Professional Documents
Culture Documents
Available at www.sciencedirect.com
Research Paper
article info
Experimental testing and calibration of new machines and fertiliser can be costly and time
Article history: consuming. Standard centrifugal fertiliser spreaders are designed to achieve a uniform
Received 11 January 2011 distribution across the entire field. Fruit trees, however, are planted in rows and it would be
Received in revised form advantageous to concentrate the fertiliser closer to the tree trunk and hence the root
19 April 2011 system. This could reduce the amount of fertiliser needed and will also reduce the negative
Accepted 21 April 2011 environmental impact. A discrete element method (DEM) model of a centrifugal fertiliser
Published online 8 June 2011 spreader was developed. A sensitivity study was used to determine the DEM parameters by
comparing the results to experimental results. The effects of the disc speed, feed position,
feed rate and vane angle on the spread pattern were investigated experimentally and
compared to the DEM results. A deflector plate was designed to separate the flow into two
paths along the tree lines. The DEM results of the deflector were compared to the experi-
mental results. It was shown that the DEM model can be used to make both qualitative and
quantitative predictions of the spread pattern under different spreader settings. The model
could also accurately predict the effect of the deflector on the spread pattern. This indi-
cates that DEM can be a powerful tool in the development of new spreader concepts.
ª 2011 IAgrE. Published by Elsevier Ltd. All rights reserved.
Fig. 2 e Experimental setup showing the hopper, disk and collector trays.
310 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Fig. 3 e (a) Vane dimensions, (b) definition of the vane angle and (c) the definition of the feed position.
Spread patterns can be measured by placing a row of col- studies. Standardised tests can be conducted in spreading
lecting trays in the field. They are usually placed perpendic- halls (Reumers, Tijskens, & Ramon, 2003b). Collecting trays
ular to the driving direction. As the tractor passes over the are usually used in these halls, but the external effects are
trays they collect the fertiliser and a measurement of the eliminated and the measurements are more repeatable. Some
transverse distribution is obtained. This is the most realistic halls are also air-conditioned to maintain the atmospheric
method to measure the spread patterns but it has several conditions (Van Liedekerke et al., 2008). Spreading halls,
drawbacks (Van Liedekerke et al., 2008). Conditions such as however, also suffer from a few drawbacks (Reumers,
wind and humidity and other field conditions cannot be Tijskens, & Ramon, 2003a): the tests are time consuming,
controlled which make it difficult to carry out comparative halls are large and expensive to build and operate, and large
amounts of fertiliser are needed. Also after spreading the Olieslagers, Tijskens, and Ramon (2004) and Aphale et al.
market value of the used fertiliser is low due to damage and (2003) developed analytical models based on basic equations
the mixing of different products. of motion to predict the behaviour of a single particle on
Reumers et al. (2003b) used a hybrid strategy to estimate a rotating disc with a straight radial vane. The total spread
spreading patterns. The experimental equipment they used pattern was determined statistically by determining the
was cheaper and more compact than spreading halls and it trajectories for a number of single particles, each with different
consumed less fertiliser. A cylindrical measuring device with initial conditions (Dintwa, Van Liedekerke et al., 2004). This type
a set of compartments is used to intercept and collect the of model cannot take particleeparticle interaction into account
particles immediately after leaving the disc. The result is the which is experienced when the particles flow from the hopper,
so-called cylindrical distribution pattern. Using photographic move on the disc along the vane or during free flight after
images, the horizontal velocity vectors of the particles can be leaving the disc. Villette, Cointault, Piron, and Chopinet (2005)
estimated and used to calculate the ballistic flight which then developed a similar model for a single particle on a concave
results in approximations of the complete transverse and disc with pitched straight vanes. Olieslagers, Ramon, and De
static (stationary) distribution patterns. Baerrdemaeker (1996) developed an analytical model for
Centrifugal spreaders have been modelled by several conical discs with straight and curved vanes. Particle interac-
researchers to predict and to determine the influence of various tion for particles flowing out of the hopper until they reach the
parameters on the spreading pattern. Dintwa, Van Liedekerke, disc was taken into account by changing the initial particle
Table 1 e Comparison between experiments and simulations with disk speed u [ 400 rpm, feed rate [ 523 g sL1, feed
position [ (35, L50) mm and vane angle q [ 0 .
