Grains3D, a flexible DEM approach for particles of arbitrary convex shape - Part II: Parallel implementation and scalable performance

Rakotonirina, Andriarimina Daniel; Wachs, Anthony

doi:10.1016/j.powtec.2017.10.033

Cited by 39 publications

(30 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Reasonable particle counts which yield insight on the flow behaviour range from 100, 800, over 3,125 to 6,000 . Experiments with spherical or analytical particle descriptions, which are not the focus of the present paper, in contrast work with 100,000 particles upwards and can even 2·10 9 particles in total with more than 15,000 particles per rank/node . Our experimental setup thus covers the regime of the state‐of‐the‐art.…”

Section: Resultsmentioning

confidence: 99%

“…Given the predominance of spatial decomposition schemes, the realisation of MPI parallelisation (logically distributed memory) of DEM is well‐understood . Though many roadmaps predict that the gain of performance in future supercomputers will stem from an increase of (shared memory) cores, literature on shared memory parallelisation in the DEM context however is rare.…”

Section: Related Algorithmic Conceptsmentioning

confidence: 99%

See 1 more Smart Citation

A multi‐core ready discrete element method with triangles using dynamically adaptive multiscale grids

Krestenitis

Weinzierl

2018

Concurrency and Computation

View full text Add to dashboard Cite

Summary The simulation of vast numbers of rigid bodies of non‐analytical shapes and of tremendously different sizes that collide with each other is computationally challenging. A bottleneck is the identification of all particle contact points per time step. We propose a tree‐based multilevel meta data structure to administer the particles. The data structure plus a purpose‐made tree traversal identifying the contact points introduce concurrency to the particle comparisons, whilst they keep the absolute number of particle‐to‐particle comparisons low. Furthermore, a novel adaptivity criterion allows explicit time stepping to work with comparably large time steps. It optimises both toward low algorithmic complexity per time step and low numbers of time steps. We study three different parallelisation strategies exploiting our traversal's concurrency. The fusion of two of them yields promising speedups once we rely on maximally asynchronous task‐based realisations. Our work shows that new computer architecture can push the boundary of rigid particle computability, yet if and only if the right data structures and data processing schemes are chosen.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Related Algorithmic Conceptsmentioning

confidence: 99%

A multi‐core ready discrete element method with triangles using dynamically adaptive multiscale grids

Krestenitis

Weinzierl

2018

Concurrency and Computation

View full text Add to dashboard Cite

show abstract

“…The computing cost is high, but we claim that this is the price to pay to accurately represent a non-convex shape. Interestingly, and as pointed in Grains-Part II [33], since contact detection is a serial operation and its cost for non-convex bodies is markedly larger than spheres or convex bodies, parallel computations of granular systems made of non-convex particles scales extremely well (with a weak scaling factor close to 1). In fact, the MPI communication overhead is literally negligible compared to the contact detection phase per sub-domain, i.e., per process.…”

Section: Gjk-based Contact Detection For Non-convex Particlesmentioning

confidence: 93%

“…Our main objective is to elaborate on the construction of a simple glued/clumped convex method to model non-convex shapes. Our method is based on existing and validated tools already introduced in the two first chapters of this trilogy of papers on Grains3D [39,33]. Our method shares some features with the method used in the Bullet Physics library [5], developed for video games, virtual reality and movies, to handle contacts between convex and non-convex bodies.…”

