You are on page 1of 8

Soil & Tillage Research 126 (2013) 34–41

Contents lists available at SciVerse ScienceDirect

Soil & Tillage Research


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

A discrete element model for soil–sweep interaction in three different soils


Ying Chen a, Lars J. Munkholm b, Tavs Nyord c,*
a
Department of Biosystems Engineering, University of Manitoba, Winnipeg, Manitoba, Canada
b
Department of Agroecology, Faculty of Science and Technology, Aarhus University, DK-8830 Tjele, Denmark
c
Department of Engineering, Faculty of Science and Technology, Aarhus University, Blichers Allé 20, DK-8830 Tjele, Denmark

A R T I C L E I N F O A B S T R A C T

Article history: Soil–tool interactions are at the centre of many agricultural field operations, including slurry injection.
Received 3 February 2012 Understanding of soil–tool interaction behaviours (soil cutting forces and soil disturbance) is important
Received in revised form 23 August 2012 for designing high performance injection tools. A discrete element model was developed to simulate a
Accepted 24 August 2012
slurry injection tool (a sweep) and its interaction with soil using Particle Flow Code in Three Dimensions
(PFC3D). In the model, spherical particles with bonds and viscous damping between particles were used
Keywords: to simulate agricultural soil aggregates and their cohesive behaviours. To serve the model development,
Sweep
the sweep was tested in three different soils (coarse sand, loamy sand, and sandy loam). In the tests, soil
DEM
cutting forces (draught and vertical forces) and soil disturbance characteristics (soil cross-section
PFC3D
Force disturbance and surface deformation) resulting from the sweep were measured. The measured draught
Disturbance and vertical forces were used in calibrations of the most sensitive model parameter, particle stiffness.
Calibration The calibrated particle stiffness was 0.75  103 N m1 for the coarse sand, 2.75  103 N m1 for the
Validation loamy sand, and 6  103 N m1 for the sandy loam. The calibrated model was validated using the soil
disturbance characteristics measured in those three soils. The simulations agreed well with the
measurements with relative errors below 10% in most cases.
ß 2012 Elsevier B.V. All rights reserved.

1. Introduction In this work, a modelling approach was used to study soil–tool


interaction, as measurements of soil dynamic behaviours are costly
Many studies on soil–tool interaction have been carried out in and time consuming. Analytical models based on the passive soil
the past with the intention of improving the design of soil engaging failure theory for predictions of soil cutting forces and the amount
tools. Soil–tool interactions are usually characterised by two of soil disturbed were among the first attempts in the literature.
phenomena: forces arising at the soil–tool interface (draught and This method was developed and modified by several researchers
vertical forces) and displacement of soil particles (soil disturbance) (Hettiaratchi et al., 1966; Godwin and Spoor, 1977; McKyes and Ali,
(Conte et al., 2011). They are the two most important performance 1977) and later summarised by McKyes (1985). The analytical
indicators of a slurry injection tool. High draught force require- method has proved to be simple to use and fairly accurate.
ment is the major limitation that prevents producers from using However, its applications are limited to simple blades and assumed
large implements for slurry injection. Soil cross-sectional area soil failure patterns. Although the analytical theory was extended
disturbance is one of the key factors which determine the amount to more complicated shapes of tools (disc and mouldboard plough)
of slurry which can be injected into the soil without overflowing to more recently (Godwin and O’Dogherty, 2007), the assumption of a
the soil surface (Mckyes et al., 1977; Chen and Tessier, 2001). Soil soil failure pattern still remains a limitation. Numerical models,
disturbance also affects how well the injected slurry is spread in such as Finite Element Method (FEM), have also been used to
soil. Slurry spreading in a larger soil volume promotes better model soil–tool interactions for more complex tool shapes and soil
aerobic condition and may decrease denitrification losses. The behaviours in the literature (Shen and Kushwaha, 1998; Plouffe
resultant soil surface profile of an injection tool is a measure of et al., 1999; Abo-Elnor et al., 2004). A thorough review on the use of
injection quality, as it affects how well the injected slurry is the FEM for soil–tool interaction has been done by Upadhyaya et al.
covered by soil (Rahman et al., 2005), and thereby how effectively (2000). Soil particles in the FEM were connected and acted like one
ammonia and odour emission is reduced (Chen and Tessier, 2001; object, while in reality large soil fractures may form and particles
Hansen et al., 2003). may move independently with a large displacement during a field
operation. In such cases, there may be numerical convergence
problems with the FEM (Abo-Elnor et al., 2004). The common
* Corresponding author. Tel.: +45 87157637.
limitation of the analytical method and FEM is that they cannot
E-mail address: Tavs.Nyord@agrsci.dk (T. Nyord). predict soil behaviours at particle levels. The analytical method

