Powder Technology: Jiahe Shen, Craig Wheeler, Dusan Ilic, Jian Chen

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

PTEC-14628; No of Pages 15

Powder Technology xxx (2019) xxx

Contents lists available at ScienceDirect

Powder Technology

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

Application of open source FEM and DEM simulations for dynamic belt
deflection modelling
Jiahe Shen, Craig Wheeler, Dusan Ilic, Jian Chen ⁎
Centre for Bulk Solids and Particulate Technologies, School of Engineering, The University of Newcastle, NSW 2308, Australia

a r t i c l e i n f o a b s t r a c t

Available online xxxx Belt deflection due to the belt weight and bulk solid loads is essential not only for reducing the flexure resistance
but also for extending the service life of belt conveyors. Current theoretical analysis and numerical modelling
Keywords: have failed to fully consider the interaction between a belt and bulk solid and therefore are not suitable to inves-
Dynamic belt deflection tigate belt deflection under dynamic conditions. The aim of this study was to develop a numerical model for
FEM predicting dynamic belt deflection by coupling the Finite Element Method (FEM) and Discrete Element Method
DEM (DEM). The open source software Code_Aster (FEM) and LIGGGHTS (DEM) were used for the coupling through
Coupling
an interface programmed in Python.
Python
Simulation results in terms of the longitudinal and transverse belt deflection were discussed and compared to the
results obtained from the experimental measurements. Influences on the dynamic belt deflection were analysed
by varying the operating conditions including the belt speed and tension. Furthermore, the belt pressure distri-
bution attributed to the bulk solids was investigated based on the developed coupled model. Good agreements
between the modelling and experiments were observed, indicating the validation of the coupled FEM and
DEM model.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction A variety of FEM and DEM coupling methods have been undertaken
in the field of soil engineering, mechanical engineering and other inter-
The deflection of a conveyor belt occurs when transporting a bulk disciplinary fields [1–9]. These methods can be divided into one-way
material through successive idler sets. While the weights of the belt and two-way coupling. The one-way coupling means the obtained
and bulk material lead to the belt deformation, the interaction at the data from the DEM would be transferred into the FEM unidirectionally.
belt and material interface has significant influences on the dynamic It is not necessary to exchange the simulation data mutually as the neg-
belt deflection. Complexity increases due to this mutually interactive ligible deformation of parts would not affect the boundary conditions in
process where the belt deflection determines the material pressure dis- the DEM significantly. A number of researchers have applied the one-
tribution which, in turn, affects how the belt deflects. Therefore, it is dif- way coupling for a wide range of applications, including shot peening
ficult to investigate the dynamic belt deflection. [1], roll compaction process [2] and many others. Contrarily, the
Numerical simulations are becoming a powerful tool with the im- two-way coupling requires both the FEM and DEM to exchange their
provement of computer power. The Finite Element Method (FEM) and data bidirectionally and update each domain simultaneously. This
Discrete Element Method (DEM) are two of the most widely used method is widely utilised for simulations with large changes of the
methods in the field of bulk material processing and handling. The input conditions [3,4].
FEM (originally based on the continuum theory) is developed to analyse Depending on how the FEM and DEM involve the entire analysis,
solid deformations caused by external load, while the DEM (designed Strnsky [5] classified the FEM and DEM coupling strategies as surface,
for an assembly of particles) is aiming at modelling the behaviour of volume, multiscale and contact coupling respectively. The surface cou-
each individual particle. However, for the dynamic belt deflection that pling, as relatively the most straightforward method, is suitable for
involves both belt (continuum) and particles (discrete), neither the analysing particles interacting with solid components. Onate [6], Villard
FEM nor DEM is able to simulate independently. Therefore, a coupled [7], Fakhimi [8] and many others have applied this method to analyse
method of the FEM and DEM is required to be developed for this specific geomechanics problems. Furthermore, the investigations of tyre and
problem. soil contact were undertaken by Michael [3] and Zhao [9] using the sur-
face coupling. The ground was modelled as a bed of spherical particles
⁎ Corresponding author. using DEM, while the rubber tyre was modelled with FEM to determine
E-mail address: jchen17@uon.edu.au (J. Chen). the deformation. Similarly, Dratt [4] developed a coupled model for

https://doi.org/10.1016/j.powtec.2019.08.068
0032-5910/© 2019 Elsevier B.V. All rights reserved.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
2 J. Shen et al. / Powder Technology xxx (2019) xxx

