WATER & ENVIRONMENT

TRAINING

EXERCISES

Improving the world through accurate flow modeling

Guide 1: CFD Project Workflow Guide
 FLOW-3D Version 11
CFD Project Overview

CFD Project Workflow Guide

Contents
Select a problem with known results for proof-of-concept testing ............................................................. 1
Set up and run a coarse test case ................................................................................................................. 2
Select and calibrate numerical methods ...................................................................................................... 3
Minimize & quantify the mesh dependency................................................................................................. 4
Quantify the validation uncertainty.............................................................................................................. 5
Calibrate mesh-dependent input parameters .............................................................................................. 6
Model the case of interest ............................................................................................................................ 7
Report results................................................................................................................................................ 7

This project workflow guide is not an exhaustive decision-tree for all elements of a project. Rather, it is
meant to be a checklist with a middling level of detail. Some possible project elements have been
omitted (generating geometry, for example). The Flow Science support staff hopes you find this guide
helpful. Please send suggestions to [email protected].

Select a problem with known results for proof-of-concept testing

Whenever you must model a new class of problems using CFD, it is good practice to check whether you
can set up a CFD model that gives expected results. To check, you will need a case that is representative
of the problem of interest and that has known results. Once you have tested one or more representative
cases for a problem type you can skip this step for new problems of the same type.

 Check the online resources for similar cases

Some classes of problems (ogee spillways, for example) have already been thoroughly validated
against physical experiments with FLOW-3D. You can search to see how other users have
modeled the same type of problem:

 www.flow3d.com > Resources > Bibliography > html search (Ctrl-F) for keywords
 www.flow3d.com > Site Search (upper right input box) for keywords
 users.flow3d.com > Conference Proceedings, Validations, Example Simulations,
Technical Notes (for cases related to specific physics models), and Training (for tutorials)

 Compare the test case to the case of interest

 How were the test case experimental results collected?
 Can experimental uncertainty be quantified for the test case?

Page 1 of 7
  Are the test case results from a reliable source?
 Does the test case include the same physical processes as the case of interest?
 Does the test case have similar physical and time scales as the case of interest?
 Are the dimensionless numbers (Re, Fr, We, etc.) similar between the cases?

A test case may still be valuable even if it doesn’t meet all of the above criteria, but there will be
relatively less certainty in how to parameterize the case of interest.

Set up and run a coarse test case

 Follow the simulation setup guide using the coarsest possible mesh
This setup mesh is purposely kept as coarse as possible to make it run as fast as possible. After
completing the simulation setup guide (on this site) you will have:

 An annotated diagram with all necessary information in a consistent unit system,

 A simulation folder with:
o all .stl files (checked and repaired)
o a ready-to-run prepin file
 A mesh that resolves the thinnest flow sections with 3 or 4 cells,
 A mesh that follows best-possible-practices,
 Rationales for all specified physics models and coefficients.

 Run and post-process the coarse case

If deadlines are pressing, only run a portion of the case. Answer the following questions:

 Are the solver warnings acceptable?

 Do the solver text and runtime plots indicate that the time step sizes, pressure
iterations, mean and turbulent kinetic energies, and fluid volume are appropriate?
 Does the model run in a reasonable amount of time?
 Is it possible to get the necessary outputs from post-processing?
 Check 3-D and 2-D plots. Do the results look right?
 Check Probe and Text outputs. Do the results follow the expected trends?

If the answers to the above questions are yes, the simulation is configured adequately. If any
answers are no, consider adjusting the simulation setup with the following checklist:

 Troubleshoot the coarse case if necessary

 Does the mesh follow best-possible-practices?
 Are geometry files checked and repaired and well-resolved by the mesh?
 Are the boundary conditions appropriate and realistic?
 Are the initial conditions appropriate and realistic?
 Do the boundary conditions and initial conditions match?
 Are the active physics models appropriate?
  Are physics model and component coefficients correct? Check units and values carefully.
 If all of the above considerations have been checked carefully and the problem is not
resolved, refer to the User Manual > Troubleshooting chapter.

Continue adjusting and re-running the coarse case model until you are confident in the
simulation and ready to commit to longer run times.

Select and calibrate numerical methods

In this step, you will test the model sensitivity to different numerical methods. Whenever a higher-order
method gives significantly different results, you must make a choice as to whether or not to use it. This
step is optional but recommended whenever the coarse test runs quickly and there is sufficient time in
the budget. Test as many of the following as you can.

 Check sensitivity to the VOF method

Models with free surfaces have several options for volume-of-fluid methods. The default option
includes the most robust surface reconstruction method, but higher-order methods are
available that may be more accurate. In particular, if you have a curved free surface that moves
(e.g., waves), a higher-order option may be desirable. A rule of thumb is to try Split Lagrangian
VOF and compare the results to the default One Fluid, Free-Surface VOF. If your output doesn’t
change significantly, then use the more robust and faster default option.

 Check sensitivity to the momentum advection method

Higher-order momentum advection methods are generally more accurate, but may be less
stable and/or generate more noise in the solution.
nd
 If your model has a free surface, try 2 -order, monotonicity-preserving momentum
nd
advection. If there is no free surface, try pure 2 -order momentum advection.
st
 Compare the results against the default 1 -order method, and if the results are
significantly different use the higher-order method.

Special considerations:
st
 1 -order should be used with wave boundaries (along with Split Lagrangian VOF).
st
 There is some anecdotal evidence that 1 -order momentum advection may give more
nd
accurate diffusion for highly turbulent flow than 2 -order, monotonicity-preserving
momentum advection. This has not been demonstrated for all cases.

 Check sensitivity to recommended implicit methods

Implicit methods are indicated whenever they will result in a significant speed-up of the run
time. The mentor, for example, suggests an implicit method when the speed-up will be at least a
factor of 10. If an implicit method will not result in a significant speed up, then explicit methods
are preferred because they are more reliably accurate. Some physics models (e.g., Moving
Objects) may need implicit methods for stability.
  Compare results from indicated implicit methods to results from explicit methods.
 If the results are sufficiently similar and the implicit method(s) are noticeably faster,
consider using the implicit method for the rest of the project.

 Check sensitivity to boundary conditions

 Move the boundary location (mesh block extents) away from the region of interest and
re-run the coarse case until the results do not vary significantly.
 Each expansion should be an even multiple of the cell size in that direction. Check for
each expansion that the cell size remains the same.
 Take the smallest model that is free from boundary effects as the working case.
 Turbulence at inlet boundaries can affect some models. Try a large turbulence intensity
(e.g., 5%) at the boundary to see if it has an effect. If it does, use values that correspond
to physically-realistic turbulence intensities at the boundaries.

 Check results sensitivity to physics model coefficients

Physics models may include numerical coefficients (as opposed to physically-based input values)
that affect the results. The default settings of these coefficients are often acceptable, but can be
calibrated to better match experiment.

 For example, the two-equation turbulence models use a purely numerical maximum
turbulent mixing length parameter (TLEN). Test the sensitivity of the results to constant
vs. dynamic TLEN. Start with 7% or 10% of the controlling length scale (often max fluid
depth) and increase the value (e.g., to 15%, 30%, 50%, 100%) until the results cease to
vary significantly. Take the setting that gives results that are closest to the experiment.
 Repeat the process for any other empirically-based physics models that have numerical
knobs that can be shown to affect the model results.

Minimize & quantify the mesh dependency

Your model is working as expected and has been proven to run. Now you can begin running the case on
more accurate meshes. Running the same case with different mesh resolutions demonstrates that the
solution converges when the mesh and time step are refined. This step is sometimes called solution
verification and is discussed at length in ASME (2009) and (ASCE) 2009. This step is optional, but highly
recommended when time and resources allow it.

 Add copies or restarts with refined meshes

The easiest way to do this is to refine the cell sizes of all mesh blocks, in all directions, by the
same factor. A refinement factor of 2 is common (increasing the cell count by a factor of 8), but
any ratio larger than 1.3 is acceptable. A minimum of three grids is recommended.

 Create a copy or a restart copy of the simulation. If it ran to steady state, then a restart
copy with a shorter finish time is recommended for faster results.
 Refine the mesh uniformly in all directions by the selected refinement factor.
  Adjust the finish time (if desired). The adjustment (if any) depends on (1) how long it
took the coarse case to reach steady state and (2) if the new simulation is a restart.
 Repeat the process to create a third case with an even finer mesh. The third case may
be set up as a restart from the second case even if the second case has not run yet.

 Run and post-process the finer-mesh cases

 Set Simulate > Set Max Simultaneous Simulations = 1 if the refinements are restarts.
 Add all the simulations to the run queue and run them in order.
 Post-process the results. Use the same variables of interest as in the coarse test case.

 Calculate the mesh-sensitivity of the results

 The results should be converging on a constant value. Demonstrating this means the
solution is “verified”. See ASCE (2009) and ASME (2009) for details and calculations.
 A simple rule of thumb: the model is “mesh-converged” when refining the mesh causes
the output to change less than 3% (for output with magnitude greater than 1000) and
less than 1% (for output magnitude less than 1000). This is a general guideline, and not
always appropriate. Use more rigorous methods if in doubt.
 The design mesh is either the converged mesh or the finest mesh that is feasible.

Quantify the validation uncertainty

This step quantifies input, numerical, and experimental uncertainty for cases that involve experimental
data. This step is optional. It is included here because it results in 95% and standard-deviation
confidence intervals which are more defensible than simple error comparisons. The interested user is
referred to ASCE (2009) and ASME (2005, 2009) for guidance and details.

 Numerical uncertainty
The results from mesh-dependency study described above can be used to calculate numerical
uncertainty, order of convergence, Grid Convergence Index, and extrapolated results.

 Input uncertainty
Input parameters represent physical reality, which is to some extent unknowable. The way to
deal with this is to quantify the uncertainty of the CFD results due to the uncertainty in the input
parameters. This means estimating the distributions of likely input values, picking values from
the distributions, and recording and interpreting the results when those values are used. This is
time-consuming. It is preferable to find input uncertainty using the design mesh, but using a
coarse mesh is much faster.

 Experimental uncertainty
You must make some estimate of the experimental data uncertainty whenever you want to
compare CFD results to experimental results. Standard methods exist for quantifying
experimental uncertainty; most involve estimating measurement uncertainties and combining
 them to find result uncertainties. Detailed experimental uncertainties are ideal, approximate
evaluations (e.g., ±2%) are common in practice.

 Validation uncertainty
Validation uncertainty is the estimated standard deviation of the parent population of the
combination of errors (ASME 2009), that is, a function of numerical, input, and experimental
data uncertainty. The uncertainty uval can be compared to the disagreement |E| between the
CFD model and the experiment:

 Wherever uval is greater than or close to |E|, the disagreement is within the level of
“noise” of the experiment, and the model can be considered “good”.
 When the disagreement |E| is greater than uval, it suggests that the model setup could
be improved (e.g., by calibrating mesh-dependent parameters as described below) or
that the CFD itself is of limited reliability (e.g., turbulence models).

Calibrate mesh-dependent input parameters

CFD models often include some variables that must be calibrated after the design mesh is selected. An
example is component surface roughness: when the physical roughness is used the results become
realistic as the mesh is refined, but usually require resolutions that are not time-feasible. The usual
working method is to instead use either the converged mesh or the best possible mesh and then
calibrate component roughness to whatever value minimizes the difference between CFD and
experiment. This value may not be physically realistic!

 Select parameters to calibrate or keep fixed

When possible, adjust only one parameter for each physics model (e.g., surface roughness for
turbulence, Critical Shields Parameter for sediment scour, rate coefficient for air entrainment).
Numerical options that were justified in previous steps should stay fixed.

 Vary the calibration parameters and record the results

When possible, adjust only one parameter per run. Use restart simulations to minimize the run
time, if possible. Record the results and compare them to the experiment.

 Check for trends

This helps reduce the amount of guesswork that goes into selecting the next calibration value.
For example, results for three roughness values can be fit logarithmically to predict the value
that will minimize the model/experiment disagreement.

 Keep the values that minimize disagreement

It’s important to remember that these design values are valid only for the given mesh,
+
geometry, flow velocity, and dimensionless numbers (e.g., Re, Fr, We, k ). The parameters may
need to be recalibrated for each new case, unless they are similar. If the calibrated values are far
from physically-realistic values, be prepared to explain why (e.g., the physics model used is
 mesh-dependent, and a physically-realistic input value requires a mesh resolution that is not
feasible for the project time constraints).

Model the case of interest

Use everything you have learned from the test case.

 Start with the coarsest-possible mesh

 Follow the simulation setup guide (on this site) and prove the case runs

 Use the numerical options and input parameters from test case

 Refine the mesh to minimize error

 If the test case and case of interest are of similar scales and dimensionless numbers, use
the design mesh from the test case immediately.
 If no test case was run first, or if the test case is significantly different from the design
case, then perform a mesh dependency study as described above for the test case.

 Post-process the results and check that they appear reasonable

Report results
 Describe the physical situation and why CFD modeling is needed

 Describe the test case & report results

 Describe the physical situation and experimental data.
 List any non-standard numerical methods (and note if only default values were used).
 Describe the selection of boundary conditions and their locations.
 Describe the mesh dependency study, including sizes used and convergence results.

 Describe the case-of-interest & report results

 Report model options, whether they were the same or different as in the test case.
 Report results for the variables of interest from the design mesh resolution.
 Justify the design mesh resolution (e.g., mesh dependency results or prior experience).
 Report any conclusions drawn or lessons learned from the simulation.

