You are on page 1of 6

J. agric. Engng Res.

(1997) 68, 231—236

A Numerical Model for Flow of Granular Materials in Silos.


Part 2: Model Validation
S. C. Negi; Z. Lu; J. C. Jofriet
School of Engineering, University of Guelph, Guelph, Ontario, Canada N1G 2W1

(Received 16 September 1996; accepted in revised form 4 August 1997)

A hybrid numerical model based on finite-element and end has a hopper with an adjustable wall slope. The two
discrete-element methods was verified for modelling the side walls are made of 0·5 mm galvanized steel sheet
flow of granular materials in silos or bins. Laboratory supported by horizontal steel angles spaced 0·3 m apart.
experiments involving gravity flow of soybean from The two end walls are constructed of 5 mm clear acrylic
a parallel wall bin with length-to-width ratio of three sheet to facilitate observations. The bin is supported by
(state of plane strain) were carried out for different outlet four legs, one at each corner. The height from the outlet
widths. Computer simulations successfully matched most to the top of the bin is 1·76 m and the rectangular part is
of the observed features of the physical tests. The results 0·4 m wide by 1·2 m high, so that the height-to-width
showed that the maximum wall pressure occurs near the ratio of three approximated a deep bin. The thickness of
top of the hopper in the transition region from parallel to the experimental arrangement, i.e. the depth perpendi-
convergent flow. The ratio of the dynamic pressure to the cular to the paper is 1·25 m. The particulate solid used for
static pressure at the level of transition ranged from 2·5 to the tests was soybean.
3·6 depending upon the width of the outlet. Wall pressures were measured at two locations, one
( 1997 Silsoe Research Institute 0·06 m above and the other 0·06 m below the intersection
of the rectangular bin and the hopper. There were two
1. Introduction strain gauges (Micro-Measurements type: CEA-09-250
UW-120) at each location in the vertical direction. These
A numerical model coupling finite-element and strain gauges were uniaxial sensors and measured the
discrete-element methods was developed by the authors vertical bending strains in the silo wall. Both upper and
in Part 1 of this paper.1 In this paper, experimental results lower set of strain gauges were in the DEM domain. The
of wall pressures exerted by soybean on a model silo are strain gauges were connected to a data logger (HP 3964
compared with the predictions of the proposed hybrid A) equipped with a high-speed card (D/A convertor)
model. The physical properties of soybean for use in the which was set up to record ten readings per second. Four
numerical model were either experimentally determined or strain-tube load cells were installed, one under each of
obtained from the literature. Because the discrete-element the supporting legs of the bin, and connected to the data
method is extremely computation-intensive, requiring logger. The data logger was linked to a computer con-
a large number of particles and a very small simulation time trolled by a data acquisition program, which recorded
step, the computer simulations of granular flows were con- and analysed the data in real time.
ducted for only a few seconds after the initiation of flow To calibrate the strain gauges and the load cells, the
when the peak wall pressures occur. Statistical analyses silo was filled with water (held in a plastic bag) to 1·6 m
were performed to verify the degree of agreement be- above the outlet. The strain readings were recorded at
tween the observed and predicted data on wall pressures. 0·1 m increments of depth for both filling and emptying of
the bin. For each depth a strain gauge reading was
2. Experimental investigation related to the hydrostatic pressure of water. The com-
plete series of calibration test results provided a calib-
2.1. Model bin tests ration curve for each gauge, relating strain with normal
wall pressure at the level of the strain gauge. A calib-
The experimental arrangement is shown in Fig. 1. The ration curve for the lower set of strain gauges is shown in
model bin is rectangular in cross-section and at the outlet Fig. 2.

0021-8634/97/110231#06 $25.00/0/ag970197 231 ( 1997 Silsoe Research Institute


232 S. C. N E G I E¹ A ¸.

obtained from the output of the four force transducers,


which monitored the loss of weight with time.
Five outlet widths of 50, 60, 70, 80 and 90 mm were
chosen for the tests. Because the length of the sloping
walls of the hopper was fixed, only one variable could be
selected out of the hopper angle and the outlet width.
However, the hopper angle decreased with an increase in
the width of the outlet. Each test was replicated three
times resulting in a total of 15 tests.