Section: Introductionmentioning

confidence: 99%

Grains3D, a flexible DEM approach for particles of arbitrary convex shape—Part III: extension to non-convex particles modelled as glued convex particles

et al. 2018

Self Cite

View full text Add to dashboard Cite

Large-scale numerical simulation using the Discrete Element Method (DEM) contributes to improve our understanding of granular flow dynamics involved in many industrial processes and geophysical flows. In industry, it leads to an enhanced design and an overall optimisation of the corresponding equipment and process. Most of DEM simulations in the literature have been performed using spherical particles. A limited number of studies dealt with non-spherical particles, even less with non-convex particles. Even convex bodies do not always represent the real shape of certain particles. In fact, more complex shaped particles are found in many industrial applications as, e.g., catalytic pellets in chemical reactors or crushed glass debris in recycling processes. In Grains3D-Part I [39], we addressed the problem of convex shape in granular simulations while in Grains3D-Part II [33], we suggested a simple through efficient parallel strategy to compute systems with up to a few hundreds of millions of particles. The aim of the present study is to extend even further the modeling capabilities of Grains3D towards non-convex shapes, as a tool to examine the flow dynamics of granular media made of non-convex particles. Our strategy is based on decomposing a non-convex shaped particle into a set of convex bodies, called elementary components. We call our method glued or clumped convex method, as an extension of the popular glued spheres method. Essentially, a non-convex particle is constructed as a cluster of convex particles, called elementary components. At the level of these elementary components of a glued convex particle, we employ the same contact detection strategy based on a Gilbert-Johnson-Keerthi algorithm and a linked-cell spatial sorting that accelerates the resolution of the contact, that we introduced in [39]. Our glued convex model is implemented as a new module of our code Grains3D and is therefore automatically fully parallel. We illustrate the new modeling capabilities of Grains3D in two test cases: (i) the filling of a container and (ii) the flow dynamics in a rotating drum.

show abstract

“…Wachs et al used the Gilbert‐Johnson‐Keerthi distance algorithm to handle arbitrary convex body assemblies. Recently, the parallel scalability of this method has been investigated with simulations containing up to 1 × 10 8 particles on up to 768 cores . Polyhedral representations of particles using surface meshes can be used to model particles in a fashion similar to complex geometries.…”

Section: Uncertainty and Limitationsmentioning

confidence: 99%

Experimental Methods in Chemical Engineering: Discrete Element Method—DEM

et al. 2019

View full text Add to dashboard Cite

The discrete element method (DEM) is a time-driven simulation technique based on a Lagrangian description of particle motion that predicts the flow of granular matter and fine powders in conveying, mixing, drying, and heterogeneous gas-(liquid)-solids reactors. Powders flowing out of bins form bridges, they segregate in suboptimal pharmaceutical v-blenders, and a stream may split into large gulf streams as they enter fluidized bed reactors from standpipes and diplegs. To reduce the uncertainty in scaling up these and other powder process unit operations, researchers apply DEM. It integrates Newton's second law (acceleration equals the sum of the forces) for each particle and models contacts between the particles with springs and dashpots (dampers). It is computationally intensive since it calculates the trajectory of all particles. The availability of open source codes, commercial software, and parallel computer architectures has accelerated its adoption in pharmaceutical, agro-industrial, and mineral processes, and geophysics. The accuracy of DEM models depends on how well researchers calibrate the contact model expressions and their parameters: friction coefficients and the coefficient of restitution. Systems exceeding 1 × 10 8 particles can require weeks of computational time on large computer clusters. Current research targets non-spherical particle interactions and multiphysics problems including heat transfer, mass transfer, and chemical reactions within the particles. The field has grown to 750 indexed articles in WoS in 2017. A bibliographic analysis recognized four research clusters: granular materials, behaviour, particle shape, and deformation; flows, fluidized beds, and computational fluid dynamics; particles, impact, and validation; and granular flow, dynamics, and segregation.

show abstract

Grains3D, a flexible DEM approach for particles of arbitrary convex shape - Part II: Parallel implementation and scalable performance

Cited by 39 publications

References 46 publications

A multi‐core ready discrete element method with triangles using dynamically adaptive multiscale grids

A multi‐core ready discrete element method with triangles using dynamically adaptive multiscale grids

Grains3D, a flexible DEM approach for particles of arbitrary convex shape—Part III: extension to non-convex particles modelled as glued convex particles

Experimental Methods in Chemical Engineering: Discrete Element Method—DEM

Contact Info

Product

Resources

About