predicting the belt deflection based on the commercial software belt geometry calculated by the FEM was used as the updated boundary
ANASYS and PFC3D and extended the model with LIGGGHTS to consider condition in the DEM, while the bulk material loads on the belt obtained
the deformation of a metal sheet that endures the impact from a flow of from the DEM were the input for the FEM. The entire coupling starts
a bulk material. with the DEM modelling and ends up with a steady belt profile after
Contrary to commercial software, the open source software offers convergence. A steady belt profile means the belt deflection would
the possibility to modify the models implemented within source codes vary negligibly (≤0.001 mm) with an increased iteration. For a typical it-
and the flexibility to adapt simulations specifically for different applica- eration, the DEM simulation will run for a certain number of DEM
tions at the parameter level. In this study, a coupled FEM-DEM model timesteps, while the FEM simulation will only be performed once. In
was developed for dynamically analysing the interaction of a conveyor this study, the DEM timestep, ΔtDEM (5e-5 s) is dramatically smaller
belt and a bulk material and quantifying the deflection of the belt ac- than the FEM timestep, ΔtFEM (0.25 s). Thus, the duration of each cou-
cordingly. While the Code_Aster program was used for the FEM, the pling iteration is governed by ΔtFEM which is integrated every five thou-
LIGGGHTS program was employed for the DEM. Based on the surface sand ΔtDEM's.
coupling strategy, the modelling was divided into the FEM and DEM do- The proposed and implemented flow chart of the coupling is shown
mains and analysed by suitable methods in each. The main challenge for in Fig. 2. In the first iteration, an initial undeformed STL belt mesh file
coupling was how to create the links between the two independent created by triangular elements using meshing software (GMSH) is
simulations. Rather than wrapping the two simulation codes at a source imported into the DEM simulation as a rigid boundary. After reaching
level, a coupling interface was used, enabling each simulation to process the iteration time ΔtFEM, the DEM simulation pauses. The contact forces
the analysis independently and bidirectionally transfer the simulation of the interacted finite elements are averaged and exported as an output
data between the FEM and DEM. file (VTK). Then, these element forces are used in the FEM analysis and
Furthermore, the dynamic deflection of the belt with the same con- transferred into mesh nodal forces automatically via the Code_Aster
figuration, loaded bulk material and operating conditions were mea- built-in algorithm. The deflection of the belt is calculated using a static
sured and compared with the simulation results. The comparisons non-linear solver due to a large amount of the belt deformation. Follow-
indicated the validation of the coupled FEM and DEM model for ing is the update of the DEM boundary where the deformed belt profile
analysing dynamic belt deflection. obtained from the FEM is assigned in the DEM. Based on the new
boundary condition, the forces acting on elements of the belt are re-
2. FEM-DEM coupling calculated and exported for the use in the subsequent FEM simulation.
The coupling cycle continues until the convergent criterion that indi-
2.1. Coupling interface cates a minimal belt deformation variation, as explained in section
2.3.4 is satisfied.
The scheme of the FEM-DEM coupling is illustrated in Fig. 1, showing
the relationship of the coupling interface and controlled simulation pro-
grams. The coupling interface essentially incorporates the functions of 2.3. Coupling implementation
invoking the simulations, controlling the modelling process and ex-
changing the obtained data. The interface is programmed by a Python 2.3.1. FEM and DEM models
script which utilises the subprocess module to call external programs. As illustrated in Fig. 2, the FEM modelling follows the typical analysis
Thus, both the Code_Aster and LIGGGHTS programs can be invoked procedures including the modelling environment initialization, creation
via the terminal of the Linux operating system. The LIGGGHTS program of the belt FE model, assignment of boundary conditions and loads,
can be simply invoked using an input file in a Linux terminal command. solver analysis and result processing and output. In the investigated ex-
However, starting the Code_Aster is more complicated as the input file ample following, the FEM model consisted of a belt having a length of
only contains the FEM analysis commands. A control file which includes one idler spacing. Typically, in a belt conveying system, belt cross sec-
the FEM commands, geometry meshes, output information and work- tional profiles between any two successive idler sets are generally iden-
ing directory needs to be created and used to start FEM simulations tical. The entire belt was initially modelled as a troughed profile
from the terminal. This is also done by the coupling interface. corresponding to a three-roll idler set. A radius of 50 mm was used for
the side to centre section junctions according to the measurement of
2.2. Coupling procedures the cross-section of the belt without loading. The dimension of the
belt model is shown in Table 1.
The FEM and DEM coupling was conducted in an iterative manner In this study, two types of fabric belts with different thickness of
[4]. For each iteration, two types of data were exchanged: the deformed 6 mm and 12 mm were investigated. Due to the relatively small ratio
of the belt thickness to its overall dimensions, the belts were modelled
by orthotropic thin shell elements to consider the varying elastic moduli
of the belt in the longitudinal (x-axis) and transverse (y-axis) directions
as described in the work of Wheeler [10]. Based on Kirchhoff's theory
[11], the main stresses that may occur are the elastic stresses σxx and
σyy and the in-plane shear stress τxy. The concise form of the strain-
stress matrix can be expressed as

2 3
1 vyx
− 0
0 1 6 Ex Ey 7 0 1
ε xx 6 7 σ xx
6 7
@ εyy A ¼ 6 −vxy 1
0 7 @ A
6 Ex Ey 7 ∙ σ yy ð1Þ
γxy 6 7 τ xy
4 1 5
0 0
Gxy

where the Poisson's ratios and elastic moduli are subject to the
Fig. 1. The FEM-DEM coupling scheme based on a Python programmed interface.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 3

Fig. 2. The flow chart of the FEM-DEM coupling.

following restriction. As documented in [12–14], the Poisson's ratio νxy was assumed to be
similar to the value of a pure rubber as the rubber composition accounts
vxy vyx for the large portion of the belts. The Poisson's ratio νyx can be calculated
¼ ð2Þ
Ex Ey by (2 correspondingly. Thus, the in-plane shear modulus Gxy can be es-
timated by the equation below.
The longitudinal and transverse elastic moduli were measured by
pffiffiffiffiffiffiffiffiffiffi
the three-point bending test that involves supporting a belt sample E x Ey
Gxy ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffi ð5Þ
from the bottom at its ends and applying a vertical load at its middle. 2 1 þ vxy vyx
The relevant Young's modulus E is derived from.

3 The belt properties of the 6 mm and 12 mm thickness fabric belts


F∙l
D¼ ð3Þ were summarized in Table 2, based on which the element properties
48EIXX
in the FEM were assigned.
In order to define the boundary conditions, the belt mesh compo-
where D is the sample displacement; F is the applied force; l is the length nents, such as points, lines and surfaces are grouped and named. As
of the sample; IXX is the area moment of inertia of the belt samples and shown in Fig. 3, the belt is divided into two belt sides, one central belt
can be calculated by. and two junction belts. The contact between the idlers and the belt is
simplified by assuming the belt will conform the idler trough profile
3
b∙h without sliding. Therefore, the right end of the belt is fully fixed,
IXX ¼ ð4Þ
12 whereas the other end is partially restrained and allowed to move lon-
gitudinally when belt tension is applied. The belt outer edges are free of
where b is the belt sample width; h is the belt sample thickness. restraints.

