	SSP 1995 project summary:

[EPCC home] [SSP home] [2001 projects] [2000 projects] [1999 projects] [1998 projects] [1997 projects] [1996 projects] [1995 projects] [1994 projects] [1993 projects]

Discrete Particle and Continuum Modelling of Particulate Solids

This research is to serve as part of a three year project supported by the EPSRC. The aim of the project is to compare discrete particle and continuum computational models for the behaviour of particulate solids in silos. To ensure that the findings have wide acceptance, it is being undertaken as a major international collaboration effort, with many discrete element and continuum (finite element) modellers around the world attempting a series of standard problems. Thus, many different formulations of each model type will be used, chiefly because there is little agreement between research groups on which features are necessary to achieve a reasonable modelling.

Practical application of discrete element method (DEM) has been hampered with traditional sequential processing computers. Realistic discrete element simulations for particulate flows in silos require considerable amount of computer time. This is particularly true in three dimensional discrete element analysis that can use up enormous amounts of computer time. Many researchers are devoting to achieving a more efficient discrete element computation. Significant speed-up is possible by exploiting the inherent parallelism in the time integration of the discrete element models and the use of parallel computers.

Existing DEM programs

Two sets of DEM code written in F77 are available in the Silos Research Group, one based on grid search contact detection algorithm and the other on linked-list algorithm. Both have been developed for sequential processing computers.

Proposal

The most time consuming part of the operation, contact detection, can be parallelised. This can be achieved by using the inherent parallelism in discrete element analysis. When the computational domain of interest is subdivided into a number of subregions, these subregions can be processed in parallel. In this form of parallel implementation, each subregion is assigned to one processor. All the operations within the subregion, including the contact detection, are performed in parallel. The data needed for computations in each processor are either available in that processor or in the processors representing the surrounding subregions. This type of nearest neighbour data transfer makes the best possible use of inter-processor communication.

The practical objective of the scholarship is to port one of these DEM modelling programs from Fortran 77 to a suitable 'parallelised' version (e.g. HPF) (3-5 weeks). From here it would be used to calculate results for the various standard examples being used as part of the international project.(3-5 weeks). A final (optional) step would be the porting of the second program to a similar 'parallelised' version and generating results for the standard models using this second system.

Foreseeable significance and applications

As part of a major on-going research project, co-operating with many leading researchers throughout the world, the proposed project for parallelism of the existing DEM codes will be very meaningful for enhancing the capability of existing techniques. There will be a very good chance to expose the advent of the effective use of the supercomputer for civil engineering problems.

The modified discrete element simulation code will allow a major step forward to the more realistic simulations of particulate solids in silos. It will also allow this university to be at the leading edge of the research field and make significant contribution to the particulate flows and silo structure research.

Thomas Schroder worked on this project.

Compressed PostScript of the project's final report is available here (138633 bytes) .

Webpage maintained by mario@epcc.ed.ac.uk