the corner. Even though
temperature for the 3-D base case (simulation 1). Fig. 4a shows there are gaps between the packages and the floor and the
the 3-D streamlines of the airflow inside the warehouse as wall as well as between themselves, the major portion of the
spatial curves originating from nine representative starting flow goes through the open zone of the aisle at a slow velocity
points distributed on a 3  3 matrix on the blowing opening of (about 0.1e0.2 m s1). There are also flows induced by natural
the CU. These streamlines are numbered from 1 to 9 with convection due to temperature difference. These flows move
color-coded legend for tracing their paths easily. Fig. 4b in the gap next to the back wall as well as in the gaps between
presents the distribution of air velocity by displaying respec- the packages.
tive interpolated filled color on orthogonal slice planes. The Fig. 4c presents the isosurface plot for pressure distribu-
slice planes are selected in such a way that can reveal the tion. The value of air pressure is the same on an isosurface. It
structure of the volumetric data and include the planes can be observed that most of the isosurfaces are almost flat,
perpendicular to y- and z- directions and cut through the well-layered, and perpendicular to the vertical direction. This
center of the products. Fig. 4a and b can be examined simul- pattern implies that vertical flow due to natural convection
taneously to construct the image of the flow field in the dominates the airflow field around the packages. The domi-
domain. The air is cooled by the coil on the suction side of the nation of forced convection can be observed only in the region
CU and pulled by the fan that blows into the space at full close to the outlet opening of the CU and the short circuit zone
velocity (0.5 m s1). The colder air that has higher density goes where the isosurfaces have high curvature.
down smoothly as seen in both Fig. 4a and b. Most of this main Fig. 4d presents the slice planes plot of temperature
flow drops down toward the floor but soon gets pulled back to distribution for simulation 1. Wherever the air velocity is high,
feed to the suction side of the CU under the effect of lower such as in the main airflow or in the circulations close to the
pressure there created by the fan. This forms a short circuit for exhaust opening, the temperature is lower due to low
most of the streamlines. The air velocity in the short circuit temperature in the supply air itself or by well mixing it with
zone is the highest in the whole domain (0.2 m s1 and more). the heated air inside the room. The strong short circuit airflow
creates a low temperature zone around and under the CU
where the cold air blows into the room and has not picked up
much heat from the room yet. The first pile of product pack-
Table 2 e Simulation cases. ages benefit from this zone by heat transfer from the products
to the neighboring cold air resulting in a low temperature of
Simulation # CU X, m Z, m V, m s1 Model Notes
location about 1  C in this pile. Similarly, the packages also have a low
temperature on the aisle side because of lower temperature
1 Central 1.1 3.3 0.50 3-D Base there (around 2  C) due to the mixing effects by circulations of
case
the velocity field. Since there is almost no flow near the floor
2 " 1.1 3.3 0.50 2-D "
3 " 1.1 3.3 0.25 " CU fixed,
under the packages, there is a high temperature zone there
V varied that includes the maximum temperature. However, the air
4 " 1.1 3.3 0.30 " " gap between the floor and the first stack of packages quite
5 " 1.1 3.3 0.40 " " effectively prevents this high temperature on the floor to
6 " 1.1 3.3 0.75 " " directly affect the products. While the maximum temperature
7 " 1.1 3.3 1.00 " "
on the floor is about 8e9  C, the highest temperature on the
8 NE 1.3 3.5 0.50 " V fixed,
bottom of these packages is only about 4e5  C, which is still in
CU
moved the practical range for refrigerated warehousing. The heated
9 NW 0.9 3.5 0.50 " " up air on the floor forms some natural convection flows inside
10 SE 1.3 3.1 0.50 " " the vertical gaps, thus transports some heat from the floor to
11 SW 0.9 3.1 0.50 " " the higher stacks and rises the temperature up to a few
12 N 1.1 3.5 0.50 " " degrees although not significantly since that is not capable of
13 S 1.1 3.1 0.50 " "
spreading widely into the products. The still air under the
14 E 1.3 3.3 0.50 " "
ceiling toward the back wall results in a zone of higher
15 W 0.9 3.3 0.50 " "
temperature where its pattern shows that heat transfer is
1022 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5
Fig. 4 e Distributions of air velocity, pressure, and temperature for simulation 1 (3-D base case: X [ 1.1 m, Z [ 3.3 m).
mainly by conduction. It should be noted that even though the
forced circulation does not create significant movement of
cool air in the gaps between the packages, it maintains a low
ambient temperature of the overall air space to allow for the
dissipation of heat and to prevent large temperature
variations.
Fig. 5 presents the air velocity field and the temperature
distribution for the 2-D base case (simulation 2). In Fig. 5a, the
streamlines of the airflow are plotted on the filled background
with color representing the air velocity. The fan of the CU pulls
the air through the coil banks to cool it down and blows from
its outlet into the refrigerated space. This cold airflow has
higher density compared to the surrounding warm air and
therefore it tends to go down and feeds back to the CU inlet
because of a low pressure there created by the fan. This forms
a major short circuit circulation under the CU. Because of the
initial momentum, a part of the flow sweeps over the top of
the stacks and then returns to the inlet. A minor part of the
flow finds its way in the gap between the floor and the pack-
ages, reaches the back wall and gets boosted up by the natural
convection of hot air at the back wall to return to the CU inlet
along the ceiling. In comparison to the 3-D model, these
observations predict the flow pattern reasonably accurately
such as the short circuit flow and the flow along the back wall.
Fig. 5b shows the distribution of air pressure for simulation
2 in iso-pressure contours. It can be observed that this 2-D
pressure contour plot looks just like a cross section of the 3-D
isosurface pressure plot in Fig. 4c cut by a plane perpendicular
to the y-direction. This supports the similar pattern of the
solutions from 2-D and 3-D models.
In Fig. 5c, the temperature distribution inside the refriger-
ated space is presented using a filled color plot that shows
different level of temperature in different regions of the
computational domain. The circulation formed by the
combined effects of forced convection (due to the forced flow
at the outlet and negative pressure at the inlet) and natural
convection (due to buoyant force because of temperature-
dependent variable density of air) creates a well mixed
region under the CU with uniform low temperature as the
result. There is heat transfer through the front wall, but the
circulation of cold air is strong enough to sweep that wall and
effectively removes the heat from it, leaving only the floor and
the ceiling corners at higher temperature. Since the second
and third groups of circulation cannot reach the ceiling with
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5 1023