Table 1 Table 2
Dimension of the FEM belt model. Parameters of fabric belt properties.

Description Value Belt properties 6 mm belt 12 mm belt

Belt thickness, T (mm) 6 or 12 Longitudinal Young's modulus, Ex (N/mm2) 38.3 64.9


Belt width, B (mm) 600 Transverse Young's modulus, Ey (N/mm2) 18.8 40.2
Idler roll set spacing, S (mm) 1200 Principle Poisson's ratio, vxy 0.45 0.45
Idler trough angle, β (°) 35 Poisson's ratio, vxy 0.23 0.28
Belt radius over idler junction, R (mm) 50 In-plane shear Modulus, Gxy (N/mm2) 9.9 18.9
Idler roll length, L (mm) 200 Density, ρ (kg/m3) 1365 1365

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
4 J. Shen et al. / Powder Technology xxx (2019) xxx

Fig. 3. FEM modelling boundary conditions.

The external forces were applied in the sequence of the belt self- external loads. ζ is the dimensionless convergence value which, in this
weight, pre-tension and bulk material loads corresponding to the labo- case is defined as 1e-4. The FEM analysis ended up with a deformed
ratory experiments. While the belt pre-tension was varied in the range belt profile that can be used subsequently for the DEM modelling.
of 5kN, 10kN and 15kN, the bulk material loads were obtained from the Correspondingly, the DEM modelling contained a bulk material
DEM calculation. As large displacement and rotations occur, a non- loaded with a full burden profile corresponding to a 100% loaded cross
linear static analysis was used which provided an iterative algorithm section as defined in CEMA [Ref], travelling on the belt boundary that
to update the boundary conditions and stiffness matrix. Various effects is identical to the belt in the FEM. Two different bulk materials, coal
attributed to the geometric and boundary nonlinearity can be consid- and iron ore were simulated in this study. The belt speed was set to be
ered to calculate more accurate results. The non-linear solver basically consistently 1 m/s due to its minor influence on dynamic belt deflection
used an iterative algorithm, in which the loads or displacements [16]. Periodic boundaries were introduced at both ends of the belt to en-
would be applied incrementally and eventually reach the final values. sure a continuous conveying process where particles that left one belt
The modified Newton-Raphson method was employed to solve equilib- end would be automatically reinserted from the other end. Fig. 4
rium equations for each calculation step [11,15]. Equilibrium iterations shows the DEM model.
for a given load step can proceed to the next if the solution is conver- The Hertz Mindlin model was used to determine the pairwise poten-
gent. The convergence criteria are in a relative basis and are given by tial of particles themselves and particles to boundaries. As shown in
  Fig. 5, from Ilic [17], for two contact particles i and j, the contact forces
ep  are typically resolved into a normal force and a tangential force, each
ξ¼ ð6Þ
kRk of which has two components. The normal force is independent of the
tangential force. The normal contact force consists of the forces resulted
where ||ep|| is the square root of the sum of the squares of the load im- from a spring and a damper. Similarly, the tangential force also contains
balances; ||R|| is the square root of the sum of the squares of the applied a spring force (the shear force) and a damping force.

Fig. 4. DEM model for bulk materials.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 5

The particle rolling was considered based on the simplified elastic


plastic spring dashpot, EPSD model, as described in the work of Ai
et al. [18] in which the rolling resistance torque results from a mechan-
ical spring and can be calculated by.

kr ¼ kt ∙R2 ð16Þ

ΔM kr ¼ −kr Δθr ð17Þ

Mkr;tþΔt ¼ Mtr;t þ ΔMtr ð18Þ


 
 k 
M r;tþΔt ≤Mm
r ð19Þ


Mm
r ¼ μrR Fn ð20Þ

Fig. 5. Hertz-Mindlin contact model [17].


where kr is the rolling stiffness; kt is the tangential stiffness; R* is the
particle equivalent radius. Δθr is the incremental relative rotation be-
tween the two particles. Δt is the DEM timestep; μr is the coefficient of
Therefore, the resultant contact forces can be expressed as. the rolling friction; Mmr is the full mobilisation torque and is determined

    by timing the normal force by the rolling friction coefficient. It caps the
F ¼ kn δnij −γ n vnij þ kt δt ij −γ t vtij ð7Þ maximum value of the torque attributed to the spring.
As the bulk material was conveyed continuously on the belt, the par-
where, the first item is the particle normal force in which kn is the nor- ticle and belt contacts were detected while the contact forces acting on
mal elastic coefficient; γn is the viscoelastic damping coefficient; δn is each triangular belt element were calculated and averaged. By using the
the particle overlap; and νn is the particle relative velocity in the normal Velocity-Verlet integration algorithm [19], the positions and velocities
direction. The second item is the particle tangential force. The associated of particles were computed for the update of the next timestep or stored
symbols have the similar meanings as the symbols of the normal force. in a restart file for retrieving the modelling subsequently.
The subscripted t is specific for the tangential force. DEM calibration of particle parameters is necessary to achieve accu-
The normal particle overlap δn is given by. rate modelling of real bulk material behaviour. The main particle pa-
rameters are the interparticle friction μp, the friction between particles
δn ¼ r−d ð8Þ and the boundary mesh μw and the rolling friction μr. These parameters
for the modelled coal and iron ore materials were calibrated according
where, r is the summation of the two particle radiuses; d is the distance to a range of DEM tests, including the angle of repose, slump plane
of the two particle centres. The elastic coefficient for the normal contact angle and inclining wall friction tests. The adhesion of the modelled par-
kn is expressed as. ticles was ignored due to the low contents of both moisture and fine
particles. Moreover, parameter settings regarding the DEM contact
4  pffiffiffiffiffiffiffiffiffiffi
kn ¼ E R δn ð9Þ models were assumed based on previous research outcomes
3
[17,20,21]. These parameters have been proven to be suitable for coal
where E* is the equivalent Young's modulus; R* is the equivalent parti- and iron ore particles in DEM modelling. The input parameters specified
cle radius. in the DEM modelling are summarized in Table 3.

