biosystems engineering 101 (2008) 225238

Available at www.sciencedirect.com

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

Research Paper: SEdStructures and Environment

Simulation of three-dimensional airflow in grain

storage bins
O.A. Khatchatouriana,b,*, M.O. Binelob
a
Department of Physics, Statistics and Mathematics, Regional University of the Northwest, Rio Grande do Sul,
R. Sao Francisco, 501, 98700-000 IJUI, RS, Brazil
b
Computer Science Department, University of Cruz Alta, R. Andrade Neves, 308, 98025-810 Cruz Alta, RS, Brazil

article info
A mathematical model and software were developed for the three-dimensional simulation of
Article history:

airflow through high capacity grain storage bins by considering the non-uniformity of the seed

Received 21 August 2007

mass. To validate the proposed model, empirical relationships between air velocity and static

Received in revised form

pressure drop were obtained for compacted layers of several storage depths for soya bean,

29 May 2008

maize, rice and wheat mass. The software was written in ANSI C which is transferable to

Accepted 4 June 2008

a variety of platforms. For the construction of 3D geometry and the generation of meshes

Published online 21 August 2008

free-of-charge software was used. The solver software generated a system of linear algebraic
equations using the finite -element method. Three iterative processes were carried out: (1) calculation of a local permeability coefficient, using the pressure distribution in the immediately
previous iteration step, (2) search for the system design point, located in the performance curve
of the aerator fan, and (3) adaptation to refine the mesh. A local criterion to estimate the efficiency of complex aeration system in storage bins was proposed. The simulations showed
good performance. It was considered that the method could be applied to optimise the performance of existing grain stores and lower the engineering costs of new grain stores.
2008 IAgrE. Published by Elsevier Ltd. All rights reserved.

1.

Introduction

Aeration is widely used in grain stores to cool the grain mass,

to avoid humidity migration, to temporarily conserve the humidity of grains, to remove scents from the grain mass, and to
apply fumigation.
The resistance to the airflow in an aeration system depends
on the airflow parameters, on the characteristics of the product surface (i.e. rugosity), on the form and size of any extraneous impurity in the mass, on the configuration and size of the
interstitial space in the mass, on the size and amount of broken grains, and on the depth of the grain.
The research carried out by Shedd (1953), Brooker (1961,
1969), Brooker et al. (1982), Bunn and Hukill (1963), Pierce and

Thompson (1975), Haque et al. (1981), Ribeiro et al. (1983), Jayas

et al. (1987), Maier et al. (1992), Weber (1995), Khatchatourian
et al. (2000), Navarro and Noyes (2001), Khatchatourian and
Savicki (2004), and Khatchatourian and de Oliveira (2006) has
examined the influence of some of these parameters on airflow pattern in seeds storage. A recent review of the reported
mathematical models of airflow through grain mass was presented by Gayathri and Jayas (2007).
With increasing depth of grain storage, the mass of grain
can no longer be assumed homogeneous. Non-homogeneity
can significantly alter the physical parameters involved in
the aeration process, such as air velocity and static pressure
drop. However, there is no research relating compaction of
the grain and the airflow pattern under these conditions.

* Corresponding author. Department of Physics, Statistics and Mathematics, Regional University of the Northwest, Rio Grande do Sul,
R. Sao Francisco, 501, 98700-000 IJUI, RS, Brazil.
E-mail addresses: olegkha@unijui.edu.br (O.A. Khatchatourian), mbinelo@yahoo.com.br (M.O. Binelo).
1537-5110/$ see front matter 2008 IAgrE. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.biosystemseng.2008.06.001

226

biosystems engineering 101 (2008) 225238

Nomenclature
A
a
b
C
c
G
H
i
k
L
LX
M
m
n
n
P
Pe
Q
q
R
S

surface area, m2
product-dependent constant
product-dependent constant
compaction function, dimensionless
product-dependent constant
product-dependent constant
bed depth, m
order number of corresponding inlet
permeability coefficient, m3 kg1 s
bed depth, m
full length of a trajectory, m
total number of experimental points
grain mass, kg
product-dependent constant; inlet number
unit vector normal
pressure, Pa
air entrance or exit pressure in Pa;
global airflow rate, m3 s1 kg1
local specific airflow rate, m3 s1 kg1
product-dependent constant
empirical coefficient

To simulate the aeration of grain, with any type of air distribution systems, it is necessary to develop software to predict the distribution of the parameters, because obtaining
empirical data is very difficult and costly. Most research
on airflow simulation in grain stores is related to onedimensional, two-dimensional or axisymmetric cases; although the flow is usually three-dimensional. Even when
the grain mass distribution is two-dimensional or axisymmetric, the airflow inlets do not satisfy these conditions.
Also, the aeration of large grain stores is frequently carried
out separately in different segments.
The principal objectives of the present work were as follows:
(a) to create a mathematical model, algorithm, and software,
to calculate the static pressure, streamlines, and airflow
velocity distribution in three-dimensions under nonhomogeneous conditions;
(b) to determine the variation in compaction factor for several
depths of grain;
(c) to study the relationship between the air velocity and the
pressure gradient as a function of the compaction factor;
(d) to develop and incorporate into the software a criterion
for system performance based on estimating threedimensional air distribution in grain storage bins; and
(e) to carry out numerical simulations of real and hypothetical
grain stores with aeration to detect areas of operational risk.

2.