there without any significantly strong air stream to remove it.

In comparison to the distribution of temperature for 3-D
model in Fig. 4d, it can be seen that the respective patterns are
very similar. The 2-D distribution of temperature is very
similar to that on one y-direction cross section of the 3-D
model.
For assessing the cooling effectiveness and uniformity of
temperature in the whole refrigerated space, we consider the
following parameters: maximum temperature, mean tempera-
ture, and standard deviation of distributed temperature around
the mean temperature. The first two express the effectiveness of
the whole system. The last parameter represents the uniformity
of temperature which is also an important factor in refrigeration
storage. Generally, the lower the temperature (maximum and
mean) and the standard deviation, the better it is. The results of
these parameters for simulation 1 (3-D) and simulation 2 (2-D)
are given in the first two rows of Table 3. The maximum
temperatures for both are almost the same. The mean temper-
ature for the 2-D model is lower than that for the 3-D model,
whereas the standard deviation for the 2-D model is signifi-
cantly higher than that for the 3-D model. This can be explained
by the fact that the 2-D model implies that the packages are
extended from one plane of symmetry to another one while the
cold air supplied to both cases have the same flow rate which
leaves for the 2-D model less room volume to cool and therefore
a lower mean temperature. However, the 2-D model has no
cooling from the side (aisle side) as the 3-D model does, resulting
in less uniformity or higher standard deviation of temperature
distribution.
Fig. 6 shows the distributions of maximum, mean, and
standard deviation of temperature over the refrigerated space
as functions of blowing air velocity for the simulation cases
2e7 with the CU at its base location. The maximum temper-
ature occurred at the region on the floor under the third pile
(the second one from the back wall, Fig. 2a). It can be observed
that a higher blowing air velocity will give lower maximum
and mean temperatures and standard deviation of tempera-
ture distribution. Therefore, it is better to have a higher
blowing air velocity as it improves the temperature distribu-
tion within a warehouse. However, in practice, the blowing air

Table 3 e Effects of the location of the cooling unit (CU) on

temperature distribution.
CU Max. Mean Std.
location temp.,  C temp.,  C deviation,  C

Central (3-D) 8.80 2.17 1.13

Central (2-D) 8.79 2.10 1.50
NE 9.20 2.58 1.53
Fig. 5 e Distributions of air velocity, pressure, and NW 9.25 2.87 1.53
temperature for simulation 2 (2-D base case: X [ 1.1 m, SE 8.71 1.47a 1.20b
SW 9.10 1.78 1.51
Z [ 3.3 m).
N 9.08 2.63 1.50
S 9.47 1.60 1.26
E 8.47c 2.07 1.28
that much effectiveness, especially in the part toward the W 9.10 2.16 1.53
back wall, a still air region remains there. In this region, heat
conduction dominates with its pattern shown in Fig. 5c. High a Lowest value of mean temperature.
b Lowest value of standard deviation.
temperature regions are found at back wall e floor corner and
c Lowest value of maximum temperature.
its neighborhood as the heat flux from outside comes through
1024 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5

