Simulation of complex underground mine workings

[RyanConway91]

I want to simulate the mining and post-closure conditions of a large complex underground mine. The mining portion is easy - just map in drains. However, I am unsure of how to simulate the post mining period. I want to make predictions on how the workings will fill through time after the pumps are shut off. Post mining, the workings are void space, so one option would be use TVK/TVS to map in "void space" (high k, sy = 1)? However this would require a very high resolution model to the void volumes and connectivity right. Another option I thought of would be to use buried LAKs and SFRs? Another option would be CLN is MF6 had that...

have others simulated this situation? If so, how?

[cnicol-gwlogic]

Hi Ryan, re the tvk/s option, you can just do volumetric (hgu/void-based) upscaling to remove the need for fine spatial res. We've done that a bit.

[cnicol-gwlogic]

Yep that's right, along those lines. We found some differences between fine and upscaled models in the ensemble stats (upscaled had wider inflow bounds), but that could have been user error and/or something we didn't think of. I think if you very carefully thought through the local fine-->upscaled local geometries and internal cell conductances of each in all directions, you would get it pretty bang on in the upscaled version.

Note that upscaled heads won't necessarily be fully drained, because it'll be a weighted average over void (dewatered) and rock (partially saturated) space. The more void in the big cell volume, the closer heads would be to drained conditions. So you may need to consider whether that suits your objectives or not - probs depends on whether this is inflows, pressures, or both.

[RyanConway91]

Thanks! Another random question, how do you typically map complex 3d geometries into your model? We usually get .msh or .obj files for mines. I've tried different approaches for intersecting the grid and volumes, but they are all pretty clunky. Current best workflow involves voxelizing the model grid and 3d geometry and finding 3d geometry voxel points inside the grid. Straight forward volumetric intersections in Paraview or pyvista seem to always throw "non-manifold" errors. To get the Sy correct in the workings, I would like a more robust method for the volumetric intersections

[cnicol-gwlogic]

@neilmanewell is best placed to answer that one.

[neilmanewell]

Hi Ryan,
RE: the differences in the uncertainty of inflows was because there was an ore zone I wasn't properly upscaling (it was 3 times thinner than the model cell). I was properly upscaling void zones/workings/tunnels and faults, but I was lazily upscaling the ore by lowering mean, stdev and max sy. Proper upscaling fixed it. So don't be lazy is the lesson. ha.
RE: complex 3D geometries - I actually ended up using Pyvista and Monte Carlo point sampling within volumes, then vtkSelectEnclosedPoints to make the selection. The Monte Carlo sampling with thousands of points did seem to do a pretty nice job. Was fairly happy in our case, but maybe we were lucky.

[RyanConway91]

Thanks - your approach for intersection sounds similar to mine - I create fixed density voxel points within my model cells, where the voxel ids can be mapped back to model cells, and then use vtkSelectEnclosedPoints to find the number of voxels inside the model cell. This gives the volume/volumetric proportion of the 3d object in each model cell. Would be nice if we could just directly intersect the 3d model grid and the 3d volume though!.... not sure why this seems so hard.