Discrete Element Methods in STAR-CCM+

Petr Kodl

CD-adapco

Engineering numerical methods used to simulate motion or large number of interacting discrete objects

Comparable to short range force MD simulations in methodology

Established by P.A. Cundall, O.D.L. Strack: A discrete numerical model for granular assemblies. Geotechnique, 29:47–65, 1979

Classical mechanical method

Mesh free

CPU intensive

– Transient

– Explicit schemes

Provides detail resolution other methods can not achieve

Used to describe wider class of methods but in terms of STAR-CCM+ we focus on granular flows

Bulk state results from particle interactions – no constitutive relation is used

Introduction to Discrete Element Methods (DEM)

Anisotropy

– Stress chains

– Large spatio-temporal fluctuations

Persistent contacts

Shear resistance

Jamming and arching

Reynolds’ dilatancy

Granular materials and their specific properties

Sand

Food particles

Metal particles

Capsules and pills

Slurries

Grains

Soil

Granular materials

When does it make sense?

– Highly loaded particulate flows

– Collisions are important

– Particle shape is important

– Details of collisions are important

– Typical granular flow properties are studied – jamming, shearing

What are the limits for practical problems?

– Fine grain particles (<1e-4)

– Achievable but the CPU time can be prohibitively expensive for industrial

problems

– The collision details are typically not critical outcome

– Very large particles (>1m) where the local deformation is important and the

contact law small deformation assumption is not valid

DEM applications

Implemented within Lagrangian framework

– Reuses known concepts

• Lagrangian phase

• Injectors

• Boundary interactions

• Sub stepping of the solution

Extends concept of Material particle

Additional tracking of

– Orientation

– Angular motion

– Inter-particle collisions

Soft particle model (penalty function based force evaluation)

Not statistical – 1 parcel = 1 particle

DEM in STAR-CCM+

5.06 - 28 Oct, 2010

– Initial DEM release

– Hertz Mindlin contact model

– Spherical and composite particles

– Moving walls via applied velocity

condition

– Stationary mesh and MRF

6.02 - 28 Feb, 2011

– Rigid mesh motion

– Phase specific boundary behavior

– Drag laws suitable for highly

loaded flows

• Ergun equation – Gidaspow

Timeline of DEM in STAR-CCM+

6.04 - 1 July 2011

– Walton-Braun linear hysteretic

contact model

– Parallel bonds

– Flexible / breakable particle

clumps

– Lattice injectors

– Charged particles

6.06 – October 2011

– Cohesive particles

– Improved particle tracking code

– User controlled time steps

– Additional drag coefficients

• Haider Levenspiel

– Two way coupling for charged

particles

Timeline of DEM in STAR-CCM+

7.02

– Randomized position injectors

– Porosity injection limits

– Improved particle-flow interaction

through fast estimate of projected

area and length

– Contact data sources, reports and

visualization

7.04

– Particle trapping walls

– Improved randomization of initial

particle distribution

– Performance optimizations both in

serial and parallel

Timeline of DEM in STAR-CCM+

– Comparison of contact force

models for the simulation of

collisions in DEM based granular

flow codes,

Alberto Di Renzo, Francesco Paolo

Di Maio, 2004, Chemical

Engineering Science

– Aluminum oxide spheres shot

against glass plate with varying

impact angle

– Apparent coefficient or tangential

restitution, rotation rate and

rebound angle compared to

laboratory experiment and

reference implementation

Validation –contact mechanics

– Discrete Particle Simulation of

Solid Flow in Model Blast Furface,

Zongyan Zhou, Haiping Zhu

ISIJ Vol 45, 2005

– Studies solid flow patter in blast

furnace

– STAR-CCM+ compared to

experiment and reference results

Validation – granular flow pattern formation

– STAR-CCM+ solution compared to

Ergun equation

– Tested case – porous bed with

periodic walls

– Analytic solution pressure drop ~

108Pa

Validation – pressure drop

DEM Solutions EDEM

– Mature industry focused code

– STAR-CCM+ will be compared to most frequently in terms of DEM

physic/features

– Founded 2002

– First release of the code in 2005

– First industrial grade release - 1.2 – May 2007

– Second generation solver and internal architecture code released as version

2.0 - 9 May 2008

– Current release EDEM 2.4 - September 16, 2011

Competitive analysis

STAR-CCM+

– Distributed memory (MPI)

• Domain decomposition

• Cluster friendly

– 2d, 3d

– Volumetric representation

• + Allows to solve coupled problems

• - Extra work required for meshing

– Rich, multi physics framework

EDEM

– Shared memory (OpenMP)

• Loop parallelism

• Single workstation

– 3d

– Surface representation

• + Almost no surface preparation

• - Makes coupling difficult

– Single purpose solver code

Competitive analysis – basic characteristics

STAR-CCM+ EDEM

Spherical particles x x

Rigid composites x x

Breakable flexible clumps x Custom coding

Hertz Mindlin x x

Hysteretic model x x

Parallel bonds x x

Cohesion x x

Linear spring Can use hysteretic model x

JKR Can use cohesion model x

Electrostatics 2 way coupled Limited

Particle/flow interaction 2 way coupled No longer supported

Competitive analysis

STAR-CCM+ EDEM

Heat transfer particle-particle, particle-flow,

particle-particle radiation

Particle-particle

Interfaces General Parallel planes

Particle shape editor

x x

Moving geometry Rigid body motion Rigid body motion Easy to setup – no meshing

required Transient post processing Track files Full solution replay

Competitive analysis

Conclusion

– Competitive in terms of implemented features

– Advantage for complex physics

• Reuse of feature implemented for general Lagrangian framework

• Ability to implement more complex physics due to the background FV discretization

– Further improvements

• Simplify the workflow for complex moving geometries

• Transient post processing and solution history

Competitive analysis

Not easy to quantify – depends on characteristics of particular case

– Packing structure

– Distribution of particles in the computational domain

– Amount of physics

– Coupling

– Overall case size

• Overhead of the STAR-CCM+ framework – mostly affecting small cases

• Large cases become memory bound when running on single machine mostly due to

irregular memory access patterns

Performance and scalability

CPU time vs number of particles

– Naively O(N^2)

– Ideally O(N)

• Good collision detector should linearize the

detection time

– Example

• CPU time / solver step vs # of particles

• # of particles up to 150000

• Densely packed

• Credit: Phillip Morris Jones, London Office

Performance and scalability

Solver time vs # of CP

– 3d Hopper

– 100 000 spherical particles

– Well distributed

– Credit: Lucia Sclafani

Performance and scalability

Physics

– Liquid bridges, capillary forces, free surface-particle interaction in VOF

– Mass transfer, drying, coating

– Smooth simulation physics decomposition DEM, FEA, EMP

– Surface only DEM

Performance and scalability

– Improved cache coherency for single workstation runs

– Dynamic particle centric load balancing

GUI and usability

– Transient post processing and solution snapshots

– CAD import and interpolation of particle shape by sphere trees

Future development

Examples

Thank you

Questions?