DEM properties Statistical analysis
Reference Figure Particle Particle-wall Particleeparticle Damping Relative Average Average Angular Radial
stiffness kn Friction mw [] friction mp [] ratio x [ ] deviation angle a Radius Width Width
[N m1] RD r [m] b c [m]
Fig. 6 e Development of the spread pattern from the (a) stationary spread pattern to the (f) moving transverse distribution.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5 313
x
Fig. 9 e Reference experimental stationary spread pattern
Fig. 8 e Schematic showing the ballistic flight of a particle with disk speed u [ 400 rpm, feed rate [ 523 g,sL1, feed
and the landing position. position [ (35, L50) mm, vane angle q [ 0 .
314 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Fig. 10 e Stationary spread patterns using different numbers of particles in the DEM model.
were used in the DEM model. Although Van Liedekerke (2007) inefficient. Instead, a vertical rectangular tube with the same
proposes a non-linear Hertz-Kono-Kuwabara contact model, cross section as that of the hopper opening was used.
in this study a linear contact model was used with viscous Approximately 1000 particles were injected down the tube and
damping assumed at the contacts (Itasca, 2003). allowed to settle under gravity. The coordinates and proper-
Fig. 7 shows the two forms of DEM model used. Fig. 7a ties of each particle was then stored. Using the tube cross
shows the basic model without the deflector. Modelling the sectional area, the bulk density and the required flow rate
complete hopper with thousands of particles would be (g s1), the downward velocity of the particles could be
Fig. 11 e DEM stationary spread patterns for different particle stiffness values and mw [ 0.3, mp [ 0.3 and x [ 0.5.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5 315
Fig. 12 e DEM stationary spread patterns for different particle-wall friction coefficients and kn [ 1 3 104 N mL1, mp [ 0.3 and
x [ 0.5.
Fig. 13 e DEM stationary spread patterns for different particleeparticle friction coefficients and kn [ 1 3 104 N mL1, mw [ 0.3
and x [ 0.5.
Fig. 14 e DEM stationary spread patterns for different contact damping ratios and kn [ 1 3 104 N mL1, mw [ 0.3 and mp [ 0.3.
316 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Fig. 15 e Experimental and DEM stationary spread patterns for different disk speeds with feed rate [ 523 g sL1, feed
position [ (35, L50) mm, vane angle q [ 0 .
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5 317
2 2 2
d z dz dx dz
0.07 ¼ g K þ (2)
dt2 dt dt dt
0.06
0.05 where
0.04 3Cd ra
K¼
0.03 4rp d
0.02
0.01
with Cd [] the particle drag coefficient, ra ¼ 1.2 kg m3 the
density of air, rp [kg m3] the particle density, d [m] the
0
particle diameter and g ¼ 9.81 m s2 the gravitational accel-
0.1 500 rpm eration. The coordinate system used is defined in Fig. 8. In the
test case, deleting the particles decreased the calculation time
0.09 RD = 10.2%
from 16.0 h simulation time per second to 2.7 h per second.
0.08 Deleting the particles could however not be done when the
0.07 deflector was modelled. The particles were allowed to
0.06 complete the whole trajectory and the landing positions were
0.05 recorded in PFC3D. A typical deflector simulation took 14 h of
calculation time per second.
0.04
The spherical particles were generated using the measured
0.03
size distribution as shown in Table 2. The bulk density of the
0.02 fertiliser was measured as 1013 kg m3 and the particles in the
0.01 model were given a density of 1800 kg m3 which resulted in
0 a bulk density within 1.5% of the measured value.
-5 -4 -3 -2 -1 0 1 2 3 4 5
The constant drag coefficient Cd ¼ 0.44 of a sphere was used
Transverse direction [m] in all the simulations. This value was used by Olieslagers et al.
(1996), Aphale et al. (2003), Grift and Hofstee (2002) and Van
Fig. 16 e Experimental and DEM moving transverse Liedekerke, Tijskens, & Ramon, 2009. Aphale et al. (2003) also
distributions for different disk speeds, with feed showed that for a fertiliser particle of diameter 2 mm, the
rate [ 523 g sL1, feed position [ (35, L50) mm, vane angle transition to a constant drag coefficient is typically around
q [ 0 . a Reynolds number of 50, i.e., equivalent to a particle speed of
0.4 m s1. A typical particle will leave a 300 mm diameter disc
rotating at 540 rpm at a velocity of roughly 10 m s1, which
justifies the use of a constant Cd value. The other DEM
calculated. All particles were given this initial velocity. When
parameters (particle friction and particle stiffness) were
all the particles reached the end of the tube, another batch of
determined through a sensitivity analysis which will be
the stored particles was created inside the tube using
described in the next section.
a vertical offset. The position of the highest particle in the tube
was used to calculate the offset to avoid overlapping of the
batches. This resulted in a steady flow of particles from the
tube onto the disc and was within 1% of the measured flow 4. DEM simulation calibration
rate. Each new batch of particles generated was given
a different colour for visual observations (Fig. 7a). Fig. 7b The effect of the DEM parameters on the stationary spread
shows the form of the DEM model including the deflector. pattern was compared to the experimental result, as shown in
318 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Table 3 e Comparison between experiments and simulations with the particle stiffness kn [ 1 104 N,mL1, particle-wall
friction mw [ 0.3, particleeparticle friction mp [ 0.3 and the damping ratio x [ 0.5.
