You are on page 1of 15

Granular Matter (2017) 19:49

DOI 10.1007/s10035-017-0728-3

ORIGINAL PAPER

Coupling of FEM and DEM simulations to consider dynamic


deformations under particle load
Mathias Dratt1 · André Katterfeld2

Received: 29 November 2016 / Published online: 23 June 2017


© Springer-Verlag Berlin Heidelberg 2017

Abstract Coupled FEM–DEM simulations enable the direct Dpart Particle distribution, mm
analysis of the load, the deformation and the stresses inside E Young modulus of the steel sheet samples,
machine parts which interact with bulk materials. The N/mm2
analysis of large deformations of elastic parts is interesting as E part Particle young modulus, N/mm2
the deformation will significantly influence the bulk material f Element force vector
behaviour. In this paper a bidirectional coupling method for fp Volume load vector
the FEM software ANSYS Classic and the DEM software h max Transfer height, mm
LIGGGHTS is presented. The coupling algorithm was Im Mass flow rate, t/h
verified and validated using a modified draw down test rig. K Element stiffness matrix
The results from the experimental investigations and the L ELEM Finite element edge length, mm
FEM–DEM simulations are compared. A very good correla- lLMP Measurement positions, mm
tion between experiments and simulations could be found. lout Length of the outlet, mm
m total Bulk material total mass, kg
Keywords DEM · FEM · Coupling · Deformation · pk Vector for the element node loads
Validation · Draw down test p Pressure, N/mm2
t Time, s
telem Element thickness, mm
List of symbols tKL Flap opening time, s
tsample Thickness of the steel sheet samples, mm
A Area, mm2 tsim Simulation time, s
bout Width of the outlet, mm v displacement vector
dk Grain size distribution, mm v Poisson ratio, -
vc Contact velocity, m/s
vexp Maximum deformation during the experi-
This article is part of the Topical Collection on: Understanding mental test, mm
granular media - from fundamentals and simulations to industrial vsim Maximum deformation during the analogues
application.
FEM–DEM simulation, mm
B André Katterfeld tDEM DEM timestep, s
andre.katterfeld@ovgu.de tFEM FEM timestep, s
1
v Average of deformation, mm
IBAF GmbH - Institute for Construction Machinery, Drive
ε Porosity, -
Systems and Materials Handling, Listemannstr 10a, 39104
Magdeburg, Germany μP Coulomb friction coefficient, -
2 μR Rolling friction coefficient, -
Chair of Conveying Technology, Institute of Logistic and
Material Handling Systems, Otto von Guericke University μw,perspex Wall friction coefficient between bulk mate-
Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany rial and Perspex, -

123
49 Page 2 of 15 M. Dratt, A. Katterfeld

μw,sample Wall friction coefficient between bulk mate- the coupling is presented in this paper. For the validation, the
rial and steel sheet, - load dependent sag of a cantilever beam which is positioned
ρ Material density of the steel sheet beams, in the centre of a bulk material stream is measured and sim-
kg/m3 ulated. The comparison of the experimental and simulated
ρb Bulk material density, kg/m3 data should prove the developed coupling method.
ϕb,stat Angle of repose, ◦
1.2 Literature review

