Sie sind auf Seite 1von 16

TECHNICAL ARTICLE

TECHNICAL ARTICLE

DISCRETE ELEMENT METHOD (DEM):


BASIC THEORY AND UTILITY

Saurabh Sarkar, Ph.D.


Applications Engineer, Rocky DEM

1. INTRODUCTION
Over the past 50 years, the Discrete Element Method (DEM) has emerged as a powerful first
principle simulation tool for a wide range of industries dealing with bulk particulate solids or granular
media. DEM’s power is derived from the underlying simplicity of concept: each particle in the granular
assembly is treated as an individual entity for which Newtonian equations of motions are solved. State-
of-the-art DEM tools like Rocky provide a unique platform for scientists and engineers to achieve
enhanced process understanding by mapping the dynamic interplay of process, material, and geometric
variables.

DEM data lends itself to exhaustive post-processing, empowering the user to gain insight not
readily obtained from experiments. Rates such as power consumption, material breakage, physical
boundary wear, temperature, mass transfer, and more can be calculated, measured, and evaluated using
DEM. These results help scientists and engineers design better, more efficient engineering equipment
and processes to meet their needs without the added time and expense of real-world experimental
designs. Some examples of industries using Rocky include:

• Mining & Metals: Predicting flow patterns and liner wear in transfer chutes and mills Metal
Industry: Charge flow in blast furnaces

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

• Pharmaceutical: Tablet coating and granulation scale up and mixing optimization

• Food: Seasoning and package filling of snack items

• Chemical Process: fluidized bed and mixers

• Heavy Equipment:

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

The purpose of the current article is to provide an introduction to the basic theoretical concepts
in DEM. This article is intended to help the reader evaluate whether DEM is fit for his/her needs, and
if it is, then (a) what are the criteria to choose a DEM software, and, (b) provide some best practices
and guidelines for optimally setting up a DEM case and executing it.

2. Basic Model Description


2.1 Granular Flows : DEM vs. Continuum Approaches

Granular media flow can be quite complex, as these flows are known to exhibit solid-like, fluid-
like, or a combination of both behaviors. For example, sand in an hourglass behaves like a fluid while a
stockpile of sand can have a solid-like stress-strain response. Traditionally, continuum-based methods
like Computational Fluid Dynamics (CFD) have been used to simulate granular flows, which were
achieved by ignoring the individual particles and treating the granular phase as a continuous phase.
In these approaches, the entire domain is meshed into discrete cells, and conservation laws for mass,
energy, momentum and constitutive stress-strain relationships are solved in each. However, these
constitutive relationships are often empirical in nature and hard to derive, and the application is
restricted to specific cases. Furthermore, discretization of the flow field places an additional constraint
on these approaches, especially when the particles are much larger in size relative to the characteristic
length scale of the flow.

In contrast, DEM is essentially a mesh-free method that solves Newton’s equations of motion
for each particle. The most important consideration for ensuring a high fidelity simulation is to account
for all the relevant forces acting on each particle. This in turn depends upon several factors including
accuracy of the inputs and the implementation of physics models, and the validity of assumptions
made to simply the simulation to generate meaningful results in a practical time frame.

2.2 Basic DEM Algorithm

In practice, the user passes the following inputs to start a simulation: (a) Geometry: Specific
equipment and particle geometries (shape, diameter (D)). (b) Process variables: Motion for moving
parts, feed rates etc. (c) Material properties for particles and geometry. These include properties like
density (ρ), Young’s modulus (E), thermal conductivity (k), and Poisson’s ratio (ϑ). The interaction of

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

particles with other particles and equipment walls are captured through interaction parameters like
friction and restitution coefficients, and adhesion parameters. The simulation time step ∆t is calculated
based on the material properties of the particles and boundaries. Particles are then initialized within
the domain and assigned initial grid coordinates. At this point, all contacts and detected. Then, all
the body and surfaces forces acting on the particle at a given time t are added, which are then used
to compute the particle acceleration. This is followed by numerical integration with time over ∆t to
obtain particle linear and angular velocities, and after another integration, we get new positions. Thus
after time t+∆t, new positions and velocities for all the particles are obtained. If the end of this time
step does not correspond with the end of the simulation, all contacts are detected in the updated bed
configuration and the same cycle is repeated. This continues till the end of the simulation. It is thus
possible to visualize and predict the temporal and spatial evolution of the entire granular system over
the course of the simulation.Figure 1 highlights the flow chart for a general DEM algorithm.

SETUP: User imports geometries, sets up particle groups, and determines particle-to-particle and particle-to-boundary interactions for
the simulation project.