 Report validation uncertainty (optional)

 Report estimated numerical, input, and experimental uncertainty and methodology.
 Report any additional calculated verification and validation results (e.g., GCI).
 If no uncertainties were estimated, justify the omission. Rigorous uncertainty estimation
is an ideal that is not always necessary or feasible in practice (e.g., when CFD is used to
qualitatively compare potential designs).
Guide 2: CFD Simulation Workflow Guide
 FLOW-3D Version 11
Simulation Setup

Simulation Setup Guide

Table of Contents
Before starting .............................................................................................................................................. 1
On the Model Setup: General tab ................................................................................................................. 4
On the Model Setup: Physics tab .................................................................................................................. 5
On the Model Setup: Fluids tab .................................................................................................................... 5
On the Model Setup: Meshing & Geometry tab ........................................................................................... 5
On the Model Setup: Output tab ................................................................................................................ 10
On the Model Setup: Numerics tab ............................................................................................................ 10
Final Check after setup ............................................................................................................................... 10
Run the simulation ...................................................................................................................................... 11

This setup guide is not an exhaustive decision-tree for all elements of a simulation. Rather, it is meant to
be a checklist with a middling level of detail. Some possible model elements have been omitted
(electromechanical physics models, for example). The Flow Science support staff hopes you find this
guide helpful. Please send suggestions to [email protected].

Before starting
 Draw and annotate a diagram of the physical problem
The diagram should include:

 A consistent unit system (e.g., SI, CGS, slug/foot/second) that matches the scale of the
problem (very small and very large numbers can increase rounding error),
 Geometric dimensions,
 Locations of fluid and free surfaces,
 Notes on body forces (like gravity) and boundary forces (like shear stress),
 All heat and fluid fluxes and sources,
 All solid moving objects and their velocities, locations, and travel paths
 Unknown variables and the locations where you want to them.

The completed diagram should be an sketch of the physical problem that includes all of the
relevant engineering information.

Page 1 of 11
 Determine the general modeling approach
Decide how to approach the problem using the annotated diagram as a guide. Record your
selections on the diagram as you proceed.

 Label the fluids in the problem (e.g., air, water, oil),

 For liquid/gas systems, can local gas velocities be ignored?
o Yes: consider a one-fluid model: decide if the gas pressure and temperature
change with time (meaning one of the bubble & phase change physics models is
needed).
o No: consider a two-fluid model and decide if the gas can be treated as nearly
incompressible or must be treated as fully compressible (costs more resources).
 Can a portion or all of the flow be treated as depth-averaged? If yes, consider using
shallow water physics in one or more mesh blocks
 Does a fluid have a sharply-defined interface (free surface)?
o Yes: consider free surface or sharp interface tracking
o No: consider no sharp interface tracking (fluid separation is diffuse or the whole
domain is submerged)
 Is the flow viscous? If yes, consider using viscous flow physics (almost always used).

 Determine the physics packages that are part of the modeling approach
Now decide which physical mechanisms are important. Add the relevant engineering
information for each of these options to the diagram. Try to justify why physical mechanisms
were included or ignored – this helps catch mistakes early. A common checklist is given below.

 Can gravity and reference-frame motion be ignored? If not, consider gravity and non-
inertial reference frame physics.
 If heat transfer can’t be ignored, consider heat transfer physics, and make fluid-to-solid
heat transfer selections below:
o Can all solid components be treated as having a constant, uniform temperature?
If yes, consider no fluid-to-solid heat transfer.
o Can all solids be treated as having a “lumped” (spatially-uniform) temperature?
If yes, consider uniform component temperatures.
o If some solids must have temperature that vary in space due to initial condition
but are constant in time, consider the non-uniform, constant temperature
option.
o If some solids must have temperatures that vary in both space and time,
consider the full energy equation option.
 If a fluid varies in density (e.g., due to temperature gradients, salinity gradients,
contaminants, or entrained air), consider variable density physics.
 If a fluid contains other phases (e.g., contaminants, particles, or tiny bubbles):
o consider scalar physics if the contaminant(s) can be treated as continuous
concentrations,
 o if a contaminant is added at free surfaces consider air entrainment physics or
defect tracking physics,
o if contaminants should be treated as a finite number of particles (with or
without drag) and not as continuous concentrations, consider particle physics,
o if they should be treated as continuous concentrations that represent many
discrete particles (with drag effects), consider drift-flux physics,
o if they should be treated as packable (dense) concentrations that can take solid
forms, consider granular flow physics, sediment scour physics, or dissolving
object physics.
 Are there two pure-phase liquids with different densities and viscosities along with a
free surface and a gas phase? If yes, consider using the two-liquids and free-surface
workaround (set RHOFS & MUS as the density and viscosity of the second liquid in the
prepin file namelist $PROPS, and then set all initial and boundary condition densities as
liquid mixture densities between RHOFS and RHOF).
 Estimate the Weber number. If We is less than or close to 1, consider surface tension
physics (computationally expensive).
 Estimate the Reynolds number Re. Decide if turbulence physics are necessary. If yes,
+
decide which turbulence model to use, estimate the roughness Reynolds number k , and
decide if surfaces should be hydraulically smooth (ROUGH = 0) or rough (ROUGH > 0).

 Draw a Cartesian or Cylindrical coordinate system on the diagram

 Decide whether to use Cartesian or Cylindrical coordinates based on the flow geometry.
 Decide where to place the origin.
 Draw the axes. Gravity is usually in the negative z direction, because some boundary
conditions and hydraulic data output require it. Try to avoid placing axes that are mostly
diagonal to the flow.

 Draw the computational domain on the diagram

Computational domains (mesh blocks) are 3-D rectilinear or axisymmetric cylindrical wedges.
Their edges are always parallel to the coordinate system. They may be nested or linked.

 Sketch the block edges where they are not in regions of rapid change (high gradients of
pressure, velocity, or temperature).
 If possible, sketch a single block. This minimizes interpolation error. Consider using
automatic subdivision (available in v11) to locally resolve the mesh.
 Minimize the number of cells where there will be no flow. Consider adding domain-
removing components to your sketch to turn off unused cells.
 If you know already that linked or nested blocks will be necessary to resolve geometry
or flow features, sketch them now. Use as few blocks as possible.
 Special consideration: if you (a) have more than one mesh block, and (b) plan on using
regularly-sized cells in each (no mesh stretching), then place the edge coordinates of
 each block so that the x, y, and z distances between them are even multiples of the
largest cell length of the same direction in all blocks. This will make perfectly aligned
gridlines between blocks and minimize interpolation error.
 Write the edge coordinates of the blocks.

 Draw the boundary conditions on the diagram

Write which boundary condition type best represents the physical situation at each boundary.
If none of the available boundary condition types are a reasonable approximation of the physical
situation at a boundary then that boundary must be moved elsewhere.

 Generate, check, and repair all .stl files

Check all .stl files for errors and repair them using MiniMagics, Admesh, netfabb Studio,
MeshLab, and/or similar .stl tool programs. Common errors include inverted normals, zero-area
facets, non-manifold edges, and holes. These may not be detected by FLOW-3D but can cause
bad solutions.

On the Model Setup: General tab

 Add some notes

Notes help explain what was done and why for future users or for later reference. Describe the
problem, the purpose of the simulation, the case number, the approach, etc.

 Choose solver precision and number of processors to use

A good practice is to leave at least one core (either 1 or 2 processors, depending on whether
hyper-threading is active) for the operating system and background programs. Another common
practice is to use an even number of cores to reduce computational overhead.

 Select number of fluids, interface tracking, and flow mode options

Use the annotated diagram as a guide.

 The number of fluids refers to where momentum equations are to be solved. In one-
fluid models momentum is solved in regions where F > 0. In two-fluid models, F = 0
represents Fluid 2, and momentum is solved in all open and partially-open cells.
 The interface tracking options define whether changes in the fluid fraction should be
sharp or diffuse. The sharp interface option tracks the location and orientation of the
fluid interface within cells, and permits free surfaces in one fluid models. The no sharp
interface option treats the interface between fluids as a mixture, and forces all open
volume to be full of fluid at all times.
 The flow mode option applies only to two-fluid problems. It treats regions where F = 0
as either compressible (ideal gas law) or incompressible.
  Define the finish condition(s)
Set the finish time. Additional finish conditions can be based on time, fill fraction, or steady-
state criteria.

 Define how to restart the simulation from existing results (optional)

Restart options allow initialization from an existing results file. The restart options define what
information is taken from the results and what information is reset to the current simulation
initial conditions.

On the Model Setup: Physics tab

 Activate the relevant physics models based on the annotated diagram

 Be aware of dependencies: some physics models can’t be activated without others.
 On the diagram or an attached sheet, list all empirical values and coefficients used.
 Have a rationale for each value on the list.
 Make notes of any values that you are unsure of.

On the Model Setup: Fluids tab

 Define the properties of fluid 1 and (if used) fluid 2

Define the appropriate physical properties for the active physical models using the annotated
diagram as a guide. Parameters that are associated with inactive physics models will be grayed
out.

 Define any necessary properties of the interface

Define the properties of the interface between regions of F = 1 and F = 0. These include the
properties for surface tension physics, phase change physics, and two-fluid diffusion.

 Record values their rationales/sources/citations

As before, note any values on the diagram that you are uncertain of.

On the Model Setup: Meshing & Geometry tab

 Import and create geometry

 Add subcomponents to create the solid geometry described in the annotated diagram
using .stl files and primitives. Include any shapes that will be mass sources.
 Apply transformations to the subcomponents: magnifications, rotations, and
translations. The transformations will be applied in the order just described. Consider
defining transformation centers for subcomponents that need multiple transformations.
  Associate the subcomponents with components. Define components so that all their
subcomponents share the same material properties (e.g., surface, solid, and motion
properties). Give descriptive names to each component.
 Define domain-removing components to block off open regions where flow will never
occur, and where the open void does not need to vent to a boundary condition. This
reduces the active cell count, and makes the simulation faster. If in doubt, skip this step.

 Define the mesh

 Following your diagram, create the mesh blocks and specify their extents.
o Specify the Cartesian or Cylindrical coordinate system.
o Check that linked blocks have edges that align exactly.
o Check nested blocks are inside their containing blocks (they may share edges).
o Add mesh planes in containing and linked blocks to match the edge extents of
their nested or neighboring blocks.
o Check that inter-block boundaries are not located in high flow-gradient regions.
 Specify target cell sizes or counts for each block and direction. Consider the following:
o Cubic cells are the most accurate.
o The ratio between the longest and shortest side of each cell should be less than
3:1. Exceptions increase uncertainty in the results, but may be necessary.
o All solid cross-sections should be resolved by at least 2 or 3 cells.
o Fluid cross-sections should be resolved by at least 3 or 4 cells. Accurate results
often require 10 to 12 cells across the thinnest flow section.
o If necessary to get these resolutions, add mesh planes, mesh plane cell counts,
and mesh plane cell sizes. This will make the mesh irregular (stretched). The
stretching ratio between adjacent cells should be less than 1.25 or 1.3.
o As much as possible, the cell face size in nested blocks should be ½ that of the
cells in the containing block. This minimizes interpolation error.
o Adjust mesh block edge locations so that the side-to-side total length of each
block is an even multiple of the largest expected cell length in all blocks in the
same direction. This ensures that the gridlines of regular (not stretched) cells in
neighboring blocks will line up exactly, minimizing interpolation error.

 Check the mesh and geometry resolution

 Check the mesh:
o Visually check the grid line matching between neighboring mesh blocks.
o Check Mesh Information for each block, especially total number of cells,
maximum adjacent cell size ratio, and maximum aspect ratios.
o Check Mesh Block Summary Information for total number of real cells in all
blocks. Check the following rules of thumb (for single-machine parallel
processing):
 cell counts in the hundreds will run in seconds
 cell counts in the thousands will run in minutes
  hundreds of thousands of cells run in hours or days
 millions of cells take days or weeks to run
 FAVORize the domain and check the geometry. It should appear well-resolved,
especially around gaps and thin sections.
 If there are multiple mesh blocks, preprocess the simulation and check Diagnostics >
Preprocessor Summary (prpout.xxx) for the search term ‘mismatch’ to find the
difference in resolved open area on either side of mesh block interfaces. The total error
should usually be much less than 5%.
 Adjust and check the mesh until the following criteria are satisfactory:
o geometry resolution
o resolution of flow sections
o total cell count (as a function of desired run time)
o grid line matching between neighboring blocks
o aspect ratios for single cell sides
o aspect ratios between neighboring cells (due to stretching)
o aspect ratios between cells in neighboring blocks
o Open area mismatch (from the preprocessor)
 Make note of the meshing methodology and rationale.
 If moving solids with prescribed motion are used, consider running the simulation
without fluid to check the movement definition.

 Define all necessary component properties

Follow the annotated diagram and update it where necessary.

 Go through the list of Component Properties for each component and identify the
available selections. Grayed-out selections require physics that have not been activated.
 Hold the mouse cursor over selection boxes to find the variable code for unfamiliar
available selections. Search the User Manual’s Input Variable Summary chapter for
the variable code to identify the definition, default value (if left blank) and units of the
selection.
 Note a rationale for each value entered or left at default setting. The goal is to be able
to explain each property value used by the selected physics of the simulation.
 Define mass source flow rates and properties and moving object motion types and
properties, if present.