Mathematical model

The problem of incompressible viscous isothermal flow is described by the system of equations of continuity [Eq. (1)] and of
NavierStokes [Eq. (2)]:

t
U
V
V
Xi
x
y
z
3
r
DP
m

time, s
intermediate argument
velocity vector, m s1
velocity, m s1
product-dependent constant (i 1, 2, 3)
coordinate located in floor plan, m
coordinate along airflow axis, m
coordinate located in floor plan, m
porosity factor, dimensionless
density, kg m3
pressure drop, Pa
dynamic viscosity, Pa s

Subscripts
a
air
b
bulk
e
entrance, exit
g
grain
i
order number of corresponding inlet
k
kernel
L
local
X
in point X(x, y, z)

div V 0;

DV
grad P mV2 V;
Dt

(1)

(2)

where V is the velocity vector in m s1; r is the density in

kg m3; t is the time in s; P is the pressure in Pa; m is the dynamic viscosity in Pa s.
The solutions of this system (usually reduced to the nondimensional form) depend on the effective Reynolds number
(calculated on apparent velocity taking into account the porosity of the grain mass) and relate to the pressure and velocity distributions in each point of the integration domain for
each moment in the form of a vector-function V f (grad P),
where the components u, v and w of velocity V and P are primitive variables of the initial system.
However, experimental data show that the relationships
between the velocity and pressure gradient are different for
each type of grain, even for the same Reynolds number. This
is probably caused by the factors that cause airflow resistance
to vary, e.g. the geometrical form of the particles since grains
are not spherical they are distinct for different products; zones
within the grain mass exist where there is limited porosity
and there are differences in the rugosity of particle surface.
There are also other factors, e.g. grain layer compaction, variation of humidity content, and the presence of impurities
that create differences between the measured values and
those calculated by solutions of the system described by Eqs.
(1) and (2). This implies that attempts to simulate the airflow
through the grain mass using the equations of continuity
and NavierStokes whilst contributing to our theoretical understanding of the problem are far from practical. The local
air velocity in an aerated grain storage can vary over a wide

227

biosystems engineering 101 (2008) 225238

range depending on the cross-sectional area and on the design

of the aerator. Grain stores can have regions of laminar, turbulent and transition flows. This complicates the creation of
mathematical models based on the use of the NavierStokes
equation.
For small velocities corresponding to laminar flow, a proportional relationship exists between the air pressure drop and the
air velocity (i.e. the HagenPoiseuille or BlakeKozeny equation):
dP=dyfV0V kdP=dy;

(3)

where k is coefficient of proportionality; V jVj is the absolute

value of velocity, m s1.
Applying logarithms and taking the derivative produces
dln V
1:
dlnjgradPj

(4)

For the turbulent regime that corresponds to the larger

values of air velocity, the pressure drop is proportional to
the velocity squared (i.e. the BurkePlummer equation) where
1=2

dP=dyfV2 0V kjdP=dyj

(5)

can produce significant errors. For large grain storage bins,

especially with aeration in sections, there are regions with
the increased air velocity and regions where air velocity can
be practically zero. In these cases for calculations of the distribution of pressure and velocity the variation in flow conditions has a significant influence.
It is difficult to describe precisely airflow by means of these
relationships (which depend only on two constants) for all flow
regimes (laminar, transition and turbulent flows). If coefficients
a and b are chosen to accurately describe the transition regime,
the influence of velocity in limiting situations (laminar or
turbulent regime) will be too strong. If limiting regimes are
well described, then, the relationship for the transition regime
is insufficiently exact. Moreover, these relationships when
applied to two-dimensional and three-dimensional cases are
difficult to analyse.
Finding the derivative dln V=dlnjdP=Lj from both Eqs. (7)
and (8) respectively:
1 bVln1 bV
dln V
;

dlnjdP=Lj 21 bVln1 bV  bV

(9)

Thus, for turbulent flow

(6)

For the transition flows the relationship between the air

pressure drop and air velocity lies between linear and square
law dependency.
There are a large number of nonlinear motion equations in
the literature to describe airflow in porous media (Scheidegger, 1960; Bear, 1988). In most of these equations the gradient
of pressure is expressed as function of velocity by second-order
parabola without a free term, i.e. as the sum of dependences
for the laminar and turbulent regimes.
The fullest recommendations for estimating static pressure requirements are given by Navarro and Noyes (2001)
and the basic results of works for pressure-drop modelling
in stored grain masses are complied in ASAE (2000). In both
these studies the equation of Hukill and Ives (1955) has been
adapted to calculate the pressure gradient:
DP
aV2
;

ln1 bV
L

(8)

where R and S are product-dependent constants.

For example, this equation was used for simulation of airflow through packed bed of grain in works of Haque et al.
(1981), Haque et al. (1982), and Hunter (1983).
Eqs. (7) and (8), i.e. two-parameter models, present good results to simulate static pressure drop in silos when the velocity is similar in all points of a silo and if this velocity pertains
values where constants a and b (or R and S ) have been accurately determined for a particular grain. When calculating
the distribution of air in large grain storage bins with significant variations in cross-sectional area, when regions of laminar, turbulent and transition flow exist, these relationships

(10)

Both these expressions satisfy to limiting conditions for

laminar and turbulent flow conditions, i.e.




dln V
dln V
1 and lim
0:5:
(11)
lim
V/N dlnjdP=Lj
V/0 dlnjdP=Lj
This means that Eqs. (7) and (8) are capable of describing the
airflow through grain mass for all flow regimes (laminar, transition and turbulent flows). Also Fig. 1, where data from Shedd
(1953) and this study are presented, indicates that the data
predicted by Eq. (9) for recommended a and b values show
a significant divergence from experimental data and excessive
dominance by the transient regime. Estimation of Reynolds
number shows that deviation from Darcys law occurs after
Re 10, and the transient regime occurs where 10 < Re < 60.

(7)

where a and b are constants used to describe a particular

grain. However, in addition to Eq. (7), Navarro and Noyes
(2001) also recommend the use of the following equation:
DP
RV SV2 ;
L

1 RS V
dln V

:
dlnjdP=Lj 1 2RS V

1.0

d(lnV)/d(ln|gradP|)

dln V
0:5:
dlnjdP=dyj

0.9

Laminar-flow
condition

0.8
Transient
regime

0.7

Turbulent
condition

0.6

0.5
-2

10

ln|gradP|, Pa m-1
Fig. 1 Observed and predicted variation of derivative
dln V=dlnjgradPj[fgradP for airflow through soya bean
mass (blue points and curves) and wheat mass (black points
and curves): ,, 6, Shedds (1953) data ; -, :, authors data;
predicted by Eq. (12); - - -, predicted by Eq. (9).