2.2 Material properties

The use of the numerical model1 requires a number of


physical properties of the material that is being repre-
sented. Physical characteristics of soybean, such as aver-
age particle diameter, bulk density, particle density and
moisture content were measured in the laboratory.
Material properties required for the discrete-element
analysis, such as coefficient of restitution, stiffness and
damping coefficient were also experimentally deter-
mined. The coefficient of restitution, e, between particle
and wall was determined by dropping soybean on a steel
plate from height H and measuring the rebound h with
a high-speed video camera such that e was then com-
puted from
Fig. 1. Experimental arrangement

S
h
e" (1)
H
An auger was used to fill the model bin with soybean.
The bin walls had been ‘‘broken-in’’ during an earlier The stiffness and damping coefficient of a single par-
investigation with soybean.2 Measurements of wall pres- ticle were indirectly obtained by measuring the natural
sures were taken prior to opening the outlet, i.e. under frequency of a bunch of soybean particles. If the soybean
static conditions, as well as during discharge. The output grains are tightly packed in a tray, the mass density
from the strain gauges was recorded on tape at 0·1 s approaches the particle density, and the natural fre-
intervals and the wall pressures calculated from the pres- quency obtained from a tray full of soybean is not signifi-
sure—strain relationship. At each location, the strains cantly different from that of a single particle. In general,
measured by the two gauges were averaged for the analy- the higher the stiffness of a solid body, the larger the
sis of data. The average flow measurements were natural frequency. Also, when the input frequency is
equal to the natural frequency of the body, the pheno-
menon of resonance occurs.3 A vibrational method was
used to determine the resonance frequency. A sinusoidal
signal provided by a heterodyne analyser (B and K 2010)
was amplified by a power amplifier (B and K 2706) and
fed to a vibration exciter (B and K 4809) with a frequency
range 0—2 kHz. A plastic tray 70 mm in diameter by
20 mm deep was tightly packed with about 60 g of
soybean and placed on the table of the vibration exciter.
The soybean was covered with a very thin plastic sheet to
which an accelerometer (B and K 4344) was attached for
recording the output amplitude. Another accelerometer
was attached to the top of the exciter table to record the
input frequency. Both the output and input signals were
Fig. 2. Calibration curves for lower strain gauges plotted on a recorder chart. The natural frequency was
MO D EL V AL ID AT I ON 233

Table 1 Poisson’s ratio, coefficients of interparticle friction and


Input values for numerical simulation particle—wall friction were obtained from the literature.5–7
Parameters Particle— Particle— References
particle wall
3. Results and discussion
Stiffness K , N/m 11 703 23 406 Present paper
n
Stiffness K , N/m 9362 18 724 Rong et al.7
s 3.1. Constitutive relationship
Damping D , N m/s 27·28 29·19 Present paper
n
Damping D , N m/s 0 0 Rong et al.7
4 In order to use the secant constitutive relationship
Coefficient of
restitution, e 0.13 0.45 Present paper in the finite-element analysis, the material property
Coefficient of parameters in Eqn (19) of (Ref. 1) were experimentally
friction, f 0.33 0.35 ACI5 determined.4 Experimental results showed that the rela-
Poisson’s ratio l 0.33 Jofriet et al.5 tionship between equivalent strain e6 and equivalent stress
Bulk density o, p6, can be modelled by a family of exponential functions
kg/m3 771 Present paper and that the relationship between the equivalent strain
Particle density o ,
s e6 and the plastic volume strain ep by a family of relation-
kg/m3 1033 Present paper v
ships consisting of a quadratic function and a straight
line. As well, the elastic equivalent strain limit e6 was
s
found to increase linearly with increase in confining pres-
obtained from the frequency axis at the point where the sure P . The secant elastic shear modulus G is also
3
ratio of output-to-input vibration amplitudes was a a function of confining pressure P .
3
maximum. The experiment was repeated ten times and Zhang and Jofriet’s material models4 were modified by
resulted in a mean natural frequency of 650 Hz with Meng et al.2 to allow more convenient use in the finite-
values ranging from 500 to 750 Hz. element analyses. In the elastic range, a family of straight
The normal stiffness K and damping coefficient lines was used to represent the relationship between
n p6 and e6 . In the plastic range, the relationship was left
D were then calculated as follows:3
n unchanged. The modified relationship can be expressed
K "m u2 (2) as
n n
2m u p6"GeN eN )eN
D " n (3) s
n 2J(y/x)2!1 pN "¼ (1!e~w1e1 ) e6 'e6
0 s
where m is the mass of an assembly, u the natural ¼ (1!e~w1e1 s )
n G" 0 (6)
frequency, y the output vibration amplitude and x the e6
input vibration amplitude. s
The normal stiffness and damping coefficient obtained where for soybean
from Eqns (2) and (3) were taken as the corresponding
¼ "1·555 P
values for an individual particle. When two particles 0 3
(1 and 2) or particle (1) and wall (2) are in contact the ¼ "0·0498!0·00035 P , P )70 kPa
1 3 3
equivalent normal stiffness K and equivalent normal
¼ "0·0253 P '70 kPa
damping coefficient D are given by 1 3
The relationship between plastic volume strain ep and
K K v
K" 1 2 (4) equivalent strain e6 can be expressed as
K #K
1 2 ep"0 0)e6 )e6
v s
D"D #D (5)
1 2 ep"A e6 #A e2!A e6 !A e2 e6 )e6 )e
l 1 2 13 2s s 1
Equations (4) and (5) were used to calculate the particle—
ep"B #B e6 !A e !A e2 e )e6 (7)
particle and particle—wall normal stiffnesses and damp- l 1 2 1s 2s 1
ing coefficients reported in Table 1. in which for soybean
The material parameters in the constitutive relation-
ship used for the finite-element analysis were determined A "!0·243!0·00094 P
1 3
by Zhang and Jofriet,4 who carried out approxima- A "0·00563!0·000043 P
tely 150 triaxial tests on corn and soybean. Other material 2 3
properties for use in the numerical model, such as B "0·1098#0·0000125 P
2 3
234 S. C. N E G I E¹ A ¸.