 Add and define springs and ropes and valves and vents
 Spring/rope objects connect moving components to each other or to stationary anchor
points. Identify attachment locations for each spring/rope.
 Identify the spring/rope type to be used (see User Manual for details)
 Identify the Hookean spring coefficient for each.
 Give each spring/rope a descriptive name, and mark it on the diagram.
  Valves allow gas (aka “void”) bubble pressure to escape the domain in a realistic way.
Identify the location point of each. They must be in open areas, not solids.
 Estimate the valve loss coefficient and external pressure for each valve (see User
Manual).
 Give each valve a descriptive name, and mark it on the diagram.

 Add and define baffles and flux surfaces

Follow the annotated diagram to locate solid (flow-deflecting) baffles and measurement planes.

 Each baffle region has one definition and as many limiters and transformations as
necessary.
 Baffles will be moved to the nearest grid line during preprocessing. Try to place baffles
normal to the axes to avoid “stair-stepping”.
 Specify baffle porosity (0 = solid to 1 = transparent to flow).
 If the baffle is porous (0 < porosity < 1), specify baffle drag coefficients.
 Specify baffle heat transfer coefficients (when heat transfer physics are active)
 Activate the flux-surface option for baffles that will measure flow properties.
 If particle physics are activated, define particle-counting bins manually in the prepin file
namelist $PARTS (see User Manual > Input Variable Summary and Units chapter >
Particle Setup: Flux Surface and Sampling Volume Data section).

 Add and define history probes and sampling volumes

Follow the annotated diagram for locations to measure output variables and/or check known
variables.

 Give each probe and sampling volumes a descriptive name for post-processing.
 Define locations and coupled motion parameters for history probes.
 Define particle counting bins for sampling volumes (see flux surfaces section above).

 Add and define mass-momentum sources

Mass-momentum sources are boundary conditions located inside the mesh (as are mass
sources, valves, and vents). Use the annotated diagram as a guide.

 Specify each mass-momentum source’s shape. The source is defined around the origin,
with flow vector pointing toward z-max.
 Rotate the mass-momentum source to get the desired flow direction normal.
 Translate the mass-momentum source to the desired location in the domain.
 Specify source, sink, and flow rate properties.
 Specify motion properties only if the source moves during the simulation.

 Define all boundary conditions

Follow the annotated diagram. Check the following:
  Boundaries should usually be located far enough from the region of interest that
increasing the boundary distance further does not change the results in the region of
interest. This can be checked later with multiple versions of the same simulation set up.
 Boundary types should be appropriate for the situation. For example:
o Pressure boundaries set reference pressure and free surface elevation,
temperature, contaminant concentration, etc. These flow parameters are
communicated into the domain. In free-surface flows, Pressure boundaries are
appropriate where the Froude number Fr < 1. They do not specify velocity
(except for tangential components).
o Outflow boundaries do not communicate any information into the domain,
unless they unintentionally reflect angular momentum. Free-surface elevation
cannot be specified at Outflow boundaries, which are appropriate where Fr > 1.
o Velocity and Volumetric-flow-rate boundaries apply a uniform velocity below a
free surface elevation, and do not specify pressure.
o Grid-overlay boundaries are used to interpolate a flow solution from a previous
simulation as a spatially-varied, time-constant velocity profile. They do not
specify pressure.

 Add and define fluid and void/gas initial conditions

Initial conditions prescribe the solution everywhere in the domain at t = 0, and set up the entry-
values for all equations: flow velocity, heat transfer, electric potential, contaminant
concentration, etc.

 Follow the annotated diagram and specify all available initial conditions.
 If gravity physics are active, consider the hydrostatic pressure distribution option, which
will also apply to boundary conditions.
 Check that fluid initial conditions, boundary conditions, and component initial
conditions match. Check velocity, free-surface elevation, temperature, and turbulence.
 Check that the initial conditions are good approximations of reality.
 Consider adding initial fluid regions to minimize splashing on solid surfaces.
 Note the rationale for each initial condition, including default values.
On the Model Setup: Output tab
 Choose the basis for output frequency (time, fill, or solidified fraction)

 Choose additional output of interest

 Choose selected data to output more frequently

 Define output intervals for all data types

The default intervals are fractions of the finish time: for restart and long print
data and for selected, history, and short print data. Specified intervals are in
units of time between outputs.

On the Model Setup: Numerics tab

The numerical options in FLOW-3D are intended for advanced users and give significant control over
numerical methods used to solve the governing equations. When used improperly some of these
options can cause problems with the solution. In general, do not change these settings without
understanding what the adjustment does and having justification for the change. Some calibrations are
discussed in the General Project Workflow Checklist. See User Manual > Troubleshooting for more
information on adjusting numerical methods for specific problems.

Final Check after setup

 Check that the simulation pre-processes without crashing

 Check the pre-processed output in 3-D

 Do the fluid and geometry locations and resolutions “look right”?
 Are all specified baffles in the expected location? Are they acceptably resolved?

 Check the pre-processed output in 2-D slices

 Are fluid and geometry initialized in the correct locations?
 Is geometry well-resolved? Are there sliver cells?
 Are all initial fluid values (e.g., depth, velocity, pressure, density) as expected?
 Do initial fluid values match the boundary conditions?
 Will there be unnecessary splashing on solid surfaces?

Adjust the model setup if necessary and repeat the check. The preprocessed simulation results
should give you a good idea of how the run will start.

 Check the pre-processed text output files

 Were any errors noted in Diagnostics > Preprocessor Errors (prperr.xxx)?
  Check each section in Diagnostics > Preprocessor Summary (prpout.xxx). Some
commonly checked items are given for demonstration here:
o In section /OBS/, do specified component surface area and calculated
component surface area agree reasonably well? If not, consider adjusting the
mesh resolution.
o In section summary of open areas at inter-block boundaries, is the percent
difference less than 1% (ideal), 3% (typical) or 5% (usually the max acceptable)?
If not, consider adjusting the mesh resolution.
o In section /PARTS/, were any fluid history probes eliminated unexpectedly? If
yes, consider adjusting the history probe location.
o In section /PARTS/, is the total number of particles generated less than the
maximum number of particles requested (max possible)? Will the simulation run
out of available particles during the run?

Run the simulation

 Check solver text output and solver warnings

 What are the warning messages?
 Is the time step reasonable? Will the simulation finish in your lifetime?
 What solution is limiting the time step? Does the mentor suggest an implicit method?

If necessary, refer to troubleshooting.

Exercise 1: Running & Post-Processing an Example
 Water & Environment Training on FLOW-3D v11
Exercise 1: Running & Post-Processing an Example Simulation

Exercise 1: Running & Post-Processing an Example

®
In this exercise, you will learn how to use FLOW-3D and its native post-processor to:

1. Create a new workspace and add an example project file included with the software,

2. View the geometry and mesh,

3. Select and add parameters to be output more frequently than the default,

4. Preprocess and run the simulation and interpret the runtime output,

5. Load results files,

6. Understand the options for post-processing in 1-D, 2-D, and 3-D,

7. Reflect results around a plane of symmetry,

8. Make .avi videos and .bmp images of the results,

9. Make a simulation copy and perform a restart.

The example problem is of flow over a bench-scale sharp-crested weir. Only half of the weir is modeled,
and symmetry along the centerline of the weir is assumed.

Create a New Workspace

® ®
1. Launch FLOW-3D by double-clicking the FLOW-3D icon on your desktop.

2. On the Simulation Manager tab, create a new workspace by selecting File > New Workspace
from the menu at the top.

3. Enter the workspace name: Hydraulics Examples. Keep Create subdirectory using workspace
name checked so that the workspace and all its simulations will be under the same directory.

Page 1 of 24
Add an Example Simulation
1. Select File > Add Example… from the top menu or right-click the workspace name and select
Add Example… The dialog shown below will appear.

2. Select Flow Over a Weir from the alphabetically-ordered examples list and Open. Keep the
default name and Create subdirectory using simulation name option. Click OK to finish
importing.

3. The project has now been imported into the Hydraulics Examples workspace. Load the project
by double-clicking its name:
 Water & Environment Training on FLOW-3D v11
Exercise 1: Running & Post-Processing an Example Simulation

Description of the Primitive Geometry

The model is a very old validation case of a sharp-crested weir. It was built using FLOW-3D primitives
(simple shapes). Some of the shapes (subcomponents) are solids, and some are holes to cut away the
mitered edge. All the subcomponents are visible in this view, regardless of whether they are solids or
holes. Later we will view the model as the solver will see it, and it will appear as half of a sharp-crested
weir. Models are easier to visualize during setup using .stl geometry files, which will be covered in a
later exercise.

Mouse Modes & Viewing Options in Display Panes

4. Select the Model Setup tab and then the Meshing & Geometry tab below that. Locate the large
display window showing the model and domain illustrated below.

2. Left Button – Rotate. Click and hold the left-mouse button and move the mouse in the
Meshing_Geometry window. The model will rotate accordingly.

3. Middle Buttom or Wheel – Zoom. Click and hold the middle-mouse button or wheel while
moving the mouse vertically, or rotate the wheel, to zoom.

4. Right Button – Pan. Click and hold the right-mouse button and move the mouse in the window.
The model will move with the mouse.

5. Experiment with the buttons above the view pane. The first four
buttons reset the view. The last button toggles between orthographic and perspective views.

6. Experiment with probing surface coordinates with Shift+click. When you hold the Shift key a
black cross target appears (if it doesn’t try releasing and re-pressing Shift). When you click the
mouse button, the XYZ coordinate of the solid surface at that point appears in the lower-left
corner of the GUI. You can access a list of probed points by selecting View > Probe Points List… in
the menu above the view pane.

7. Experiment with setting a pivot point with Ctrl+click (or toggle the button) A gray lozenge
marks the pivot point, which can be hidden by toggling View > Hide (or Show) Pivot Point in the
menu above the view pane.
8. Close the Probe Points list and toggle the button to turn off the pivot point option. Click

the button to reset the view and then toggle the button until the x-axis points to the
right and the z-axis points up. You can toggle the axis and control its location by selecting the
options in View > Draw Axis on the toolbar menu above the view pane.

9. Global Transparency: The Global Transparency slider in the toolbar controls the transparency
of all objects in the display window.

5. Component- and Subcomponent-specific Transparency: these can be set in the Geometry tree

(toggle ). Right-click on a component name or open the Subcomponents list and right-click
on a subcomponent to adjust the transparency. Try out the transparency options.

6. Other View Options: experiment with the following. The button prints the current view,
which can be used with a .pdf printer program to generate images of the mesh and geometry
setup (hint: use the Print Screen keyboard button and a Paint program as an alternative). The

button allows you to draw a box and places a pivot point in its center before zooming to

the selection. The button fits the domain and geometry to the view window.

7. Mesh Display: It’s important to examine the mesh to assess resolution. It’s also important to
be able to turn it off to see the geometry. The mesh can be viewed from the Mesh menu item.
Check the Mesh > Show option so the mesh is displayed, and select Mesh > View Mode > Grid
Lines.
 Mesh Planes (user-specified grid lines) are displayed in a different color and/or weight than the
regular Grid Lines, which are generated automatically by FLOW-3D.

8. Color Options: Colors can be assigned to components, subcomponents, and mesh blocks. The
colors are to make identifying parts of the simulation easier when editing: they do not affect the
solution.

A. Open the Mesh tree ( to toggle) and expand the tree with the + icon.

B. Click on the color palette next to Mesh Block 1 and experiment with different grid line and
mesh plane colors.

C. Select Tools > Options from the menu above the display pane. On the Color tab, select the
Background color swatch and make the background white. Adjust your mesh colors if
necessary to be able to visualize both mesh planes and grid lines.
Assessing Mesh Resolution with FAVOR
One of the most important aspects of simulation setup is choosing an appropriate computational mesh.
If the mesh is too coarse the geometry and flow may not be resolved enough and the simulation will not
give good results. If the mesh is too fine the runtime may be unnecessarily long. The goal of mesh setup
is to use just enough cells to resolve the geometry and the flow features of interest.

There are two ways of judging how well a computational mesh resolves the geometry. One way is to run
the preprocessor, and this gives the most information short of running the simulation. A quicker way is
to FAVORize the geometry. FAVOR embeds the geometry and initialized fluid in the current
computational mesh using the same routines as the preprocessor.

1. Click the icon in the window toolbar. The FAVOR dialog will appear.

2. Check that Geometry Surface = Solid Volume and ISO Surface Value = 0.5 (more or less). Click
Render to view a gray iso-surface fitted to the solver-embedded solid geometry.

3. The checkboxes associated with individual Components allow all or some components to be
FAVORized. There is only one component, so ignore the checkboxes in the middle of the dialog.

4. At the bottom of the dialog, activate the Show Issues checkbox. Cells where FAVORize will have
known problems will be highlighted in orange. Check that there are no known issues when the
relative Tolerance = 0 and 100.

5. The component/subcomponent list at the bottom of the dialog lets you overlay the geometry
that was imported with iso-surfaces that are fit to the embedded geometry. First de-select
Geometry Surface 0 to turn off the FAVORized iso-surface. Then highlight Subcomponent 1 and
Subcomponent 3 to see the solid part of the geometry. The remaining subcomponents are holes
that cut bevels into the visible subcomponents. Now turn Geometry Surface 0 back on to see
the contrast between the embedded geometry iso-surface and the original defined
subcomponents. Play with Transparency and viewing the mesh Grid Lines while you’re looking
 at this view. Increasing mesh resolution will usually improve the match between iso-surface and
original geometry. It’s important to understand that the way the solver will “see” the embedded
geometry most accurately viewed in 2-D plots (which you’ll do later). The 3-D FAVORize view is
not precise because the iso-surface fitting is imperfect and approximate.