228

biosystems engineering 101 (2008) 225238

Where Re > 60 turbulent flow occurs. These results are in

agreement with the data of Wright (1968) and Bear (1988).
Therefore the use of Eq. (7) is limited, since it is very difficult to simultaneously achieve good results for both Eqs. (7)
and (9), using only two constants. The same problem concerns
Eqs. (8) and (10). Also, Eqs. (9) and (10) depend on only one constant b and S/R, respectively. To improve accuracy different
values of factors a and b are usually adopted for different intervals and this can be very inconvenient.
Khatchatourian and Savicki (2004) proposed the formula to
describe the variation of the derivative d(ln V)/d(ln(jdP/dyj)) for
all the three flow conditions corresponding to the laminar,
turbulent and transition flows:
dln V
3 arctanU

;
dlnjgrad Pj 4
2p

(12)

where U(P) a ln(jgrad Pj) b is an intermediate argument;

a > 0 and b are constants.
Evidently, when jgrad Pj / 0, U / N,
lim 3=4
u/N
arctanU=2p 1, which corresponds to the laminar flow; and
when jgrad Pj / N, U / N, lim 3=4 arctanU=2p 0:5,
u/N
which corresponds to the turbulent flow, i.e. Eq. (12) satisfies to
limiting conditions Eq. (11).
Fig. 1 shows reasonable agreement between the curve calculated by the Eq. (12) and the experimental data.
Integrating Eq. (12) in relation to the logarithm of the pressure gradient gives the expression for the velocity:



 
(13)
ln V ln 1 U2  2U arctanU p 3U 4a c;
where c is a constant of integration.
As will be shown, this equation, depending on three constants (a, b and c), describes well the experimental data in all
regions. In addition, unlike Eq. (7), explicit dependence of velocity on a pressure gradient in Eq. (13) essentially simplifies
its use together with the equation of continuity for problems
formulated in two-dimensional and three-dimensional.
Finally, the mathematical model of the airflow in the particular media for the three-dimensional case consists of a system of two equations:
div V 0;

where the y coordinate in m corresponds to vertical direction,

the x and z coordinates are located in the perforated floor plan.
Substituting Eq. (17) in Eq. (14), the nonlinear partial differential equation is obtained:






v
vP
v
vP
v
vP
k

k

k
0:
(18)
vx
vx
vy
vy
vz
vz
The boundary conditions for the problem considered have
the form:
P Pe

of the silo;

(15)

The scalar equation (14) is the continuity equation for incompressible fluid. The vector equation (15), which has replaced
the NavierStokes equation, shows that the velocity vector
and pressure gradient are collinear in all points of the airflow
domain and that the ratio of the absolute values of these vectors is a function of the pressure gradient. Expressing the coefficient of proportionality k by




 
k exp ln 1U2 2U arctanU p3U 4ac jgrad Pj;
(16)
and using Eq. (15), the velocity components u, v and w for the
three-dimensional case can be expressed in the form
vP
vP
vP
; v k ; w k ;
vx
vy
vz

20

where Pe is air entrance or exit pressure in Pa; and n is unit

vector normal to the wall or floor surface.
Eqs. (16)(18) along with the boundary conditions, Eqs. (19)
and (20), describe the steady-state pressure and velocity distributions in a cross-section of an aerated grain storage.

3.

Software description and development

The nonlinear partial differential equation for pressure Eq.

(18) was solved by the finite-element method (Segerlind,
1976) using an iterative process to calculate the permeability
coefficient k using Eq. (16) in each point of the integration domain and using the pressure distribution from the immediately previous iteration step.
The software, developed in ANSI C, consisted of tools
for geometry construction, mesh generation, generation of
system matrix, solver of obtained system of linear algebraic
equations and tool for results three-dimensional presentation
and analysis. Since commercial tools can be costly, free-ofcharge software was used when possible.

(14)

V

u k

(19)

n  grad P 0 Neumann condition on the walls and floor

3.1.


 
grad P
exp ln 1 U2  2U arctanU p
jgrad Pj


3U 4a c :

Dirichlet condition for air entrance and exit;

(17)

Geometry construction

The geometry of the system can be constructed in any system

CAD, CAE, or any three-dimensional modelling software package that can export data to a standard format. In this work
Blender3D was used (http://www.blender.org). This software
is available at no-cost under General Public License (GPL). It
is three-dimensional modelling software aimed at artistic
works, but it proved to be very efficient for constructing the
three-dimensional geometry of the storage bins. A user-interface was developed in Lazarus (http://sourceforge.net/projects/lazarus/) to create geometry choosing basic storage bin
dimensions.
The storage bin geometry data were exported to smash file
format, which is a format that can be read by Tetgen (http://
tetgen.berlios.de/). A Perl script (http://www.perl.org/), used
for exporting data, was modified to include exporting face materials. Different face materials were used in order to recognise the surfaces with different bounding conditions, such
as inlets and outlets.

biosystems engineering 101 (2008) 225238

3.2.

Mesh generation

For mesh generation Tetgen, available under a GPL license

was used. It generates quality tetrahedral meshes using
Delaunay algorithms. Firstly, a coarse mesh was generated.
To refine the obtained mesh, a quality mesh file was generated
by the solver, then, Tetgen was used to refine the mesh
according to the parameters indicated in this file. The best results were obtained by dynamic adaptive refinement of the
mesh, based on a tetrahedron size selection in inverse proportion to the tetrahedron pressure gradient. Each tetrahedron
not satisfying the user specified ratio was recursively decomposed into eight new tetrahedral elements according to the
method shown in Liu and Joe (1996).

3.3.