0167-1987/$ – see front matter ß 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.still.2012.08.008
Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41 35

describes soil failures in forms of soil body (e.g. a crescent or a modelling of soil–tool interactions, Shmulevich et al. (2009) and
wedge) and the FEM accounts for total deformation not only, in Shmulevich (2010) pointed out that although the literature
individual soil particles. reported good correlations between simulations and measure-
This study used a newer numerical method, the Discrete ments, there is still a lack of robust methods for DEM model
Element Method (DEM), in dealing with discontinuous soil parameter calibrations.
particles. It aims at modelling soils at particle levels without The objectives of this study were to (1) develop a discrete
limitations on magnitudes of particle displacement and shapes of element model to simulate soil–sweep interaction using PFC3D, (2)
tool. In the DEM, soil is treated as an assembly of individual inversely calibrate a model parameter using experimental data of
particles and each particle interacts with its neighbouring particles soil cutting forces from three soils, and (3) validate the model
under external forces, such as tillage action. As a result, forces arise results using experimental data of soil disturbance from those
at the contact points between particles, causing displacement of three soils.
particles. The magnitude of the contact force is determined by the
particle properties (such as stiffness) and the overlap between 2. Methodology
particles in contact. The motion of a particle follows Newton’s
second law. To produce dynamic attributes of each particle in every 2.1. Experiment
calculation cycle, additional computation time is needed to store
and compute the information as compared with the FEM. 2.1.1. Description of the equipment
To date there are only a few studies in DEM simulations of soil– The sweep tested (Fig. 1a) was 180-mm wide with a sweep
tool interaction in applications of bulldozing (Momozu et al., 2003; angle of 808. The sweep and shank (46 mm wide) were mounted on
Franco et al., 2007; Shmulevich et al., 2007; Obermayr et al., 2011). a toolbar at a rake angle of 158. A force transducer was installed
These studies dealt with simple earth moving tools (wide blades) between the shank and the toolbar to measure the soil cutting
operated in cohesionless soils. Whereas agricultural soil engaging forces of the sweep. The toolbar was specially designed for testing
tools are of more complicated shapes and agricultural soils are soil engaging tools. It consisted of a three-point hitch and two
cohesive. Among the few literature studies on agricultural soil, van trailing wheels (Fig. 1b). The trailing wheels were adjustable to
der Linde (2007) simulated a powered subsoiler using a commer- control the tool working depth. Nyord et al. (2010) provided a
cial DEM code, Particle Flow Code in Three Dimensions (PFC3D) detailed description of the toolbar.
(Itasca Consulting Group, Minneapolis, MN). That study dealt with
only one type of soil (a sandy soil). Later, Mak et al. (2012) 2.1.2. Description of the testing facility
simulated a simple soil engaging tool (a narrow blade) using PFC3D. The testing facility for sweep tests included outdoor soil bins
That study dealt with two contrasting soils (coarse and fine soils), (Fig. 2), located at the Foulum Research Centre (DK-8830 Tjele,
and the model was not calibrated with measurements but with Denmark). The bins contained three types of Danish soils: coarse
predictions of draught forces from an analytical method. In sand soil from Jyndevad [Ortic Haplohumod, texture: 4% clay
addition, the model was not validated in that study. All the (<0.002 mm), 5% silt (0.002–0.063 mm), 17% fine sand (0.063–
aforementioned soil–tool interaction simulations, regardless of 0.2 mm), 72% coarse sand (>0.2 mm), and 2.0% organic matter],
applications, focused on soil draught forces of simple tools, and loamy sand soil from Foulum (Typic Hapludult, texture: 9% clay,
presented only limited qualitative information on soil disturbance. 24% silt, 32% fine sand, 33% coarse sand, and 2.5% organic matter),
Calibration of model particle properties (termed as micro- and sandy loam soil from Rønhave (Typic Agrudalf, texture: 18%
properties) is the major challenge of DEM simulations. Asaf et al. clay, 31% silt, 29% fine sand, 20% coarse sand, and 2.5% organic
(2007) calibrated PFC2D (two dimensional) model parameters for matter). The soil bins were 40 m long, 2.7 m wide, and 1.5 m deep.
soil tillage applications. Soil particles were modelled by clumps of For the past 13 years, the soils were cultivated with conventional
two discs and soil was assumed as cohesionless. Calibrations were tillage and seeded with mainly cereal crops. All bins had barley
based on in situ field sinkage tests using different soil penetration stubble (height: 60–70 mm) at the time of the experiment. The
tools. The calibrated model parameters and similar particle shape automatic open–close roof of the facility allowed controlling the
were used by Shmulevich et al. (2007) in simulation of interaction amount of rain which the soils were receiving.
between soil and a wide cutting blade using PFC2D. In simulating a
powered subsoiler, van der Linde (2007) calibrated PFC3D model 2.2. Sweep tests
parameters for a sandy soil using laboratory direct shear and
compression tests. Sadek et al. (2011) calibrated PFC3D parameters The sweep was tested on September 22, 2011. Each soil bin was
for a sandy soil with different moisture contents and bulk divided into three plots (0.7 m wide) along the entire length of the
densities. In reviewing the general concepts of the DEM and bin (Fig. 2). To allow for replications of treatment and comparisons

Fig. 1. Equipment for tests: (a) sweep and force transducer; (b) toolbar set up.
36 Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41