6. Close the dialog to return to the normal setup view.

The image above on the right shows the weir structure. The sharp crest of the weir is visible and it
appears to be adequately resolved. There is a rounding at the weir corner that could be examined in 2-D
plots to see if it’s an artifact of the iso-surface approximation or actually poor geometry resolution.
Preprocessing the Simulation
The FAVORize process gave you some good information, but there is more you want. 2-D plots are more
accurate than the 3-D rendering because they do not include the approximately fitted iso-surfaces
drawn by FAVORize and 3-D rendering. 2-D plots use cell-by-cell area and volume fractions to give the
precise surface locations computed by the solver.

1. Select Simulate > Preprocess Simulation > Local from the top menu bar.

2. The interface will prompt you to Save the FAVORize selections you made (click Yes) and will
then switch back to the Simulation Manager and the preprocessor will run. It should complete
within a few seconds, and a message will appear in the Solver text indicating “Preprocessor
Done”.

3. In the Solver text, scroll up and find the total, fluid, and solid sub-domain cell counts. The total
and fluid counts are both around 50,000 cells. This is important information, since the number
of active cells directly relates to the simulation run time.

Loading a Results File

1. The preprocessor generates a results file named prpgrf.”project name”. To load the results file,
select the Analyze tab.

2. If no results file is currently loaded you will be prompted to select a file. Keep the default
Custom option. Custom file types are flsgrf.xxx (created by the solver during a run) and
prpgrf.xxx (created by the preprocessor) that require you to select output options yourself.
Existing file types are flsplt.xxx and prpplt.xxx which contain non-alterable plots that are pre-
defined in the prepin file (i.e., the simulation instructions). The prp prefix means a file is
preprocessor output, and the fls prefix indicates flow solver output.

3. Select prpgrf.Flow_Over_A_Weir in the dialog box, and OK to load the file.

Viewing Preprocessed Geometry and Initial Fluid Configuration in 3-D and 2-D
The Analyze panel will now be displayed. Although all sub-tabs are available, typically only 2-D and 3-D
plots are necessary to check the model setup.
The 3-D subtab will be displayed initially when the Analyze tab is selected. First, you’ll generate the
same display that was generated by the FAVORize function.

1. On the Analyze > 3-D tab, find the Iso-surface dropdown. Select Complement of Volume
Fraction so the surface that is plotted is the interface between solid geometry and open space.

2. Find the Color Variable dropdown, and select None. You do not need to color the surface by a
flow parameter right now.

3. Select Render in the lower right corner of the screen. The tab at top will change to the Display
tab, and will show the same image as you saw in the FAVORize display.

The next step will be to generate a 2-D plot along the weir centerline to show the initial fluid pressure
and location with the mesh.

4. Select the Analyze > 2-D tab.

5. Select the X-Z radio button in the Plane group. Note that the sliders in the center of the screen
reset so that the full extents of the X and Z domain is plotted, and the Y sliders come together
so that by default only one slice is displayed (at y = 5.25, the geometric center of the y-domain).

6. Move both of the Y-direction sliders to the left-most position. The numbers on either side of
the slider show you the Cartesian coordinate of the slider. We want both sliders to be at Y =
0.25, the cell center of the first cell inside the domain. Note that the cell number J = 2. This
indicates that the first cell in the domain is actually the second cell in the mesh block created by
 the solver: the first cell at J = 1 is a ghost cell (outside of the domain) used to compute boundary
conditions.

7. Select the Mesh checkbox to overlay the mesh on the results.

8. Click Render in the lower left to generate the graphics. You will see the image shown below.

Zoom in and out with the mouse roller and pan with the right mouse button. The sharp crest of the weir
indicates the mesh resolution is probably adequate for capturing the important geometric features, and
that the rounded edges in the FAVORize view were due to poor iso-surface fitting. The initial fluid
pressure (the color variable) and velocity (the vectors) are shown. They are important for checking the
setup. If other flow quantities such as density or scalar concentration had also been initialized, they
could be checked by selecting them in the Contour Variable dropdown list back on the Analyze > 2-D
tab.

Requesting Additional Output Detail from the Solver

Before running the simulation, there are a few details that should be checked and added:

 Check the Simulation Units are specified to allow labels on output plots
 Request Hydraulic Data (includes Froude number, fluid depth, free-surface elevation,
near-bottom and depth-averaged velocity)
 Specify Selected Data output for more frequent plots to see details and make movies

1. Check the Simulation Units: Select the Model Setup > General tab. On the right-hand side you
will see a group box named Units. CGS (centimeters, grams, seconds) is selected for the
Simulation Units. There is no Temperature Unit, and none are needed since there are no Heat
Transfer physics models activated on the Physics tab.

2. Request Hydraulic Data: Select the Model Setup > Output tab. At the right, under Additional
Output, select the checkbox for Hydraulic Data as shown below. This will cause the fluid
elevation, fluid depth, and Froude number to be computed and stored in the results file. These
data are not computed unless this option is selected because they are secondary parameters
derived from other values.
 3. Specify Selected Data output: Selected data are user-specified output parameters that are
written to the results more frequently than Restart data. By default Selected data and Restart
th th
data are output every 1/100 and 1/10 of the simulation finish time, respectively. Selected
data is useful for creating smooth animations and seeing detailed flow patterns develop and
change.

On the Model Setup > Output tab, select the following Selected data: Fluid Fraction, Fluid
Velocities, Hydraulic Data (which is available now because you selected it as Additional
Output), and Pressure.

Rule of Thumb

Only specified selected data variables will be written to the results file. If you decide later
that you need a variable which was not specified in the Selected Data list, the simulation
will need to be re-run to get the output you need! On the other hand, too many selected
outputs will increase the size of your results file. It’s usually better to select too many than
not enough.

Running the Simulation

1. Save the simulation from the File menu at the top. Note before you save that there is an
asterisk following the simulation title at the very top title bar of the screen. The asterisk means
there are unsaved changes you’ve made to the simulation.

2. Start the simulation by selecting Simulate > Run Simulation > Local from the menu bar at top.
The view in the interface will automatically switch to the Simulation Manager tab and the
preprocessor will start, followed by the runtime solver. Simulations that are running or waiting

to run can be paused , terminated or rearranged in the queue in the lower left window:
Checking and Diagnosing the Running Simulation
The Simulation Manager tab can be thought of as a dashboard for the run. The efficiency and accuracy
of the solution are indicated by the runtime plots and solver text messages. Familiarize yourself with
the various runtime diagnostic plots available in the dropdown list above the plot window:

th
The plots are generated at every History Data output step, which is, by default, 1/100 of the simulation
finish time. All the data shown here is also available as General History data in the output flsgrf file. You
can draw a box to zoom in on the plots by holding the left mouse button and dragging a zoom window.
Use Ctrl+left-click to zoom back out.

1. Stability limit & time step size: Compares the time step stability limit (the smallest time step
required in any cell to meet any explicit solver that is active) and dt, the actual time step being
 used. Ideally the time step dt is the same as the stability limit but it may be smaller when there
are too many pressure iterations or a stability limit is unexpectedly exceeded (e.g., due to
splashing).

2. Time-step size: A simple plot of the dynamically-selected time step over time.

3. Epsi & maximum pressure residual: Epsi represents the pressure iteration convergence criteria
that the solver uses to determine if the pressure solution is accurate. At every time step,
pressures and velocities are solved in all cells with fluid until the residual Ū (a measure of
potential error) in each cell is less than some small cutoff value epsi. The max residual is for the
cell with the maximum value of Ū after the pressure solution has either converged or reached
the maximum allowed number of iterations. The pressure iteration fails if the max residual is
larger than epsi after the maximum number of pressure solver iterations, otherwise it is
converged.

4. Pressure convergence ratio: the ratio of maximum residual to epsi. Less than one means the
pressure solution converged, greater than one means it did not. An “ideal” solution maintains
res/epsi less than one at all times. A few iteration failures are unlikely to damage the simulation
accuracy, especially when they are isolated or occur at the beginning of the run. When the
maximum residual is much lower than epsi, it means that the solution is converging very quickly
(the solution is “easy”). When it is close to epsi, it implies that the convergence criteria selected
is an efficient choice. Non-default options for controlling the pressure solution are rarely
needed or desirable, but are available on the Model Setup > Numerics tab and will be discussed
in a later lecture.

5. Pressure iteration count: The number of iterations of the pressure solver. Different pressure
solvers have different “best” values. In general, less than ten is a desirable value.

6. Convective volume error (% lost) and Multi-block volume error (% lost): Represents the
amount of fluid gained (negative value) or lost (positive value) due to advection errors or inter-
block interpolation errors, in percent. Usually much less than 1%, and should always be less
than 3%.

7. Fill Fraction, Volume of fluid 1, and Fluid 1 surface area: self-explanatory. In this simulation,
3 2
the units of volume are cm and surface area is in cm . Constant fluid volume and surface area
are two indicators that the simulation has reached steady state. Fill fraction is the dimensionless
ratio of fluid volume to total open (non-solid) volume in all blocks of the domain.

8. Mass-averaged mean kinetic energy and Mass-averaged mean turbulent energy: the domain-
averaged, time-mean, kinetic and turbulent kinetic energies. These are two more indicators of
the steadiness of the flow. Average TKE is not shown in this case because the flow is being
modeled as laminar according to the selections on the Model Setup > Physics tab.

9. Other plots are available depending on the physics models used: these include thermal energy,
number of particles, and other output.

Take a few moments to look over the available runtime plots for the simulation. See if they make sense
to you.
Also look at the solver text output results at the bottom of the screen. This is a copy of the hd3msg file
that the solver writes to the simulation directory while it runs. Even if the user interface (GUI) crashes,
the solver will continue to run and write all the important output files. The text output lists a summary
of the simulation at each Short Print Data Interval. Each line gives:

1. output time t (usually in seconds, refers to time elapsed in the model)

2. output cycle (the number of time steps solved to date, including the current time step),

3. actual time step delt used during the cycle (less than or equal to dtstbl),

4. limiting time step required for stability by any explicit solver process dtstbl,

5. solver process that requires the limiting time step, as a two-letter code (cx = x-
direction convective fluid flux, for example)

6. number of pressure iterations during the cycle iter,

7. the ratio res/epsi of pressure solution maximum residual (any cell) to convergence criteria epsi

8. volume of fluid in the simulation domain,

9. the known %loss due to convective flux error and inter-block interpolation error

10. the fraction of the open domain volume that is filled with fluid frac,

11. elapsed real-world time and computer clock time el_time and clk_time (hh:mm:ss),

12. multi-processor efficiency ratio %PE, as percent

Loading Results While the Solver is Running

The results of a simulation can be viewed while the simulation is running or after the simulation is
complete. It is often useful to visualize the results early to ensure that it is running correctly.

1. Select the Analyze tab. A message appears indicating that the prpgrf file no longer exists. The
prpgrf file was generated during preprocessing and was deleted to save space when the
simulation ran. The flsgrf file is the simulation output file and contains the same information as
the prpgrf file, plus the output of the run. Select Continue and a file selection dialog will be
presented.

2. Select Custom output files and the flsgrf file, and then select OK.

Understanding Plot Types and Data Sources

The Analyze tab will now be displayed. There a number of ways in which to visualize the results of a
simulation. The available plot types are:
 Custom: a rarely-used way to get “under the hood” of the post-processor. Text commands used for
this tab are given in the User Manual chapter on Post-Processing.

Probe: Displays values for individual cells as well as non-cell data like integrated global values,
boundary condition values, and component-specific integrated output. Plots charts or text of values
vs. time.

1-D: Flow parameters and other data can be viewed along a line of cells in the X, Y, or Z direction.
Extents (limits) can be applied both in space and in time.

2-D: Results can be viewed in an X-Y, Y-Z, or X-Z plane. Plot limits can be applied both spatially and
in time. Solids are always displayed, and fluid is colored by selected flow parameters. Velocity
vectors and particles (if present) can be displayed. Solid surfaces and fluid free surfaces are drawn
using all available data, and are more accurate than in 3-D plots.

3-D: Both fluid and solid surfaces can be visualized together or separately. The fluid surface can be
colored by flow quantities, which show the state at that surface. Additional information such as
velocity vectors, particles (if present), and streamlines can be added. Plot limits can be applied both
spatially and in time. The surfaces are drawn using a 3-D iso-surface method that does not locate
them as accurately as in 2-D plots.

Text Output: Restart, Selected, and Solidification data can be written to text files for every cell
within selectable extents in space and time. Use this tab to get many cell-by-cell numerical values. If
just a few locations are of interest, it is easier to place History Probes before the run and get their
output from the Probe tab.

Neutral File: Restart and Selected Data can be interpolated to give results at a set of user-defined
points. This output is useful for plotting time-varied results along a flow path that curves within the
mesh.

FSI/TSE: Only used for visualizing deforming solids output from the finite-element stress-analysis
physics package.

Once a plot type (tab) is selected, the next step is to choose the data source. There are five sources of
data in FLOW-3D:

Restart: All cell-by-cell flow variables necessary for a complete solution, plus any Additional Output.
th
Output frequency = 1/10 of the finish time by default.
th
Selected: User-selected cell-by-cell flow variables. Output frequency = 1/100 of the finish time by
default.