1 Introduction The amount of papers to the topic of coupled FEM–DEM


simulation is quite limited. Most work focuses on the devel-
1.1 Motivation opment of an in-house code and implement FEM and DEM
algorithms in one software. This may be sufficient to solve
Nowadays numerical simulation methods have become an one problem. But the possibility of a reuse for different appli-
essential engineering tool that is widely used in mechanical cations is low. The alternative is the coupling of existing
engineering and more and more increasing in the field of bulk simulation software via specific protocols. Of course, there
material handling. The Finite Element Method (FEM) is usu- is no general coupling interface found in most (commercial)
ally used for the analysis of component deformations due to software packages. Hence, the most general approach is the
external loads. The definition of load assumptions is a diffi- indirect data exchange via binary or ASCII files. The advan-
cult problem, if the complex interaction between machine tages of this approach are, that the features and the usability
components and bulk material should be analysed. Often of existing programs can be used and the coupling strategy
highly idealised and simplified load assumptions based on can be applied for different software packages. Hence, this
continuum models are used. The use of the Discrete Element approach was used for the presented coupling.
Method (DEM) is an interesting alternative for this problem One of the most important publications for the direct cou-
due to its capability to consider the granular behaviour of bulk pling of DEM and FEM is the pioneering work of Munjiza
materials at particle level. The DEM is used primarily for the [3]. Although the title of Munjizas book “Combined finite-
optimisation of bulk material flow problems. However, this discrete Element Method” seems to be very close to the topic
method also allows the determination of the realistic load pat- of this paper, Munjizas work has a totally different focus.
terns by detecting the contact forces between the simulated His aim is to model single or many deformable solids or
bulk material and the interacting part. Nevertheless, when solids which can break into many fragments. To allow this,
applied alone, the DEM is not capable to consider the defor- Munjiza merges “Finite element-based analysis of contin-
mation of the part. To realize a simulation with both, realistic ua… with discrete element-based transient dynamics, contact
load assumptions and deformed machines components, the detection and contact interaction solutions” [3]. Hence, the
coupling between FEM and DEM is necessary. Combined finite-discrete Element Method may be used for
Different requirements on the developed FEM–DEM the simulation of mining processes like rock blasting or rock
coupling result from the deformability of the machine com- cutting [4,5] or the simulation of structural collapse in earth
ponent. If the deformations of the stressed component are quake scenarios [6] and for impact simulations with military
negligibly small, the behaviour of the bulk material is not background. However, Munjizas approach cannot be used
significantly influenced. Applications with negligible defor- for the analysis of the interaction of particles and machine
mations of the machine components are e.g. chutes, impact components.
walls, scraper blades etc.. Therefore, it is possible to neglect Papers with the same focus like the presented work are
the feedback of the component deformations on the entire published by Villard et al. [7], Wellmann and Wriggers [8,9],
bulk material flow. In contrast, the bulk material flow is Stránský and Jirásek [10] and Michael et al. [11]. These
significantly influenced for large deformations. A typical works describe the interaction of particles with boundary
application with large deformations is the load analysis for structures based on coupled FEM–DEM simulations using
idler bearings at a belt conveyor. The idler load depends on in-house codes. Wellmann and Wriggers describe the defor-
the belt pretension, conveying speed and the belt deflection mation of a rubber membrane filled with particles under
due to the bulk load. In this case the deformation of the belt vertical load and the deformation of a bended shell caused by
plays an important role and cannot be neglected [1,2]. the load of a particle bed. The deformation of a membrane
The FEM–DEM coupling presented in this paper con- which is in contact with particles is also described by Villard
siders both options. Hence, the coupling should allow the et al. [7]. Michael et al. [11] uses coupled FEM–DEM simula-
analysis of (high-) dynamic deformation processes for indus- tions for the analysis of the tire behaviour on the ground. The
trial applications such as scraper or belt conveyors. Beside ground was modeled by a bed of particles. In all mentioned
the description of the coupling method also the validation of publications [7–11] the transfer of the contact forces from the

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 3 of 15 49

Fig. 1 Coupling strategy.


Calculation cycle of the
FEM–DEM coupling with
ANSYS Classic and
LIGGGHTS visualised on the
example of a loaded fixed beam

DEM simulation to the FE model is realized using equivalent In the first step an extended stl-file is derived from an
node forces. But the implementation of the finite elements is arbitrarily meshed part in the FEM program ANSYS Clas-
limited to linear triangular shell elements and/or linear tetra- sic (step in Fig. 2). The stl-file contains beside the actual
hedral elements, thus finite elements without midside nodes. triangulated surface elements also the number of the orig-
The reason for this limitation is probably the use of planar inal finite element. The extended stl-file can be imported
triangulated surfaces in most DEM algorithms. Hence, such in the used DEM program LIGGGHTS and defines the
a FEM–DEM coupling on equation level requires the same deformable machine component which interacts with the par-
meshing of the structure in FEM and DEM and therefore, ticles (step in Fig. 2). After a certain time tFEM (which is
the use of these basic finite element types. However, it is a multitude of the much smaller DEM time step tDEM ) the
very well known that these basic finite element types cause time averaged contact forces from the particle load on each
a relatively stiff behavior of the FE model which will under- wall element are written in an ASCII file marked with the
estimate the occurring deformations. Hence, such elements file extension “*.fem” for the import into the FEM software
are rarely used and the more sophisticated FEM codes will (step in Fig. 2). The DEM loads can be assigned to the
not support their use. right finite element in the FEM program due to submitted ID
Two commercial DEM programs EDEMTM and Rocky of the finite element (step in Fig. 2). A very important task
offer the coupling to FEM (ANSYS Workbench) and prob- is the calculation of the node forces based on the load for
ably the use of more sophisticated finite elements. But these each element. This calculation is trivial for primitive meshes
codes only allow a one-way coupling (unidirectional). Hence, but it requires some additional thoughts if the method should
applications with large deformations of the machine compo- work for modern shell and volume elements of higher order.
nent and its influence of the bulk material flow cannot be Hence, this issue is addressed in the following subsection.
simulated. An unidirectional coupling would end after the calculation
The FEM–DEM coupling presented in this work allows of the deformation in the FEM program (step in Fig. 2).
a bidirectional coupling as well as the use of state of the art For a truly bidirectional coupling the new deformed mesh
finite element types with a squared displacement approach. is exported into a new extended ASCII file (step in Fig. 2).
This time not only the new coordinates of the nodes are trans-
2 Developed coupling ferred to the DEM but also the interpolated node velocities
which are calculated based on the time step tFEM for each
2.1 General coupling strategy load step (LS). Hence, the file is saved in the more general
csv-format. The node velocities allow a direct deformation of
The general coupling idea is visualised in Fig. 1 on the exam- the meshed wall structure in LIGGGHTS over time. With
ple of the sag calculation of a loaded fixed beam. The main this approach the particles do not loose contact to the walls
idea is the data exchange between FEM and DEM simulation and the shear forces (based on an incremental calculation of
via ASCII files. For better understanding Fig. 2 visualises the the tangential overlap) do not get lost. The coupling cycle
timetable of the coupling. continues with the calculation of the new loads on the mesh