PROCESS: User chooses to begin processing the simulations. For each individual particle, the DEM program does the following:
• Locates all neighboring particles and boundaries with which the particle will come into contact.
• Calculates the sum of all forces and moments (Euler equations) acting upon the particle: ‡

∑ Fnet = ∑ Fbody + ∑ F surface = m dv


dt

MOVE: The DEM program uses the current particle position, velocity, and timestep information to move the particle to its next lotion
in the simulation:
∑ Fnet dt
t+ ∆t

vnew = vold + ∫
t
m
t+ ∆t

xnew = xold + ∫
t vnew dt

REPEAT:
YES Simulation end time not reached and particles still in range?

NO

STOP

Figure 1: Flow chart for a general DEM algorithm

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

2.3 Contact Detection and Contact Dynamics

As mentioned above, at the heart of DEM implementation is accurate computation of the forces
acting on every particle in the simulation. This requires the following: (a) A robust contact detection
algorithm for accurate detection of particle-particle and particle-wall contacts.

(b) An accurate contact dynamics model to resolve oblique contact forces acting on the particles
into and tangential and normal directions.

Contact Detection: While contact detection can be trivial for spheres, it is rather complicated
for arbitrarily shaped particles. Contact detection and henceforth the ability to handle shaped particles
accurately is a critical differentiator between DEM codes. Figure 2 presents two versions of a tablet
shape, the first represented with glued spheres as provided by most DEM codes and the second
represented by a polyhedral shape of the type provided by Rocky. The advantages and disadvantages
to each kind of shape representation are listed in Table 1.

Glued spheres Polyhedral shape


Standard DEM Codes Rocky sharp edges

Figure 2: Tablet shape represented by glued spheres (left) and exact polyedra (right)

Table 1: Comparison between using inexact glued-sphere shapes vs. exact polyhedral shapes

Glued Spheres Polyhedra

Contact Detection Easy contact detection Difficult contact detection

Accuracy Low accuracy High accuracy


Large aspect ratio, sharp edges, Large aspect ratio, sharp corners and edges
Aspect Ratio
corners are a limitation are accurately represented
Cannot break unless bonded
Breakage Can break and have different models
spheres are used

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

Contact Dynamics: In soft sphere DEM models, the particles undergoing contact are assumed
to be deformable; the extent of deformation is dependent on then material properties of the particles
and the contact forces on the particles are a function of the particle deformation quantified through
mutual overlap (Figure 3). A contact dynamics model essentially provides a mathematical formulation
to approximate this behavior for two impacting solids. In most DEM codes, contact dynamics models
approximate the particles in contact to be masses at the end of a spring in both the normal and
tangential directions. The spring stiffness is a function of the material properties (E, D and ϑ) of the
contacting particles and boundaries.

Figure 3: Soft sphere DEM showing particle deformation and computation of overlap

Figure 4 provides an illustration of a general contact model used to compute normal forces
between two contacting particles. As particles contact, the spring with a spring stiffness (KL) is said
to “load” (Figure 4a). The particles continue moving towards each other resulting in deformation and
overlap δ. The spring load and the normal contact force Fη on the spring increases with increasing δ.
There are variations amongst contact models in computing Fη, but most use a relation of the type
Fη�K_L.δη At the point of maximum overlap, which ideally should be a very small fraction of the
particle diameter (<<5%), repulsive forces dominate and the particles start to recede, referred to as
“unloading” of the spring with a spring constant KU (Figure 4b). A residual overlap (δres) may remain
after the particle contact forces drop to 0, in which case further unloading would bring back the
particles to their initial starting configuration. Loss of energy during the collision can be approximated
by either using a viscous dashpot attached to the spring, or relating KL and KU to the coefficient of
restitution (e). Figure 4c shows a typical force-displacement plot for evolution of the normal forces
during loading and unloading cycle for the Hysteretic Linear Spring model, which one of the several
contact models implemented within Rocky. Different contact models are expected to have different
force-displacement curves.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

Just like normal forces, there are several contact models used to compute tangential forces.
Most of these models use the computed normal force and friction coefficients to capture the frictional
nature of granular flows. In addition, they may also include some dependence on the loading history.

Rocky features several well-referenced contact dynamics models. While a detailed discussion
of contact dynamics models is beyond the scope of the current article, readers are referred to some
excellent technical reviews and lectures for more information [1, 2].