General History: Non-cell-specific data. Only available on the Probe sub-tab. Includes global output
like time step size, epsi, and mean kinetic energy. Also includes output for moving objects, history
probes, flux planes (measurement windows), control volumes, history particles, and more.

Mesh Dependent History: Only available on the Probe sub-tab. Includes boundary condition
information like flow rates.
 Solidification: Only available if the solidification model is active.

Examples of some of the available plot types will be generated in the next step.

Plotting 3-D Graphics

Select the Analyze > 3-D tab. There are 4 choices which always need to be made on this tab.

1. Iso-surface: The surface (interface) to draw in three dimensions. Select Fraction of Fluid to plot
the fluid free surface.

2. Color variable: The selection here will determine what variable to color the fluid surface with.
Select Pressure.

3. Iso-surface options > Component Iso-Surface Overlay: options for plotting the interface
(surface) of solids. Select Solid Volume to plot the solid components (weir).

4. Time: Click and drag the Left (Min) Time Frame Slider so that Min = 0 and Max = 1.25 seconds.

5. Click the Render button to generate a series of 11 plots between t = 0.0 and 1.25 seconds which
contain the weir and fluid surfaces, with the fluid surface colored by pressure. If the geometry
 isn’t visible, use to reset the view. Step through the images by selecting time steps from
the list at the right. The first and last output times should be similar to these:

6. Return to the Analyze > 3-D tab. Choose Selected data as the Data Source to access the more
frequent data specified earlier in the exercise.

7. Set the Min Time Slider all the way to the left (t=0.0)

8. Now click the Render button. The view will switch to the Display tab and 101 plots will be listed
in the Available Time Frames list. Click Next to advance through the frames.

Reflecting Results Across a Symmetry Boundary

The simulation was set up with a symmetry plane as a boundary condition along the centerline of the
weir. Only half of the weir structure is being modeled and only that half is displayed. For presentations
it would be helpful to show both halves of the weir and explain the assumption of symmetric flow.

1. Return to the Analyze > 3-D tab and select the Open Symmetry Boundaries checkbox.

2. Click Render. The fluid surface should now appear open (transparent with sharp edges) at the y-
min symmetry boundary, which is at y = 0 cm.
 3. To mirror the results across the y=0 plane, select Tools > Symmetry from the menu above the
display and select the Y Direction checkbox in the dialog. Click Apply and Close to exit the
dialog. The display show a full weir structure as shown below.

Creating a 3-D Animation

The next step will be to create an animation of the 3-D fluid surface. Animations are movies created
from the frames in the Available Time Frames list. To improve the visual effect of animations, it is
recommended that a common color scale be applied to all frames.

1. Return to the Analyze > 3-D tab and select the Global radio buttons under Contour Limits.

2. Make sure you have selected all time steps with the Time Frame Sliders and click Render.

3. Again, select Tools > Symmetry > Y-Direction to mirror the results across the Y=0 plane. Select
Apply and Close. Pan (using the right mouse button) until the weir is just below and to the right
of the color bar.

4. To create the animation, select Tools > Animation > Rubberband Capture. Instructions will
appear to explain the next step; select OK to continue.
5. Click and hold the left mouse button somewhere to the upper-left of the weir and color bar, and
drag the selection box that appears so it encloses the weir and color bar. Repeat the process if
necessary. Click the Capture button to accept the selection.

6. A dialog will appear allowing the animation to be named. The default name for animations is
out.avi. A more descriptive name is always recommended. The default frame rate is 10 frames
per second. This may be too fast for some simulations, so enter frame rate = 5. Keep the option
to delete the individual frame images after they’ve been combined into a movie.

7. Each time frame will be rendered to the Display window and bitmaps will be written (and then
deleted) in the simulation directory. The camera view may wobble a little during this process:
this is a known issue and will be addressed in a later exercise on using FlowSight post-
processing. Once the rendering is complete, the following dialog will appear. Click OK to begin
the next step of the process. The Video Compression dialog will open. Note that the message
below stays open: do not click OK on it again.
 8. The default compression for animations is Full Frames (Uncompressed) and is not
recommended since the file size can be quite large and difficult to load in video players and slide
show presentations. Select Microsoft Video 1 since the codec is universally available on
Windows platforms. Unselect the Data Rate checkbox so that the quality of animations is not
limited by the data rate.

9. Click OK to begin the compression process. When the compression is complete, the following
dialog will appear. Click OK to continue.

10. To play the animation, use the operating system to browse to the directory where the
simulation is located. You can do this quickly by going to the Simulation Manager tab and
clicking on the path listed for Simulation Input File. Find the .avi file you created and play it by
double-clicking on it.

Plotting Options in 2-D

1. Go to the Analyze > 2-D tab.

2. Choose Selected data. Note that the Particle Type option grays out because there is no particle
data in the Selected data set (it wasn’t activated on the Output tab).

3. The most useful plane to view results for this simulation is the centerline of the weir. Choose
the X-Z radio button in the Plane group and set both of the Y Limits sliders = 0.25 cm, which is
 the location of the cell centers closest to the centerline at Y = 0. Set the Time Frame sliders to
include all output times.

4. Select Contour Variable = Velocity Magnitude. Geometry is always displayed on 2-D plots, so
nothing needs to be selected to view it.

5. Click Render to generate a time sequence of 2-D plots of pressure in the plane. Graphics similar
to following will appear on the Display tab.

6. Make an overlay of the free surface with time: select Overlay from the dropdown plot type list
in the upper right:

Click on several of the early time steps to make an overlay plot:

 7. Return to Single plots and click on the Format button in the upper right-hand corner.

Experiment with the options there, especially the Vector Size options. Click Apply to change the display
without closing the dialog. Take some time and experiment. When you are done, select Reset, then
change the Background to White, and then click OK to close the dialog.

Plotting Time History Results in Probe

1. Select the Analyze > Probe tab. Time history plots of variables are created here. There are four
types of time-dependent data that you’re likely to use. Solidification data is not discussed here.

A. Restart data and Selected data: cell-specific values of flow variables, plotted for a single
spatial point over time when using Probe history.

B. General history data: non-cell-specific quantities which vary with time. Typical examples
are global quantities and component-specific quantities. Many General History variables are
integrations of cell-by-cell values. Examples include integrations over the whole domain,
over a solid surface, or over the plane of a baffle.
 C. Mesh dependent data: time-dependent parameters computed at mesh boundaries. Typical
quantities are flow rate and fluid height. Note that some parameters only report the
specified value (like fluid height), while others report the computed value (like flow rate).
You can usually tell the difference because computed values fluctuate, while specified
values are constant.

2. Select the General History radio button under Data Source. Note that the Data Point (location)
sliders gray out because the values for this source are global.

3. Select Mass-Averaged Fluid Mean Kinetic Energy from the list and Render. You can use the
mouse to zoom and pan.

4. Return to the Analyze > Probe tab and select Text as the output form. Then Render again.

5. The text output can be written to a file by selecting the Save As button (don’t forget to name
the file). Click Continue to close the text output dialog.

Plotting Text Output

Text Output (on the Analyze > Text Output tab) is very similar to the text which can be output from the
Probe tab except that output for many cells in a 3-D rectangle can be output from Text Output, while
only one cell at a time can be output from Probe spatial data. Only Restart and Selected data can be
output as Text Output. The default spatial extents are set to the entire domain, and the default time
extents are the last time step only. The spatial and time extents should be selected with care as the text
files can become very large.

1. Experiment with outputting a single time step. Plot Free Surface Elevation and Fluid Depth. You
can save the text file with Save As. Click Continue to close the text output dialog.

2. Experiment with outputting two or more time steps. Note the section breaks between time
steps: these often must be removed outside of FLOW-3D for additional post-processing.
Exercise 2: Setting Up Geometry
 Water & Environment Training on FLOW-3D v11
Exercise 2: Setting Up Geometry

Exercise 2: Setting Up Geometry

Problem Description & Objectives

The next three exercises focus on building a sharp-crested weir simulation. An ellipsoidal obstacle lying
on its side will be embedded in the channel downstream of the weir. Topography will be added to the
downstream channel bottom. This setup will be used in future exercises to simulate flow over the weir
as well as outfall from pipes at the base of the weir.

Learning Objectives
In this exercise you will learn techniques to:

1. Check and fix stereolithography (.stl) geometry files using netfabb Basic and pyAdmesh,
2. Add .stl geometry to a simulation,
3. Add simple geometry to a simulation using "primitives",
4. Create holes in geometry (Boolean operations),
5. Transform geometry (rotation, translation, magnification).

Page 1 of 16
Create a New (Blank) Simulation File

1. Launch the FLOW-3D interface by double-clicking the FLOW-3D icon on your desktop.

2. If you have not done so already, create a workspace for this training titled Hydraulics Examples
following the instructions in Exercise 1, p.1.

3. Go to the Simulation Manager tab.

4. Select (click to highlight) the workspace Hydraulics Examples in the Portfolio.

5. Create a new simulation named Weir Geometry.

a. Select File > Add New Simulation… from the dropdown menu at the top of the screen.
b. Enter Weir Geometry as the simulation name.
c. Leave Create Subdirectory Using Simulation Name checked.
d. Select OK.

Check & Fix the .stl File

The starting point for developing a new simulation is making the geometry. Simple geometry can be
build with primitives in FLOW-3D. Complex geometry must be imported in stereolithography (.stl)
format. You will use .stl files in this exercise for a weir plate and some simple downstream topography.

.stl files are exported from computer-aided design (CAD) programs and contain a closed surface of
triangular facets. Each triangle (facet) which makes up the surface is described by three Cartesian
vertices and a normal vector pointing outward from the enclosed volume. In the author’s experience, all
CAD packages always add errors and artifacts during .stl creation. Simple errors can be fixed quickly with
.stl repair tools. Systemic errors require correcting and re-exporting the file from CAD. Typical simple
errors are missing facets (holes), zero-area facets, non-manifold vertices, and reversed or incorrect unit
normals. Geometry files must always be checked for errors and errors must be repaired. netfabb
Basic, MeshLab, and qAdmesh (which comes with FLOW-3D) are free tools for repairing .stl files.

1. Install and/or launch netfabb Basic (http://www.netfabb.com/basic.php).

2. Remember from Exercise 1 how to open the simulation directory from inside the Simulation
Manager tab? Do so, and copy the geometry files weir1.stl and Topo_Example_04_fixed.stl to
the new Weir Geometry simulation folder. It’s recommended (but not required except in this
exercise) that you keep your .stl files in your project folder.

3. Check the .stl file using netfabb Basic:

a. Open Windows Start Menu > All Programs > netfabb Basic > netfabb Basic.
b. Select Part > Add and navigate to Libraries > Documents > FLOW-3D > FLOW-3D Projects >
Hydraulics_Examples > Weir_Geometry > Weir1.stl. If you have trouble finding the file, ask
your instructor for help.
 c. The image in netfabb Basic indicates that there are errors. Select Extras > New
Analysis > Standard Analysis. A new branch of the tree appears in the Context Area at the
right. You can switch between views and information by clicking on different branches of
the tree.
d. Read the analysis in the Context Area and find where it reports that: (a) the surface is not
closed, and (b) there is 1 hole with 3 boundary edges and 0 bad edges.
e. Check the volume and surface area of the part. Note that mm just m eans ‘unit o f length’ :
most .stl tools assume you are making .stl files for 3-D printing.
f. Rotate (right mouse button) and pan (center mouse button/wheel) the view until the front
face of the weir is visible. Note the red facet on the left side of the crest. This is the missing
facet.

4. Repair the .stl file using netfabb Basic.

a. Select Extras > Repair Part. Select Automatic Repair at the bottom of the Context Area.
Choose Default Repair and then Execute.
b. Click Update and note how the number of holes is changed to zero.
c. Click Apply Repair and then choose to Remove Old Part.
d. Don’t use the Save option. Instead, choose Part > Export Part > as STL. Name the file
weir1_repaired.stl, and make sure it’s being saved into the Weir Geometry project folder.
Click Save.
e. Repeat the process for Topo_Example_04_fixed.stl. There should be no errors in that file.
You can load multiple parts at once and toggle between them using the tree to check that they are
oriented correctly and in the same coordinate system. There are many other features of netfabb
Basic that you can explore on your own.

A second .stl tool, called qAdmesh, is included with FLOW-3D. The default mode of operation for
qAdmesh is to check a file and report any errors found, without fixing the errors. qAdmesh fixes normals
more precisely than netfabb Basic, but generally does not fill holes as well. It is recommended that you
use qAdmesh after netfabb Basic as a final check.

5. Re-check and repair the .stl file using qADMesh:

a. In FLOW-3D, go to the Model Setup > Meshing & Geometry tab.
b. Select Tools > qAdmesh.
c. Select Browse (next to the Input box) and then find your local copy of weir1_repaired.stl.
The simulation folder should already be open in Windows; look at the address bar at the top
of that window to guide your browsing. Select Open when you’ve selected
weir1_repaired.stl.
d. Under Output, select the Binary STL radio button. The Output location box is now
activated.
e. Change the name of the output file (at the end of the path) to weir1_repaired2.stl.
f. Click Apply to Default check (and fix) the file.
g. Check the Messages pane and carefully review the details of the repair.
The information in the Messages pane indicates the results. The status listed under Final indicates no
errors, so the file has been fixed.

