Björn Kaiser

The workflow suggested by Javier is one way of doing that. There's more than one way to skin a cat. Another way is to delete the slices 2-10 and re-insert them. FEFLOW 7.0 provides the option [b]Insert slice(s) between...[/b]. If you use FEFLOW <= 6.2 you have to use an unrealistic large distance between these slices exceeding the layer thickness (e.g. 999999 m). The 3D Layer Configurator will distribute the slice equidistantly.

If you have one single-seat license available you cannot run several models at the same time.

Your second question cannot be answered with yes or no, because other dependencies (e.g. model settings, total number of cores etc.) play an important role. Accordingly, in one configuration the answer would be yes, while in another configuration the answer would be now.

An extreme example: Let's say you have a large number of cores (e.g. 48) and you have only few computational mesh nodes (e.g. 20,000), then this large number of cores would even slow down the simulation, because the internal communication of the threads overprint the actual solution process. In this case fewer cores would be the better option.

I would expect that the average concentration of a well at a specific time is equivalent to the concentration at this node at the same time. Could you please verify that by inspecting the Average Concentration History chart and the Mesh Inspector in the Slice View. In case you have a Multilayer Well (MLW) please assure you use the Mesh Inspector at the lowest node of the MLW.

Yes, that's correct. If three slaves are defined then these three slaves will solve the the model over and over again based on different parameters sets. If you use the default FEFLOW Global Settings then "the slave computer" will decide how to make use of other available local CPU cores to solve the FEM. In a similar fashion, if the the number of slaves is less than the maximum number of CPU cores available then FEFLOW will run inevitably run in parallel.

The functions I have shown above are formulated in C/C++, but can be easily adopted in Python. Please let me know if you have further questions.

A random parameter field generator within FEFLOW is planned for the future. It has not been developed yet.

Accordingly, you need to use an external generator like the one you indicated. You could use the output of the generator as input in FEFLOW to parametrize your model.

One option is to convert the generator output into the ASCII based dat-file format readable by FEFLOW. This file can be used as a map for parameter linking (regionalization). The disadvantage is the manual work. I would expect that this task is tedious and time-consuming particularly if you have to carry out a large number of simulations.

Another possibility is the development of a script (Python) or plugin (C/C++). I suggest to use Python, where Kernel Control Methods are provided. Generally, Kernel control methods enable the user to start/stop the kernel, to load/save documents, and to control the simulation. You could use kernel control method [font=courier][color=blue]doc.startSimulator()[/color][/font] to start the simulator within a loop ranging from 0 to n_number_of_simulations-1. Before you start the simulator you could assign the random parameter field to the mesh elements

In this context, please note that available regionalization methods are not available for Python. Instead you need to assign the K-values via
[font=courier][color=blue]
extern void IfmSetMatXConductivityValue3D (IfmHandle, int, double);
extern void IfmSetMatYConductivityValue3D (IfmHandle, int, double);
extern void IfmSetMatZConductivityValue3D (IfmHandle, int, double);
[/color][/font]

You could derive the K-values of each element by taking advantage of using the FEFLOW API and GIS API (e.g. arcpy) within a single script. Since the mesh do not change over time (I assume you do not use the free & moveable method) you need to export the element centroids as a point shapefile from the Data panel only one time (not from the script).

Within each loop you may convert the parameter field into a GIS raster file or TIN (Triangulated Irregular Network). To transfer (project) the K-values of the raster / TIN to the shapefile attributes you could use the arcpy function [font=courier][color=blue]AddSurfaceInformation_3d (in_feature_class, in_surface, out_property, {method}, {sample_distance}, {z_factor}, {pyramid_level_resolution}, {noise_filtering})[/color][/font]. In former times, the function was also called Surface Spot.

After the K-values are written to the input feature's attribute table you may use the FEFLOW API function [font=courier][color=blue]IfmFindElementAtXYZ (IfmHandle, double x, double y, double z);[/color][/font] to derive the element ID's which correspond to the points. As soon you derived the element ID you may finally assign the K-value based on the three assignment functions I have shown above.

In general the workflow seems a bit complicated, but the script would only have a few lines of code.

Are you sure that you only want to create random parameter fields without optimizing the parameters (PEST/FePEST)?

According to my knowledge an example for 3U square geothermal piles does not exist. However, the general principles of spatially discretizing geothermal piles are the same as with any other spatial discretization. First of all, the geometrical features of interest must be represented within the finite element mesh. Moreover, domains where sharp contrasts of material properties and where strong gradients of process variables characterize the system need to be sufficiently resolved by the mesh. A smooth gradation between fine refined areas to coarse areas may contribute to a stable numeral solution. Last but not least, all process variables are solved at mesh nodes, while material properties are assigned on mesh elements. Accordingly, a fine spatial discretization within impermeable regions (concrete piles, aquitard) is required to account for its impervious nature. The FEFLOW book (chapter 16.6.4) addresses fully discretized BHE 3D models which could provide some additional hints on how to discretize geothermal installations.

To represent the water column by means of the hydrostatic pressure I suggest to assign a fixed Hydraulic-head BC to all nodes of the first slice. Use a value which corresponds to the water-table.

If you deactivate elements all connected nodes which are [b]completely[/b] surrounded by inactivate elements are also deactivated. As soon as a node is connected with a single active element then this node will participate in the numerical simulation.