You are on page 1of 3

9/4/23, 4:56 PM QA00000026975e - DS Support Knowledge Base / Question & Answer

Knowledge Base
Information

MY FAVORITE CONTENT
Best practices for an Abaqus/Explicit Discrete Element QA Article: QA00000
Applicable From: 6.13 Gold
Method (DEM) analysis (6.13-1)
Last Update Date:
Portfolio / Domain: SIMULIA Abaqus Unified FEA / SIMULIA Abaqus Unified FEA Rating:
Product: SIMULIA Abaqus Analysis Tokens Views: 1061 view

QUESTION What are the best practices for a Discrete Element Method (DEM) analysis?

ANSWER (The following applies to Abaqus 6.13 and higher.)


The Discrete Element Method (DEM) in Abaqus/Explicit is intended to simulate granular media interactions. Each
particle is modeled using a single PD3D element. The PD3D element has a single node and behaves like a rigid sphere.
The discrete element method cannot be used in conjunction with the smoothed particle hydrodynamics (SPH) or
coupled Eulerian-Lagrangian (CEL) methods.

Checking the total energy (ETOTAL) and other energy quantities like kinetic energy (ALLKE) will help to determine if
the specified stable time increment is acceptable. ETOTAL must be constant throughout the analysis whereas for a
system that has no frictional dissipation, ALLKE must also be conserved.

Some of the recommended practices in using DEM are:


1. Automatic generation of particles
Beginning with Abaqus 2016, particles can be automatically generated. For more details, please see
QA00000041123.
a. When a particle generator is used, the analysis can be run only on a single domain. See Item 7 below, Improving
the performance of the analysis.
b. If we have only the total mass of the granular material to begin with, then the maximum number of DEM
particles to model this granular material can be approximated as the total number of particles of uniform size
(this size is the mean of the probability distribution function of the sizes of the particles) whose total mass is
equivalent to that of the granular material.
2. Pressure-overclosure relationship
Particle deformation is approximated via a contact definition which allows penetration of the rigid PD3D elements.
The bulk response of a granular media is therefore a consequence of the inter-particle contact property. Hence a
hard contact definition (*SURFACE BEHAVIOR, PRESSURE-OVERCLOSURE=HARD), which is default behavior for
pressure-overclosure definition, is not meaningful for DEM applications.
In Abaqus, the discrete particles are modeled as spheres with unit surface area. Therefore, the pressure-overclosure
relationship defined by the user is actually equivalent to a force-overclosure relationship. One may choose any type
of pressure-overclosure relationship: linear, tabular or exponential. A tabular force-overclosure relation may be used
to specify Hertz-type contact between particles. Use a maximum penetration between 0.1% to 0.05% of the radius
of largest discrete particle to construct the tabular force-overclosure data based on the Hertz formula.
Beginning with Abaqus 2016, use *SURFACE BEHAVIOR, PRESSURE-OVERCLOSURE = HERTZ to let Abaqus
compute the pressure-overclosure relationship directly using the material properties of the particles. You must
define the limiting slope of the pressure-overclosure curve: a typical limiting value would be a slope equivalent of
0.1% to 0.05% penetration (implying that no particle sees a penetration more than this chosen limit in the
analysis).
3. Stable time increment
A small time increment is essential for stability of explicit integration as well as accurate contact tracking. If a
model has only PD3D elements and rigid bodies, then you have to provide a time increment for the analysis. If
deformable elements are present along with PD3D elements, Abaqus/Explicit will be able to compute a default
stable time increment for the analysis. However, if the velocity of the discrete particles is very high this time
increment may be too large for accurate tracking, and, scaling down the time increment may be required for such
problems. In cases where you need to specify the time increment for the analysis, a general guideline is to use a
time increment (or a fraction of it) that is the minimum of:
a. time required to avoid numerical instability: 0.1*SQRT(m/k) to 0.4*SQRT(m/k), where m and k represent the
particle mass and contact stiffness, respectively.
b. time required for the particle with highest initial velocity to cover 0.01% radius of smallest particle.
4. Stable time estimator: method
When the model has both discrete particle elements and deformable elements and if variable automatic time
incrementation (default in Abaqus/Explicit) is used, it is recommended that you use element-by-element stable
time increment estimation (*DYNAMIC, EXPLICIT, ELEMENT BY ELEMENT).
5. Initial overclosures
Large initial overclosures between the discrete particles are stored as offsets without generating any contact force.
Therefore, in clusters (a group of rigidly connected overlapping discrete elements) the initial overclosures will not
lead to large contact forces.
6. Use of CONTROLS parameter
(Undocumented in 6.14 and earlier) Contact is tracked within a box-like region whose limits are determined by the
initial geometry of the model. Typically, a model would have a regular finite element mesh with confined DEM

https://kb.dsxclient.3ds.com/mashup-ui/page/resultqa?from=search%3fq%3ddem&id=QA00000026975e&q=dem 1/3
9/4/23, 4:56 PM QA00000026975e - DS Support Knowledge Base / Question & Answer
particles; the default box computed by Abaqus within which contact is tracked is usually sufficient. However, if the
model consists of only DEM particles, then the box limits may not enclose the region in which contact takes place.
You may increase the size of the bounding box as follows:
*DISCRETE SECTION, CONTROLS = controls name
*SECTION CONTROLS, NAME = controls name
data line 1
data line 2
X1,Y1,Z1, X2,Y2,Z2