Under some conditions, .stl tools may have difficulty fixing a file. If multiple repairs do not solve the
problems then re-export the .stl file from CAD. Different .stl tools use different repair algorithms and
may produce different results. It’s recommended that you repeat the process of checking and fixing files
until both netfabb Basic and qAdmesh report no errors.

If an .stl file is larger than about 420 MB, neither software will be able to open it. In that case use a 64-
bit software like MeshLab or a professional version of netfabb. The use of MeshLab is discussed in a
separate exercise on topography.

Import the Fixed Weir .stl File

1. Go to the Model Setup > Meshing & Geometry tab.

2. Select Subcomponent > Geometry File(s) from the dropdown menu above the display pane.
a. Add to Component = New Component (1) since there are currently no components.
b. Click the Add button.
c. Select weir1_repaired2.stl and click Open.
d. Set Subcomponent Name = weir plate and then click OK.
e. Give the new Component 1 Name = weir and select OK to finish adding it as a Solid
component.

3. Rotate (left mouse button), zoom (mouse wheel), and pan (right mouse button) the camera to
view the weir.

4. Save the simulation.

Determine Weir Subcomponent Dimensions and Unit System

Sharp-crested rectangular weirs are a relatively well understood method of measuring flow in channels.
They measure flow accurately only when certain approach and downstream conditions are met and
when their construction follows established guidelines. You need to know the dimensions of the plate
itself, the width and elevation of the crest notch, and the angle of the blade edge to guarantee proper
rd
flow measurement. The US Bureau of Reclamation (USBR) Water Measurement Manual (3 edition
rd
revised 2001) and Water Measurement Structures (3 edition revised 1989) by M.G. Bos are guides. Now
you will use FLOW-3D to find the dimensions of the weir, although you could do so more easily back in
netfabb Basic.

1. Toggle the Geometry tree using the button.

2. Expand tree branches to open Geometry > Component 1: weir > Subcomponents >
Subcomponent 1: weir plate > Min/Max. Examine the Extents, Min, and Max coordinates of
the weir subcomponent. The weir is 40 units wide (centerline at y=0), 17 units tall (base at z = 0),
and about 1.1 units wide (upstream face at x = -0.1).
3. Toggle the button until the y-axis vector points to the left.

4. Select View > Draw Axis > Origin from the menu above the display to set the axis at the origin.

5. Set the Transparency slider all the way to the left so the weir is opaque.

6. Zoom and pan the view so the crest fills the screen as shown below. Do not rotate the view.
7. Select View > Probe Points List … from the display pane menu.

8. Left-click the mouse button and then press and hold Shift. Repeat if necessary until the mouse
arrow becomes a cross-hairs when you hold Shift.

9. Carefully Shift-click on the edges of the notch at the points shown.

a. No point is recorded when you click on blank space.

b. Take multiple readings to estimate the uncertainty of your mouse location.
c. Zoom in and pan if necessary to get more precise readings (rotating is not recommended).
d. The most recent point probe value is also shown in the bottom left margin of the FLOW-
3D
window. When the probed point is in empty space, the X-value displayed there is an
estimate (around -42) of the domain extent.

10. Use your measurements to find the edge coordinates of the crest notch. The notch is open from
y = -5 to 5 (10 units wide), and begins at z = 12 and continues to the top of the plate at z = 17 (5
units tall).

11. The list remains populated until you select Clear Probe Points. Do this now.

12. Click on the icon to switch to a profile view from the side. Pan and zoom to view the weir.
13. Use the Transparency slider to increase transparency until you can see the blade edge. You may
see two, in which case the one of interest is the lower one.

14. Repeat the measuring process to find the coordinates of the blade endpoints.

15. Find the blade angle using trigonometry:

a. tan θ = ∆x/∆z = (0.995378 + 0.0970524)/(11.9914 – 10.3757) = 0.676.
-1
b. Tan (0.676) = 34°, the internal angle of the blade.
c. Subtract the internal angle from 90° to find the external angle = ~ 66°.
d. The external angle must be at least 45° as shown in the schematic below.

16. Zoom in very close to the upper edge of the blade and identify the flat crest bevel. Bos (1989)
and others recommend bevel thickness between 1 and 2 mm (0.03 to 0.08 feet). A series of
closely-zoomed and careful measurements show that the blade crest is almost exactly 0.08 units
wide in the x direction. This strongly suggests that the .stl file was made in units of feet.
However, the total thickness of the plate is about 1.1 feet, which is much wider than
recommended for sharp-crested weirs. The total thickness may be a design error.

Schematic enlargement of rectangular sharp-crested weir (Bos 1989)

Scale Weir Subcomponent From Feet to Meters
1. Clear and close the probe points dialog. Open the Geometry > Component 1: weir >
Subcomponents > Subcomponent 1: weir plate > Transformations > Magnifications branch of
the tree in the Geometry Window.

2. Set Global = 0.3048 to scale the subcomponent to meters.

3. Check in the Min/Max tree for the subcomponent as before: the new extents show the scaling.

4. Save your work.

Specify the Simulation Units as SI
1. Go to the Model Setup > General tab.

2. The default Simulation Units are Custom, which means they are unspecified (units can be
anything, as long as the system is consistent for all parameters).
a. Change the Simulation Units to SI.
b. When prompted for the Temperature Unit, select Kelvin. This selection is unimportant since
there will be no heat transfer in this simulation.

3. Save your work, and return to the Model Setup > Meshing & Geometry tab.

Add the Upstream Channel as Primitive Boxes

The upstream channel will be added using box primitives. The floor will be defined according to the
sketch drawn below and using the dimensions determined from the Min/Max Y coordinates and the
Min X coordinate of the weir subcomponent found in the tree.
1. Select Subcomponent > Box from the view pane menu.

2. Enter the values shown below in the dialogs and click OK to accept the values. Note that the
floor will be part of a new subcomponent, so it can have a different surface roughness than the
weir. Name the subcomponent channel bed and the new component channel.

3. The Geometry tree now has two components, each with one subcomponent. Note that the new
box has been added as Subcomponent 2: the subcomponents are numbered in the order they
appear in the tree, regardless of which component they are part of. Next you will add the walls.
4. Add the first wall as another solid box like before but with the following specifications:
a. Add the box to Component 2: channel.
b. X low = -31, X high = 0.3048.
c. Z low = 0, Z high = 6.
d. Y low = -4, Y high = -3.

5. Expand the Min/Max tree under Subcomponent 3 and verify that the values match what the
input above. If there are any mistakes, fix them with the following steps:
a. To adjust the X,Y, and Z extents, change the values under the Limiters branch.
b. To associate a subcomponent with a different component, right-click the subcomponent
name and select Reassign.
c. Locate these two options now for future reference.

6. Double-click the wall subcomponent in the view to select it (it will turn green). Right-click the
selected subcomponent to bring up a context menu. The title of the menu should be
Subcomponent 3.
a. Select Copy…
b. Make sure that Add to Subcomponent = 2: channel.
c. Set Position = As Last Subcomponent.
d. Click Transform, and set Translate Y = 7.
e. Click OK, check that you’re adding the new wall to Subcomponent 2, and then OK again.
f. Double click anywhere in blank space to deselect the wall.

7. Save your work.

Import the Downstream Bed .stl File
1. Practice the .stl import process by importing Topo_Example_04_fixed.stl as follows:
a. Make it a solid Subcomponent 5 named topo.
b. Make it part of a new solid Component 3, also named topo.

2. Check the Geometry tree to make sure the topography is part of Component 3. If it is, it will also
be a unique color in the display. If it’s not, right-click the topo subcomponent and Reassign it to
a new component.

3. Save your work so far.

Add Subcomponent Holes to Represent Pipe Outfalls
We will now add the final pieces of the geometry by creating two cylindrical holes in the weir face.
These holes will represent outflow pipes; you will specify the flow rate later using Mass/momentum
sources. The sources will add flow without needing to extend the pipes to a domain boundary, so you
don’t need to make the pipes extend very far into the weir and upstream floor.

Tip
Holes replace solids in their own component. Holes can only be made as subcomponents.
They remove volume from other subcomponents that come before them in their component
tree. Use more than one hole subcomponent to cut through more than one component.

1. Add the hole subcomponent.

a. Select Subcomponent > Cylinder from the menu above the display.
b. Enter the values shown below, including Add to Component. Cylinders are defined around
the z-axis. The next step will be to rotate and move it.

2. Check your work.

a. Use the Transparency slider above the view pane to set transparency to about 75% so you
can see the cylinder.
b. Locate the new Subcomponent 2 added to Component 1 in the Geometry tree.
c. Check that the Subcomponent Type = Hole. Change the type if you made a mistake.

3. Rotate and move the cylinder.

a. Expand the Transformation branch and then the Rotations and Translations sections.
b. Enter the values shown below to rotate and translate the pipe.
4. Check your work. If you made a mistake defining the cylinder size, use the Limiters branch of
tree to fix it. If you made mistakes rotating or translating, use the Transformations branch.

5. Save your work so far.

6. Create the second hole as a copy of the first one:

a. Right-click the name Subcomponent 2: Outfall 1 in the tree and select Copy…
b. Name the new subcomponent LH hole.
c. Add it to component 1: weir.
d. Assign the Position: As Last Subcomponent.
e. Make sure the Subcomponent Type is Hole.
f. Select OK to create the subcomponent copy.

7. Practice finding the Translation branch of the new subcomponent. Change the Y Translation
from negative to positive 1. Check your work against the image below.

8. The holes will only be “drilled” through the weir plate (Component 1). Now re-do the holes to
drill the rest of the way into the bed (Component 2).
a. Practice: make a Copy of both holes by right-clicking either the tree or the shape.
b. Set Add to Component = 2: channel for both copies.
c. Set Position = As Last Subcomponent for both copies.
d. Check that you have 9 subcomponents in 3 components, and that #2, #3, #7, and #8 are the
holes, as shown below. Use right-click Reassign if necessary to match the picture.

9. Save your work again.

Exercise 3: Meshing & Boundary Conditions
 Water & Environment Training on FLOW-3D v11
Exercise 3: Meshing & Boundary Conditions

Exercise 3: Meshing & Boundary Conditions

Problem Description & Objectives

In this exercise a computational grid will be created with boundary conditions. The grid (mesh) defines
the domain: the region in which fluid calculations take place. The goal of meshing is to resolve the flow
features and geometry adequately to get good results. Finer meshes (smaller cell sizes) give more
precise results but require longer run-time and more memory. There is a limit to how fine the mesh
should be; beyond this limit (called the converged mesh) reducing the cell size does not improve results.
An efficient mesh may be developed by some or all of the following methods:

 Picking cell sizes to resolve the geometry and flow features according to rules of thumb,

 Specifying grid lines to resolve the surfaces of flat geometry,

 Adding domain-removing components to eliminate unnecessary cells,

 Using symmetry boundaries at the centerline to halve the domain size,

 Using nested, linking, and conforming mesh blocks to selectively resolve flow regions,

 Checking the geometric resolution of the mesh with multiple methods,

 Mesh dependency studies: reducing cell sizes in multiples of 2 until the results converge.

Page 1 of 21
FLOW-3D Learning Objectives
In this exercise you will learn techniques to:
1. Import an existing simulation,
2. Make linked, nested, and conforming meshes,
3. Check mesh quality with FAVORize, mesh block summary tool, mesh information tool and
preprocessor output files,
4. Add mesh planes to resolve flat geometry,
5. Check the number of computational cells in the mesh,
6. Deactivate empty cells with domain removing components to improve simulation speed, and
7. Add appropriate boundary conditions.

Import an Existing Simulation File

The starting point for this exercise will be the simulation created in Exercise 2. You could just continue
working on the same prepin file, but for practice you will copy the completed project and add it to the
Portfolio. This method is useful when receiving a simulation via email.

1. Launch FLOW-3D®, go to the Simulation Manager tab, and select the Hydraulics Examples
workspace.

2. Make a new project folder:

a. Click the Workspace File path in the upper right of the screen to open the workspace folder
or navigate to it manually in the operating system. Note that there are two sub-folders, one
for each simulation you’ve run so far.
b. Right-click in the folder and select New > Folder to create a third folder. Name it
Weir_Mesh.

3. Copy the completed simulation into the new folder:

a. Open the Weir_Geometry folder.
b. Hold Ctrl+left-click to select prepin.Weir_Geometry, weir1_repair2.stl, and
Topo_Example_04_fixed.stl.
c. Press Ctrl+C to copy them to the clipboard.
d. Open the new Weir_Mesh folder and use Ctrl+V to paste the three files into the new
simulation folder. Make sure you moved the repaired .stl files.

4. Rename the prepin file:

a. At the top of the folder window select Organize > Folder and Search Options > View tab.
Uncheck the Hide Extensions for Known File Types option. This allows you to edit the
simulation name, which is the extension to the prepin.xxxx file.
b. Right-click the prepin.Weir_Geometry file and Rename it to prepin.Weir_Mesh. Click Yes
when prompted to change the extension.

5. Load the file on the Simulation Manager tab:

a. Select the Hydraulic Examples workspace and select File > Add Existing Simulation… from
the menu.
 b. Browse to the Weir_Mesh folder and open prepin.Weir_Mesh. The simulation will appear
with the old name (which is stored as text in the prepin file itself).
c. Right-click on the new (highlighted) Weir Geometry and select Rename. Name it Weir
Mesh.
d. Select File > Save All from the menu at top to save the new name in the prepin text and in
the Workspace file that specifies the project paths.