Draught and vertical forces of the sweep were measured using a


transducer (Fig. 1a) (extended octagonal ring) with rated capacities
of 62.0 kN and 85.2 kN for draught and vertical forces, respectively.
A radar mounted on the toolbar monitored the travel speed. Forces
and speed were recorded using a MGCplus datalogger (Hottinger
Baldwin Messtechnik) at a sampling rate of 600 Hz. Nine sets of
force and speed data were obtained for the nine plots.
Soil cross-section disturbance by the sweep was measured
following the tests using a 390 mm wide profile meter consisting of
130 free-dropping wooden pins. The measurement followed the
steps: (1) excavation of loose soil in a furrow by hand; (2)
placement of the profile meter across the furrow; the pins
automatically adjusted for their vertical positions in accordance
with the contour of the furrow (Fig. 3a); and (3) tracing of the
profile of the pins on a paper (Fig. 3b). The soil surface profile was
measured using the similar procedure, except that the first step
was omitted. Measurements were performed at four random
locations for soil cross-section and six random locations for surface
Fig. 2. The testing facility showing soil bins, roofing system, soil surface conditions profile in each plot. In total, 36 soil cross-section profiles and 54
prior to the tests, and the layout of the test plots within a bin.
surface profiles were obtained. Profiles were later analysed for
characteristics of soil disturbance.
among the soils, a constant working depth (100 mm) and a
constant tractor travel speed were used for all plots. This gave 2.4. Data processing and analysis
three replicates within each soil bin. During the tests, the tractor
wheels travelled on two metal rails outside the bin (Fig. 1b), Fig. 4 shows a typical data set from the force transducer and
allowing for a precise control in the spacing of sweep paths within speed radar sensor. Data points between 0 and 3 s were taken as
the bin. offsets which were subtracted. Data points for the draught force,
vertical forces, and speed were taken from 10 to 45 s, and the
2.3. Measurements averages of those data points (21,000 data points) were used in
statistical analyses to examine effects of soil type on the cutting
At the time of the sweep tests, soil cores (60 mm diameter and forces and in model calibrations as well.
100 mm long) were taken from undisturbed areas in the bins. A Soil profiles were characterised using cross-sectional area,
total of 30 soil cores were taken in three soil bins at random width, and soil surface profile. Soil cross-sectional area disturbed
locations with 10 cores from each soil bin. Soil cores were brought was the area under the soil surface (A) (Fig. 5a) and it was
to the laboratory, weighed, oven-dried at 105 8C for 24 h, and measured using a Digitizing Area-line Meter (Model: Super PLANIX
weighed again to measure the soil moisture content (dry basis) and b, Tamaya Technics Inc., Japan). The disturbance width (W) was the
dry bulk density. furrow opening width at the soil surface level. Soil surface profiles
In situ soil frictional angle and cohesion were measured using a were quantified by areas above and below the soil surface, as
torsional shear box (shear ring diameter: 100 mm and height: illustrated in Fig. 5b, The total area above the soil surface ðAþ ¼
P þ
50 mm). Before the measurements, the top 0–25 mm soil layer was ai Þ was named as positive disturbed area and that below the
P 
removed; then the ring was inserted into the soil at a depth of surface ðA ¼ ai Þ was named as negative disturbed area, and
75 mm below the original soil surface, so that the shear ring was they were measured also using the Digitizing Area-line Meter.
positioned in the middle of the tillage layer. A test was performed
for five different normal loads in the range of 7.3–32.3 kPa. Tests 2.5. Development of the discrete element model
were performed at two random locations in each soil bin, giving a
total of 6 measurements in three bins. 2.5.1. Model particles
In situ soil shear strength was measured using a vane shear Model particles were an assembly of balls, basic PFC3D
tester (vane diameter: 33 mm and height: 50 mm). The vane tester elements, representing soil particles. Note that the terms of ball
was inserted into the soil at 75 mm depth. A total of 30 and particle are used interchangeably hereafter. The contacts
measurements were performed at random locations in three bins between model particles were considered to follow the behaviours
with 10 measurements per bin. described by the PFC3D parallel bond model (PBM) (Potyondy and

Fig. 3. Soil disturbance measurement: (a) the profile meter in an excavated furrow; (b) traced profile on a paper.
Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41 37

4
Force (kN) or speed (m/s)

3
Travel spee d
2
Draught force
1

0
Vercalforce
-1
0 5 10 15 20 25 30 35 40 45 50 55
Time (s)

Fig. 4. A typical data set recorded during a test run of the sweep in the sandy loam;
the negative vertical forces indicate the downward direction.

Cundall, 2004). In the PBM, particles are bonded together with a Fig. 6. Cylindrical bond between two balls in the PFC3D parallel bond model (PBM).
cylindrical ‘‘cementicious’’ material (Fig. 6). The bond can
withstand forces and moments; their magnitudes depend on the
micro-properties (e.g. shear and normal strengths) of the bond.
Determination of these micro-properties is described later in the
paper. Particles connected with bonds form clusters of particles,
which mimics soil aggregates and the cohesive soil forces between
soil particles. Bonds can break when external forces exceed the
strengths of the bonds, which mimics the tillage action of a soil
engaging tool. Although irregular particle shapes can be con-
structed using balls in PFC3D, it is less important when the PBM is
used, as balls exist in forms of clusters in the PBM (van der Linde,
2007).

