This site is accessible to all versions of every browser. However, this browser may not support basic Web standards, preventing the proper display of this site. If you experience any display problems, please upgrade your browser to a newer, standard compliant, version.
EPFL
ROSO
EPFL Dr Jean-Albert Ferrez
english only

I have left EPFL !

My research activities are still on these pages, but the rest, in particular Linux ressources, are now on my new site hosted at IDIAP.



Full version, AVI, 24294K

Dynamic Triangulations for Efficient
3D Simulation of Granular Materials

This research project involves using dynamic weighted Delaunay triangulations to support efficient large scale distinct element simulations of granular materials in 3D.

The Distinct Element Method (DEM) is used to perform computer simulations of highly discontinuous phenomena. DEM considers every particle in the system independently: each follows its own trajectory and interacts with the environment and other particles. One of the problems with DEM is the high computational cost associated with the localization of the interactions between the particles.

If the particles are discs (2D) or spheres (3D), an efficient collision detection method based on dynamic triangulations can be used, that reduces the number of potential collisions to O(n).

For more details, see "further reading", below.

Examples

VRML files are a good way to see the result of small 3D simulations. You can get a plug-in (Cortona for Windows or Cosmo, for SGI, Windows and Mac) to display VRML in your web browser or a stand-alone viewer (VRML for Linux).

Larger cases need more elaborate - but less interactive - tools. Most movies are in QuickTime or in AVI format for now. Both format should display fine on any platform. When both formats are available, the QuickTime version is usually better in quality but much larger in size, this is due to the tools I used to create them, but mostly to my limited knowledge about digital movies.

As of October 2001, there is about 205MB of pictures, movies and VRML files on this page!
Bold lines or links are those that I consider more interesting and are good for a first overview.

Various animated VRML files of trivial simulations:

  • 2 balls rolling on each-other (VRML, 13K - JPEG, 500x500, 31K).
  • 10 balls falling down in an hourglass. The color of each ball changes with its velocity. (VRML, 100K - JPEG, 500x500, 29K).
  • Packing of ~100 spheres in a cube (VRML, 485K - JPEG, 500x500, 40K).
  • Packing of 1 big, 6 medium and 173 small balls (VRML, 687K - JPEG, 500x500, 55K).
  • About 270 balls falling down around a cylinder (VRML, 360K - JPEG, 500x500, 40K). Check out the bottom view of the VRML.
  • A red cylindrical captor among grains (shown with some transparency) (VRML, 1038K - JPEG, 500x500, 57K).
  • Some large, medium and small grains floating around. The size ratio are 100:10:1 and the small grains are indeed hardly visible unless your window is at least 1000x1000 (VRML, 3235K - JPEG, 500x500, 65K).
  • Another packing, enforced by tapping (VRML, 2038K - AVI, 321K - JPEG, 500x500, 31K).

Some files showing the triangulation used to detect contacts:

  • A VRML file of a static triangulation (VRML, 34K - JPEG, 761x900, 147K), and a video showing it from various angles (QuickTime, 6991K - AVI, 2469K).
  • The flip in 3D used to maintain the triangulation. Flipping is swapping between the two possible tetrahedral decompositions of a non-degenerate 5 points set (VRML, 100K).
  • 8 balls in a cube showing the dynamic triangulation used for the collision detection (VRML, 1111K).
  • Another similar example (VRML, 1252K).
  • 7 balls falling down, also showing the dynamic triangulation used for the collision detection (VRML, 405K).
  • This movie shows a larger set, but the quality is rather poor (QuickTime, 3631K).
  • I borrowed from Didier an example of the similar 2D case: movie (QuickTime, 3728K - AVI, 1395K), initial frame (JPEG, 401x401, 75K) and final frame (JPEG, 401x401, 73K). The rest of Didier's simulations are still online here.