MY FAVORITE CONTENT
where, X1, Y1, Z1 and X2, Y2, Z2 are the coordinates of the endpoints of the longest diagonal of the box.
7. Improving the performance of the analysis
(Applicable to Abaqus 6.14 release and later): DEM particles can now be distributed across multiple domains.
However, if clusters of DEM particles exist in the analysis, then all the DEM particles are pulled into the same
domain affecting performance significantly.
(Applicable to Abaqus 2016 release and later): If a particle generator is used in the model, then the analysis can be
run only in a single domain. The performance, however, can be improved by splitting the generation phase and the
usage phase into two separate analyses such that the generation phase is run on a single domain while the usage
phase is run on multiple domains.
a. Generation phase: Run the particle-generation step to generate all the particles (please see QA00000041123 for
more details) on a single domain. Use the utility plugin that will create an input file with current nodal
coordinates and element connectivity using only the active nodes and elements currently shown in the viewport
of the Visualization module of Abaqus/CAE (QA00000023036). Open the .odb of the generation-phase analysis,
and render only DEM particles to be active in the viewport. Using the plugin, create an input file that has *NODE
(current coordinates) and *ELEMENT (mesh connectivity) data of the DEM particles.
b. Usage phase: Include the input file that was created in the above-mentioned procedure in the input file of the
usage phase. This usage phase can now be run on multiple domains for faster performance.
(Applicable to Abaqus 2016 release and later): If *SURFACE BEHAVIOR, PRESSURE-OVERCLOSURE = HERTZ is used
and the model contains DEM particles of different materials, then the analysis can be run only on a single domain.
To circumvent this issue, use a tabular pressure-overclosure relationship (*SURFACE BEHAVIOR, PRESSURE-
OVERCLOSURE = TABULAR) instead.
If the particles are of the same material, but different sizes then the sizes of the particles can be defined through a
distribution table. A single *DISCRETE ELASTICITY definition must be used to define the material properties of all
DEM particles along with *SURFACE BEHAVIOR, PRESSURE-OVERCLOSURE = HERTZ. In this case the analysis can
be run on multiple domains.
At least 10,000 DEM particles per domain is suggested to achieve good scalability when multiple domains are used.
Otherwise, performance will be downgraded significantly because the DEM particles will be decomposed to each
subdomain and the communication costs between the subdomains will increase dramatically.
8. Using clusters of DEM particles to model a particle of an arbitrary shape
Currently, only spherical shapes are allowed for DEM particles. If an arbitrarily shaped particle is needed, it can be
modeled as an assembly of smaller spherical DEM particles (called a cluster). Please see QA00000041124 for more
details. Note that there will always be some regions within a cluster that have mass contributions from two or
more DEM spherical particles, potentially making the cluster heavier than the particle it is modeling. So it is
possible that the total mass of the DEM particles may exceed the total mass of the granular material if the number
of particles was calculated based on the mass of the arbitrarily-shaped particle and the total mass of granular
material. Also clusters of DEM particles limit a DEM analysis to run only on a single domain.
9. Coefficient of restitution
If we know the coefficient of restitution of the impact between the particles, a suitable contact damping
(*CONTACT DAMPING, DEFINITION=DAMPING COEFFICIENT) can be defined to model this dissipation of energy
such that the required coefficient of restitution is achieved. For more details, please see QA00000040673.
10. Drag and buoyancy forces
(Applicable to Abaqus 2016 release and later ): User-subroutine VDLOAD can be used to define drag and buoyancy
forces on DEM particles. For more details, please see QA00000041127. Please note that the underlying
assumptions in applying the drag force are (a) particles in the fluid are too few in number to have any influence on
each other, and (b) the fluid has no turbulence.
11. Visualizing Particles
When there are particles of different radii in the model, field output variable PRAD should be requested. This field
output will be useful to visualize the particles, but if the radius difference is not so big then the PRAD field output
may be not well visualized. The PRAD field output variable is not supported in Abaqus/CAE and it has to be
manually added in the input deck. At the moment, PRAD is only designed to render DEM particles.

KEYWORDS DEM, particles

SUBSCRIBE TO
CHANGES

RATING On a scale of 1-5, how would you rate the technical content of the article?

Please rate this article...

LET US KNOW
IF THIS ARTICLE
NEEDS TO BE
ENHANCED

UNCLEAR MISSING INFO DUPLICATE OUT OF DATE ERROR DETECTED

https://kb.dsxclient.3ds.com/mashup-ui/page/resultqa?from=search%3fq%3ddem&id=QA00000026975e&q=dem 2/3
9/4/23, 4:56 PM QA00000026975e - DS Support Knowledge Base / Question & Answer

See Comments (1)

MY FAVORITE CONTENT

https://kb.dsxclient.3ds.com/mashup-ui/page/resultqa?from=search%3fq%3ddem&id=QA00000026975e&q=dem 3/3

You might also like