10

8
Temperature, °C

Max. temperature
6
Mean temperature
Standard deviation
4

0
0.2 0.4 0.6 0.8 1
Supply air velocity, m/s

Fig. 6 e Effects of blowing air velocity on temperature

distribution.

velocity is limited by equipment specifications and more

importantly, capital and operating costs. It can also be noticed
that the assumed face velocity of air from the CU in the
simulations presented is less than half the face velocity typi-
cally found in larger warehouses, since the latter requires the
higher air velocity to make up for the loss of refrigeration load
resulted from infiltration and daily operation such as product
loading/unloading activities. There is also a limitation on the
coil face velocity that assures that no accumulated water on
the coil surface is blown off and entrained into the air stream.
In large installations, face velocities of 400e550 ft/min
(2e2.8 m/s) are quite common.
The simulation results for the base case of the present
model can be qualitatively compared with simulation model
and experimental data for a large refrigerated facility reported
by Runsey (2008) in order to show similarity. Air distribution
and airflow patterns throughout warehouses clearly show
a typical pattern, which is the result of thermal stratification.
The simulation results from Runsey (2008) showed
a maximum temperature variation of 4.7  F. He also reported
that from actual recorded measurements over a seven-day
period, the maximum temperature variation across the
product was 3.4  F. The simulation results for the base case of
the present study show a maximum temperature variation of
3.3  C (5.8  F) in the product packages. This result of maximum
temperature variation is higher but comparable to that from
Runsey (2008); particularly considering the difference in
blowing air velocity.
Table 3 (from the second row down to the end showing the
results of 2-D simulations only) demonstrates how the CU
location affects the thermal effectiveness of the system, again
with maximum and mean temperatures and standard devia-
tion. It can be observed that the location SE (simulation 10 in
Table 2) gives the lowest mean temperature and also the
lowest standard deviation, whereas the location E (simulation Fig. 7 e Effects of the location of the cooling unit on
14) gives the lowest maximum temperature. temperature distribution.
Fig. 7 presents the maximum temperature, the mean
temperature, and the temperature standard deviation as location. These functions are found by applying the cubic
functions of the CU location (X, Z ) in the form of isotherm spline interpolation method on the obtained data. It is
contours which show the tendencies of these temperature observed that the maximum temperature, the mean temper-
parameters as the CU moves away from the central (base) ature, and the standard deviation decrease as the CU moves to
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 3 3 ( 2 0 1 0 ) 1 0 1 5 e1 0 2 5
locations E, S, and SE, respectively. Generally, the compro-
mised direction to SE can be the best choice. Therefore, the
location of the cooling unit is very important in determining
the overall performance of the refrigerated warehouse.
Air velocity and temperature simulations prior to warehouse
construction can achieve significant reductions in refrigerated
warehouse energy costs. These results can be obtained with
a better temperature profile throughout warehouses, primarily
due to air distribution. Computerized simulations at initial
design stage will help to build better energy efficient refrigerated
facilities. Using the methods presented in the paper, simula-
tions can be done for the design blowing air velocity using the
actual dimensions of the warehouse. In operating a warehouse,
it is best to stack product packages higher and closer to the
cooling units keeping within the limits for safety and materials
handling.
5. Conclusions
Numerical modeling can be used to predict fluid flow and heat
transfer and to assess thermal uniformity in refrigerated
warehouses. Instead of doing simulations by using expensive
3-D models, proper 2-D ones can be used to reduce computing
cost while still producing useful results that are accurate to
some reasonable degree. This replacement is critical where
many design parameters and their interactions involve,
resulting in a large number of required simulations with
limited time budget and computing resources that make the
use of only 3-D simulations impractical. As blowing air
velocity from the CU increases, better cooling effectiveness