and

(A !B )2 B !A A
B "! 1 2 e " 2 1 e6 "! 1
1 4A 1 2A s 2A
2 2 2
In Eqns (6) and (7) all stresses, G and P are in kPa and
3
all strains are in 103 m/m. The confining pressure, P , is
3
taken as positive.
The soybean had a bulk density of 771 kg/m3, a par-
ticle density of 1033 kg/m3, and the diameter of the
particles averaged 6 mm. The moisture content of
soybean was 12% w.b. The other input parameter
values used in the computer simulations are given in
Table 1.
Fig. 3. Comparison of numerical and experimental results of wall
pressures for an outlet width of 50 mm: (#) Exp lower; (*) Exp
3.2. Comparison of numerical and experimental results upper; (—) DEM lower; (—) DEM upper

Five computer-aided analyses were carried out with


the numerical model outlined in the Part 1 paper,1 It can be seen in Figs 3—5 that the numerical results
matching the experimental conditions described earlier agree well with the measured values. Both simulation and
as to bin geometry and material. The domain included experimental results show that wall pressures at the two
two DEM subdomains, one at the intersection of the strain-gauge locations, one 60 mm above and the other
hopper and vertical walls and one at the outlet. The 60 mm below the intersection of the rectangular bin and
remainder of the domain was divided into 18 elements the hopper, were not significantly different under static
and 93 nodes for one-half of the symmetric silo. The size conditions. However, after the gate was opened, the wall
of the finite-element mesh was selected after a sensitivity pressures were considerably higher at the lower location
analysis in which the results obtained with different ele- than those measured at the upper location. It can also be
ment sizes were compared with the computing times. It seen that the maximum wall pressure occurred soon after
was determined that 36 elements were sufficient to repro- the initiation of flow, generally within the first 0·5 s.
duce the model test conditions accurately. Very little The experimental results in Fig. 3 show that for an
difference in results was obtained when the number of outlet width of 50 mm, the maximum pressures of 4·8 and
elements was doubled, whereas the computing time in- 2·3 kPa occurred at the lower and upper locations,
creased dramatically. Also, finer meshes were used to
represent critical areas of interest. The elements in the
vicinity of the intersection and the outlet were assigned
half the height of elements in the remainder of the do-
main. The layout of elements and of DEM domains is
similar to that in Fig. 4 of the Part 1 paper.1
The temporal variations over the first 4—5 s of meas-
ured wall pressures from the tests with outlet width of 50,
70 and 90 mm are shown in Figs 3, 4 and 5, respectively.
These results were compared with predictions made us-
ing the computer model. The points plotted before time
zero represent the static pressures recorded before the
outlet gate was opened. The experimental results re-
ported are the average of the two sets of data obtained at
the same level. As mentioned earlier, because of the
enormous computational requirement only about 3 s
were executed by numerical analyses, which was suffi-
cient to predict the maximum wall pressures that occur
immediately after discharge begins. Each of the five nu- Fig. 4. Comparison of numerical and experimental results of wall
merical analyses took about 30 h on a Silicon Graphics pressures for an outlet width of 70 mm: (#) Exp lower; (*) Exp
IRIS-4D computer. upper; (—) DEM lower; (—) DEM upper
MO D EL V AL ID AT I ON 235

Fig. 5. Comparison of numerical and experimental results of wall


pressures for an outlet width of 90 mm: (#) Exp lower; (*) Exp
upper; (—) DEM lower; (—) DEM upper Fig. 7. Experimental and simulated heights of stored material as
a function of time for three outlet widths as shown