(a) (b)

(c)
Figure 4: Contact dynamics model showing the (a) loading of spring as particle approach each other
and deform (b) unloading of the spring as the particle recede away (c) typical force-displacement curve for the
Hysteretic Linear Spring Model.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

Simulation Time Step: Based on the brief description of the contact models, another critical
simulation parameter must be addressed- the simulation time step ∆t which principally dictates how
fast a simulation would run. Consistent with the spring mass analogy, ∆t is related to the time period
of a simple harmonic oscillator, ∆t �√(m/K) where m is the mass attached to the spring and K is the
spring constant. K is dependent upon the material properties (D,E,ν ) of the contacting particles. The
smaller the time step, the longer the run time. From the proportionality described above, it can be
easily understood that small time steps can be obtained for particles with high stiffness (large E) or
small sizes.

2.4 Advanced Models

Often enough, different applications merit incorporation of different physics models to accurately
describe particle-level phenomena. These could be incorporation of forces additional to the contact
forces, or body forces (weight) fundamental to DEM, like fluid or cohesive forces. Alternately, stresses
on geometry due to particle flow may be desired. Users may also be interested in modeling heat
transfer, geometry wear, and breakage phenomena.

Rocky DEM provides the following advanced models, which are discussed briefly here:

1) Breakage Models: In Rocky, breakage of non-round particles that conserve both mass and
volume are accomplished using two approaches of the instant fragmentation model: the Ab-T10 model
[4] and the Professor Tavares model [5-7]. These models require further input of material properties like
energy required to break a particle of a reference size. Both of these are shape-independent methods
based on a Voronoi fracture particle subdivision algorithm and a probabilistic breakage method based
on the collision energy.

2) Multiphysics Modeling: Rocky is integrated within ANSYS® Workbench™ and couples


seamlessly with ANSYS® Fluent™ and ANSYS® Mechanical™ to run coupled DEM-CFD and DEM-
FEA simulations, respectively. In a two-way DEM-CFD coupled simulation, In the CFD-DEM method,
the fluid flow is obtained by the conventional continuum approach, providing information to calculate
the fluid forces acting on individual particles while the motion of the particle is obtained using DEM.
For FEA coupling, geometry loads from Rocky simulation are exported and passed to the FEA solver
to evaluate structural stresses on the geometry. Figure 5 shows snapshot from a DEM-FEA coupled
simulation to predict transient loads on a bucket excavator.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

Figure 5: DEM- FEA coupling to predict structural loads on a bucket excavator

3) Adhesion Models: Rocky provides multiple models to simulate cohesive flows, often
associated with wet materials. In Rocky, cohesion/adhesion of a bulk solid is a function of the stress, so
the linear force adhesion contact model captures this physical phenomenon by scaling the cohesion/
adhesion with the force of contact. Several different adhesive force models are featured in Rocky to
help replicate different kinds of adhesion situations. Figure 6 shows processing of cohesive particles
in an asphalt plant.

Figure 6: Particulate Behavior inside an Asphalt Plant Equipment: mixing section of the rotary dryer – Particle
with adhesive and cohesion forces included

4) Rolling Resistance Models: In DEM, rolling resistance refers to a moment that opposes the
rolling motion of a particle. This moment is usually incorporated in the model as a practical way to
represent the effect of non-sphericity and /or irregularities on the surface of particles which restricts

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

rolling Two rolling resistance models are available in Rocky: a constant model and an elastic-plastic
one, corresponding to models of type A and type C, respectively, in the classification proposed by Ai
et al. [3].

5) Energy Spectra: This tool collects energy for normal and tangential collisions gathered
between user-specified (experimental or literature-based) minimum and maximum energy (a) per
particle type per time step, and (b) per collision type, which accounts for all combinations of particle–
particle and particle–boundary collisions. These calculations are used to predict power draw, process
throughput, and milling efficiency. Please refer to this blog for further details.

6) Thermal Models: In Rocky, an additional equation for the thermal energy balance can be
solved along with the equations governing the motion of the particle. The temperature variation of a
particle can be obtained over time by taking into account the total particle heat transfer rate, which
itself accounts for the heat transfer that occurs during the contact with other particles or walls, and,
if fluid flow is considered, the convective heat transfer between particles and fluid. Figure 7 shows the
variation of temperature for a powder bed within a rotating drum.

Figure 7: Heat conduction within a rotating drum

7) Flexible Fibers: Rocky allows simulation of fibers, which implements the model proposed by
Guo et. al [9]. Flexible fibers are created by connecting aligned sphero-cylinders by means of virtual
bonds, which have elastic and viscous properties. Relative displacement between them may produce
linear and angular deformations of these virtual bonds leading to bending and flexion of the fiber.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

8) Geometry Wear: Rocky implements the validated Archard’s Wear model [8] to compute
equipment wear. The incremental loss in depth is computed based on a wear parameter and shear
work. As an example shown in Figure 8, the evolution of liner wear for a SAG mill with time can be
easily studied using Rocky.