2.5.2. Soil domain and virtual sweep Fig. 7. The discrete element model developed using PFC3D: particles, box, and
sweep.
Dimensions of the soil domain were set as
0.4 m  0.3 m  0.9 m (width  depth  length). Particles were
generated in the domain and were confined in a box built using
five PFC3D walls (Fig. 7). Given the fixed dimension of the soil the sweep through the particle assembly to simulate the operation
domain and the computing time limitation, the model particle of the sweep in soil.
sizes were chosen to follow a uniform size distribution between 6
and 16 mm, meaning that all particle sizes between 6 and 16 mm 2.5.3. Micro-properties of model particles
were generated with an equal probability. This size range is in The PFC3D PBM requires eight particle micro-properties as
accordance with the aggregate size range of 1–32 mm measured in model inputs for simulations. They are the ball friction coefficient:
several Danish soils by Munkholm and Kay (2002), Munkholm m, the ball normal and shear stiffness: Kn and Ks (N m1), the bond
et al. (2002), and Munkholm et al. (2007). Discrepancies caused by normal and shear stiffness: K̄n and K̄s (Pa m1), the bond normal
the particle shape and size would be compensated with calibra- and shear strength: s̄ and t̄ (Pa), and radius multiplier of bond: R̄m
tions of particle micro-properties. This compensation approach (dimensionless). The radius multiplier is the ratio of the diameter
seems acceptable at the stage of the DEM development today of cylindrical bond and that of the smaller ball of two balls in
(Shmulevich, 2010). contact. The value of R̄m varies from 0 (no bond at all) to 1 (the bond
A virtual sweep (Fig. 7) was constructed with seven PFC3D walls radius equal to the diameter of the smaller ball in contact). In this
to represent the sweep used in the soil bin tests. The basic study, some model parameters were related with measured
geometry of the virtual sweep was developed to closely match that fundamental soil properties, which is an important approach for
of the actual sweep. There were two major differences between the determinations of DEM model parameters (Shmulevich, 2010). The
virtual sweep and the real one. The virtual sweep did not have particle friction, m was considered as the soil internal friction
curved surfaces and corners, and its shank was approximated by coefficient measured in the soil bins, and the bond shear strength, t̄
small segments of walls. The key features maintained included the of model particles was considered as the vane shear strength
sweep length, width, sweep angle, rake angle, and shank width. measured in the soil bins. To reduce the number of parameters to
The virtual sweep was initially positioned at one end of the soil be calibrated, the following assumptions were made: s̄ ¼ t̄ ; K̄s ¼
domain at the desired working depth, prior to simulation. PFC3D K̄n ; K s ¼ K n ; and R̄m ¼ 0:5 (Asaf et al., 2007; van der Linde, 2007).
allows specifying a speed to walls. This feature was used to move Thus, only Kn and K̄n were required to be determined for the model.

Fig. 5. Soil disturbance characterisation: (a) soil cross-section disturbed by the sweep (A = cross-sectional area; W = width at the soil surface); (b) soil surface disturbance
(a+ = positive disturbed area; a = negative disturbed area).
38 Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41