respectively, resulting in maximum dynamic pressure to


static pressure ratios of 3·2 and 1·4. The higher pressure
ratio ranged from 2·5 to 3·6 for hopper openings of with time are shown in Fig. 7. Three outlet widths, 50, 70
50—90 mm (Figs 3—5). and 90 mm, were chosen for presentation. It took about
The peak wall pressures for the five different outlet 13 s to empty the material with the 50 mm opening, and
widths are plotted in Fig. 6, together with values pre- 7·5 s and 6 s with 70 and 90 mm openings, respectively.
dicted by the numerical model. The magnitude of maxi- Both experimental and numerical results indicate that
mum pressure increased steadily with an increase in the the decrease in height was linear with time in the vertical
opening width. The agreement between the measured part of the silo during discharge. The rate of decrease was
and predicted peak values is excellent for all outlet widths 0·135 m/s with a 50 mm opening, 0·224 m/s with 70 mm,
with the error in the range 3—5%. The measured peak and 0·243 m/s with 90 mm and the corresponding flow
pressures increased from 5 to 5·6 kPa corresponding to rates ranged from 0·07 to 0·12 m3/s. For a particular
an increase in opening from 50 to 90 mm. This is because outlet width, the rate of discharge above the intersection
the acceleration of particles increased with an increase in is independent of the height above the opening because of
hopper opening, resulting in a greater impact at the top the effect of wall friction. However, the rate of decrease in
of the hopper where the transition from parallel to con- height increases when the surface falls below the top of
vergent flow occurs. the hopper (0·56 m above the outlet) owing to the de-
The changes in height of the stored material (distance crease in cross-sectional area of the hopper. The rate of
from the outlet to the free surface of the stored material) lowering in the material surface determined numerically
compares very well with the experiment for the 70 and
90 mm openings. For the 50 mm opening the numerical
rate appears to be about 5% greater.
Statistical analyses were performed to verify the degree
of agreement between the experimental and numerical
results of wall pressures. The mean and the variance or
standard error of observed data were calculated from
three replications of the tests. Based on these data, a con-
fidence interval was determined to check if the predicted
values fitted between the upper and lower limits. Figure 8
shows the upper bound and lower bound lines for wall
pressures at the lower location (60 mm below the top of
the hopper) plotted as a function of time, along with the
results of computer simulations for the 60 mm hopper
opening. It can be seen that the values predicted by the
Fig. 6. Maximum wall pressures at different outlet widths; (#) numerical model lie in the 95% confidence interval.
Exp; (—j—) DEM Large fluctuations of the upper and lower bound lines are
236 S. C. N E G I E¹ A ¸.

measured at 0·06 m above the intersection. The magni-


tude of maximum pressure increased from 5 to 5·6 kPa
corresponding to an increase in outlet width from 50 to
90 mm. The dynamic pressures exceeded the static pres-
sures by a factor of 2·5—3·6 just below and 1·3—1·5 just
above the top of the hopper for openings of 50—90 mm.
The height of stored material decreased linearly with
time in the vertical part of the silo during discharge. The
rates of decrease varied from 0·135 to 0·243 m/s corres-
ponding to openings of 50—90 mm. The statistical ana-
lyses showed that the predicted wall pressures lie within
the 95% confidence interval computed from the observed
data.

Fig. 8. Statistical analysis of wall pressures for an outlet width of


60 mm: (—) upper and lower bounds; (*) numerical results References
1
Lu Z; Negi S C; Jofriet J C A numerical model for flow of
granular materials in silos: Part 1, Model development. Jour-
noticeable in the first half second or so. This is because nal of Agricultural Engineering Research, 1997, 68, 223—229
2
the wall pressures varied substantially from the onset of Meng Q; Jofriet J C; Negi S C Finite element analysis of bulk
flow until a steady state was reached (see Figs 3—5). All solids flow. Part 1: Development of a model based on
calculations from the hybrid model for the outlet widths a secant constitutive relationship. Journal of Agricultural
Engineering Research, 1997, 67, 141—150
(not shown) did fall inside the 95% confidence-interval 3
Mohsenin N N Physical Properties of Plant and Animal
estimate. Materials, 2nd ed. New York: Gordon and Breach, 1986
4
Zhang Y; Jofriet J Equivalent stress-strain relationships for
soybean and corn. ASAE Paper No. 93-4506, ASAE, St.
4. Conclusions Joseph, MI, 1993
5
Jofriet J; Meng Q; Negi S C; Finite element analysis of bulk
solids flow. ASAE Paper No. 934064, ASAE, St. Joseph,
Good agreement was found between the wall pressures MI, 1993
6
measured in the model silo and calculations from the American Concrete Institute Recommended practice for de-
numerical model based on finite-element and discrete- sign and construction of concrete bins, silos and bunkers for
element methods. The peak pressures were recorded soon storing granular materials. ACI Standard 313—77, 1977
7
Rong G H; Negi S C; Jofriet J C Simulation of flow behav-
after the start of flow (within the first 0·5 s) at 0·06 m iour of bulk solids in bins. Part 1: Model development and
below the intersection of the rectangular bin and hopper. validation. Journal of Agricultural Engineering Research
This pressure was generally twice the wall pressure 1995, 62, 247—256

You might also like