229

4.
Validation of the mathematical model for
non-homogeneous conditions in a grain mass
To validate the proposed mathematical model, the empirical
relationships between air velocity and static pressure drop
were obtained for compacted layers with several grain storage
depths. The coefficients a, b and c presented in the mathematical model were obtained experimentally for soya bean,
maize, rice and wheat grains. In large storage bins, due to
compaction, grain mass is a non-homogeneous medium and
the permeability coefficient varies as a function of the grain
layer depth as well as pressure gradient. Therefore the influence of the grain mass compaction factor on the permeability
coefficient was investigated.

Problem solving and representation

4.1.

The developed code is cross-platform and can be compiled in

any ANSI C compatible compiler. The input files to the
solver software are the output files from Tetgen which describes nodes, faces and tetrahedral elements, and generates
a file describing the boundary conditions and precision requirements. Firstly, the solver software generates the local
matrix for each tetrahedron applying the finite-element
method. Using the local matrix information, the global system
matrix was generated. Since the system order was large and
the matrix was very sparse, a special class was created to handle the matrix, optimising memory and also optimising the
time to access the elements. Instead of using standard sparse
matrix classes, the in-house programming of the special class
enables full advantage to be taken of the system peculiarities,
optimising both memory usage and processor time. The
successive over-relaxation (SOR) method (Hageman and
Young, 1981) was used for resolving the system of linear algebraic equations. The developed solver was shown to have
good performance.
The software executes three iterative processes: (1) it calculates the permeability coefficient in each point of the integration domain, using the pressure distribution in the
immediately previous iteration step, (2) it searches the system
design point, located in the performance curve of the aerator
fan, and (3) it adaptively refines the mesh according to the tetrahedron size per pressure gradient ratio.
After the system is solved, an output file is generated in
VTK (Visualization toolkit, http://www.vtk.org/) format. This
file includes the nodes and tetrahedral elements. For each
node the value of pressure and for each tetrahedron the velocity
vector is exported. Paraview software (http://www.paraview.
org/), which is open source and available free-of-charge, was
used for to visualise the results.
The velocity vector for the used scheme of a finiteelement method is constant inside the simplex element
(tetrahedron). Using theory of consistent conjugate approximation (Oden and Reddy, 1973) velocities in all vertices of
the tetrahedrons were calculated, i.e. a continuous vector
field was obtained. Further, for each vertex, the full airflow
trajectory length (from inlet up to outlet) was calculated.
The received values were then used to calculate the local criterion introduced in this work to estimate the ventilation
system performance.

Experimental equipment

To simulate the aerated grain storage characteristics, the

equipment, described by Khatchatourian and Savicki (2004),
was used to experimentally determine the grain mass compaction factor caused by the weight of layers above. The grain
mass porosity varied as a function of the layer depth. The influence of compaction on the relationship between the airflow
velocity and the static pressure drop was analysed.
Fig. 2 shows the experimental equipment which consisted
of a centrifugal fan, an orifice-plate and small silo composed
of a polyvinyl chloride tube (inside diameter of 0.2 m and
height of 1 m) or a steel tube (inside diameter of 0.11 m and
height of 1 m). To model the conditions at the bottom of a grain
store, a compacting device was developed with a lever, which
made it possible to apply enough force to simulate the depth
up to 50 m. In the tests, soya beans, maize, and wheat had
a moisture content of 1213% and rice had a moisture content
of 10%. Impurities were less than 2%, as determined by the
Laboratory of Seeds Analysis, Department of Agrarian Studies,
Regional University of the Northwest, Rio Grande do Sul UNIJUI, Brazil.

4.2.

Experimental results

The experimental results, presented in Fig. 3, show the relationship between airflow velocity and static pressure drop in
the soya beans, shelled maize, rice and wheat mass. Table 1
shows the values of empirical model coefficients a, b and c,

Fig. 2 Sketch of the experimental equipment.

230

biosystems engineering 101 (2008) 225238

0.5

Velocity, m s-1

ln(dP/dy), Pa m-1

0.4
-1

-2

-3

0.3

0.2

0.1

0.0
-4

500

1000

lnV, m s-1
Fig. 3 Relationship between air velocity (V) in m sL1 and
air pressure drop (dP/dy) in Pa mL1; ,, soya bean,
coefficient of correlation R2 [ 0.9954; 6, shelled maize
R2 [ 0.9982; B, rice, R2 [ 0.9934; >, wheat, R2 [ 0.9972; d,
predicted by Eq. (13).

obtained by minimising the residual error between experimental and simulated data. The simulations based on these
coefficients satisfactorily described the experimental data
(Fig. 3).
Experimental data in Fig. 4 show the significant influence
of the storage layer depth on the aerodynamic resistance of
the grain mass over the studied depths (from 1 m up to 50 m).
Fig. 5 presents a reduction of the measured porosity factor 3
with storage depths for soya bean, maize and rice, where 3 is
the ratio of the void volume to the total bed volume. Experimental porosity values were measured using a specially developed and adjusted pycnometer. The relationship between the
reduction in porosity and layer depth H can be presented as
H n
3
eS50 ;
30

(21)

where 30 is the porosity factor for H 1 m dimensionless. The

empirical coefficients S and n, which were obtained by leastsquares method, are presented in Table 2.
The analysis of the measurements of the porosity factor 3
for various bed depths indicated that the effective velocity increase due to reductions in the porosity factor in the deepest
layers was not sufficient to explain and calculate the pressure
losses under these conditions. The values in Fig. 4, calculated
by using a porosity reduction for H 50 m, are significantly
different from the corresponding experimental points.

1500

2000

2500

3000

dP/dy, Pa m-1
Fig. 4 Influence of bed depth (H ) on the air velocity (V) in
m sL1 as function of air pressure drop (dP/dy) in Pa mL1
(one-dimensional storage), shelled maize: -, H [ 1 m; ,,
H [ 10 m; C, H [ 20 m; B, H [ 30 m; :, H [ 40 m; 6,
H [ 50 m; d predicted; $, predicted by porosity reduction
for H [ 50 m; - - -, predicted by Eq. (25).