and uniformity (lower maximum and mean temperatures and
lower spatial standard deviation) can be achieved. The effect
of the location of the CU on temperature distribution is
complicated with the interaction of positions in x- and z-
directions. The best temperature range and uniformity can be
obtained with the CU moved toward the SE location. The
analyses presented here and the results can be useful for
designing and operating refrigerated warehouses.
references
ASHRAE, 2002. ASHRAE Handbook: Refrigeration. American
Society of Heating, Refrigerating and Air-Conditioning
Engineers, Atlanta, GA.
Baird, C.D., Gaffney, J.J., 1976. A numerical procedure for
calculating heat transfer in bulk loads of fruits or vegetables.
ASHRAE Transactions 82, 525e535.
Baird, C.D., Gaffney, J.J., Talbot, M.T., 1988. Design criteria for
efficient and cost effective forced air cooling systems for
fruits and vegetables. ASHRAE Transactions 94 (1),
1434e1454.
Celik, I.B., Ghia, U., Roache, P.J., Freitas, C.J., Coleman, H., Raad, P.
E., 2008. Procedure for estimation and reporting of uncertainty
due to discretization in CFD applications. Journal of Fluids
Engineering 130 (7), 078001.
1025
Chourasia, M.K., Goswami, T.K., 2007. Simulation of effect of
stack dimensions and stacking arrangement on cool-down
characteristics of potato in a cold store by computational fluid
dynamics. Biosystems Engineering 96 (4), 503e515.
Fletcher, C.A.J., 1984. Computational Galerkin Methods. Springer-
Verlag, New York.
Foster, A.M., Barrett, R., James, S.J., Swain, M.J., 2002.
Measurement and prediction of air movement through
doorways in refrigerated rooms. International Journal of
Refrigeration 25 (8), 1102e1109.
Foster, A.M., Swain, M.J., Barrett, R., James, S.J., 2003.
Experimental verification of analytical and CFD predictions of
infiltration through cold store entrances. International Journal
of Refrigeration 26 (8), 918e925.
Foster, A.M., Swain, M.J., Barrett, R., D’Agaro, P., James, S.J., 2006.
Effectiveness and optimum jet velocity for a plane jet air
curtain used to restrict cold room infiltration. International
Journal of Refrigeration 29 (5), 692e699.
Foster, A.M., Swain, M.J., Barrett, R., D’Agaro, P., Ketteringham, L.
P., James, S.J., 2007. Three-dimensional effects of an air
curtain used to restrict cold room infiltration. Applied
Mathematical Modeling 31 (6), 1109e1123.
Freitas, C.J., 2002. The issue of numerical uncertainty. Applied
Mathematical Modelling 26 (2), 237e248.
Hardenburg, R.E., Watada, A.E., Wang, C.Y., 1986. Agriculture
Handbook No. 66: The Commercial Storage of Fruits,
Vegetables, and Florist and Nursery Stocks. United States
Department of Agriculture (USDA), Washington, DC.
Hoang, M.L., Verboven, P., De Baerdemaeker, J., Nicolaı̈, B.M.,
2000. Analysis of the air flow in a cold store by means of
computational fluid dynamics. International Journal of
Refrigeration 23 (2), 127e140.
IIR (International Institute of Refrigeration), 1966. Practical Guide
to Refrigerated Storage. Pergamon Press, Oxford.
Kays, W.M., Crawford, M.E., Weigand, B., 2005. Convective Heat
and Mass Transfer, fourth ed. McGraw Hill, New York.
Nahor, H.B., Hoang, M.L., Verboven, P., Baelmans, M., Nicolaı̈, B.M.
, 2005. CFD model of the airflow, heat and mass transfer in
cool stores. International Journal of Refrigeration 28 (3),
368e380.
Nicoulin, C.V, Jacobs, P.C., Tory, S., 1997. Computer modeling of
commercial refrigerated warehouse facilities. In: Proceedings
of the 1998 ACEEE Summer Study on Energy Efficiency in
Industry, pp. 15e27.
Rodi, W., 1984. Turbulence Models and Their Application in
Hydraulics: State-of-the-art Paper, 2nd revised ed.
International Association for Hydraulic Research (IAHR).
Runsey, I., 2008. Temperature and Air Distribution in Refrigerated
Warehouses e A New Dimension to Energy Savings. FS: Food
Safety, Food Supply, Food Solutions 213. GDS Publishing Ltd.
<http://www.us-fs.com/pastissue/article.asp?
art¼270885&issue¼213>.
Smale, N.J., Moureh, J., Cortella, G., 2006. A review of numerical
models of airflow in refrigerated food applications.
International Journal of Refrigeration 29 (6), 911e930.
Tressler, D.K., Van Arsdel, W.B., Copley, M.J., Woolrich, W.R.,
1968. The Freezing Preservation of Foods. AVI Publishing
Company, Westport, CT.
USDA, 2008. Capacity of Refrigerated Warehouses 2007 Summary.
National Agricultural Statistics Service (NASS), United States
Department of Agriculture (USDA).
Woolrich, W.R., Hallowell, E.R., 1970. Cold and Freezer Storage
Manual. AVI Publishing Company, Westport, CT.