Preliminary simulations showed that model particle behaviours The experimental results provide important information for
were not sensitive to K̄n , but very sensitive to Kn. Therefore, a producers, regarding the performance of the sweep working in
constant value of 5  107 Pa m1 (van der Linde, 2007; Sadek et al., these soils. Positive soil surface deformation may favour slurry
2011) was used in simulations in this study, while Kn was being covered by soil, if the disturbed areas are located above the
calibrated. injected slurry. In contrast, negative soil disturbed area may pose a
risk for slurry being exposed to the air due to a lack of soil cover.
2.5.4. Damping of the model particles Based on the measured soil cutting forces, producers can
In PFC3D, users can choose either or both of local damping and determine the ballasting for appropriate sweep penetration and
viscous damping, to introduce energy dissipation to particles, so the number of sweeps which can be pulled by a given tractor.
that particles become more stable. The damping formulations are
described in detail in Itasca (2008). Basically, the local damping 3.2. Model results
acts to suppress motion of particles, and the viscous damping acts
at contacts between particles to oppose relative motion of two In generating particles, particle porosities were matched to the
particles. The local damping coefficient (a) ranges from 0 to 1, The measured soil bulk densities (Table 1) with an average soil particle
viscous damping coefficient (b) is specified by the ratio of damping density of 2.65 mg m3 (Campbell, 1985). With these porosities
constant to the critical damping constant b (a function of mass and the model generated between 10,266 and 13,665 balls within the
natural frequency of undamped system). A system is said to be specific soil domain dimensions, depends on the soil. Additional
underdamped when b < 1, overdamped when b > 1, and decays to model inputs included stiffness of the virtual sweep (assumed as
zero at the most rapid rate when b = 1. The default values are the stiffness of steel: 1  109 N m1) and friction coefficient
a = 0.7 and b = 0. Damping effects are relevant to both soil between particles and the virtual sweep which was assumed to
disturbance and cutting forces. Selections of damping coefficients be 0.41, a typical value for the friction between soil and an
are discussed in the section of model results. engaging tool (Godwin, 2007). The porosities and other model
inputs for three different soils are summarised in Table 2.
3. Results and discussion
3.2.1. Model behaviours as affected by damping coefficients
3.1. Experiment results Model behaviours were examined under different damping
coefficients which affect the model dynamics in soil cutting forces
Significantly lower soil bulk density was found for the loamy and soil particle displacement. Under assumed particle stiffness,
sand soil than for the other soils (Table 1). The moisture content of the model was run first using the default the PFC3D damping
the sandy loam was two times that of the coarse sand, and that of coefficients (a = 0.7; b = 0). PFC3D allows monitoring forces acting
the loamy sand was intermediate. This statistical trend also held on walls in both horizontal and vertical directions. This feature was
true for the results of soil vane shear strength and soil cohesion. used to monitor soil cutting forces of the virtual sweep. Typical
Soil internal friction coefficients did not correlate to soil type. The outputs of draught and vertical forces from the model are shown in
average tractor speed was 0.89 m s1 (3.19 km h1). Operating the Fig. 8a. As the sweep enters the soil domain, draught and vertical
sweep at this speed in the coarse sand required 0.292 kN draught forces increase rapidly, and then fluctuate about a fairly constant
force. Comparatively, the draught force was 50% higher in the value. Forces decrease to zero as the sweep comes out of the soil
loamy sand, and twice as high in the sandy loam. The vertical force domain. The fluctuating forces reflect breaking of bonds between
also followed the same statistical trend. These results may imply individual particles as the sweep travels. A bond breaks when
that soil with higher shear strength resulted in higher draught external forces (from the sweep) exceed the strengths of the bond,
force. In all three soils, the sweep disturbed similar cross-sectional and the forces between particles become zero when the bond is
areas. However, the disturbed width at the soil surface was broken, resulting in fluctuating of the force, which in turn affect the
significantly smaller for the coarse sand soil. Results of the positive total forces arising at the soil–tool interface. The phenomenon of
disturbed area showed that the sweep threw more soil above the force fluctuation reflects the dynamic nature of soil particles
surface in the sandy loam soil. Conversely, in this soil a lower observed in the soil bin tests (Fig. 4).
negative disturbed area was observed. With the PFC3D_EV (Enhanced Visualisation) used in this study,
one can examine many dynamic attributes of particles at any given
time, including position, translational and rotational velocities,
Table 1 mass, and which neighbouring particles are in contact with the
Means of soil properties, cutting forces, and disturbance characteristics measured in
the soil bin tests.
particle. The attributes of particle translational velocities are of
interest as they reflect soil disturbance resulting from the sweep.
Variable Soil type As the sweep travels, individual particles are being impacted,
Coarse sand Loamy sand Sandy loam resulting in variable velocities of those individual particles. This
Soil property was captured by screenshots of velocity contours (Fig. 8b). The
Dry bulk density (kg m3) 1.41aa 1.33b 1.41a contours show that the highest particle velocities are at the
Moisture content (d.b.) (%) 8.98c 14.84b 18.20a interface of the sweep (from the tip to the shank) and particles.
Vane shear strength (kPa) 9.12c 15.22b 21.68a Particles are dislodged over the top of the domain. Over time,
Cohesion (kPa) 15.7c 25.2b 36.0a
Internal friction coefficient 0.51a 0.67a 0.63a
impacted particles settle down behind the sweep and velocities of
Soil cutting force particles eventually decrease to zero as the calculation continues.
Draught force (kN) 0.292c 0.430b 0.585a However, with the default values of damping coefficients, model
Vertical force (kN) 0.101c 0.152b 0.178a particles move higher or farther than expected according to the
Soil disturbance
authors’ experience, as illustrated in Fig. 8b. This occurred
Cross-sectional area (mm2) 24600a 26300a 25400a
Disturbance width (mm) 339b 359a 358a regardless of the Kn value.
Positive disturbed area (mm2) 5140b 13390b 57440a To minimise the excessive particle movement, viscous damping
Negative disturbed area (mm2) 2299a 1722ab 884b was added to the model particles. Trials showed that with smaller
a
Means followed with different letters within the same row are statistically viscous damping coefficient, particles were less stable, while with
different at P < 0.05 according to Duncan’s multiple range tests. larger viscous damping coefficient, particles moved as large
Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41 39

Table 2
Inputs for simulations of three soils using the discrete element model.