   
1 1−v1 2 1−v2 2 2.3.2. Data exchange
¼ þ ð10Þ Successful data exchange is essential to couple the FEM and DEM
E E1 E2
modelling. It involves mutually extracting and assigning the FEM and
1 1 1 DEM simulation data: the bulk material loads and the belt mesh. With
¼ þ ð11Þ
R R1 R2 the assistance of Python programming, this process is implemented au-
tomatically without the interference of users.
where v1and v2 are the Poisson ratios; R1 and R2 are the radii of the two The bulk material loads from the DEM need to be assigned to the
particles. FEM. The bulk material loads are derived from the summation of parti-
The normal viscoelastic damping coefficient γn can be derived by. cle contact forces with respect to each individual belt element. The DEM
rffiffiffi modelling exports the forces as stresses (forces/area) in the output
5 pffiffiffiffiffiffiffiffiffiffiffi
γ n ¼ −2 β m Sn ≥0 ð12Þ
6
Table 3
Specifications of the DEM parameters.
where β is a constant. m* is the equivalent mass; Sn is the normal stiff-
ness. Parameters Coal Iron ore

  Inter-particle friction, μp 0.6 0.6


β ¼ InðeÞ=sqrt InðeÞ þ π2 ð13Þ Rolling friction, μr 0.5 0.7
Wall friction, μw 0.5 0.6
pffiffiffiffiffiffiffiffiffiffi Particle density, ρp (kg/m3) 1400 3500
Sn ¼ 2E R δn ð14Þ Bulk density, ρb (kg/m3) 800 2000
Coefficient of Restitution, εres 0.2 0.2
1 1 1 Young's Modulus, E (N/m2) 1e7 1e7
¼ þ ð15Þ Shear Modulus, G (N/m2)
m m1 m2 1e7 1e7
Poisson's Ratio, ν 0.3 0.3
Timestep ΔtDEM (s) 5e-5 5e-5
where, m1 and m2 are the particle masses.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
6 J. Shen et al. / Powder Technology xxx (2019) xxx

(VTK) file. The most straightforward method to address the material In contrast, the interaction of the belt and bulk material can be
loads is to directly apply the element stresses to the corresponding reversed if the deformation of the updated belt is less than that in the
belt elements without any data calculation or transformation. However, previous iteration. To overcome the variation of the material loads re-
there are 2352 belt elements included in the belt mesh, making it im- sulted from the update of the belt boundary, the DEM simulation is set
practical to assign the stresses manually. Therefore, Python program- to proceed for 0.25 s (in 5e3 timesteps) that is long enough for the
ming is employed to extract the stresses from the DEM output data, bulk material to reach a stable state before the load data is transferred
identify the related belt elements and automatically map the material to the FEM.
loads. One restriction associated with the update of the DEM belt boundary
Due to the software compatibility, a belt element may have different may rise due to the loss of the contact shear history. The old belt bound-
IDs in the Code_Aster and LIGGGHTS programs. To identify the elements ary is totally replaced by the new deformed belt. As a consequence, the
and assign correct stresses, a detection algorithm is developed and im- contact information of the particles and old belt boundary will not be
plemented based on element spatial positions, as shown in Fig. 6. For preserved for the subsequent modelling in the following iteration.
an arbitrary belt element, the centroid can be calculated by averaging While the normal component of the contact force can be restored im-
the coordinates of the element vertices. If the coordinates of a centroid mediately by reading the particle positions and computing the contact
calculated by the DEM is in line with an element centroid the FEM, the overlaps accordingly, the shear component of the contact force based
element stress obtained from the DEM will be assigned to this FEM ele- on the history effect cannot be retrieved as the tangential overlap is cal-
ment accordingly. A tolerance (5 mm) is introduced to take into account culated by integrating the relative tangential velocity at the contact
the round-off effects of the software. The data exchange of the material point over time. Therefore, it is necessary to investigate whether the
loads is completed by looping all the belt elements and mapping the shear history will affect the bulk material loads significantly. Two DEM
corresponding stresses. simulations are performed with and without the consideration of the
With the modelling proceeding, the belt boundary in the DEM is up- shear history. Special attention is paid to the duration of the first 1.2 s
dated, leading to the deviations of the nodal coordinates of the belt ele- (2.4e4s timesteps) after the belt boundary has been updated. For
ments between the FEM and DEM. To overcome this, connections that every 0.2 s, the bulk material forces are exported for comparisons.
link the FEM element IDs to the corresponding elements in the DEM Both the loads resulted from the coal and iron ore materials are simu-
are created after they are matched up in the first iteration. Therefore, lated, and the results are presented in Fig. 9. The material loads are re-
for the following iterations, the forces acting on the DEM belt elements ferred to as the summation of the forces on the belt and dissolved into
can be assigned to the correct FEM elements via these ID connections. three force components in the belt transverse (X-axis), longitudinal
As an example, Fig. 7 shows the assignment of the loads in the FEM (Y-axis) and vertical (Z-axis) directions. The dash lines stand for the
simulations for coal and iron ore materials in the first iteration. The total simulations without the shear history, whereas the solid lines are re-
forces applied to the belt in the FEM are 317kN and 840kN for the coal lated to the forces calculated with the shear history. The X and Y force
and iron ore respectively, which correlate to the loads obtained from components are considerably negligible compared to the Z force com-
the DEM. ponent with approximately two orders of magnitude. They are highly
For the boundary update in the DEM, the deformed belt mesh ob- overlapped with the horizontal “zero” coordinate in Fig. 9, making it dif-
tained from the FEM is inserted and defined as the new boundary ficult for all the lines to be clearly presented.
after deleting the previous belt mesh. However, Code_Aster does not For the coal, it is evident that the force components are almost con-
output mesh files as STL format that is accepted by LIGGGHTS for sistent for both shear history on and off scenarios. The deviation of the
importing external boundaries. Therefore, conversion of the belt mesh bulk material loads is not distinct, giving a maximum value of only
file is required. The meshing program GMSH is invoked to read the 0.5%. A similar trend is observed for the iron ore, indicating the minimal
Code_Aster output files (MSH file format) and convert them into STL variations of the loads with the shear history.
files, which is also conducted automatically through the Python cou- It can be concluded that the loss of the shear history due to the up-
pling interface. date of a new belt boundary will not result in significant influences on
The bulk material may agitate immediately after the belt boundary is the calculation of the overall bulk material loads. Therefore, the belt
updated. Fig. 8 shows the belt experiences a larger deformation in the mesh data exchange method is proper and can provide relatively accu-
current iteration. At t = 0.02 s of the iteration, the material particles rate simulation results in terms of the bulk material loads. Indeed, the
contact with the belt ends where less belt sag occurs, as shown in feasibility of this method is later validated by comparing the belt deflec-
Fig. 8 (a). Then, the loads concentrate in the middle of the belt at t = tion predicted by the coupled model to experimental measurements.
0.04 s when all particles drop down and impact on this area. From this
point on, the material particles are, to some extent bouncing on the 2.3.3. Iteration continuation
belt, and the loads shift towards the belt ends at t = 0.06 s. A relatively In order to model the belt deflection as a dynamic and continuous
stable state of the bulk material is eventually achieved at t = 0.08 s, as process, the involved iterations are required to be connected. This
shown in Fig. 8 (d). means the results obtained from the current iteration should be