123
49 Page 4 of 15 M. Dratt, A. Katterfeld

Fig. 2 Time schedule for the developed FEM–DEM coupling

in the DEM (step in Fig. 2) and the export of a new load triangle areas. The vectors resulting from the contact forces
file for the FEM part (step in Fig. 2). of particles and walls FID(n) and FID(n+1) is summed up to
the load vector Fi . This is not needed for the corresponding
2.2 Calculation of node forces for arbitrarily shaped area of a triangle element; its load vector results directly from
meshes the contact force. As published by Dratt in 2010 [1,15], the
contribution of the global x, y and z-components of the load
FEM breaks problems down to small elements that are con- vector is assigned to equivalent node forces. The required
nected at nodes and solves them individually. In applications weighting factors are related to the used finite element types
that model thin walled components in three-dimensions the (line, surface, or volume elements) and the degree of poly-
use of shell elements is preferred. This special type of ele- nomial of their shape functions.
ment unites the membrane stiffness of membrane elements The node displacements of a 8-node rectangle element are
as well as the bending stiffness of plate elements [12,13]. described by the following eight shape functions Eqs. (1)–
Typical examples of these elements are three-dimensional (3) used with dimensionless, natural coordinates ξ and η in a
square 6-node triangle and 8-node rectangle elements. The range of ξ, η = ± 1. These equations correspond to the unit
term square relates to the polynomial degree of the shape displacement functions of the 8-node master square as shown
function of the element type that describes the deformation on the right side of Fig. 3 on which every 8-node rectangle
of those elements. Compared to linear elements this type has element is mapped with Cartesian coordinates. This helps to
extra nodes in the middle of its sides that allows adapting avoid changing shape functions used for multiple element
the edges to a square function. The increase in the degrees geometry in the Cartesian coordinate system [14].
of freedom results in higher accuracy of the results while
keeping the number of elements low [14].
To develop a realistic coupling approach, it is important to 1
Nk (ξ, η) = − (1 + ξk ξ ) (1 + ηk η)
identify the corresponding elements in the finite and discrete 4
element simulations. To verify this, the finite element mesh (1 − ξk ξ − ηk η) für k = I, J, K , L (1)
with its element nodes I to L is transferred into an stl-format. 1  
Nk (ξ, η) = (1 + ξk ξ ) 1 − η2 für k = N , P (2)
By breaking the 8-node rectangle elements into a pair of tri- 2
angle areas it is guaranteed that the derived areas are planar, 1 
Nk (ξ, η) = 1 − ξ 2 (1 + ηk η) für k = M, O (3)
which is needed by the DEM contact laws. The same happens 2
for 6-node triangle elements that are transferred into a trian-
gular area. If the FEM mesh is mixed and contains rectangle
and triangle elements together, it is important to verify that The shape function can be summed as a matrix as shown in
one pair of triangle stl-areas is corresponding to one rectan- Eq. (4):
gle element and a single triangle element to only one triangle
stl-area. This is ensured by exporting the element number
 
as additional information in the stl-file. Figure 3 shows the NI 0 NJ 0 NK 0 NL 0 NM 0 NN 0 NO 0 NP 0
N= .
basic principal of the DEM–FEM coupling in the case of the 0 NI 0 NJ 0 NK 0 NL 0 NM 0 NN 0 NO 0 NP
ith 8-node rectangle element and its related pair of derived (4)

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 5 of 15 49

Fig. 3 a Basic principal of the DEM–FEM coupling by using a 8-node rectangle square and a 6-node triangle element. b Structure of a 6-node
master triangle and a 8-node master square

According to Bathe [16], the elastic potential of a general 


P

finite element is given by Eq. (5): p= pk Nk (ξ, η) = N · pK mit k = I . . . P. (7)


k=I

1 T Derived from the volume loads the general form of the force
= v Kv − vT f. (5)
2 vector is:

In Eq. (5) v is the node displacement vector, f is the element  z 2 y2 x2
force vector and K is the element stiffness matrix. The con- fp = N N · pK dV =
T
NT N · pK dx dy dz. (8)
tribution of the elements weight is the force vector resulting (V ) z 1 y1 x1
from the volume load fp .
The boundaries and differential of the volume integral in
 Eq. (8) is positioned in the Cartesian coordinate system while
fp = NT p dV (6) the shape function approach Eqs. (1) and (3) is in natural ξ-
(V ) η-coordinates. The coordinate’s transformation is completed
using the reduced two-dimensional Jacobi-matrix:
According to Bathe [16], the load p is interpolated using    ∂y