Parameter Coarse sand Loamy sand Sandy loam Source

Porosity of particle assembly 0.47 0.50 0.47 Derived from bulk density
Particle friction coefficient 0.51 0.67 0.63 Soil friction coefficient
s̄ and t̄ (Pa) 9120 15,200 21,680 Soil shear strength

clusters and became less realistic. Therefore, both underdamping strength, making the effect of particle stiffness on the forces more
and overdamping should be avoided. Trials also showed that it was dominant.
necessary to reduce the local damping in order for the sweep to To find the particle stiffness which results in the best match
disturb a realistic volume of particles. Through numerous between simulations and measurements, the agreement between
preliminary trials, one observed that a local damping coefficient simulations and measurements was assessed using the relative
a = 0.5 and a viscous damping coefficient b = 1.0 appeared to be errors defined as follows:
the best possible combination to reflect realistic dynamic  
100 jM d  Sd j jM v  Sv j
behaviours of soil particles. This is demonstrated in Fig. 9a where REf ¼ þ (1)
2 Md Mv
particles are more stable when compared to those in Fig. 8b. It was
noticed that changing these damping coefficients also changed the where REf is relative average error for soil cutting force (%); M
magnitudes of forces. When compared Fig. 9b with Fig. 8a, one stands for measurement and S stands for simulation; subscripts d
found that adding viscous damping doubled the draught forces. and v stand for draught and vertical respectively.
Table 3 also demonstrates the calibration process. For example,
3.2.2. Model calibrations under the coarse sand, with a series of assumed values of Kn from
With the selected damping coefficients (a = 0.5; b = 1.0), the 5.0  102 to 1.75  103 N m1, the predicted values vary from
particle stiffness (Kn) was inversely calibrated through matching 0.245 to 0.395 kN for draught force, and 0.088 to 0.127 kN for
the soil cutting forces between simulations and measurements. For vertical force. When comparing these predicted values to the
each soil, a series of assumed Kn were used in running the model to experimental data in Table 1, the relative errors vary from 4 to 31%.
predict soil cutting forces. Predicted soil cutting forces of each The calibrated Kn is taken as 0.75  103 N m1 as it causes the least
simulation run were averaged over the stable portion of the curve relative error. Similarly, the calibrated Kn are 2.75  103 and
(Fig. 9b). Table 3 demonstrates how the value of Kn affects the 6.0  103 N m1 for the loamy sand and sandy loam respectively.
predicted soil cutting forces. Draught force increases at larger Kn
value for all soils. In general, effects of Kn on the vertical force 3.2.3. Model validation
follow the same trends, but less pronounced. The increasing trend The calibrated model was validated through comparing
is more obvious in the case of the coarse sand, followed by the simulated soil disturbance characteristics with measured ones.
loamy sand, and then the sandy loam. This correlates inversely to To catch the soil cross-section disturbance, the particle assembly
the bond strength. The sandy loam had the highest bond strength, was ‘‘sliced’’ along the sweep travel direction at different
which may have suppressed the sensitivity of particle stiffness on simulation times (Fig. 10). Over the course of a simulation run,
the forces. In contrast, the coarse sand had the lowest bond the impact from the sweep on particle velocities is a progressive

Fig. 8. Simulation results with the PFC3D default damping coefficients: (a) an
example of simulated draught force (top curve) and vertical force (bottom curve; Fig. 9. Simulation results with modified damping coefficients: (b) instantaneous
direction: downward); X-axis is calculation step; Y-axis is force in N; (b) particle particle velocity; the velocity unit is m s1; (a) draught force (top curve) and vertical
assembly (side view) showing instantaneous particle velocity contours as the force (bottom curve; direction: downward); X-axis is calculation step; Y-axis is
sweep travels; the velocity unit is m s1. force in N.
40 Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41

Table 3
Calibration process and effects of particle stiffness (Kn) on soil cutting forces.

Soil type Assumed Draught Vertical REfa (%)


Kn (N m1) force (kN) force (kN)

Coarse sand 0.50  103 0.245 0.088 14


0.75  103 0.297 0.099 2
1.00  103 0.299 0.095 4
1.25  103 0.365 0.119 21
1.50  103 0.389 0.124 28
1.75  103 0.395 0.127 31

Loamy sand 2.00  103 0.412 0.128 10


2.25  103 0.413 0.130 9
2.50  103 0.436 0.131 8
2.75  103 0.449 0.141 6
3.00  103 0.452 0.128 10 Fig. 11. Screenshots of simulation showing the resultant cross-section disturbed by
Sandy loam 5.00  103 0.561 0.184 4 the virtual sweep.
5.50  103 0.575 0.185 3
6.00  103 0.584 0.185 2
6.50  103 0.598 0.187 4
examined when all particles settled (particle velocities became
7.00  103 0.605 0.195 6
zero). An example of soil surface profile is illustrated in Fig. 11. The
Bold values represent the calibrated value with the lowest relative error between simulated soil deformation is not quite the same. For example, the
measured and simulated forces.
a
REf = relative error between measured and simulated forces.
virtual soil seems to be more plastic, and simulations give more
pronounced negative disturbed area than the positive disturbed
process. As the sweep travels, particles which initially in rest near area. Numerical values of soil disturbance were obtained also using
the tip of the sweep are pushed (Fig. 10a); then the velocity field is the PFC3D_EV measurement tools.
extending to the surroundings of the sweep as it advances (Fig. 10b Simulated soil disturbance characteristics in three soils were
and c), and soil disturbance reaches a stage when there is no more compared with those measured in the soil bins (Table 1). The
expending in the number of impact particles (Fig. 10d). At this agreement between simulations and measurements was evaluated
stage, the impact area is considered to be equivalent to the soil with the relative error defined as:
cross-sectional area measured in the soil bins. Impact areas are of a
 