Fig. 6. Mesh element identification based on the triangular centroid.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 7

Fig. 7. Applied loads check via element reaction forces: (a) coal; (b) iron ore.

assigned as initial conditions for the next. The modelling variants, values As shown in Fig. 10, the solid line in each graph indicates the initial
and states of the current iteration need to be preserved and restored at position of the belt, whereas the dash line indicates the solved deformed
the outset of the next iteration. Therefore, the next iteration can start belt profile in the current iteration. Fig. 10 (a) shows the undeformed
from the point where the previous iteration ends. This process is belt without any stresses at the beginning of the coupling. After the
recognised as the iteration initialization of both the FEM and DEM bulk material loads, belt self-weight and pre-tension are applied, the
aspects. belt deflects and gives a displacement of D1 in the vertical direction.
For the FEM initialization, an initial state that essentially consists of The belt reaches the equilibrium where the internal stress increases to
the stress fields, displacements and local variables in the previous calcu- σ1 to balance the external forces. The following iteration is illustrated
lation can be defined for the non-linear analysis solver. The definition in Fig. 10 (b). The belt self-weight and pre-tension are consistent with
involves reading the database created previously and extracting the re- the first iteration, whereas the bulk material loads are calculated by
sults required from the database. The process of the FEM analysis inte- the DEM modelling and equal to P2 (P2 = P1+ ΔP2). Then, the FEM anal-
grated in the coupling can be illustrated by the 2D graphs in Fig. 10. ysis continues with the displacement D1 and the stress σ1 based on the

Fig. 8. The variation of the coal material pressure distribution after the update of the belt geometry with larger belt sag for the current iteration.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
8 J. Shen et al. / Powder Technology xxx (2019) xxx

Fig. 9. Effects of the shear history on the acting force: (a) coal; (b) iron ore.

first iteration assigned to the belt as the initial state. Therefore, the only information of particles and other DEM settings for continuing the
imbalance force is the variation of the bulk material loads ΔP2. The modelling. The particle information includes the parameters such as the
belt deforms accordingly, giving the displacement of D2 = D1 + ΔD2 densities, frictions, pair styles and other coefficients. The DEM settings
and stress of σ1 = σ1+ Δσ2. Similarly, for the third iteration shown define the units and atom style, simulation box size and shape and
in Fig. 10 (c), the solution from the second iteration is defined as boundary settings. These variables are not required to be specified
the initial state. The variations of the displacement ΔD3 and stress again when a new iteration starts. However, contact pair styles may not
Δσ3 correspond to the change of the bulk material loads ΔP3. The cou- restart exactly due to the time difference when particle contact forces
pling terminates in the final iteration n, as shown in Fig. 10 (d) where are computed. For uninterrupted LIGGGHTS modelling, forces are calcu-
the FEM produces the displacement Dn that can be approximated as lated based on particle velocities in the middle of each DEM timestep. If
the displacement Dn−1 of the previous iteration. Therefore, the final the modelling stops and restarts, forces will be initially evaluated with
dynamic belt profile can be obtained when the belt deformation vari- particle velocities given at the end of the previous timestep. Nevertheless,
ation is negligible. the variations of the forces due to this effect are not significant. The par-
The initialization of the DEM can be achieved by retrieving the state ticle behaviour at the outset of a new iteration can be idealised to be con-
recorded in a LIGGGHTS restart file. The restart files store the necessary sistent with the behaviour at the end of the previous iteration.

