We are curious if it is important to include Pillars in the CFD models for tanks that have pillars. We studied three cases, all in 3D. We attempted to study these effects in 2D (to run the cases more quickly), but the results in 2D are rather different from those in 3D, so, for this study, the 2D results had to be discarded.

This is a 160 x 160 feet, 4 MG tank. When pillars are present, they are circular columns, 18 inches in diameter, separated by 16 feet. The inlet and outlet are 40 feet from the nearest parallel wall. The flow is simultaneous inflow and outflow at a rate 8.85 MGD. The inlet and outlet are both 2.1' x 2.1' square ducts (with mean flow velocity is 3.3 ft/sec). The height of the water is 19.7 feet. All displays shown here are from above (Plan View). All simulations are isothermal.

To provide a reference to other cases, mixing times can be compared to the time it takes to fill this tank (at 8.85 MGD) to the depth this case simulates: ~10.2 hours.

Square Tanks with no Pillars
Mesh. A multiblock strucutured grid with 201,920 elements. (18K gif)
Steady State Velocity. (30K gif)
Mixing Statistics (low res). (4K gif) or Mixing Statistics (high res). (11K gif). Mixing time for this case is 49% of the time to fill the tank to the simulated depth.
Tracer Distribution Animation for 4 hours of tank time. ([480x360] QuickTime 570K). This shows the effect of the mixing new (i.e. red) fluid from the inlet with the existing (i.e. blue) fluid in the tank. This can also be interpreted as viewing the injection of a tracer into the tank. The color shows the time-varying local concentration of the tracer.

Square Tanks with two Pillars at the Inlet and one Pillar at the outlet

Mesh. A multiblock strucutured grid with 210,606 elements. (21K gif)
Steady State Velocity. (28K gif)
Mixing Statistics (low res). (4K gif) or Mixing Statistics (high res). (11K gif). Mixing time for this case is 54% of the time to fill the tank to the simulated depth.
Tracer Distribution Animation for 5 hours of tank time. ([480x360] QuickTime 650K).

Square Tanks with 100 Pillars evently distributed throughout the tank

Mesh. An unstrucutured hex grid with 263,556 elements. (38K gif)
Steady State Velocity. (32K gif)
Mixing Statistics (low res). (4K gif) or Mixing Statistics (high res). (11K gif). Mixing time for this case is 90% of the time to fill the tank to the simulated depth.
Tracer Distribution Animation for 5 hours of tank time. ([480x360] QuickTime 670K), or for more resolution ([640x480] QuickTime 1.2M).

An additional movie (QuickTime 1.6M) shows how the mixing progresses during 22 hours of tank time.

Comparison of Outlet Tracer Concentration for all three cases
History of Outlet Concentration (850x710 GIF 13K).

This plot shows the history of the tracer concentration at the outlet for three simulation cases (no pillars, 3 pillars, and 100 pillars; simultaneous fill and draw mode. The differences in the histories for these three cases is quite small. For reference, the history of the tanks mean concentration (for the no pillar case) is also shown. For each case, the outlet history is quite close to the tank's mean concentration history.

The impression from the outlet data alone is that the three cases have very similar mixing characteristics. As it turns out, this impression is incorrect. The mixing statistics (see mixing statistics above) show that the 100 pillars' case mixes much more slowly that the other two cases.

Note: During the first 5 hours of tank time, the run was done with a 3 second time step (requiring ~75 CPU second per time step on a SGI Octane). The flow field becomes "psuedo" steady state after ~3 hours. The remainder of the simulation was done with a 30 second timestep (requiring ~86 CPU seconds per time step). The total run time for each case is 7.2 CPU days.