Some simulations deal with packings of spheres of various sizes, in particular improving their density by tapping. Here are videos showing this on an example with 383 grains of 3 different sizes:

  • Front view of a 2 seconds case (QuickTime, 4621K - AVI, 380K).
  • Same case, right view (QuickTime, 4762K - AVI, 380K).
  • Same case, back view (QuickTime, 4844K - AVI, 380K).
  • Same case, left view (QuickTime, 4947K - AVI, 381K).
  • Similar case but lasting, 5 seconds, tour view at full speed (QuickTime, 2987K - AVI, 2089K) or in slow motion (QuickTime, 12545K - AVI, 4848K).

These simulation show the influence of the tapping frequency on the packing process. Final views of 5575 similar grains after 10 seconds of vibrations followed by 2 seconds without vibrations.

  • At 2Hz, the initial layers are still distinct: single view (JPEG, 400x400, 14K) or tour view (QuickTime, 3132K - AVI, 1871K).
  • At 10Hz the initial layers are gone: single view (JPEG, 400x400, 16K) or tour view (QuickTime, 3327K - AVI, 1897K).

Pouring grains in a cylinder:

  • Falling down from the top (VRML, 1297K).
  • Flowing out from the side, short version at actual speed (VRML, 3180K), or in slow motion (VRML, 3180K)..
  • Same flow, but the grains are colored with respect to their velocity, actual speed (VRML, 4066K), or in slow motion (VRML, 4066K).
  • Longer version: low quality (MPEG, 1043K) or high quality (QuickTime, 2450K).

Hourglass flow:

  • The grains are colored with respect to their velocity: low quality (MPEG, 1137K) or high quality (QuickTime, 16555K).
  • The grains are colored with respect to the initial layers: low quality (MPEG, 1151K), high quality small size (QuickTime, 7872K), or high quality full size (QuickTime, 24960K).
  • The full version of the raytracing-enhanced hourglass shown on the top of the page (AVI, 24294K).

When some edges of the triangulation are mandatory, it becomes possible to construct clusters of grains of arbitrary shape. Here are some basic examples:

  • 5 tetrahedra each composed of 4 spheres (VRML, 164K - JPEG, 500x500, 46K).
  • 5 rods made of 28 spheres (VRML, 1007K - JPEG, 500x500, 43K).
  • Rods and tetrahedra (VRML, 2453K - JPEG, 500x500, 65K).
  • A very rough approximation of a flat disc built with 128 spheres (VRML, 761K - JPEG, 500x500, 31K).
  • A larger view with various shapes (JPEG, 500x500, 58K).

Didier could show the forces among the grains in his 2D simulations. I tried to do the same in 3D, but it is harder to actually see something... The forces are represented as solid rods, blue for internal grain-grain forces and red for external grain-wall forces. The diameter of those rods grows with the magnitude of the force. The grains themselves are almost transparent.

  • Static VRML file of 209 grains in a cylinder (VRML, 590K).
  • Front (JPEG, 750x751, 404K), side (JPEG, 750x751, 406K), top (JPEG, 750x751, 382K) and bottom (JPEG, 750x751, 365K) views of the same set.

Again in the goal of extending Didier's 2D simulations, here is a first attempt at reproducing the impact of a rock falling down on a granular bed.

  • Vertical impact: (AVI, 182K).
  • Oblique impact: (AVI, 174K).
  • The individual images for both impacts can be seen on a separate page.
  • Obtaining the granular bed: (AVI, 195K).
  • Large view of the asymmetric crater of the oblique impact (JPEG, 480x480, 174K).

Here is a detailed view of the rearrangement of medium and small spheres that takes place when they leave space for a larger one. Bottom view of an experiment involving 1762 spheres:

  • Movie (AVI, 3085K - QuickTime, 7627K).
  • Snapshots: initial situation (JPEG, 500x500, 249K) and final situation (JPEG, 500x500, 243K).

Further reading


©2002 ROSO-EPFL, 1015 Lausanne, Jean-Albert Ferrez
update: 27 October 2002