Fig. 10. The process of the FEM analysis for belt deformation: (a) the first iteration; (b) the second iteration; (c) the third iteration; (d) the final iteration.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 9

Fig. 11. Boundary settings for the DEM modelling.

The main concern associated with the DEM initialization is how to within the simulation box while the fictitious walls stay in the same
define the simulation box boundaries. For LIGGGHTS, the two periodic positions. Particles that travel beyond the belt will be conveyed by the
boundaries at both belt ends are fixed and cannot be adjusted during fictitious walls. Moreover, these fictitious walls can also mitigate the im-
modelling. Due to the applied pre-tension, the belt will extend longitu- pact of particles when exiting and entering the periodic boundaries and
dinally and create an excess to the simulation domain, leading to the therefore be considered as buffer zones.
termination of the DEM simulation. To overcome this, two fictitious
walls that are independent to the belt mesh are added to the ends of 2.3.4. Coupling convergence
the belt, as shown in Fig. 11. The periodic boundaries are set at the far- Due to the interaction of the belt and bulk material, the pressure act-
thest ends of the fictitious walls. Thus, the belt is allowed to extend ing on the belt may vary and distribute differently with time, leading to

Fig. 12. Maximum belt deflection versus coupling iteration: (a) 6 mm thickness belt with coal; (b) 12 mm thickness belt with coal; (c) 6 mm thickness belt with iron ore; (d) 12 mm
thickness belt with iron ore.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
10 J. Shen et al. / Powder Technology xxx (2019) xxx

Fig. 13. Simulation results for the 6 mm thickness belt with the iron ore and 5kN tension: (a) bulk material particle movements in the DEM modelling; (b) belt deflection in the FEM
modelling.

a variation of the belt profile for each coupling iteration. Therefore, the in Fig. 12 (a), (b) and (d). With the same belt tension, the 6 mm thick-
convergence of coupling is determined by a criterion based on the max- ness belt with the iron ore material shows the largest deflection, while
imum longitudinal belt deflection that not only has considerable effects the deflection of the 12 mm thickness belt with coal is minimum. For
on determining the belt conveyor trampling resistance [22], but also is a each scenario, it is evident that the maximum longitudinal belt deflec-
crucial consideration when designing belt conveyor systems [23,24]. If tion decreases with the increase of the belt tension.
the maximum longitudinal belt deflection in the current iteration varies The convergence criterion was applied to determine the completion
b0.5% of the magnitude in the previous iteration, the coupling is consid- of the coupling for each case. As shown in Fig. 12, it can be observed that
ered to be convergent, after which the results are output. the coupled FEM and DEM modelling will be convergent within eight it-
Fig. 12 shows the maximum deflection versus the coupling iterations erations for the belts and bulk materials investigated in this study. The
for the 6 mm and 12 mm thickness belts loaded with coal and iron ore deflection deviation of the last two iterations is in the range of 0.1–
are presented. It can be observed that the maximum belt deflection 0.5%. The coupling time can vary according to the iterations. Overall,
varies with the coupling iteration although the variation is not distinct the computation time is in the range of 4–6.5 h for each individual

Fig. 14. Maximum belt deflection vs velocity in simulations: (a) 6 mm belt with coal; (b) 12 mm belt with coal; (c) 6 mm belt with iron ore; (d) 12 mm belt with iron ore.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 11

case. While the DEM accounts for 90% of the total simulation time, the evident that the particle velocities in the conveyed direction (y-axis)
FEM only contributes 10%. vary due to the deformation of the belt. With larger deformation, parti-
After the coupling is convergent, simulation results including the cles travelling on the belt tend to have a higher velocity component in
particle movement and the belt deflection in the last iteration of cou- the vertical direction (Z axis). Moreover, it is observed that the maxi-
pling were visualized using post-processing software - Paraview and mum belt deflection occurs approximately in the middle of the idler
GMSH. Fig. 13 shows an example of the simulation results of the spacing. Further investigation of the dynamic belt deflection is pre-
6 mm thickness belt with the iron ore under 5kN belt tension. It is sented in the following section.

Fig. 15. Comparison of the experiment and simulation results of the longitudinal deflection for (a) 6 mm belt with coal; (b) 12 mm belt with coal; (c) 6 mm belt with iron ore; (d) 12 mm
belt with iron ore.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
12 J. Shen et al. / Powder Technology xxx (2019) xxx

Fig. 16. Comparison of the experiment and simulation results of the transverse deflection for (a) 6 mm belt with coal; (b) 12 mm belt with coal; (c) 6 mm belt with iron ore; (d) 12 mm belt
with iron ore.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 13

Fig. 17. Coal load distribution on 6 mm belt: (a) longitudinal and (b) transverse directions.

Fig. 18. Coal load distribution on 12 mm belt: (a) longitudinal and (b) transverse directions.

Corresponding to the modelling, 6 mm and 12 mm thick belts, coal


3. Results and discussions and iron ore were used for the experiments. The belt tensions were var-
ied in the range of 5kN to 15kN at equivalent belt speeds of 0.25 m/s to
3.1. Validation of the coupled FEM and DEM model 1.5 m/s.
Based on the experiments, it was found that the dynamic belt deflec-
Dynamic belt deflection was measured using photogrammetry to tion was not significantly affected over the range of equivalent belt
convert 2D images into a 3D spatial profile. The experiments were con- speeds tested, although belt speed may affect the interaction of the
ducted on an experimental test apparatus that replicates a belt conveyor belt and bulk material [10,25]. A similar trend is also suggested by the
by moving several idler roll sets underneath a fixed belt. Details of the coupled modelling, as shown in Fig. 14. The maximum deflection re-
dynamic belt deflection measurements, including the principle, proce- mains approximately consistent with varied belt speeds while higher
dures and data analysis method are documented in Shen [16]. belt tension tends to reduce the belt deflection. Thus, the belt deflection