trapezoid shape that was also observed in the soil bins (Fig. 5). To jM  Sj
REs ¼ 100 (2)
obtain the simulated soil cross-sectional area, the model was run M
for three soils and particle velocity contours were captured at the
stage as shown in Fig. 10d. The boundary of the impact area was where REs is relative average error for soil disturbance character-
identified on the velocity contour through the differences in istics (%); M is measurement; S is simulation. The simulated cross-
particle velocity. The area of the boundary was obtained using the sectional areas and widths were in close agreement with those
PFC3D_EV measurement tools of Distance and Area. Similarly, the measured in the soil bins, as indicated by the low relative errors (1–
simulated soil disturbance width at the soil surface was obtained 4%) (Table 4). Higher relative errors (5–26%) were found for soil
from the top view of the particle velocity contour (Fig. 10e). surface deformation between simulations and measurements. The
Both cross-sectional area and width were taken from instanta- discrepancies may be attributable to several facts. As the
neous particle velocity contours. However, soil surface profile was calibration was done through matching the cutting forces, not

Fig. 10. Particle assemblies showing instantaneous particle velocity field as the sweep travels: (a–d) side view; (e) top view.
Y. Chen et al. / Soil & Tillage Research 126 (2013) 34–41 41

Table 4
Simulated soil disturbance characteristics for three soils and errors relative to the experimental data.

Soil disturbance characteristics Coarse sand Loamy sand Sandy loam

Value REsa (%) Value REs (%) Value REs (%)

Cross-sectional area (mm2) 24,900 1 26,500 1 26,100 3


Disturbance width (mm) 327 4 348 3 346 3
Positive disturbed area (mm2) 65 26 139 14 605 25
Negative disturbed area (mm2) 210 9 180 5 75 15
a
REs = relative error between simulated and measured soil disturbance characteristics.

