MFIX-DEM Enhancement for Industry-Relevant Flows Email Page
email
Print This Page
print
Performer: University of Colorado - Boulder
Horizontal jet experiments particle tracking post-processing.
Horizontal jet experiments particle tracking post-processing.
Website: University of Colorado
Award Number: FE0026298
Project Duration: 09/01/2015 – 08/31/2020
Total Award Value: $3,778,002
DOE Share: $2,857,479
Performer Share: $920,523
Technology Area: Coal Utilization Science
Key Technology:
Location: Boulder, CO

Project Description

This project will improve performance of the MFIX-DEM code to enable a transformative shift for industrial use. The current simulations fall short of the O(108) particle simulations that must be completed on a timescale of days to enable simulations with physically-relevant domain sizes to be incorporated into industrial design cycles within five years. The team will accomplish this by tailoring best-in-class practices to bear on the challenges posed by the MFIX-DEM algorithm and code base. The MFIX-DEM code will be refactored to minimize data movement and synchronization between the Eulerian and Lagrangian updates. This will facilitate optimization of the particle update to expose multiple levels of parallelism, allowing the algorithm to map onto highly-parallel accelerators such as many-core architectures and GPU's, thus the code will run efficiently from the workstation to supercomputer. The proposed approach will enhance MFIX DEM by using a profiling methodology to identify numerical and algorithmic bottlenecks. Both serial and parallelization bottlenecks will be overcome via vectorization, cache utilization, algorithmic improvements, and implementation of hybrid MPI/OpenMP parallelization methods that synergize with current heterogeneous high performance computing (HPC) architectures and accelerators. Optimizing MFIX-DEM and implementing parallelization for accelerated HPC systems will enable simulations of industrially relevant problems and on machines that industry are likely to have in the coming years. The ultimate goal is to achieve a speedup of two orders of magnitude; a refined estimate will emanate from the profiling effort. Based on our preliminary findings and recent work, a realistic goal for Phase 1 is a performance improvement and demonstration on an industrially-relevant simulation involving 108 particles. Regarding the latter, the team will survey over 30 PSRI member companies during the beginning of the project to identify industrial needs. New experiments will be performed involving ~108 particles in a system of industrial relevance, and this experiment will be used to demonstrate the enhanced MFIX code. Uncertainty quantification (UQ) will also be performed by coupling the available UQ toolkit PSUADE with enhanced version of MFIX. UQ using the enhanced MFIX code on larger and industrially relevant systems will be demonstrated.

Project Benefits

The proposed approach will enhance MFIX DEM by using a state-of-the-art profiling methodology developed by our team members to comprehensively and continuously identify numerical and algorithmic bottlenecks. Both serial and parallelization bottlenecks will be overcome via vectorization, cache utilization, algorithmic improvements, and implementation of hybrid MPI/OpenMP parallelization methods that synergize with current heterogeneous high performance computing (HPC) architectures and accelerators. Optimizing MFIX-DEM and implementing parallelization for accelerated HPC systems will enable simulations of industrially relevant problems and on machines that industry are likely to have in the coming years. The ultimate goal is to achieve a speedup of two orders of magnitude; a refined estimate will emanate from the profiling effort.

Contact Information

Federal Project Manager Jason Hissam: jason.hissam@netl.doe.gov
Technology Manager Briggs White: briggs.white@netl.doe.gov
Principal Investigator Christine Hrenya: hrenya@colorado.edu

 

Click to view Presentations, Papers, and Publications