Fig. 19. Iron ore load distribution on 6 mm belt: (a) longitudinal and (b) transverse directions.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
14 J. Shen et al. / Powder Technology xxx (2019) xxx

Fig. 20. Iron ore load distribution on 12 mm belt: (a) longitudinal and (b) transverse directions.

mainly relies on the belt tension rather than the belt speed. For the 3.2. Bulk material pressure distribution
modelling validation, comparisons of the results from the simulations
and experiments were made for varied the belt tensions and a constant The above results were explored further for insight into the loads
belt speed of 1 m/s. that the bulk materials exert on the belt. The loads acting both longitu-
dinally and transversely were calculated, from which the load distribu-
tion on the belt was obtained. The belt was divided into six equal
3.1.1. Comparison of the longitudinal belt deflection
subsections of 200 mm for the calculation of the longitudinal loads
The developed FEM-DEM coupling model was validated by compar-
and 12 subsections of 50 mm each to calculate the transverse loads.
ing the longitudinal and transverse deflection to the experimental mea-
The loads on each subsection are the cumulative totals of the forces on
surements of the belts with the same dimension and operating
all elements located in the relevant area. The calculated longitudinal
conditions. The longitudinal belt deflection referred to a polynomial
and transverse forces are shown in Figs. 17 to 20.
line that best fits the results of either the simulations or the experi-
From the above results, it is evident that the forces fluctuate in the
ments. The comparisons are shown in Fig. 15.
longitudinal direction. Nevertheless, an increasing trend can be ob-
It is evident that the maximum longitudinal belt deflection occurs
served, indicating the loads on the second half of the idler spacing are
approximately in the middle of the idler spacing. Also, the longitudinal
more considerable than on the first half. This may probably be due to
belt deflection is asymmetric, showing the deflection of the first idler
the active and passive stress states induced inside the bulk material
half is lower than that of the second half. Good overall correlations
[25]. The maximum forces are located on the subsection (600-
may be observed between the experiments and simulations. Some dis-
800 mm) immediately after the centre of the span. Regarding the trans-
crepancy is observed, especially for the first half of the idler spacing (0-
verse load distribution, it is observed that the peak loading force on the
600 mm) in Fig. 15 (b), the middle idler spacing (450-900 mm) in
belt occurs either on the central section of the belt or at the belt junc-
Fig. 15 (c) and the second half idler spacing (600-1200 mm) in Fig. 15
tions (from −150 to 150 mm), while the forces on the belt side sections
(d). Nevertheless, it is suggested that the deviation is minimal with re-
taper and eventually come to zero at the belt edges. Moreover, the re-
spect to the related longitudinal belt deflection. The maximum devia-
sults also highlight that the influence of the belt tension on both the lon-
tion is only 3.12 mm for the 6 mm thickness belt with the iron ore
gitudinal and transverse loads is not significant.
under the belt tension of 5kN, giving a deviation percentage of 9%. It
In addition, the transverse load distribution also shows that the bulk
can be concluded that the coupled FEM and DEM can provide good ac-
material loads exerted on the inclined sides of the belt is in the order of
curacy in predicting the dynamic longitudinal belt deflection.
37–41%, and the loads supported by the centre belt is 59–63% of the
total loads. This finding was found to be within the range of values ob-
3.1.2. Comparison of the transverse belt deflection tained by Ilic using both experimental and simulation techniques [17].
Straight lines were used to represent the transverse belt deflection The experiments suggested that the bulk material loads exerted on
due to the limited points along the belt cross-section that were not the inclined sides of the belt and on the central belt were in the 35–
enough for creating curved line fit. The comparisons of the dynamic 45% and 55–65% ranges of the total loads respectively. Therefore, the
transverse belt deflection are presented in Fig. 16 accordingly. For the coupled FEM and DEM modelling presented shows reasonable overall
6 mm belt shown in Fig. 16 (a) and (c), the simulations show good correlation to the experimental work.
agreement with the experiments except for the edges of the belt sides
where the simulated deflection exceeds experimental measurements.
The deviation may result from the fact that the belt tension was applied 4. Conclusion
through two pulleys at both ends of the test apparatus and therefore
tensions is not evenly distributed along the belt width. In this case, the This study developed a coupled FEM and DEM model for analysing
corresponding stress at the belt edges would be greater than that in the dynamic belt deflection. Based on the proposed strategy and proce-
the middle. Moreover, the absence of the belt transition zones may con- dures, coupling was implemented with the focus on the challenges as-
tribute further to the stress concentration at the belt edges. For the sociated with the data exchange and continuation of analysis. The
12 mm belt shown in Fig. 16 (b) and (d), it is evident that the simulation coupled FEM and DEM model was used to analyse the dynamic deflec-
results correlate well to those from the experimental testing with devi- tion of two types of belts having different properties loaded with coal
ations within an acceptable range observed, especially around the belt and iron ore materials. The simulation results showed good agreement
sides and horizontal to side idler (wall) junctions. with the experimental measurements, indicating the validation of the
In conclusion, the comparison between experiments and modelling coupled FEM and DEM model. Furthermore, the bulk material loads
validates the coupled FEM and DEM approach and suggests that accu- exerted on the belt were investigated using the developed coupled
rate analysis of dynamic belt deflection can be achieved. model. This study yielded the following main findings including:

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068
J. Shen et al. / Powder Technology xxx (2019) xxx 15