Setup Statistical analysis
Reference Figure Disk Feed Feed Vane Relative Average Average Angular Radial
speed rate [g s1] position angle deviation RD angle a radius r [m] width b width
[rpm] [mm] c [m]
Fig. 9 and Table 1 (Experiment 1), with a disc speed of 400 rpm, 4.1. Number of particles used in simulation
feed rate of 523 g s1, a feed position of (x,y) ¼ (35,50) mm and
a 0 vane angle. In each experiment 90 kg of fertiliser was used which is in the
To compare the DEM results to the experimental results, order of 106 particles. It would be very time consuming to use
the following five parameters were used (Van Liedekerke, this number of particles in the DEM model, so the evolving
Tijskens, Dintwa et al., 2009). The average distribution angle stationary spread pattern was plotted at intervals to find the
was defined as: number of particles where the predicted pattern converges.
XX From Fig. 10 it can be seen that the pattern changed slightly up
a¼ fij aij (3)
j i to the point where roughly 10,000 particles were used (the
equivalent of approximately 0.6 kg). Beyond this point, there
where i and j run from one to the number of columns and rows
was no significant change in the pattern. Thus, in this paper
respectively, aij is the angle of the ij-th collector tray as defined
10,000 particles were used in each simulation. Van Liedekerke,
in Fig. 9 and fij ¼ Mij =Mtotal is the mass fraction of particles in
Tijskens, & Ramon, 2009 made use of 7000 particles.
the collector tray. The angular width was defined as:
P
Mij ðexpÞ Mij ðsimÞj 4.3. Particle-wall friction coefficient
i
RDj ðexp; simÞ ¼ (8)
Mj ðexpÞ þ Mj ðsimÞ
According to Hofstee and Huisman (1990), the particle-wall
where Mj is the total mass in the j-th row (transverse line). (particle-disc and particle-vane) friction coefficient influences
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5 319
Fig. 17 e Experimental and DEM stationary spread patterns for different feed rates with disk speed u [ 400 rpm, feed
position [ (35, L50) mm, vane angle q [ 0 .
the particle discharge velocity from the disc, the discharge tube device to measure the dynamic friction coefficient.
angle (i.e. the angle between the radial and tangential Experiments with near-spherical ammonium nitrate particles
discharge velocities) and the discharge position (period that resulted in a value of 0.31. Van Liedekerke et al. (2006); Van
a particle stays on the disc). Grift, Kweon, Hofstee, Piron, and Liedekerke, Tijskens, Dintwa et al. (2009) measured the
Villette (2006) showed that the dynamic friction coefficient is dynamic friction coefficient between NPK particles and a steel
inversely proportional to the particle diameter and found plate by letting a cluster of particles slide on an inclined plate
a mean value of 0.36 for urea fertiliser granules. Hofstee (1992) until it reached a constant velocity. The slope of the plate was
made use of a rotating plate device to measure the dynamic used to calculate the friction coefficient as 0.35.
friction coefficient between fertiliser and different surfaces. In this study, three values were used for the particle-wall
The relative velocity varied between 1 m s1 and 21 m s1 while friction coefficient, mw ¼ 0.1, 0.3 and 0.5. The particle stiff-
the normal load varied between 0.130 N and 0.522 N.The fric- ness was set to kn ¼ 1 104 N m1, the particleeparticle fric-
tion coefficient with steel was found to be between 0.34 and tion coefficient was set at mp ¼ 0.3 and the damping ratio at
0.45. Kweon, Grift, and Miclet (2007) made use of a spinning- x ¼ 0.5. The results are shown in Fig. 12 and Table 1 (DEM 3 and
320 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Fig. 19 e Experimental and DEM stationary spread patterns for different feed positions with disk speed u [ 400 rpm, feed
rate [ 523 g sL1, vane angle q [ 0 .