Figure 8: Mill Showing Surface Wear Modification

9) Particle Shapes: Rocky allows realistic representation of multiple particle shapes which
includes volumetric 3D bodies, 2D shells and fibers which can be made rigid or flexible. As mentioned
earlier, this is possible owing to a robust contact detection algorithm and effective implementation
across the solver system.

In summary, it is important to bear in mind the following from a theoretical perspective:

a) DEM is a mesh-free method and solves equations of motion on each particle in the granular
assembly. Also, DEM time steps are typically of the order of a microsecond. These make DEM a
computationally intensive method.

Figure 9: Examples of some particle shapes that can be simulated using Rocky

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

b) Accurate contact detection is extremely important. For spheres, contact detection is trivial.
However, for shaped particles, computational complexity increases manifold due to complicated
contact detection.

c) The maximum overlap allowed by contact models is a very small fraction of the particle
diameter, which makes DEM suitable for small strain problems. For applications where large
deformations and shape changes are encountered, ex- compaction of powder bed to form a tablet, or
melting of particles, DEM would not be the method of choice.

3. Practical Usage : Balancing Accuracy and


Speed
As DEM is computationally intensive, users must evaluate the benefits of physical accuracy
against the increased computational cost on a case by case basis. Some ways to run the case optimally
are:

3.1 Making Assumptions & Material Calibration

Let’s consider fine powder mixing with billions of particles, as is done in the pharmaceutical
industry. Several important questions need to be asked:

1) Is it feasible to run at a 1-to-1 scale?

One can obtain key insights even by considering a larger particle size, which has been calibrated
against known experimental measurements like the angle of repose or shear cell data. Particles can
be approximated as non-rolling spheres to simulate behavior of micronized particles. Other material
properties, like Young’s modulus and friction and restitution coefficients which are not readily known,
may be adjusted to match the calibration data while yielding a relatively larger time step. More on
material calibration can be found here.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

2) Should exact particle shape be used?

Pharmaceutical powders have arbitrary shapes and sizes and do not roll. For this case, using
non- rolling spheres is a good assumption. However, if instead of fine powder, tablet are considered,
the moment of inertia has an important effect, and exact shape must be used. As pointed out earlier
in Section 2.3, the choice of using exact shape must be balanced with computation time and physical
accuracy in mind. Unlike most codes, Rocky provides users the option of using exact shapes.

3) There is air in the blender. Should the fluid forces be accounted?

In this case, fluid forces can be neglected, which is justified given that is a collision dominated
flow that the particle density is much larger than surrounding fluid. But if a fluidized bed is used for
mixing, fluid forces must be accounted for by coupling with a CFD code. Rocky DEM is fully integrated
with ANSYS Workbench tools and allows DEM (Rocky) –CFD (Fluent) coupling.

4) How long should the simulation run?

For this application, running the simulations to steady state and tracking mixing homogeneity
would give key insights. Behringer et al [10] show a good example how a tablet-coating operation can
be optimized using DEM simulations run to steady state and extrapolating mixing times.

5) Does the entire geometry need to be simulated?

In a batch blender, the entire geometry should be simulated. In case of continuous processes,
like simulation of large mills for operational efficiency and linear wear, a section of the geometry using
periodic boundary conditions is sufficient.

3.2 Hardware

Choice of hardware is extremely important for solving practical problems in an acceptable time
frame. Options available to users are using multiple CPUs, a single GPU, or multiple GPUs (available
only with Rocky). GPUs are increasingly being used to solve large calculations that were not possible
before. Rocky’s shared parallel memory technology and multi-GPU hardware capacities have proven
to be efficient in solving large systems Figure 10 shows the massive advantage of using Rocky’s
proprietary multi GPU technology in solving a large problem.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

It must be understood that the choice of optimum hardware is more complicated than simply
choosing GPUs, if they’re available. GPUs are well suited for computationally intensive jobs (e.g.,
complicated shapes) where the same task is performed over and over again (e.g., particle topology
stays the same) and the geometry fits into the shared memory. For more information on selecting
hardware, refer to this blog.