• The simulations and experimental measurements suggest that for the [7] P. Villard, B. Chevalier, B. Le Hello, G. Combe, Coupling between finite and discrete
element methods for the modelling of earth structures reinforced by geosynthetic,
range of speeds investigated, belt deflection is not significantly influ- Comput. Geotech. 36 (5) (2009) 709–717.
enced by the belt speed. [8] A. Fakhimi, A hybrid discrete–finite element model for numerical simulation of geo-
• For the belts and bulk materials investigated, the maximum dynamic materials, Comput. Geotech. 36 (3) (2009) 386–395.
[9] C.-L. Zhao, M.-Y. Zang, Application of the FEM/DEM and alternately moving road
belt deflection occurs in the middle of the idler set spacing. The longi- method to the simulation of tire-sand interactions, J Terramech. 72 (2017) 27–38.
tudinal belt deflection profile between idler sets is not symmetric, [10] C. Wheeler, Analysis of the Main Resistances of Belt Conveyors, University of New-
showing a slightly larger deflection of the second half compared to castle, Newcastle, 2003.
[11] R.D. Cook, Finite Element Modeling for Stress Analysis, Wiley, 1994.
the first half. As expected the increased belt tension lead to a decrease [12] Y.-c. Guo, S. Wang, K. Hu, D.-y. Li, Optimization and experimental study of transport
in belt deflection with a non-linear relationship exhibited. section lateral pressure of pipe belt conveyor, Adv. Powder Technol. 27 (4) (2016)
• A slightly increased trend of the loads is found along the equivalent 1318–1324.
[13] D. Marasová, Ľ. Ambriško, M. Andrejiová, A. Grinčová, Examination of the process of
belt speed direction. Moreover, the bulk material loads on the central
damaging the top covering layer of a conveyor belt applying the FEM, Measurement
belt and belt junctions dominate the total loads. The percentages of 112 (2017) 47–52 Supplement C.
the loads acting on the inclined belt sides and central belt are in the [14] Q.J. Zheng, M.H. Xu, K.W. Chu, R.H. Pan, A.B. Yu, A coupled FEM/DEM model for pipe
order of 40% and 60%, respectively. This is found to be in good agree- conveyor systems: analysis of the contact forces on belt, Powder Technol. 314
(2017) 480–489.
ment with those previously presented by Ilic [17]. [15] J.-P. Aubry, Beginning with Code_Aster. A Practical Introduction to Finite Element,
Framasoft, 2013.
[16] J. Shen, C. Wheeler, J. O'Shea, D. Ilic, Investigation of the dynamic deflection of con-
It is expected that the outcome of this study can be directly used as a veyor belts via experimental and modelling methods, Measurement 127 (2018)
technical support for the design of belt conveyor systems. However, fu- 210–220.
ture work will focus on transferring this “quasi” dynamic coupling that [17] D. Ilic, Bulk Solid Interactions in Belt Conveying Systems, PhD Thesis The University
of Newcastle, Australia, 2013.
requires pauses of the FEM and DEM during data exchange into more
[18] J. Ai, J.-F. Chen, J.M. Rotter, J.Y. Ooi, Assessment of rolling resistance models in dis-
delicate coupling where the data is exchanged for every DEM timestep. crete element simulations, Powder Technol. 206 (3) (2011) 269–282.
Also, the coupled FEM and DEM model should further be developed for [19] W.C. Swope, H.C. Andersen, P.H. Berens, K.R. Wilson, A computer simulation method
more general and complex applications. for the calculation of equilibrium constants for the formation of physical clusters of
molecules: application to small water clusters, J. Chem. Phys. 76 (1) (1982)
637–649.
References [20] D. Ilic, C. Wheeler, Measurement and simulation of the bulk solid load on a conveyor
belt during transportation, Powder Technol. 307 (2017) 190–202.
[1] K. Murugaratnam, S. Utili, N. Petrinic, A combined DEM–FEM numerical method for [21] C. Wensrich, A. Katterfeld, Rolling friction as a technique for modelling particle
shot peening parameter optimisation, Adv. Eng. Softw. 79 (2015) 13–26. shape in DEM, Powder Technol. 217 (2012) 409–417.
[2] A. Mazor, L. Orefice, A. Michrafy, A. de Ryck, J.G. Khinast, A combined DEM & FEM [22] C. Spaans, The calculation of the main resistance of belt conveyors, Bulk Solids
approach for modelling roll compaction process, Powder Technol. 337 (2018) 3–16. Handl. 11 (4) (1991) 1–16.
[3] M. Michael, F. Vogel, B. Peters, DEM–FEM coupling simulations of the interactions [23] DIN22101, Continuous Mechanical Handling Equipment, Belt Conveyors for Bulk
between a tire tread and granular terrain, Comput. Methods Appl. Mech. Eng. 289 Materials, Bases for Calculation and Design, Deutsche Norm, 1984.
(2015) 227–248. [24] ISO5048, Continuous Mechanical Handling Equipment – Belt Conveyors with Carry-
[4] M. Dratt, A. Katterfeld, Coupling of FEM and DEM simulations to consider dynamic ing Idlers –Calculation of Operating Power and Tensile Forces, International Organi-
deformations under particle load, Granul. Matter 19 (3) (2017). zation for Standardization, 1989.
[5] J. Stransky, Open Source Dem–Fem Coupling, 2013. [25] C.A. Wheeler, A.W. Roberts, M.G. Jones, Calculating the flexure resistance of bulk
[6] E. Oñate, J. Rojek, Combination of discrete element and finite element methods for solids transported on belt conveyors, Part. Part. Syst. Charact. 21 (4) (2004)
dynamic analysis of geomechanics problems, Comput. Methods Appl. Mech. Eng. 340–347.
193 (27) (2004) 3087–3128.

Please cite this article as: J. Shen, C. Wheeler, D. Ilic, et al., Application of open source FEM and DEM simulations for dynamic belt deflection
modelling, Powder Technol., https://doi.org/10.1016/j.powtec.2019.08.068

You might also like