However, the greater porosity did not guarantee smaller resistance to airflow in the grain mass. For example, the rice in
the husk (or paddy) had a resistance greater than the maize although the rice porosity was greater. It is possible that free
volumes of air between husk and the grain increased porosity
but did not increase cross-sectional area for airflow.
It must be concluded that, besides the global non-uniformity defined by the alteration of the mean porosity factor
with the depth variation, there is local non-uniformity caused
by the seed form and that this does not significantly alter the
porosity factor value of the medium. Probably, the compaction
of non-spherical seeds creates local dense regions through
which airflow is hindered.
The experimental data presented in Fig. 6 show that for the
studied velocity and depth variation intervals, the relative
pressure gradient increment C (jgrad PHj  jgrad P0j)/jgrad P0j
can be considered as being independent of air velocity and depends only on the storage layer depth H, where H is a distance
between the upper seed surface (free surface) and the layer
under consideration. This hypothesis was confirmed by
multi-factorial analysis of variance and by a nonparametric
association test of the Spearman rank order correlation
(Table 3).
The function C C(H ) relates to the initial pressure gradient jgrad P0j, where P0 corresponds to grain depth H 1 m,
and the pressure gradient jgrad PHj for considered depth H at

Table 1 The empirical coefficients a, b and c with 95% confidence bounds for different seeds, sum squared error (SSE),
coefficient of determination (R2) and root mean squared error (standard error) for Eq. (13)

Soya bean
Maize
Rice
Wheat

SSE

R2

RSME

0.82  0.12
0.61  0.07
0.51  0.13
0.86  0.15

3.57  0.66
2.92  0.39
3.08  0.82
5.49  0.98

2.77  0.12
2.75  0.08
2.23  0.13
2.18  0.06

0.5013
0.1304
0.3526
0.1296

0.9954
0.9982
0.9934
0.9972

0.0480
0.0289
0.0525
0.0348

231

biosystems engineering 101 (2008) 225238

0.5

1.00

C=(gradP - gradP0)/gradP0

0.99
0.98
0.97
0.96
0.95
0.94

0.4

0.3

0.2

0.1

0.93
0.92

0.0
0.1
0

10

20

30

40

Bed Depth, m
Fig. 5 Porosity reduction with bed depth; -, soya bean;
:, maize; d, predicted; - -, 95% confidence bounds for
prediction.

same velocity. This function, designated as the compaction

function in this work, was presented in the form


CH G 1  eaH ;
(22)
where G and a are empirical product-dependent constants
presented in Table 4 for soya bean, maize and rice.
These constants were obtained by minimising
2
M 
X
grad Pi  grad P0
G1  eaHi 
;
(23)
min
a;G
grad P0
i1
where M is the total number of experimental points for selected grain type.
Fig. 7 shows the variation of compaction function with
layer depth for soya bean and maize. The compaction function allowed the influence of the depth H to be included in
the model through the intermediate argument U, substituting
the pressure gradient without compaction jgrad P0j for the expression jgrad PH /(1 C )j:
U a lnjgrad PH =1 Cj b:

(24)

As a result, Eqs. (13) and (22) and with the intermediate argument in Eq. (24) relate the air velocity for the storage layer located in the depth H, and the necessary pressure gradient.
Fig. 4 shows close agreement between observed and predicted
data.
To take account of the grain bulk density (and in implicit
form the bed depth) the ASAE Standards 2000 recommends
using the equation obtained by Bern and Charity (1975):

0.3

0.4

0.5

Fig. 6 Variation of the compaction function

C[(jgradPHj L jgradP0j)/jgradP0j for shelled maize with bed
depth H at various air velocities V in m; ,, H [ 10 m; C,
H [ 20 m; B, H [ 30 m; :, H [ 40 m; 6, H [ 50 m;
d predicted.

 2
r 
rb
b
V
V2
DP
rk
rk

X
;
X1 X2 


3
r 3
r 3
L
1 b
1 b
rk
rk

(25)

where DP is pressure drop, Pa; L is bed depth, m; rb is product

bulk density, kg m3; rk is product kernel density, kg m3; X1,
X2 and X3 are constants.
Eq. (25), based on the equation from Ergun (1952), represents a three-parameter model and describes the relationship
between airflow and pressure drop better than Eqs. (7) and (8).
Unfortunately, as Fig. 4 shows, it is impossible to describe the
effect of bed depth on resistance to airflow of grain for all
range of airflow with permanent values X1, X2 and X3 and
Eq. (25). The empirical coefficients X1, X2 and X3 for this simulation were obtained by minimisation of the residual error between observed and simulated data, using Eq. (21) and the
relationship
rb =rk 1  3:

(26)

Navarro and Noyes (2001) recommended calculating the average value of the grain bulk density during filling a silo by

Table 3 Influence of bed depth H and air velocity V on

compaction factor C
F-value

Probability > F

R2

Soya bean
Depth, m
Velocity, m s1

76.6
0.54

0.002
0.855

0.958
0.035

Variable

Table 2 Porosity factor 30 for H [ 1 m and empirical

coefficients S and n of Eq. (21) for different seeds

0.2

Velocity, m s-1

50

Seed type

30

Coefficient of
determination (R2)

Shelled corn
Depth, m
Velocity, m s1

221.6
5.05

0.001
0.103

0.954
0.031

Soya bean
Maize
Rice

0.43
0.44
0.61

0.0680
0.0736
0.0664

0.5261
0.4683
0.5134

0.9985
0.9997
0.9989

Rice
Depth, m
Velocity, m s1

147.2
4.03

0.001
0.138

0.958
0.045

232

biosystems engineering 101 (2008) 225238

Table 4 The empirical coefficients G and a of compaction

function C, Eq. (22), for different seeds
Seed type

Standard error
Coefficient of
(RSME)
determination (R2)