Create Upstream, Downstream, and Nested Overlapping Mesh Blocks

The region near the weir will be characterized by sharp changes in momentum (especially at the
overflow nappe) and thin geometry (the weir is only about a foot thick). This will require a fine mesh to
resolve. On the other hand, the approach channel will be characterized by relatively straight, steady
flow, and can tolerate a coarse mesh without too much loss of detail. Similarly the downstream flow
over the topography doesn’t require high resolution for this case. You will create four mesh blocks to
resolve the flow and conform one of them to the weir. The first two blocks will be the upstream and
downstream coarse grids. The third will overlap and bridge the two coarse blocks, and the fourth will be
conforming and will be added later.

1. Make sure that Weir Mesh is the highlighted project on the Simulation Manager tab.

2. Select the Model Setup > Meshing & Geometry tab and show the Mesh Window .

3. Right-click on Mesh – Cartesian in the tree and select Add a Mesh Block. New blocks are
created at the origin and are 1 m x 1 m x 1m. Find the block at the bottom of the weir.

4. Expand the Mesh - Cartesian > Mesh Block 1 tree and expand Block 1's X Direction, Y Direction,
and Z Direction subtrees.

5. The first and last Mesh Plane coordinates for each direction are the block edges. Enter values
for both Mesh Planes in the X Direction, Y Direction, and Z Direction as shown on the left side
of the image below.

6. Repeat steps 3 – 5 using the values on the right side of the image below to add the
DOWNSTEAM block 2.

7. Repeat steps 3 – 5 using the values on the bottom side of the image below to add the FINER
WEIR BLOCK block 3.

8. Check that the mesh tree matches the next page, and the mesh seen from above matches the
page after that.

a. Use the X-Y View button to view the mesh from above.
b. Turn on Mesh > View Mode > Grid Lines if necessary.
c. Experiment with different levels of Transparency.

9. Save your work!

The settings shown above define the three mesh blocks that appear on the next page. Notice that the
grid lines don’t match at all. It’s important to match grid lines between blocks or risk increasing the
interpolation error. Now you’ll work on making the cell sizes accurate.
Set Uniform Cell Dimensions
The domain now contains the region we are interested in. The resolution of both blocks is too coarse, so
now set uniform cell sizes in each block. Later you will add more Mesh Planes (user-specified grid lines)
to resolve tricky parts of the geometry and improve interpolation at the inter-block boundaries.

Rules of Thumb
1) The ideal cell shape is a cube.
2) Non-cubic cells should be shortest in the direction of the greatest gradient.
3) The long:short aspect ratio of any single cell should be no greater than 3:1.
4) Cells in regions of high gradients should be as close to cubes as possible.
5) Nested/linked blocks should have cell sizes that are even multiples of their neighbors.
6) The ideal cell size ratio across an inter-block interface is 2:1.

The nappe will have strong gradients (direction changes) in the X- and Z-directions, and some
contraction in the Y-direction. Gradients exist in all directions for the nappe. Per the fourth rule of
thumb above, cubic cells should be used in the overlapping nested block. Unmatched grid lines at mesh
block interfaces are sources of convective volume error and artificial momentum diffusion, so the cell
size of the containing block should be an even multiple of the smaller mesh block cell size to match the
grid lines.

1. Right-click on the title Mesh Block 3: FINER WEIR BLOCK and select Auto Mesh…

2. You will activate directions and fill in values as shown below. Blocks 1 and 2 will get Auto
Mesh… twice each for different vertical and horizontal cell sizes. For each step:
a. Delete the value in Total Cells (not required, but makes things more clear)
b. Select the Size of Cells radio button
c. Enter Size of Cells as shown below.
d. Select OK to accept the new criteria.
e. Repeat for each block and each direction as shown below, clicking OK after each step.
f. Check visually that all grid lines match. Toggle between views and zoom in as needed.

3. Save your work.

Check Model Resolution Using FAVOR

At this point in the geometry building process it will be useful to FAVORize the geometry to
determine if all the specified geometry is adequately resolved.

1. Practice what you remember from the first exercise.

2. Toggle Mesh > Hide All/Show All to make the geometry easier to see.

3. Rotate and zoom the view to look at the FAVORized geometry closely.

4. Check to see if there are any FAVOR issues; there should be none.

The geometry looks decently resolved. Remember that FAVORize is only an approximation because it
has to fit an iso-surface through the cells and ignores the area fractions at the cell faces. It’s a good first
test, though, and so far the mesh resolution looks adequate.
Optional: Tweak the Mesh to Minimize Inter-block error
The simulation will run with the mesh you’ve already added. However, it’s always a good idea to
eliminate inter-block grid line mismatch. This section will sharpen your meshing skill.

1. The meshes have perfect gridline matching right now. This is because the distance between all
mesh planes (within and between blocks) is an even multiple of the largest cell size, and the
largest cell size is an even multiple (x2) of the smaller cell size. This is already ideal, and fine for
all kinds of modeling.

2. An experienced CFD modeler who is used to body-fitted meshes might have a slight concern
because the weir plate faces and wall surfaces are in the middle of cells, and not exactly on a cell
face. This isn’t a problem for FLOW-3D; the partially blocked cells will get area fractions and
volume fractions that correctly locate the solid surfaces inside them.

3. An experienced FLOW-3D modeler will be concerned instead with the resolution of the weir
crest. The rule of thumb is that solids need at least 2 full cells and liquids need at least 4 full cells
across their narrowest thickness. If you look carefully at the mesh, you’ll see that the coarse
blocks do not satisfy this (they have ~ 1.5 cells across the weir), and the fine block has only 3
cells across the weir. This might affect the accuracy of the flow, though you’ll find with time that
FLOW-3D is both robust and forgiving. Still, it would be better to make everything perfect.

4. The way to do this will be to make the cell sizes such that the distances between all mesh planes
and solid surfaces in each direction (X, Y, Z) are even multiples of the largest cell size in that
direction, and then to make the smaller cell sizes in that direction exactly ½ of the larger cell
sizes.

5. Don’t worry about entering the values right now, just try to follow the ideas:
a. The weir plate is 0.33528 m thick and has surfaces at X = -0.03048 and X = 0.3048.
b. Therefore the large cell size should be ½ of the weir: 0.16764 in X, and the small cell size
should be ½ of that: 0.08382 in X.
c. The interface line between the upstream and downstream coarse blocks should be in
the middle of the weir at X = (-0.03048 + 0.16764). This is the x-min mesh plane of the
downstream block and the x-max mesh plane of the upstream block.
d. The remaining edges of all three blocks must be shifted slightly now to make all
distances even multiples of the cell size. This will make it so there’s no stretching and
grid lines match exactly in X.
e. The same idea applies in Y, but here the important distances are between the walls (6
m) and across the weir opening (3.048 m). The cell size that goes evenly into both
distances is 0.0024 m, which is far too small. Remember that cells should ideally be as
close to cubes as possible.
f. One solution for the Y-direction would be to move the walls, but that’s a trick that’s only
appropriate for testing CFD code. The real world doesn’t move so it’s numerically
agreeable. Instead, you’ll have to decide which is more important: (1) resolving the
corners of the weir opening blade exactly at cell corners, or (2) resolving the channel
wall edges exactly at cell faces.
 g. For this exercise, let’s assume it’s (2). The cell size that gives most closely approximates
a square is then dx = 0.16764 and dy = 0.16933333. You can find this by dividing the
weir opening 3.048 by the already-determined X size 0.1674 to find nearly 18 cells
across it. Divide 3.048 by 18 to find the exact cell size.
h. The Y mesh planes now need to be shifted slightly so all distances are even multiples of
0.16933333 m. The image below illustrates the idea. Don’t worry about inputting
numbers yet.

6. This is a time-consuming algebraic exercise, which is why it’s often skipped. It is perfectly OK to
skip entering the values, which are given on the next page. However, the rest of the exercise will
assume that you’ve entered the values as shown below.
Now the grid lines in X and Y match exactly and align to the most important geometry exactly. Check this
by making the geometry transparent and viewing the grid lines in the Cartesian directions. The same
process can be applied to the Z direction, where the important geometry is the bed and the weir crest.
Instead of using algebra to find the cell size, you’ll just place a mesh plane on the crest and the floor
instead.
Add Mesh Planes to Resolve Flat Geometry
Now you’ll add Mesh Planes, which are user-specified grid lines. These will be placed along geometry
surfaces to better resolve them. For this step, you will need to know the location of the edges and
interfaces. You determined these in the previous exercise; to save time the values will be provided in SI
units (remember scaling the geometry?)

Rules of Thumb

1) Use Mesh Planes as sparingly as possible because they’re likely to cause stretching.
2) Mesh Planes should only be used to define flat solids that are parallel to an axis.
3) Mesh Planes must never be used to try to resolve fluid free surfaces.
4) Mesh Planes should always be used to match nested and overlapping block extents.

1. Place the new mesh planes in the upstream block:

a. Under Mesh block 1: UPSTREAM, right-click X Direction and select Add.
b. Set the Mesh Block = 1 and the Direction = X.
c. Enter New Mesh Plane = -3.048. This is the upstream edge of the nested block.
d. Select Add. Note that the new Mesh Plane is highlighted in the display.
e. Repeat b through d to put a second plane at X = -0.03048 (the weir) in Block 1.
f. Set Direction = Y and add two planes: Y = -1.524 and Y = 1.524 for the blade corners.
g. Add two more planes in Y by copying the value from the tree (under the fine block) and
pasting it into the box: Y = -2.37066666666667 and Y = 2.37066666666667.
h. Finally, add two planes in Direction = Z: Z = 1 to match the channel bed, and Z = 3.6576 to
match the weir crest.

2. Repeat the process with Mesh Block = 2:

a. X = 0.3048 for the downstream face of the weir,
b. X = 2.98704 for the edge of the fine block,
c. Y = -2.37066666666667 and Y = 2.37066666666667 for the edges of the fine block,
and d. Y = -1.524 and Y = 1.524 for the weir corners.
e. Z = 1 and Z = 3.6576 for the bed and crest elevations.

3. Finally add matching mesh planes for Mesh Block = 3:

a. X = -0.03048 and X = 0.3048 to resolve the weir faces,
b. Y = -1.524 and Y = 1.524 to resolve the crest edges, and
c. Z = 1 and Z = 3.6576 for the bed and crest elevations.

4. Check your work and correct mistakes with the following methods:

a. Expand the Mesh Blocks in the tree and check against the image on the next page.
b. Select Mesh > View Mode > Mesh Planes to view only the specified planes.
c. Click on Mesh Planes in the tree to highlight them.
d. If there is an error, edit the Mesh Plane coordinate manually in the Mesh Block tree,
e. or right-click the incorrect Mesh Plane and select Delete, then Add it again.

5. Save your work and check it again with FAVORize.

The generalized way to determine the optimal location of grid lines (for your later reference):

 Place internal mesh planes (internal means they are not the ends of the blocks) only
where absolutely necessary to resolve geometry. In this exercise, mesh planes could be
omitted altogether, except that it’s important you learn how to use them.

 Start by selecting the largest cell sizes in X, Y, Z so that there are even multiples of the
largest cells between every set of adjacent mesh planes and important flat surfaces.
Surfaces that are not flat are well-resolved in FLOW-3D without any effort on your part!

 Move block boundaries by starting at the required mesh plane location and adding or
subtracting the largest cell size repeatedly until the value is close to the desired
boundary location. Place the boundary there. If the boundary is outside the geometry,
consider extending the geometry to the boundary.

 Continue the process in all directions until all mesh planes and important flat surfaces
are separated by even multiples of the largest cell size.

 Make cells in finer blocks exactly ½, ¼, etc. of the largest cell size.

 Repeat mesh planes in all adjacent blocks to tie the grids together precisely.
Check the Mesh Aspect Ratios with the Mesh Information Tool
1. This is a very important tool for checking your mesh quality:
a. Right-click Mesh Block 1 and select Mesh Information…
b. Check the Maximum Adjacent Cell Size Ratios (X, Y, and Z): they are close to 1 and < 1.25.
c. Check the Maximum Aspect Ratios (ZY, YZ, and ZX): they are < 3. 1 would be optimal.
d. Check the output against that shown below.
e. Close the dialog and repeat for Mesh Block 2 and Mesh Block 3.
 Water & Environment Training on FLOW-3D v11
Exercise 3: Meshing & Boundary Conditions

Make a Conforming Mesh Block

This exercise includes a lot of steps that might not be necessary to run the case, but are meant to show
you your way around the software. The conforming mesh block you’re about to create, however, is a
powerful capability introduced in FLOW-3D v11.

1. Right-click Mesh Block 3: FINER WEIR BLOCK and select Copy…

2. Keep Translations = 0 and click OK. Now there is a second, identical block on top of the fine one.
3. Open the tree for Mesh Block 4 and rename it CONFORMING MESH BLOCK.
4. Set Mesh Type = Conform to Blocked Volume.
5. Set Overlap Length = 1. The length is in meters since SI units are being used.
6. Click the Conform To Components button and uncheck boxes so only Component 1 is active.
7. Halve the cell sizes or double the cell counts to refine by a factor of 2. Type /2 or *2 after each
value in the highlighted inputs below. FLOW-3D will do the math.
8. Check that the final values for Mesh Block 4 are as shown below.
9. Save your work!