Figure 1 0: Improvement of using multi GPU over CPU

4. Weighing Your Investment & The Rocky


Difference
While the advantages and applications of DEM appear attractive, the ultimate decision on
procuring a DEM code must reflect a sound cost/benefit analysis that considers the following factors
[11]:

a. Cost and manpower: License costs, special hardware, trained personnel

b. Physics: Accuracy of force models, boundary conditions and particle shapes, code
customizability and innovation, coupling with CFD and FEM codes

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

c. Return on investment: Whether significant process insight be obtained, number of simulations


needed to get reliable results

Let’s reiterate how Rocky can provide benefits.

Compared with other commercially-available DEM codes, Rocky has many key advantages
(Table 2) that make it the particle simulation tool of choice across an increasing number of leading
companies—like PepsiCo, BMS, thyssenkrupp — across many materials handling and processing
industries.

Range of very reasonably priced licenses available; company is willing to work


License
with clients

Internal DEM expertise is not necessary; Rocky provides dedicated customer


Manpower
support and customized training
Cost and Rocky’s proprietary multi-GPU technology solves large problems at a fraction of
Manpower Hardware
the time and cost than previously possible

Rocky’s strong post processing features and Python based API makes data
Additional processing very easy and additional packages are not usually needed.
Software Furthermore, coupling with ANSYS CFD and FEA packages, and simulating
multibody motion can be achieved without the need of an additional software

Accuracy Rigorous release testing and validation of model data against experiment

One of the few commercial DEM codes to handle exact particle shapes, further
Shapes
increasing model accuracy

Willing to work closely with clients; a number of features and models in Rocky
Customizability
Physics are the result of direct client requests

Multiphysics As an ANSYS Channel Partner, Rocky provides excellent coupling with CFD and
Coupling FEM packages.

Rocky leads in DEM innovation by offering unique features like multi-GPU


Innovation
technology, complex shapes, flexible fibers, shell particles, and more

Return on
- Strong track record of saving clients both time and money
Investment

Table 2: Advantages of using Rocky for DEM calculations

Is Rocky DEM the right tool for the important work your company does? Contact your local
Rocky distributor to learn more.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com


TECHNICAL ARTICLE

References
[1] H. P. Zhu , Z. Y. Zhou, R. Yang, A. Yu, “Discrete particle simulation of particulate systems:
theoretical developments,” Chem. Eng. Sci. vol. 62, no. 13, pp. 3378-96, Jul. 2007.

[2] C. Wassgren and A. Sarkar, “Discrete Element Method (DEM) Course Module,” 2008.

[3] J. Ai, J. F. Chen, J. M. Rotter, and J. Y. Ooi, “Assessment of rolling resistance models in discrete
element simulations,” Powder Technol., vol. 206, no. 3, pp. 269-282, Jan. 2011.

[4] F. Shi and T. Kojovic, “Validation of a model for impact breakage incorporating particle size
effect,” Intl. J. Mineral Processing, vol. 82, no. 3, pp. 156–163, Apr. 2007.

[5] L. M. Tavares and R. King, “Single-particle fracture under impact loading,” Intl. J. Mineral
Processing, vol. 54, no. 1, pp. 1–28, June 1998.

[6] L. M. Tavares, “Analysis of particle fracture by repeated stressing as damage accumulation,”


Powder Technology, vol. 90, no. 3, pp. 327–339, Mar. 2009.

[7] R. M. Carvalho and L. M. Tavares, “Predicting the effect of operating and design variables on
breakage rates using the mechanistic ball mill model,” Minerals and Eng., vol. 43, pp. 91–101, Apr. 2013.

[8] X. Qiu, A. Popatov, M. Song, and L. Nordell, “Prediction of wear on mill lifters using discrete
element method,” in Proc. Intl. Autogenous and Semi-Autogenous Grinding Technology, University
of British Columbia, Vancouver, 2001, 4, 260–265.

[9] Y. Guo, C. W. Wassgren, J. S. Curtis, and D. Xu, “A bonded sphero-cylinder model for the
discrete element simulation of elastic-plastic fibers,” Chem. Eng. Sci., vol. 175, pp.118–129, Jan. 2018.

[10] P. Boehling et al., “Analysis of large-scale tablet coating: modeling, simulation and
experiments,” Eur. Journ. Pharmaceutical Sci., vol. 90, pp. 14-24, Jul. 2016.

[11] R. Bharadwaj, “Using DEM to solve bulk material handling problems,” Chemical Engineering
Progress (CEP) magazine, vol. 109, no. 9, pp. 54-58, Sept. 2012.

Rocky DEM | Next Generation DEM Particle Simulator Software www.rocky-dem.com

Das könnte Ihnen auch gefallen