Soya bean 0.6865 0.0345

Maize
0.8514 0.0155
Rice
0.7652 0.0171

0.9943
0.9858
0.9872

0.0111
0.0176
0.0155

means of known weight of loaded grain and the calculated

volume which will occupy this grain in a silo. Representing
the depth-static pressure of the grain using a nomograph is
another recommended way of estimating static pressure requirements. The basic values for loose grain were increased
as follows: 30% for wheat, 34% for maize, and 41% for sorghum
and soya beans.
For silo aeration systems this procedure gives admissible
results. For large grain storage bins, for bins with significant
variation in cross-sectional area, and for bins using aeration
in parts, such estimations can be unacceptable.
In our view, estimating the influence of bed depth on pressure increase by means of Eq. (22) is preferable to using only
one formula. Also, calculations and experiment have shown
that this dependence is very useful for inverse problem solution, when for known integral parameters of airflow (e.g. air
consumption and total pressure head) define the characteristics of grain mass which influence aerodynamic resistance.
Coefficient G in Eq. (22) can be single variable parameter since
the sensitivity of coefficient a in Eq. (22) is insignificant in
comparison with sensitivity G.
To simulate the storage bins of complex layouts, the software was operated using an iterative process which determines the equilibrium between the fan output and the
resistance of the aeration system to airflow, i.e. the operating
point of the aerator fan. The software then calculates: (1) the
pressure needed to get the required airflow rate (estimating
static pressure requirements); (2) the airflow rate, knowing

0.6
0.5

Fig. 8 Outline sketch of simulated store bin.

the initial pressure; and (3) the pressure and airflow rate in
an iterative process for the chosen fan and electric motor (by
estimating system design point).

5.

Numerical simulations

Fig. 8 shows the structural layout of real V-form floor storage

bin, used in the state of Rio Grande do Sul, Brazil. The storage
bin has a maximum width of 30 m and length of 95 m. Three
air inlet systems were analysed: (1) a central inlet system; (2)
a system with central and upper lateral inlets; and (3) a system
with central, lower lateral and upper lateral inlets. The aeration simulations in storage bins, for different layouts, were
generated using the global airflow rate of Q 9 m3 h1 t1
(2.5  106 m3 s1 kg1), which is the most commonly recommended value for aerated grain storage.
Firstly, airflow simulation in the V-form floor storage bin
was made for case 1 (air inlet ducts installed in the base of
the storage system). Although in this case the storage bin
has two axes of symmetry and it is possible to consider only
1
4 th of total storage, the simulation was carried out for complete domain, because generally symmetry conditions do not
exist.

0.4
0.3
0.2
0.1
0.0

10

20

30

40

50

Layer Depth, m
Fig. 7 Variation of the compaction function
C [ (jgradPHj L jgradP0j)/jgradP0j with bed depth (H ) in m
for soya bean and shelled maize: C, soya bean, observed
data; 6, shelled maize, observed data; , 95% confidence
bounds for prediction; - -, non-simultaneous bounds for
observation; , predicted.

Fig. 9 Surface wireframe of the tetrahedral mesh.

biosystems engineering 101 (2008) 225238

Fig. 10 Isobaric surfaces in storage bin section with

central, lower lateral and upper lateral inlet systems.

Fig. 9 shows part of the computational mesh used. The grid

had a higher density in regions where the pressure gradient
was greater. For the layout under consideration the number
of tetrahedrons was approximately 500,000.
Isobaric surfaces for storage bin section with central, lower
lateral and upper lateral inlet systems are shown in Fig. 10. It
can be seen that airflow in lower storage section was three-

233

Fig. 12 Schematic model for determination of the local

specific airflow rate.

dimensional in character. In the upper section of the storage,

the character of the airflow approached the two-dimensional
case.
The simulation results of three aeration systems under
consideration are shown in Fig. 11. Analysis of the pressure
distribution (left column) showed that the installation of

Fig. 11 Comparison of three simulated aeration systems: distribution of pressure (left column) and risk regions
(right column).

234

biosystems engineering 101 (2008) 225238

Fig. 13 Visualisation of regions with inadequate ventilation (qL < 4.5) for three air inlet systems: (1) central inlet system; (2)
central and upper lateral inlet systems; and (3) central, lower lateral and upper lateral inlet systems; Q [ 9 m3 tL1 hL1.

lateral ducts essentially equalised the airflow when compared

with the same storage bin without lateral ducts and reduced
the initial pressure head. To analyse the distribution of parameters, the software was used to show the storage sections,
which satisfied certain conditions. For example, the frameworks in Fig. 11 (right column) show only cells where velocity
is <0.015 m s1, i.e. the regions with reduced aeration
capacity.

Qa
VA
V

mg rb HA rb H

where Qa is total air flow rate in m3 s1; mg is total grain mass

in kg; V is air velocity in m s1; rb is product bulk density in
kg m3 s1; A is the cross-sectional area in m2; H is the grain
mass depth in m.
The local specific airflow rates for storage bins with the
variable cross-sectional area (Fig. 12(b)) for all internal points
X X(x, y, z) can be presented as

5.1.
Criterion for describing the efficiency of aeration
system
qL X
To estimate the efficiency of an aeration system the specific
airflow rate is usually specified. This is obtained from the ratio
between the total airflow rate and the total product mass. This
criterion is suitable for simple silo designs with constant
cross-sectional area, when the air velocity is uniform throughout the storage. If variations in the cross-sectional area are
significant, or the aeration distribution system is complex
(e.g. case 3), this criterion is not suitable.
To evaluate aeration efficiency for storage bins with variable cross-sectional area and with complex air distribution
system, a local specific airflow rate is proposed. For simple
storage layouts (Fig. 12(a)) with constant cross-sectional area
the specific volume airflow rate Q is

(27)

VX
;
rXLX

(28)

where qL(X ) is local specific airflow rate in point X(x,y,z) in

m3 s1 kg1; V(X ) is air velocity in point X in m s1; r(X ) is
product density in point X in kg m3 s1; LX is full length in
m of a trajectory on which the point X is located.
It is evident that the difference between local value of airflow rate at point P and its total average value gives a better estimation of aeration system efficiency than the comparing
local and average velocities. For constant cross-sectional
area storage bins the value of the local specific airflow rate
at all storage points is constant and equal to the value of global
specific airflow rate.