the soil disturbance characteristics; the model calibrated using soil Franco, Y., Rubinstein, D., Shmulevich, I., 2007. Prediction of soil–bulldozer blade
interaction using discrete element method. Transactions of the American
cutting forces may not fully reflect the soil disturbance behaviours; Society of Agricultural and Biological Engineers 50 (2), 345–353.
also, the damping coefficients of the model were selected through Godwin, R.J., 2007. A review of the effect of implement geometry on soil failure and
qualitative (not quantitative) observations of particle dynamics. implement forces. Soil & Tillage Research 97, 331–340.
Godwin, R.J., O‘Dogherty, M.J., 2007. Integrated soil tillage force prediction models.
The discrepancies may also be attributable to the limitation of the Journal of Terramechanics 43, 3–14.
accuracy of PFC3D_EV measurement tools for highly variable Godwin, R.G., Spoor, G., 1977. Soil failure with narrow tines. Journal of Agricultural
surface profiles. In reality, soil surface disturbance is highly Engineering Research 22, 213–228.
Hansen, M.N., Sommer, S.G., Madsen, N.P., 2003. Reduction of ammonia emission by
variable in nature as well.
shallow slurry injection: injection efficiency and additional energy demand.
Journal of Environment Quality 32 (3), 1099–1104.
4. Conclusions Hettiaratchi, D.R.P., Witney, B.D., Reece, A.R., 1966. The calculation of passive
pressure in two-dimensional soil failure. Journal of Agricultural Engineering
Research 11 (2), 89–107.
The discrete element model developed using PFC3D closely Itasca. 2008. PFC3D-Particle flow code in three dimensions, theory and background.
simulated the operation of a sweep and its interaction with soil. Itasca Consulting Group, Inc., Minneapolis, MN, USA.
Spherical particles with bonds between particle contacts can Mak, J., Chen, Y., Sadek, M.A., 2012. Determining parameters of a discrete element
model for soil–tool interaction. Soil & Tillage Research 118, 117–122.
realistically mimic cohesive behaviours of agricultural soils. McKyes, E., 1985. Soil Cutting and Tillage. Elsevier, New York.
Viscous damping between particles is necessary to reflect the McKyes, E., Ali, O.S., 1977. The cutting of soil by narrow blades. Journal of Terra-
dynamic behaviours of soil resulting from the sweep operation. In mechanics 14 (2), 43–58.
Mckyes, E., Negi, S., Godwin, R.J., Ogilvie, J.R., 1977. Design of a tool for injecting
determining PFC3D model parameters, particle friction can be organic waste slurries in soil. Journal of Terramechanics 14 (3), 127–136.
interpreted as soil friction, and particle bond strength can be Momozu, M., Oida, A., Yamazaki, M., Koolen, A.J., 2003. Simulation of a soil loosening
considered as soil shear strength. The calibration of the particle process by means of the modified distinct element method. Journal of Terra-
mechanics 39, 207–220.
stiffness was achieved using experimental data of draught and Munkholm, L.J., Kay, B.D., 2002. Effect of water regime on aggregate-tensile
vertical forces which are easy to measure, as compared to other strength, rupture energy, and friability. Soil Science Society of America Journal
dynamic behaviours, such as soil disturbance characteristics. The 66, 702–709.
Munkholm, L.J., Schjønning, P., Debosz, K., Jensen, H.E., Christensen, B.T., 2002.
calibrated particle stiffness was 0.75  103 N m1 for the coarse Aggregate strength and mechanical behaviour of a sandy loam soil under
sand, 2.75  103 N m1 for the loamy sand, and 6.0  103 N m1 long-term fertilisation treatments. European Journal of Soil Science 53,
for the sandy loam. The calibrated model was able to predict soil 129–137.
Munkholm, L.J., Perfect, E., Grove, J., 2007. Incorporation of water content in the
disturbance characteristics resulting from the sweep with relative
Weibull Model for soil aggregate strength. Soil Science Society of America
errors under 10% in most cases. Simulation of soil disturbance is Journal 71 (3), 682–691.
more challenging than simulation of soil cutting forces. Continuing Nyord, T., Kristensen, E.F., Munkholm, L.J., Jørgensen, M.H., 2010. Design of a
efforts are required towards understanding relationships between slurry injector for use in a growing cereal crop. Soil & Tillage Research 107,
26–35.
dynamic behaviours and micro-properties of DEM particles. Obermayr, M., Dressler, K., Vrettos, C., Eberhard, P., 2011. Prediction of draft forces
in cohesionless soil with the Discrete Element Method. Journal of Terramecha-
Acknowledgements nics 48 (5), 347–358.
Plouffe, C., Richard, M.J., Tessier, S., Laguë, C., 1999. Validations of moldboard plow
simulations with FEM on a clay soil. Transactions of the American Society of
The research was funded by Aarhus University Research Agricultural Engineers 42 (6), 1523–1529.
Foundation and the Danish Ministry for Food, Agriculture and Potyondy, D.O., Cundall, P.A., 2004. A bond-particle model for rock. International
Journal of Rock Mechanics and Mining Sciences 41, 1329–1364.
Fisheries. Authors wish to acknowledge technicians: Peter Nielsen, Rahman, S., Chen, Y., Lobb, D., 2005. Soil movement resulting from sweep type
Stig Rasmussen, Henrik Mortensen, and Jens Kristensen for their manure injection tools. Biosystems Engineering 91 (3), 379–392.
assistance in field work and measurements. Thanks are given to Jay Sadek, M.A., Chen, Y., Liu, J., 2011. Simulating shear behavior of a sandy soil under
different soil conditions. Journal of Terramechanics 48 (6), 451–458.
Mak for his assistance in computer programming. Shen, J., Kushwaha, R.L., 1998. Soil–Machine Interactions. Marcel Dekker Inc., New
York.
References Shmulevich, I., 2010. State of the art modeling of soil–tillage interaction using
discrete element method. Soil & Tillage Research 111 (1), 41–53.
Abo-Elnor, M., Hamilton, R., Boyle, J.T., 2004. Simulation of soil–blade interaction for Shmulevich, I., Asaf, Z., Rubinstein, D., 2007. Interaction between soil and a
sandy soil using advanced 3D finite element analysis. Soil & Tillage Research 75 wide cutting blade using the discrete element method. Soil & Tillage Research
(1), 61–73. 97, 37–50.
Asaf, Z., Rubinstein, D., Shmulevich, I., 2007. Determination of discrete element model Shmulevich, I., Rubinstein, D., Asaf, Z., 2009. Discrete element modeling of soil–
parameters required for soil tillage. Soil & Tillage Research 92 (2), 227–242. machine interactions. Advances in Soil Dynamics, vol. 3. ASABE, St. Joseph, MI,
Campbell, G.S., 1985. Soil physics with basic transport models for soil–plant chapter 5, pp. 399–433.
systems. Development in Soil Science 14, 1–11 Elsevier, New York. Upadhyaya, S.K., Rosa, U.A., Wulfshon, D., 2000. Application of finite element
Chen, Y., Tessier, S., 2001. Criterion for the design and selection of injection tools to method in agricultural soil mechanics. Advances in Soil Dynamics, vol. II.
minimize liquid manure on the soil surface. Transactions of the American American Society of Agricultural Engineers Monograph, ASABE, St. Joseph, MI,
Society of Agricultural Engineers 44 (6), 1421–1428. chapter 2.
Conte, O., Levien, R., Debiasi, H., Stürmer, S.L.K., Mazurana, M., Müller, J., 2011. Soil van der Linde, J., 2007. Discrete Element Modeling of a Vibratory Subsoiler.
disturbance index as an indicator of seed drill efficiency in no-tillage agrosys- Department of Mechanical and Mechatronic Engineering, University of Stel-
tems. Soil & Tillage Research 114, 37–42. lenbosch, Matieland, South Africa.

You might also like