∂x
the shape function approach and results in the vector for the J11 J12 ∂ξ ∂ξ
J= = ∂x ∂y . (9)
element node loads pk : J21 J22 ∂η ∂η

123
49 Page 6 of 15 M. Dratt, A. Katterfeld

Fig. 4 Assignment of the resulting nodal forces for corner, midside and shared nodes. a Regular arranged FE-mesh; b mixed FE-mesh

The differential dz corresponds to the constant imaginary by the weight factors. The equivalent node forces are then
element thickness telem . The transformation relation results summed and applied to the nodes. This is shown in Fig. 4.
according to Betten [14] as the determinant of the reduced Based on the preceding analysis we now describe, how
Jacobi-matrix: to find the weight factors for distorted, planar 8-node rect-
angle elements as shown in Fig. 5. Plane element surfaces
dV = dx dy telem = telem detJ dξ dη. (10) require all nodes I…P to be in the x-y-plane and the midside
nodes M…P are on a direct connection line of the corner
The force vector fp resulting from the volume load of the nodes I…L. Using an imaginary element thickness of telem
weight of the element results in natural ξ-η-coordinates from and a constant material density ρelem the determinate of the
the relationship: Jacobi-transformation matrix will not result in a constant
value, but rather in a polynomial in relation to ξ, η that has to
1 1 be included in the integration of the element areas, as show
fp = NT N · pK telem detJ dξ dη. (11) in Eq. (11). The weight factors are always referenced to the
−1 −1 x-y-coordinates so they cannot be directly adapted for three
dimensions. For this reason, a calculation of the weight factor
By solving the integral Eq. (11) a constant value of −1/12 is is needed that is independent of the shape function.
calculated for the corner nodes I to K and 1/3 for the midside If such an element is loaded with a volume force fp , having
nodes M to P as shown in Fig. 3a. These weight factors can a constant element thickness telem and a density ρelem and is
be applied for elements in the shape of a square, a rectangle mounted in the centre of gravity it has to be in balance. In that
and a parallelogram and can be adapted directly for three case the potential Πi for the element is calculated following
dimensions. The same approach can be used to calculate the the principal of the elastic potential at an extreme, in this
weight factors for the 6-node triangle elements, only the inte- case the minimum, according to Mueller [12]. In the current
gration boundaries have to be changed according to Fig. 3b. situation the boundary conditions for the centre of gravity of
The factors are 1/3 for the midside nodes M, N , P and 0 for the area Si(I...P) in relation to the individual areas is given by
the corner nodes I, J and K . The extension of the calcula- the following geometric relation:
tions over the whole model takes the neighbouring relations
1
of nearby elements with shared nodes into account. If a node Ai(IMSP) + Ai(NKOS) = Ai(MJNS) + Ai(OLPS) = Ai , (12)
belongs to several elements then as a first step the x-, y-, and 2
1
z-load vectors of the element are calculated and multiplied Ai(MNSP) + Ai(OPNS) = Ai(PMSO) + Ai(NOSM) = Ai . (13)
2

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 7 of 15 49

Fig. 5 Context of the areas Ai(k) in comparison to the centre of gravity of an element Si(I...P) of a planar, distorted 8-node square element

From these relations, it follows that, if the sum of opposite factors of the master triangle stay the same since the relation
individual areas Ai(k) is half of the area Ai of the planar between the individual areas and the element area is consis-
distorted 8-node element, then the sum of the weight fac- tent. When an element is wrapped or has curved sides the
tors must be −2/12 for the corner nodes I…L and 2/3 for sub area relations are used automatically as an approxima-
the midside nodes M. . .P as in the master square to ensure tion. The error that occurs from this approximation correlates
equilibrium. to the element size and can be pushed under 1% using a well-
This statement allows the determination of the weight fac- conditioned FE-mesh.
tors Wi(j,k) for the planar distorted 8-node element using the
relation of the single areas Ai(k) . Wi(j,k) is the weight factor
of the ith element and the jth component (x, y or z), at the
3 Verification
node position k = I..P.
A simple test case of a fixed beam under constant load was
1 1 
P
used to verify the coupling algorithms. As shown in Fig. 1,
N I · pk Nk detJ dξ dη · the load was generated by a particle bed. The results of the
−1 −1 k=I
coupled simulations were compared to the analytical calcu-
⎧ ⎡ 1 1 ⎤⎫−1
⎨P   
P ⎬ lations using a 1st order approach assuming small deforma-
⎣ N k · pk Nk detJ dξ dη⎦ tions. The dimensions of the beam were chosen as 1000 ×
⎩ ⎭ 100×100 mm (L×W×H). The well-known material param-
k=I −1 −1 k=I
  eters for mild steel are applied. In a first step the constant area
1 Ai(NKOS)
= Wi(j,I) = − (14) load of p = 0.005 N/mm2 was applied to the FE mesh with
6 Ai(IMSP) + Ai(NKOS) an element edge length of L ELEM = 50 mm which was also
1 1 
P the diameter Dpart of the monodisperse particle bed. The
N M · pk Nk detJ dξ dη · comparison of the simulated and analytically calculated
−1 −1 k=I maximum sag of the beam showed a difference of less
⎧ ⎡ 1 1 ⎤⎫−1 than 1%. The difference could be explained by the uneven
⎨P   
P ⎬ load distribution of the randomly generated particle bed.
⎣ N k · pk Nk detJ dξ dη⎦
⎩ ⎭ Less than 0.01% difference could be reached, if a parti-
k=I −1 −1 k=I
  cle bed with a cubical structure was generated above the
2 Ai(OPSN) beam.
= Wi(j,M) = (15)
3 Ai(MNSP) + Ai(OPSN) Further, the influence of the ratio between the element
edge length and the particle diameter was investigated by
The weight factors in Eqs. (14) and (15) left for the nodes a series of simulations with varying ratio. In [17] could
I and M are derived from the relation for the nodes given be shown, that a ratio range of Dpart /L ELEM = 1/8. . .40
in Eq. (11) and the total element area Ai and match with the did not cause significant differences in the simulated sag.
results calculated using the sub areas Ai(k) on the right. If The difference to the analytical result was still less than
using planar distorted 6-node triangle elements the weight 0.01%.