Fig. 14 Visualisation of regions with the excessive intensity of ventilation (qL > 18) for three air inlet systems: (1) central
inlet system; (2) central and upper lateral inlet systems; and (3) central, lower lateral and upper lateral inlet systems;
Q [ 9 m3 tL1 hL1.

biosystems engineering 101 (2008) 225238

235

Fig. 15 Distribution of local specific airflow rate, Q [ 9 m3 hL1 tL1 (2.5 3 10L6 m3 sL1 kgL1): (a) central inlet and (b) lower
lateral inlet.

A local criterion multiplied by aeration time has additive

properties. This allows the quality of aeration to be calculated for all parts of the storage bin even if the ventilation
is carried out separately at each of the inlets over different
periods of time.

5.2.

Numerical simulation results

To visualise risk domains in the grain storage bin, the distribution of local specific airflow rates was studied (Figs. 1319).
Fig. 13 shows the visualisation of domains with inadequate
ventilation (qL < 4.5) for three air inlet systems: (1) a central

inlet; (2) central and upper lateral inlets; and (3) central, lower
lateral and upper lateral inlets. As simulations show, the system with central, lower lateral and upper lateral inlets considerably improved the conditions of storage in regions close to
walls when compared with other inlet systems. For all cases
considered there was an area of risk in the uppermost part
of grain mass.
The regions with the raised intensity of ventilation (qL > 18)
are shown in Fig. 14. The results obtained show that the second system (central and upper lateral inlets) has a smaller volume with excessive intensity of ventilation in comparison
with others, i.e. has improved efficiency.

Fig. 16 Distribution of local specific airflow rate, Q [ 9 m3 hL1 tL1 (2.5 3 10L6 m3 sL1 kgL1): (a) upper lateral inlet and (b)
central, lower lateral and upper lateral inlets with identical initial pressures.

236

biosystems engineering 101 (2008) 225238

Fig. 17 Distribution of local specific airflow rate, Q [ 9 m3 hL1 tL1 (2.5 3 10L6 m3 sL1 kgL1): (a) central and lower lateral
inlets with different initial pressures and (b) central, lower lateral and upper lateral inlets with different initial pressures.

Additional more detailed comparative analyses of the efficiency of different aeration systems were made for the same
grain storage bin with the same global specific airflow rate
Q 9 m3 h1 t1 (2.5  106 m3 s1 kg1). The number of inputs
(from one up to three), their position (upper lateral, lower lateral and central inlets), and ratio of pressure between various
inputs were varied. Using the additive property of the local
specific airflow rate, estimations of ventilation system efficiency
were carried out separately using each of inlets during the
different periods of time. Relationships between the duration

of ventilation time through each inlet and the airflow rates

were chosen so that the global specific airflow rate Q was equal
to 9 m3 h1 t1 (2.5  106 m3 s1 kg1). The simulation results
for the distribution of local specific airflow rates in a plane of
symmetry in the grain storage bin are presented in Figs. 1518.
As the simulation results presented in Figs. 15 and 16(a)
show, if only one airflow inlet is used, there is always a large
area with superfluous ventilation. Since the airflow tends to
leave the grain mass through the line of least resistance, moving the airflow inlet from the lower position to the upper

Fig. 18 Distribution of resultant local specific airflow rate with separated functioning inlets, Q [ 9 m3 hL1 tL1
(2.5 3 10L6 m3 sL1 kgL1): (a) upper lateral, lower lateral and central inlets with equal application times (1:1:1) and (b) upper
lateral, lower lateral and central inlets with different application times (1:2:2).

biosystems engineering 101 (2008) 225238

237

Fig. 19 Visualisation of domains with the lowered intensity of ventilation (qL < 4.5; right) and with excessive intensity of
ventilation (qL > 18; left); qL was obtained by superposition of simulations for separated functioning of upper lateral, lower
lateral and central inlets with different application times (1:2:2); Q [ 9 m3 hL1 tL1 (2.5 3 10L6 m3 sL1 kgL1).

position provokes a pressure reduction and a deterioration in

ventilation uniformity. Therefore, there is area with excessive
ventilation close to the upper lateral inlet if all of three inlets
operate together with identical pressure (Fig. 16(b)).
By selecting the appropriate pressures ratio for the inlets it
is possible to considerably improve the system of air distribution in the storage bin. This is demonstrated in Fig. 17(a) for
two inlets and in Fig. 17(b) for three inlets.
In high capacity storage, the grain ventilation is usually
carried out stage by stage, serially using air inlets located in
different storage sections. Under these conditions the advantage of using of local specific airflow rates for ventilation efficiency estimation is especially great. In these cases the
resultant local specific airflow rate qL in each point of storage
bin can be calculate by the expression
Pn
ti qi
;
qL Pi1
n
i1 ti

(29)

where qi is the local specific airflow rate corresponding to ventilation with only one inlet (order number i); ti is ventilation
time with only one inlet (i); n is total number of inlets; i is order
number of corresponding inlet.
For example, Fig. 18 shows the distribution of resultant local specific airflow rates with separated operation of the upper
lateral, lower lateral or central inlets. In case (a) the application time is the same for each inlet, and the resultant local
specific airflow rate at each point of the storage bin can be calculate by
1
1
1
qL q1 q2 q3 ;
3
3
3

(30)

where q1, q2 and q3 are local specific airflow rates corresponding to upper lateral, lower lateral or central inlets.
The simulations presented in Figs. 16(b) and 18(a) indicate
the significant advantage of ventilation carried out in turn by
each of inlets in comparison with the simultaneous use of all
inlets at equal pressures. This improvement is caused because
the capacity for air to penetrate to all zones under the dominant influence of each inlet results in a more uniform distribution of qL. By varying the duration of aeration for each inlet, it

is possible to find a optimum distribution of qL for a given storage bin design.

