Simulation Idea - Is MFiX Appropriate?

MFiX Share
#1

I am working on stockpile modeling for a mining firm. The goal is to accurately predict the properties of the material coming out of the pile given the mass, properties and time of arrival of the material going into it.

I’ve been developing a simple 3D model based on large (10-100 ton) blocks of material which move according to very simple rules intended to approximate movement on a slope in excess of the angle of repose. The blocks are 1 unit on a side and stacked to form a cone with a 45 degree angle consisting of ~10,000 blocks. If a block can make a “knight’s move” of 1 unit horizontally and 2 units down, it does so. Material is added to the apex and withdrawn from the bottom, directly below it. However, the relative rates of input/output vary, allowing periods of growth and depletion of random durations.

I know these are pretty pedestrian interactions compared to what MFiX typically does but I am wondering whether it would be a useful framework for implementing this kind of scenario? I am hoping if it can support the basic model, it would allow it to be extended to replace the movement rules with gravity acting on shapes other than cubes (for example).

The reason I am pursuing such a simplistic model is that I need it to operate in real time on commercial cloud computing resources, potentially with GPU or tensor processors-- not 6800 node supercomputers running for weeks at a time. [This is when the DEM folks start laughing…] It is a serious question though.

Thoughts, ideas, suggestions? Is MFiX “too big” for this little problem? Are there other simulation packages that might be appropriate? Does it support the kind of movement rules I described? Will it run fast with MFiX because it is so simple?

Many Thanks,
eesnyder

#2

Currently MFIX DEM is only spheres. MFIX will run fast without out the fluid solver (granular flow).

I assume that the particle shape is important? That is the expensive part.

#3

For my simulation, particle shape is somewhat arbitrary. I chose a cube because it conveniently mapped to the lattice coordinates while packing gaplessly. Spheres will work and probably makes more sense as the model transitions from rule-based to physics-based.
My account is now approved for download so now I can look under the hood and see what the beast is capable of doing.
Thanks for your input. If you have other suggestions or words of advice for someone new to the software and the field in general I am all ears!
Cheers,
Eric E. Snyder, Ph.D.
Data Scientist
Freeport-McMoRan, Inc.
Phoenix, AZ 85040
esnyder@fmi.com
(480) 848-3050

#4

For this industrial-scale problem, I would suggest you try the coarse-grained DEM model （lump a number of particles into a numerical parcel to speed up） in MFIX instead of shape resolved DEM (SuperDEM). To consider the effect of irregular shapes, like sands, a rolling friction modelcan be added to CGDEM.

#5

For real time, granular flow DEM (don’t solve the fluid), and as @gaoxi said, fudge the rolling friction model to approximate shape effects. Unfortunately we don’t have a gpu version, which would be even better because you can get lots of performance in a desktop.