123
49 Page 8 of 15 M. Dratt, A. Katterfeld

Fig. 6 Modified draw down


test rig for validation tests

4 Validation the outlet has a constant value of bout = 100 mm defined by


the inner dimension of the upper chamber. The length of the
In addition to verification tests based on the analysis of the outlet can be varied from lout = 0 to 500 mm. Thus, the mass
static sag of a fixed beam, dynamic validation tests with flow rate Im can be changed. Once the outlet-dimension is
large deformations were undertaken. The test results were defined and the filling process is completed, a flap-system
compared with analogous bidirectional FEM–DEM simula- below the hand-slides can be opened (Fig. 7, right). After the
tions. For the experimental validation tests a modified draw opening, the bulk material is flowing through the outlet into
down test rig was used (Figs. 6, 7, 8). The test rig combines the lower chamber.
material flow and deformation analysis by the determina- The lower chamber can be completely removed from the
tion of the load dependent sag of a cantilever beam which is test rig and includes a firmly connected measuring frame
positioned in the centre of a bulk material stream. The aim (Fig. 8). The frame can also be removed from the lower cham-
of the validation experiment was the prove of the coupling ber after unlocking the screw connections. The frame consists
method for further use in industrial application. The main of two transverse supports which are movable in z-direction.
aim was not to increase the physical or theoretical under- The upper transverse support is used to fasten a steel sheet
standing of what is happening in the impact zone of the sample as the cantilever beam. The lower transverse support
cantilever beam. Therefore, only some general comments below the sample is equipped with a laser-optical sensor,
will be made on the fundamental characteristics of the flow which is additionally movable in x-direction. The measuring
regime. range of 200 mm is shown in Fig. 8 as a solid line passing
The following test procedure was used. A bulk material is the steel sheet sample.
filled through the hopper into the upper chamber of the draw The steel sheet deflects when the bulk material impacts
down test rig. The outlet-length can be defined by using two it. The deformation depends on the dimensions of the sam-
hand-slides below this chamber (Fig. 7, left). The width of ple, especially the thickness tsample , the young modulus E,

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 9 of 15 49

Fig. 7 Left Hand-slides and


outlet; Right flap-system

Fig. 8 Lower chamber with integrated measuring frame

the incoming mass flow rate Im and the contact velocity vc measured directly after the tests by measuring the slope of
as a result of the falling height h max . The deformation of the material which remains in the upper chamber.
the sample is being measured over the entire duration of the The calibration of the DEM parameters was done in a first
experimental test with a laser-optical sensor. In addition, for step using the angles of repose ϕb,stat in the upper cham-
an optical evaluation the tests were recorded with a high- ber (Fig. 10). For this, the static Coulomb friction μP and
speed camera. rolling friction μR of the particles were varied. In an addi-
For the validation tests four steel sheet samples with tional second calibration step the mass flow rates Im was
different thicknesses tsample = 0.75, 1, 1.5, 2 mm were used for the fine adjustment of the friction parameters. Im
used. The falling height was set to h max = 360 mm and was measured during the validation experiments as well as
is defined by the distance between the lower edge of the in a simulated parameter study of the discharge behaviour
flap-system and the upper edge of the steel sheet sam- in the upper chamber. The second calibration step was
ple. The outlet length was set to lout = 40 and 100 mm. necessary because the deformation of the steel sheets depends
The deformation measurement was undertaken for two posi- directly on the values of the incoming mass flow rates. Hence,
tions lLMP = 60 and 100 mm from the outer edge of the it is essential not only to achieve the same angle of repose
samples. but also to achieve a comparable mass flow rate similar to
Three different cohesionless bulk materials were used for the experiments.
the validation tests (Fig. 9): plastic granulate, wheat and For the particle distribution Dpart , the real grain size dis-
potash granulate. The bulk material parameters were deter- tributions dk of all bulk material samples were used. The
mined experimentally. The angle of repose ϕb,stat can be following Table 1 summarizes all experimental bulk mate-