Fig. 20 e Experimental and DEM stationary spread patterns for different vane angles with disk speed u [ 400 rpm, feed
rate [ 523 g sL1, feed position [ (35, L50) mm.
disc speed, the distribution became wider. The DEM model particles left the disc later. The measured angular widths were
was able to predict the general trends and was more accurate found to be b ¼ 4:3 ; 5:3 ; and 7:3 respectively while the DEM
with lower disc speeds. One reason for this might be the model predicted b ¼ 4:8 ; 5:9 ; and 6:7 respectively. Reumers
assumption of a constant drag coefficient. With higher et al. (2003a) made similar experimental observations using
particle speeds, the drag coefficient changes and hence a cylindrical measuring device while Van Liedekerke, Tijskens,
influences the aerodynamic effects and the landing position. & Ramon, 2009 also obtained similar DEM results in modelling
This was not included in the DEM model. The differences the same device. The DEM model was able to predict this
might also be attributed to the change in friction coefficient. It general trend with an increase in feed rate. For lower feed rates,
is well known that the friction coefficient decreases with an however, the DEM model predicted a narrower band compared
increase in relative contact velocity (Hofstee, 1992). In the DEM to the experimental results. The measured radial widths were
model a constant friction coefficient was assumed. The rela- found to be c ¼ 0:088 m; 0:094 m and 0:102 m with an increase
tive deviation along the moving transverse distribution was in the flow rate while the DEM model predicted c ¼ 0:062 m;
found to be 5.7%, 7.1% and 10.2% for rotation speeds of 0:073 m and 0:084 m respectively.
300 rpm, 400 rpm and 500 rpm respectively. Fig. 18 compares the moving transverse distributions
from experimental and DEM results for the different feed
5.2. Feed rate rates. Again, with an increase in the feed rate, the distribu-
tion became wider, especially in the transverse direction at
Three feed rates were used, 88 g s1, 250 g s1 and 523 g s1. The 2 m. For all feed rates, the DEM model predicted two
disc speed was set to 400 rpm, the feeding position was set to “shoulders” in the distribution at 2 m. In the experimental
(x,y) ¼ (35,50) mm and the vane angle q ¼ 0 . Fig. 17 shows the results, these “shoulders” only became prominent with the
stationary spread patterns as obtained from the experiments highest feed rate of 523 g s1. The relative deviation along the
and DEM simulations and the results are summarised in Table 1 moving transverse distribution were found to be 10.7%, 12.8%
and Table 3. From the experimental results it is clear that with and 7.1% for feed rates of 88 g s1, 250 g s1 and 523 g s1
an increase in the feed rate, the spread pattern widened as more respectively.
b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5 323
The vane angle was defined in Fig. 3b. All the experiments and
simulations presented up until now made use of a 0 vane
angle (radial vanes). Here the vane angle was changed to 50
and 100 . The feed position was kept at (35, 50) mm, the disc
speed at 400 rpm and the feed rate at 523 g s1. The stationary
spread patterns are shown in Fig. 20 and the results are sum-
marised in Table 1 and Table 3. It is clear form these results that
with an increase in the vane angle, the spread pattern rotated
in the direction opposite to the disc rotation. The average angle
was measured as a ¼ 106:1 ; 90:6 and 81:2 with an increase in
the vane angle while the DEM model predicted values of 97.5 ,
Fig. 21 e The deflector design showing the geometry, 92.2 and 82.1 respectively. Dintwa, Tijskens, Olieslagers, De
dimensions and the position relative to the disc. All Baerdemaeker, and Ramon (2004) obtained similar results
dimensions are in mm. using an analytical model of a single particle on a spinning disc
with different initial particle positions.
The feed position was defined in Fig. 3c. All experiments and As described in the introduction, it could be more efficient and
simulations presented up until now used a feed position of environmentally friendly if the fertiliser was spread in two
(x,y) ¼ (35, 50) mm. Here the feed position was changed while rows to accommodate fruit orchards. To accomplish this,
the disc speed was kept at 400 rpm, the feed rate at 523 g s1 a U-shaped deflector plate was designed as shown in Fig. 4. To
and the vane angle at q ¼ 0 . First the feed position was design the deflector, the stationary spread patterns (without
changed to (35, 70) mm and then to (35, 105) mm as shown the deflector and from both experiments and DEM) were
in Fig. 19. Unfortunately a large portion of the spread pattern investigated together with the fertiliser flight path as it leaves
fell outside the collecting area, but the general trend can still the disc. The DEM model made it easy to obtain the particle
Fig. 22 e Experimental and DEM stationary spread patterns using the deflector with disk speed u [ 400 rpm, feed
rate [ 523 g sL1, feed position [ (35, L50) mm, vane angle q [ 0 . RD [ 20.0%.
