psinton@aquageo.us

If the focus of your model is the larger groundwater system and not the pit then 1st kind boundary nodes set below the bottom of the pit should be ok.  Apply a constraint so that the nodes don't "inject" water into the system.  Using seepage faces implies that your focus is on the pit and its walls rather than on the much larger groundwater system.

The amount of water entering the pit is very important even if it is not an objective of the modeling because the inflow rate determines in large part the effect on the groundwater system.  1st kind bc with constraints will work fine.

It would be better to put a larger value of K in the pit rather than a small one because the boundary condition you specify will be influenced by that "pit" K.  I understand the the K of air is essentially zero, but that air is not part of the groundwater system.  For pit dewatering, I leave the bedrock K in place or would use an even larger K because the K of free water is essentially infinitely large (lakes are often modeled with very large K).  If the dewatering is insufficient to keep the pit dry, then either you need more wells, deeper wells or both.

If you have to simulate the active dewatering, I'd start with 4th-kind BC (wells) extracting water at rates that are either specified by mine operators or at a rate needed to keep the pit dry.  If the pit is dry, there won't be much in the way of actual seepage unless you are working in a very wet or very high-k area (in which case either a seepage face or a constrained 1st kind bc should be used).  I don't recommend using 3rd type bc.

Pete

Feflow is less sensitive to layer thickness variation than modflow, so using thin layers does not always result in much instability.  However, I recommend a layer thicker than the geomembrane with a thickness or volume averaged Kzz.  I've also put 1d vertical pipe elements in to mimic holes in the membrane and used the K of the pristine membrane as an alternate to an averaged K with assumed holes.

Pete

Hi Thomas,  The worthiness of a numerical implementation depends very much on the overall cost of the project. Analytical and simple numerical models are fine for small-budget projects, but not, in my view, appropriate for multimillion dollar tunneling projects.  To me, the model should also be compatible with the data collected and big projects usually come with a complex (and expensive) data set.

Cheers,
Pete

Pete

There are several examples in the documents that come with FEFLOW that show the influence of discretization in both time and space.  See for example the description of Celia et al's problem starting on page 231.  Many text books on modeling cover the subject as well, e.g. Wang and Anderson or Sptiz and Moreno.

An easy way evaluate discretization effects is to iteratively double or halve the number of time steps or elements.

scubohuntr,

FEFLOW has a two-tier termination criteria setup which is similar in some ways to the PCG solver MODFLOW uses (the FEFLOW solvers are much more sophisticated than MODFLOW's PCG solver).  The "inner loop" criterion is the termination threshold which you reduced from 1e-8 to 1e-10 (there are also controls for max iterations of the inner loop).  This allowed FEFLOW to do "more work" in solving your problem.  The "outer loop" criterion is under the "numerical parameters" part of the "problem settings" menu.  It too controls how much work feflow will do to solve the problem.  The outer loop has three controls, the error tolerance, the way error is measured and the number of iterations per timestep.  The inner loop is executed many times for every iteration of the outer loop. 

In the outer loop, if you switch from L2 to L1 or Max error, FEFLOW will do more work solving the problem at each time step.  However, if you do not give it enough iterations to do more work, the solver may terminate prematurely.  Premature termination of either the inner or outer loops may result in significantly large error in calculated total head, pressure, saturation, etc.  This error can propagate and grow with each time step and may result in a very poor result which may show up as obviously incorrect water levels or mass balance, or significant oscillations.  However, the error may not be obvious and the only way to know for sure is to runs a series of tests on solvers using different settings.

When you decreased the inner loop setting, FEFLOW was forced to do more work in solving your model.  This resulted in more complete convergence.  The large difference between the two solutions suggests that more test should be done to better understand the model's performance.

Pete

Hi king,

You could use a fixed concentration of 200 and constrain the the boundary nodes so that when set to zero they cannot discharge mass.  Compare the simulated amount of mass injected to what you know went into the ground (10 m3 at 200 mg/l).  If the conductivity is too small to accept the amount of mass you know went in the ground then either the amount is mass is too large or the conductivity is too small.  You will probably need more layers to simulate the 4 units because FEFLOW averages vertical conductivities around nodes.

In the T-list, set the 364.99 value to 0 instead of 200.

Pete

Hi Thomas,

I think one way or the other, you'll get better estimates using transient simulation rather than steady state.  However, the estimates depend heavily on the detail of information you have to work with.  If your packer tests miss the most permeable fracture, the model cannot predict large inflow rate from it.

Transient introduction of the tunnel into the model will also result in averaging of inflow rates and underestimation of gradients, errors which can only be reduced using a fine mesh and small time steps (which are in turn limited by the time and budget you have and computer resources).

I'm not familiar with Heuer equations. I imagine that initial inflows relate to turbulence of flow and damage induced during tunneling.

Pete

Hi Viven,  I recommend you read white_papers_vol1.pdf.  There are many examples of what to consider when setting up models, especially the type you are working on.  I think a 3d model would work much better than 2d, and I think you will need a dense mesh (both horizontal and vertical, and small time steps) to simulate the problem accurately.  I would use variably-saturated mode. The expected flow rate (or hydraulic gradient) are the primary consideration when setting up the mesh.  Large gradients and/or flows require denser mesh.  The tutorial has a fine example of this in which successively finer (denser) mesh is used to more accurately simulate head near a pumping well.  In your tank model, the largest flows & gradients will be at the spigot (outlet).  I would model the outlet with a fixed head set at elevation of the outlet, but I may also include material properties around the spigot to help account for turbulent flow near and in the spigot. I think 1d and 2d discrete features may be unnecessary if the mesh is dense enough.

Pete

AB/TR may work better in some situations (UZ problems for one)
Pete

Convergence problems are very problem specific, but often arise due to poorly formed elements.  Run the mesh checker (in classic); if you have more than 10% "red" elements, you need to smooth or reformulate the mesh.  Sometimes, the autostepping can get bogged down working on a small area in the model.  You can fall back to the traditional user-specified stepping in which you grow the steps using a multiplier of 1.1 to 1.3 and start from a very small value, like 1e-6.  Also, do not let your steps grow too large.  In transient models with a lot of stress variation, user-stepping can get very complicated.  When using user stepping, you must check the mass balance is ok and that the local balance is ok in areas of concern (because user stepping may not yield optimal results).  Experimentation is necessary. 

Pete