123
49 Page 10 of 15 M. Dratt, A. Katterfeld

Fig. 9 Bulk material samples. a Plastic granulate; b wheat; c potash granulate

Fig. 10 Bulk material samples in the upper chamber of the validation angle of repose after opening of the flap-system, g–i evaluation of the
test rig. Left Plastic granulate; middle wheat; right potash granulate; angle of repose after experimental test, j–l evaluation of the angle of
a–c Starting situation with even bulk material surface, d–f left and right repose after DEM simulations

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 11 of 15 49

Table 1 Summary of all DEM


Properties/bulk material Plastic granulate Wheat Potash granulate
parameters for the coupled
FEM–DEM simulation Contact model Hertz Mindlin no slip. Rolling friction model according to an adopted
elastic plastic spring dashpot model according to Wensrich et al. [18]
Bulk density,ρb [kg/m3 ] 577 743 971
Angle of repose, ϕb,stat [◦ ] 26 26 36
Uniformly distributed particle size, 3.8…5.3 3.0…5.0 2.5…6.0
Dpart [mm]
Porosity, ε [−] 0.385 0.402 0.487
Wall friction coefficient,μw,sample 0.148 0.215 0.208
[−] (bulk material vs. steel sheet)
Wall friction coefficient, 0.22 0.26 0.183
μw,perspex [−] (bulk material vs.
Perspex)
Bulk material mass, m total [kg] 13 6 10
Coulomb friction, μP [−] 0.12 0.17 0.5
Rolling friction, μR [−] 0.3 0.18 0.3
Particle young modulus, E part [Pa] 1e8
Poisson ratio, ν [−] 0.3
DEM timestep, tDEM [s] 2.5e − 5 and 1.0e − 5

rial parameters and the calibrated DEM particle parameters 0.75 mm, outlet length lout = 40 mm and a measurement
for the coupled FEM–DEM simulation. position of lLMP = 60 mm are used. The results of the laser-
The system parameters of the test rig were directly used optical measurement and the modelled displacement of the
for the simulation model. In the DEM simulations, all wall cantilever beam are presented in Fig. 11.
surfaces of the test rig were considered. For the FEM, it In Fig. 11 several characteristic points – can be iden-
is only necessary to consider the cantilever beam modelled tified in the measurement data. Fig. 12a–f shows a sequence
out of shell or volume elements. The FEM simulation is exe- of images from the high-speed camera which assigns the real
cuted as a transient analysis (time integration). The following behaviour of the experiment and the simulation to the char-
Table 2 summarizes all system parameters of the test rig and acteristic points. Peaks in the experimental data result from
the used FEM parameters. falling particles which cross the laser for a very short time.
To highlight the most important validation results the The characteristic points can be interpreted as follows:
tests configuration wheat, steel sheet thickness of tsample =
1. After t = 0.2 s the flap-system is completely open. The
Table 2 Summary of all system and FEM parameters for the coupled
first particles contact the cantilever beam at t = 0.25 s
FEM–DEM simulation
(compare point in Figs. 11, 12a). The initial mass flow
Description Value
rate (t = 0. . .0.5 s) is significantly higher than the mass
Falling height, h max [mm] 360 flow rate in steady state (t = 0.5. . .1.5 s). The reason
Outlet length, lout [mm] 40/100 for this is an initial bulk material reservoir between the
Measurement positions, lLMP [mm] 60/100 hand-slides and the flap-system.
Flap opening time, tKl [s] 0.2 2. After the opening of the flap-system the initial bulk mate-
Thickness of the steel sheet samples, tsample [mm] 0.75/1/1.5/2 rial reservoir forms a kind of “bulk material drop” which
Dimension of the steel sheet samples, L × W 340 × 100 results in an initial impulse when hitting the steel sheet.
[mm] This causes the maximum deformation during the whole
Area of locked DOF’s on the steel sheet samples, 60 × 100 test of vexp = −52.8 mm at t = 0.32 s . The simulation
L × W [mm] shows almost the identical max. deformation at this time
Young modulus, E[N/mm2 ] 185.000 (vsim = −54.3 mm).
Poisson ratio, ν [−] 0.3 3. The abrupt reduction of the mass flow rate at t =
Material density, ρ [kg/m3 ] 7850 0.42 s causes a swing back of the steel sheet sample in
Element edge length, L ELEM [mm] 10 experiment and simulation results . At this point a
FEM timestep, tFEM [s] 0.001 slightly lower deformation in the simulation can be
recognised.