324 b i o s y s t e m s e n g i n e e r i n g 1 0 9 ( 2 0 1 1 ) 3 0 8 e3 2 5
Hofstee, J. W. (1992). Handling and spreading of fertilizers: alternative for spreading hall measurements. Biosystems
part 2, physical propertries of fertilizer, measuring Engineering, 86(4), 431e439.
methods and data. Journal of Agricultural Engineering Sogaard, H. T., & Kierkegaard, P. (1994). Yield reduction resulting
Research, 53, 141e162. from uneven fertilizer distribution. Transactions of the ASAE,
Hofstee, J. W., & Huisman, W. (1990). Handling and spreading of 37(6), 1749e1752.
fertilizers part 1: physical properties of fertilizer in relation to Sokalska, D. I., Haman, D. Z., Szewczuk, A., Sobota, J., & Deren, D.
particle motion. Journal of Agricultural Engineering Research, 47, (2009). Spatial root distribution of mature apple trees under drip
213e234. irrigation system. Agricultural Water Management, 96, 917e924.
ITASCA. (2003). PFC3D, particle flow code in three dimensions, user’s Van der Linde, G.J., & Pitse, M.A. (2006). The South African
guide (2nd ed. ). Fertriliser industry. AFA Conference, Cairo, Egypt, February
Jones, J. R., Hayden, G. L., & Yule, I. J. (2008). A statistical comparison 2006.
of international fertiliser spreader test methods e confidence in Van Liedekerke, P. (2007). Study of the granular fertilizers and the
bout width calculations. Powder Technology, 184, 337e351. centrifugal spreader using Discrete Element Method (DEM)
Kweon, G., & Grift, T. E. (2006). Feed gate adaption of a spinner simulations. Phd Thesis, Katholieke Universiteit Leuven,
spreader for uniform control. Biosystems Engineering, 95(1), 19034. Faculteit Bio-ingenieurswetenskappen.
Kweon, G., Grift, T. E., & Miclet, D. (2007). A spinning-tube device Van Liedekerke, P., Piron, E., Vangeyte, J., Villette, S., Ramon, H., &
for dynamic friction coefficient measurements of granular Tijskens, E. (2008). Recent results of experimentation and dem
fertiliser particles. Biosystems Engineering, 97, 145e152. modelling of centrifugal fertilizer spreading. Granular Matter,
Moshou, D., Deprez, K., & Ramon, H. (2004). Prediction of 10(4), 247e255.
spreading processes using a supervised self-organising map. Van Liedekerke, P., Tijskens, E., Dintwa, E., Anthonis, J., &
Mathematics and Computers in Simulation, 65, 77e85. Ramon, H. (2006). A discrete element model for simulation of
Olieslagers, R., Ramon, H., & De Baerrdemaeker, J. (1996). a spinning disc fertilizer spreader I. Single particle
Calculation of fertilizer distribution patterns from a spinning simulations. Powder Technology, 170, 71e85.
disc spreader by means of a simulation model. Journal of Van Liedekerke, P., Tijskens, E., Dintwa, E., Rioual, F., &
Agricultural Engineering Research, 63, 137e152. Vangeyte, H. (2009). DEM simulations of the particle flow on
Parris, K., & Reille, L. (1999). Proceedings no. 422 Measuring the a centrifugal fertilizer spreader. Powder Technology, 190,
environmental impacts of agriculture: use and management of 348e360.
nutrients. York, UK: The International Fertiliser Society. Van Liedekerke, P., Tijskens, E., & Ramon, H. (2009). Discrete
Reumers, J., Tijskens, E., & Ramon, H. (2003a). Experimental element simulations of the influence of fertilizer physical
characterisation of the tangential and cylindrical fertiliser properties on the spread pattern from spinning disc spreaders.
distribution pattern from a spinning disc: a parameter study. Biosystems Engineering, 102, 392e405.
Biosystems Engineering, 86(3), 327e337. Villette, S., Cointault, F., Piron, E., & Chopinet, B. (2005).
Reumers, J., Tijskens, E., & Ramon, H. (2003b). Experimental Centrifugal spreading: an analytical model for the motion of
characterisation of the tangential and cylindrical fertiliser fertilizer particles on a spinning disc. Biosystems Engineering,
distribution pattern from a spinning disc: towards an 92(2), 157e164.