For example, Fig. 18(b) shows the distribution of qL in the
grain storage bin with alternate use of upper lateral, lower lateral or central inlets for durations of aeration varying as
t1:t2:t3 1:2:2. As results showed, this distribution had the
fewest regions with insufficient or excessive aeration. These
regions are shown in Fig. 19 for whole grain storage bin.
Unfortunately, where only one inlet was used there was an
inevitable increase in head pressure.

6.

Conclusions

A mathematical model of three-dimensional airflow in an aerated grain storage system was developed for non-uniform
conditions of the seed mass. Experiments were conducted to
obtain the relationship between air velocity and pressure gradient and the values of the porosity factors for different seed
types and different storage layer depths. A local criterion
was proposed to estimate the efficiency of complex aeration
system in grain storage bins.
Software was developed to determine the velocity, pressure and local specific airflow rates distributions, the global
airflow rate or initial pressure head in the grain mass store
for three-dimensional cases. The aeration system efficiency
of several stored seeds was analysed to provide the airflow
distribution uniformity and the static pressure head values
that generate the appropriate airflow rate for safe storage.
It was shown that the aeration system of grain storage bin
can be essentially improved by the use of inlets system with
different initial pressures selected for each inlet. Also, it was
shown that it is possible to optimise air distribution in a grain
storage bin by operating each inlet in turn and by selecting
a suitable aeration period for each inlet.

Acknowledgements
The authors would like to thank CNPq for the financial support for this work (process No. 464380/00-6).

238

biosystems engineering 101 (2008) 225238

references

ASAE (2000). Resistance to Airflow of Grains, Seeds, Other

Agricultural Products, and Perforated Metal Sheets. American
Society of Agricultural Engineers, St. Joseph, MI.
Bear J (1988). Dynamics of Fluids in Porous Media. Dover
Publications, New York.
Bern C J; Charity L F (1975). Airflow Resistance Characteristics of
Corn as Influenced by Bulk Density. ASAE Paper No. 75-3510.
ASAE, St. Joseph, MI 49085.
Brooker D B (1961). Pressure patterns in grain drying system
established by numerical methods. Transactions of the ASAE,
4, 7277.
Brooker D B (1969). Computing air pressure and velocity
distribution when air flows through a porous medium and
nonlinear velocitypressure relationship exists. Transaction
of the ASAE, 12, 118120.
Brooker D B; Bakker-Arkema F W; Hall C W (1982). Drying Cereal
Grains. AVI Publishing Co., Inc., Westport, CT.
Bunn J M; Hukill W V (1963). Pressure pattern prediction for nonlinear air flow through porous medium. Transactions of the
ASAE, 6, 3236.
Ergun S (1952). Fluid flow through packed columns. Chemical
Engineering Progress, 48, 8994.
Gayathri P; Jayas D S (2007). Mathematical modeling of airflow
distribution in grain bulks a review. ASAE Annual Meeting
076226, pp. 112.
Hageman L A; Young D M (1981). Applied Iterative Methods.
Academic Press, New York.
Haque E; Chung D S; Forster G H (1981). Pressure and velocity field
in airflow through packed bed of corn with fines non-Darcy
flow conditions. Transactions of the ASAE, 5, 15951604.
Haque E; Ahmed Y N; Deyoe C W (1982). Static pressure drop in
a fixed bed of grain as affected by grain moisture content.
Transactions of the ASAE, 25(4), 10951098.
Hukill W V; Ives N C (1955). Radial airflow resistance of grain.
Agricultural Engineering, 36, 462466.
Hunter A J (1983). Pressure difference across an aerated seed bulk
for some common duct and store cross-sections. Journal of
Agricultural Engineering Research, 28, 437450.

Jayas D S; Sokhansanj S; Moysey E B; Barber E B (1987). The effect

of airflow direction on the resistance of canola (rapeseed) to
airflow. Canadian Agricultural Engineering, 29(2), 189192.
Khatchatourian O A; Savicki D L (2004). Mathematical modelling
of airflow in an aerated soya bean store under non-uniform
conditions. Biosystems Engineering, 88(2), 201211.
Khatchatourian O A; de Oliveira F A (2006). Mathematical
modelling of airflow and thermal state in large aerated grain
storage. Biosystems Engineering, 95(2), 159169.
Khatchatourian O A; Toniazzo N A; Borges P A (2000). Simulacion
numerica del flujo de aire en silos para el almacenamiento de
granos. [Numerical simulation of airflow in grain silos].
Informacion Tecnologica, Chile, 11(4), 175182.
Liu A; Joe B (1996). Quality local refinement of tetrahedral meshes
based on 8-subtetrahedron subdivision. Mathematics of
Computation, 65(215), 11831200.
Maier D E; Moreira R G; Bakker-Arkema F W (1992). Comparison
of conventional and chilled aeration of grains under Texas
conditions. Applied Engineering in Agriculture, 8(5),
661667.
Navarro S; Noyes R T (eds) (2001). CRC Press LLC, USA.
Oden J T; Reddy J N (1973). Note on an approximate method for
computing consistent conjugate stresses in elastic finite
elements. International Journal for Numerical Methods in
Engineering, 6, 5561.
Pierce R O; Thompson T L (1975). Airflow pattern in
coaxial-shaped piles of grain. Transactions of the ASAE, 18,
946949.
Ribeiro C A; Fortes M; Hara T (1983). Escoamento de ar em silos.
[Airflow in silos]. Revista Brasileira de Armazenamento,
Vicosa, 891(20), 2831.
Scheidegger A E (1960). The Physics of Flow through Porous Media
(2nd ed.). University of Toronto Press, Toronto.
Segerlind L J (1976). Applied Finite Element Analysis. J. Wiley and
Sons Inc., New York, USA.
Shedd C K (1953). Resistance of grains and seeds to air flow.
Agricultural Engineering, 34(9), 616619.
Weber E (1995). Armazenagem Agrcola [Agricultural storage].
Kepler Weber Industrial, Porto Alegre.
Wright D E (1968). Nonlinear flow through granular media.
Journal of Hydraulic Division. ASCE, 94(HY4), 851872.