123
49 Page 12 of 15 M. Dratt, A. Katterfeld

Fig. 11 Comparison of the


deformation data from the
experimental test and the
FEM–DEM simulation with
wheat (m total = 6 kg), a steel
sheet sample thickness of
tsample = 0.75 mm, an outlet
length of lout = 40 mm and a
measurement position of
lLMP = 60 mm

4. The constant mass flow rate in steady state from t = 2. The mass flow rate of the particle stream is almost identi-
0.5 to 1.5 s causes an almost constant deformation of cal in experiment and simulation. This requires the right
the steel sheet sample . The average deformation in chose of the micromechanical properties of the DEM
experiment and simulation is almost identical (approx. particles. The proper calibration of these parameters
v = −26 mm). is a very important prerequisite in any DEM simula-
5. After t = 2 s a large part of the bulk material has tion. Obviously, the used calibration procedure and the
left the upper chamber. The mass flow rate and the resulting DEM parameters allow a realistic simulation of
resulting deformation of the steel sheet sample is reduced the particle behaviour.
non-linearly until t = 4.5 s . The real and simulated 3. The deformation speed is mainly dependent on the initial
deformation curves are approximately equal. impulse and requires a minimum coupling time step of
6. The final deformation state at t = 4.5 s is only influ- tFEM = 1e − 3 s. Such a coupling time step allows
enced by the bulk material resting on the steel sheet the consideration of the true interaction of steel sheet
sample. At this point a very good correlation between and bulk material stream. With typical DEM time steps
the results of the experimental test and the simulation of tDEM = 1e − 5...2.5e − 5 s this results in a ratio
can be found too. tFEM /tDEM of 40…100.
The deformation curves in Fig. 11 as well as the synchronized
image sequences in Fig. 12 show a very good correlation Figure 13 shows a more detailed look inside the material
between the experimental test and the coupled simulation. stream in the moment of the highest sag at t = 0.32 s ( ).
It is obvious, that the sag of the cantilever beam is strongly To show the particle behaviour inside the stream, a slice of
dependent on the impacting mass. The good correlation of particles with a slice thickness of two times the maximum
experimental data and simulation can be explained by the particle diameter in the x-z-plane is shown. The impact zone
following reasons: is relatively dense packed with particles (Fig. 13a). The par-
ticle stream falls with a velocity of 3 m/s onto the beam. The
1. The observed “free fall and impact” problem demand influence of air resistance as well as fluidisation effects can
the conservation of momentum in the coupled simula- be neglected due to the low velocities and the relatively large
tion approach. Due to the correct data exchange between particles. The falling particles in the stream do not hit the
DEM and FEM this fundamental physical law is fulfilled. beam directly. They hit the layer of particles which is dammed
The developed coupling algorithm works for arbitrary on the beam. The dammed particle layer show lower veloci-
meshed geometries and can be universally used. ties of less than 1 m/s. While the particles in the zone directly

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 13 of 15 49

Fig. 12 Comparison of
high-speed camera images from
the experimental test and
analogue FEM–DEM simulation
results for wheat (m total = 6 kg),
a steel sheet sample thickness of
tsample = 0.75 mm, an outlet
length of lout = 60 mm and the
measurement position of
lLMP = 60 mm

123
49 Page 14 of 15 M. Dratt, A. Katterfeld

Fig. 13 Particle behaviour in the impact zone during the highest sag velocity is shown by particle colour (a) or coloured arrows (b) to allow
of the beam at t = 0.32 s ( ). A slice of particles with a thickness an understanding of the general flow regime. In (c) the rotational speed
of 2 × Dmax in the x-z-plane is shown. The magnitude of the particle of the particles is shown via the particle colour

below the impact zone have a relatively high rotational speed, experiment and simulation is the correct calibration of the
the most particles in direct contact with the beam wall have a DEM parameters. The used calibration procedure considers
reduced rotational speed (Fig. 13c). Hence, the particle slide the angle of repose as well as the discharge time. The right
and roll along the beam (Fig. 13b+c). calibration allowed the simulation of a realistic mass flow
rate at the outlet of the upper chamber and also the realis-
tic micromechanical behaviour of the particles in the impact
5 Conclusion zone.
Hence, the developed FEM–DEM coupling between
The presented article has shown that coupled simulations ANSYS Classic and LIGGGHTS can be used for a
based on the Finite and Discrete Element Method sig- realistic analysis of the interactions between bulk materi-
nificantly improves the quality of the analysis of bulk als and machine parts [2] also for high-dynamic deformation
material handling processes with significant deformations processes. The validated coupling algorithm can be used for
of machine parts. This is possible due to the more real- the simulation of industrial applications in e.g. material han-
istic consideration of the complex load patterns resulting dling. With this simulation approach the virtual development
from the contact between the bulk material and the machine process of bulk solid handling equipment can be improved.
parts. Design variants can be easily analysed regarding the required
The validation shows a very good correlation between strength and reliability.
experimental and simulated results. Although the described
experiments and simulations where mainly undertaken to Acknowledgements The authors thank the German Ministry of
Research and Education for the financial support of the research project
prove the coupling algorithm, some fundamental observa- “SimBa” (01 IS 13 006 A).
tions could be made. The sag of the cantilever beam with a
given thickness is mainly influenced by the impacting mass
of the material stream. Hence, the mass flow rate of the mate-
rial stream determines the sag. The good correspondence of References
experiment and simulation proves that the developed cou-
1. Katterfeld, A., Dratt, M., Wheeler, C.A.: Coupling ANSYS and
pling method fulfils the required conservation of momentum. PFC3D for the simulation of the conveyor belt deflection. In: Con-
One reason for the fulfillment of this fundamental physi- tinuum and Distinct Element Numerical Modeling in Geomechan-
cal law is the proper implementation of the data exchange ics, Proceedings of the 2nd International FLAC/DEM Symposium.
between DEM and FEM, especially the developed coupling Itasca Consulting Group, Melbourne (2011)
2. Dratt, M., Katterfeld, A., Wheeler, C.A.: Determination of the
algorithm including the derivation of the node forces in bulk flexure resistance via coupled FEM–DEM simulation. In:
the FEM part which allows the use of arbitrary meshed Bulk Solids Handling, no. 3, pp. 50–58. WoMa Media, Clausthal-
geometries. Another reason for the good correlation between Zellerfeld (2015)

123
Coupling of FEM and DEM simulations to consider dynamic deformations under particle load Page 15 of 15 49

3. Munjiza, A.: The Combined Finite-Discrete Element Method, 1st 10. Stránský, J., Jirásek, M.: Open source FEM–DEM coupling. In:
edn. Wiley, Chichester (2004) 18th International Conference Engineering Mechanics, pp. 1237–
4. Labra, C., Rojek, J., Oñate, E.: Adaptive discrete/finite element 1251, Svratka (2012)
coupling for rock cutting process simulations. In: Munjiza, A. (ed.) 11. Michael, M., Vogel, F., Peters, B.: XDEM–FEM coupling of the
Discrete Element Methods. Simulations of Discontinua: Theory intersections between a tire tread and granular terrain. Comput.
and Applications, pp. 428–433. Queen Mary University, London Methods Appl. Mech. Eng. Elsevier 289, 227–248 (2015)
(2010) 12. Müller, G., Groth, C.: FEM für Praktiker Band 1: Grundlagen, 8th
5. Rojek, J., Oñate, E.: Multiscale analysis using a coupled dis- edn. Expert Verlag, Renningen (2007)
crete/finite element model. Interact. Multiscale Mech. 1(1), 1–31 13. Klein, B.: Grundlagen und Anwendungen der Finite Elemente
(2007) Methode im Maschinen- und Fahrzeugbau, 7th edn. Vieweg Verlag,
6. Lu, X., Lin, X., Ye, L.: Simulation of structural collapse with cou- Wiesbaden (2007)
pled finite element-discrete element method. In: Yuan, Y., Cui, J., 14. Betten, Josef: Finite Elemente für Ingenieure 1&2. Springer, Hei-
Mang, H. A. (eds.) Computational Structural Engineering, pp. 127– delberg (1997)
135. Springer, Shanghai (2009) 15. Dratt, M., Katterfeld, A., Wheeler, C.A.: Prediction of belt deflec-
7. Villard, P., Chevalier, B., Le Hello, B., Combe, G.: Coupling tion by coupling of FEM and DEM simulations. In: Bulk Solids
between finite and discrete element methods for modeling of earth Handling, vol. 30, no. 7, pp. 380–384. Vogel Business Media,
structures reinforced by geosynthetic. Comput. Geotech. Elsevier Würzburg (2010)
36, 709–717 (2009) 16. Bathe, K.-J.: Finite-Elemente-Methoden, 2nd edn. Springer, Berlin
8. Wellmann, C., Wriggers, P.: A coupled 3D discrete-finite element (2009)
method for the simulation of granular materials and their interaction 17. Dratt, M.: Kopplung von FEM- und DEM-Simulationen zur
with solid structures. In: Proceedings in Applied Mathematics and Analyse der Gut-Bauteil-Interaktion in der Fördertechnik. Otto-
Mechanics (PAMM), vol. 8, pp. 10353–10354. Wiley, Weinheim von-Guericke-Universität Magdeburg, PhD Thesis (2016)
(2008) 18. Wensrich, C.M., Katterfeld, A.: Rolling friction as a technique for
9. Wellmann, C., Lillie, C., Wriggers, P.: A coupled discrete-finite ele- modelling particle shape in DEM. Powder Technol. 217, 409–417
ment method modeling the interactions of granular materials and (2012)
solid structures. In: 8th Word Congress on Computational Mechan-
ics (WCCM8), 5th European Congress on Computational Methods
in Applied Science and Engineering (ECCOMAS 2008), Venice
(2008)

123

You might also like