Now the fourth mesh block will include even finer cells, but only within a distance of one meter of the
weir plate. The conforming mesh will only appear as defined in 2-D plots, like those found from pre-
processing the simulation. Note also that when the simulation is post-processed, even the unused parts
of Block 4 will contribute to the memory usage of the post-processor. When simulating, the unused
parts do not contribute to the resource demand.
 Water & Environment Training on FLOW-3D v11
Exercise 3: Meshing & Boundary Conditions

Check the Total Cell Count with the Mesh Block Summary Information Tool
1. Right-click Mesh - Cartesian and select Mesh Block Summary Information…
2. Check the text display that appears. The total number of real cells defined in all the blocks is
1,983,000 cells. More “ghost cells” that are associated with boundary conditions will be added
when you preprocess the simulation, and cells in solids and unused parts of the conforming
block will be removed, so this is a preliminary estimate.
3. Two million cells is a lot for a classroom simulation. Removing some cells is advised.

Deactivate Cells with a Domain Removing Component

Unnecessary cells can be deactivated to save memory and time, but only when they will definitely never
contain flow. It is apparent that the downstream region is not likely to be filled to the top with running
water. You can safely deactivate the cells above a reasonable Z-elevation. You could do this in two
ways:
 Deactivate cells by adding a solid component. Solids will deactivate cells unless solid heat
transfer physics are active. The solids will appear in post-processing.
 Deactivate cells by adding a domain-removing component. These will deactivate cells even if
heat transfer is on. They are also transparent when viewing results. All subcomponents of
domain-removing components deactivate cells regardless of whether they are solids or holes.

1. Start to make a new primitive box (Subcomponent menu > Box).

2. Enter the values shown below and click OK.
3. Select Domain Removing as the Type.
4. Check the component placement against the image below, and Save your work.
Add Boundary Conditions
Upstream boundaries are typically one of the following:

 Volumetric Flow Rate or Velocity type boundaries specify a free surface elevation and flow rate
or velocity. They represent a design storm or are derived from field measurements.

 Stagnation Pressure type boundaries specify a free surface elevation and represent a fluid
source with zero velocity but known depth (i.e., a reservoir). Water levels above the specified
elevation exit the domain, and levels below the specified elevation (or pressure) draws fluid in.

The downstream boundary condition depends in large part on the Froude number of the downstream
flow. There are two boundaries that are appropriate for downstream in channels.

 Pressure type boundaries are appropriate when the flow is subcritical. They represent tailwater,
and require an estimate of the downstream free surface elevation.
  Outflow type boundaries are appropriate when the flow is supercritical. They do not transmit
information back into the domain, so they are appropriate when the flow is faster than gravity
waves.

The upstream boundary condition here will be specified as pressure-type using total head H. The
downstream boundary flow is expected to be sub-critical and you will use Pressure boundaries with
tailwater depths. The top of the domain is open to the atmosphere (Pressure-type with Fluid Fraction
=
0), and the bottom of the domain is a solid surface (Wall).

The pressure in the simulation can either be gauge (atmospheric pressure = 0) or absolute (atmospheric
pressure ~ 101 kPa). The atmospheric pressure (gauge or absolute) must be specified at all Pressure-
type boundary conditions, and as the initial void pressure.

Specify the boundaries as follows:

10. Open Mesh block 1: UPSTREAM > Boundaries. Each boundary has a one- or two-character
identifier that can be clicked on to change it. The default is ‘S’ for Symmetry boundary.

11. Set the upstream boundary:

a. Click on the ‘S’ next to X Min to open the boundary dialog.
b. Select the Specified Pressure button.
c. Select Fluid Elevation and set the value = 4.2 (measured from Cartesian z = 0).
d. Specify Pressure = 0. This means that the pressure at the free surface is local “gauge”
atmospheric pressure.
e. Leave Stagnation Pressure checked. The stagnation option means that the water velocity
outside the mesh (in the ghost cells) is set to zero. This corresponds to the idea that the
free-surface elevation specified is the total head (static + velocity). Stagnation pressure is
required for upstream pressure boundaries, because it is stable. When stagnation pressure
is turned off, the boundary condition is called static pressure. Static pressure allows the
solver to calculate a velocity in the boundary ghost cells, and this can lead to run-away flow
unless there is significant back-pressure on the boundary.
f. Select OK to close the dialog.
g. Leave the rest of the boundaries in Block 1 as Symmetry (S) for now.
h. Save your work.
 12. Set the downstream boundaries:
a. Repeat the process for Block 2: set the X Max, Y Min, and Y Max boundaries to:
b. Pressure-type (with stagnation pressure option activated),
c. Pressure = 0 (at the free surface),
d. Fluid Elevation = 1.5 m.

13. Set the ceilings:

a. Go through all four blocks.
b. set Z Max = Pressure (P) with stagnation pressure option activated),
c. Keep Fluid Fraction activated instead of the elevation.
d. Set Fluid Fraction = 0 (meaning the boundary is void/air, and will remove water).

The Z-Max boundaries could be Symmetry or Wall; it doesn’t matter as long as fluid never touches
them. Sometimes a model will have splashing, and when drops hit a wall or symmetry boundary they
can stick and drip and make the slow down the solver. The F = 0 Pressure boundary approach eliminates
that possibility by removing any fluid that splashes the top of the domain. Results should always be
checked to make sure that large quantities of fluid aren’t accidentally removed this way!

14. Set all Z Min boundaries as Wall with no tangential velocities.

15. Repeat steps 4 and 5 for Mesh Block 2: FINE, which shares Z-boundaries with the containing
block.

16. Mesh Block 2 is completely enclosed by Block 1 on the X and Y sides. These boundaries will be
reset to ‘inter-block’ type by the preprocessor. Leave them as Symmetry-type.

17. Save your work.

18. Select the Boundary Condition Display button (or Mesh > View Mode > Boundaries) in
the display pane to show the boundary types in the display and check your work against the
image below. Make any necessary fixes.

19. Toggle the Boundary Condition Display button to return to the normal mesh block view.
Set Fluid Density
This step is required to be able to preprocess the model.

1. Go to the Model Setup > Fluids tab.

2. Open Properties > Fluid 1 > Density Properties.
3. Set Density = 1000.
4. Save.

Preprocess the Mesh

Preprocessing the mesh is an important final step. From this you can find many summaries, including:

 The final number of cells including ghost cells and cell deactivation effects,
 The exact location of all grid lines and cell centers,
 Adjustments made to the geometry to eliminate “sliver cells” (described in more detail in
the lectures on Numeric Options and Troubleshooting), and
 Disagreement between geometry resolution at inter-block boundaries.

1. Select Simulate > Preprocess Simulation > Local from the menu at the top. The view will change
to the Simulation Manager tab, and the solver text will show the steps as they are completed.
When it shows "Preprocessor Done", the results are ready to view.

2. Select Diagnostics > Preprocessor Errors to view the prperr.Weir_Mesh file. This file contains
important messages when inconsistencies are detected by the preprocessor. It will warn you
that different resolutions are causing different open areas between blocks. This is common
when cross-boundary aspect ratios are greater than 2:1. All the reported warnings involve Block
4, which has the same extents as Block 3 but only half the resolution. This does not matter
because the actual extents of Block 4 after it conforms to the weir will be inside Block 3, and the
ratio will be 2:1 there.

3. Select Diagnostics > Preprocessor Summary to view the prpout.Weir_Mesh file generated by
the preprocessor. You can also read the diagnostics files in any text editor.
a. There is an enormous amount of information in this preprocessor summary, which
documents every step taken by the preprocessor.
b. Click on Search at the top of the dialog and search for mesh. The first instance of the term is
at the beginning of the mesh block namelist. Namelists are sections of the prepin file; the
preprocessor goes through the namelists in order and interprets them. Press Cancel to close
the search box and scroll down to find the section labeled mesh generated in x direction:
c. Click on Search at the top of the text dialog and search for cell summary. Review the
information, which should appear as it does below. There are 144,385 active cells out of
202,498 total cells; both counts include boundary cells. The solid and domain-removing
components deactivated 29% of the cells. This correlates directly to a reduction in memory
use and a substantial runtime efficiency improvement. Fully blocked real cells are cells that
are filled with solid geometry but are still active for heat transfer: there are none here.
 cell summary:

total number of cells (active and passive) = 2221792

fluid sub-domain cells = 783824
solid sub-domain cells = 66570
cells common to both solid & fluid sub-domains = 52888

active cells include:

real cells (used for solving flow equations) = 652642
open real cells = 619659
fully blocked real cells = 32983
external boundary cells = 33336
inter-block boundary cells = 111528

d. Search again for areas at inter-block. Review the information. The inter-block area fraction
differences seems high, but most of them misleadingly apply to a conforming block (#4),
while Block 1 and Block 2 only interface directly with each other across the solid sides of the
weir plate. For this case, the inter-block area warnings can be ignored.

e. Search again for area/vol. Cells that have a large open facial area to volume ratio are likely
to have high fluxes which can limit the time step (slow down runtime). FLOW-3D normally
adjusts these by adding ‘virtual’ volume to cells with A/V ratio greater than a specified
threshold called AVRCK (by default 3.1). This speeds up the solution and makes it more
stable at the cost of a (usually) slight degree of accuracy.
3
In this simulation, FLOW-3D added 0.21 m of volume and the total open volume in the
3
mesh is 383 m . The total adjustment is less than 0.06% of the total volume: an insignificant
loss of accuracy for a probable efficiency gain.

4. Select OK to close the text file.

Exercise 4: Fluid, Physics, & Initial Conditions
 Water & Environment Training on FLOW-3D v11
Exercise 4: Fluid, Physics, and Initial Conditions

Hydraulics Exercise 4: Fluid, Physics, Initialization

Problem Description & Objectives

Continuing from the previous exercise, a fluid (water) will be specified for the simulation and fluid
regions will be initialized within the domain. Gravity and turbulence models will be activated, and
mass/momentum sources will be added to the simulation to create discharge from the pipe culverts.
Final steps before running the model will include adding some sampling locations and specifying output.

FLOW-3D Learning Objectives

In this exercise you will learn techniques to:
1. Copy an existing simulation to a new folder in FLOW-3D,
2. Activate and specify parameters for the gravity and viscosity & turbulence physics models,
3. Initialize fluid regions and boundaries with velocity and hydrostatic pressure,
4. Load fluid properties from the materials database,
5. Manually edit fluid properties, including limited compressibility,
6. Set the simulation finish time,
7. Add mass/momentum fluid sources,
8. Add history probes for point measurements, and
9. Add flux surface baffles for flow rate measurement.

The starting point for this exercise is the simulation created in Exercise 3. If you have successfully
completed Exercise 3 you may choose to simply make a new copy of that simulation as described below.
If you have not successfully completed Exercise 3, you may download or copy the file
prepin.Weir_Mesh_Completed and the geometry files to a new simulation folder and import it as an
existing simulation.

Make a Simulation Copy

1. Launch FLOW-3D®, go to the Simulation Manager tab, and right-click the Hydraulics Examples >
Weir Mesh simulation that you just finished. Select Add Simulation Copy…

2. Enter Weir Finish as the simulation name.

3. Leave the Create Subdirectory Using Simulation Name checked to keep files organized.

4. Check the boxes for weir1_repaired2.stl and Topo_Example_04_fixed.stl to copy the geometry.

5. Select OK.

Page 1 of 13
Activate the Gravity Physics Model
1. Navigate to the Model Setup > Physics tab.

2. Click on Gravity and Non-Inertial Reference Frame. The physics models are in alphabetical order
from top to bottom, left to right.

3. Enter Z Component = -9.81. The negative sign points the gravity vector downward.

4. Click OK to exit the dialog. A green check mark now appears next to Gravity.

Activate the RNG Turbulence Physics Model

The presence or absence of turbulence can be determined using Reynolds number relations; in our case
it can be assumed that the flow is turbulent. There are a number of choices for turbulence models. The
most robust and widely applicable turbulence model is the k- ε RNG (Renormalized Group) model. In
the next few steps you will activate the RNG turbulence model and specify dynamically-computed
maximum turbulent mixing lengths. The actual turbulent mixing length is the characteristic turbulent
length scale, a mathematical factor that roughly corresponds to the size of the smallest turbulent eddy
that needs to be resolved in terms of turbulent transport; eddies that are smaller than the length scale
are approximated as an average and dissipate to heat. The maximum turbulent mixing length is an
upper stability bound to prevent the turbulence model from over-estimating the length scale. It can be
global and constant (which requires some understanding of the limiting mixing length), or calculated in
each time step (dynamically) for each cell based on empirical formulations.

Rule of Thumb

A typical estimate of the actual turbulent length scale is in the range of 7% to 18%
of either the flow depth, wake width (abutments), or wake half-width (piers/jets).

We do not know the location of the greatest turbulence: is it the pipe outfall or the weir jet nappe
impingement? We can guess that the impingement point is the most turbulent region in the domain
and also controls significant flow features. Therefore, the thickness of the nappe could be used to
estimate the maximum turbulent length scale, but perhaps the most turbulent region is the pipe outfall
after all! Since we do not know enough to confidently specify the maximum mixing length, we will ask
FLOW-3D to determine the maximum turbulent length scale for us.
 1. On the Model Setup > Physics tab, select the Viscosity and Turbulence button.

2. Select the options shown below.

3. Click OK to accept the changes and Save your work.

Load Fluid Properties from the Fluids Database

Thus far, we have not specified which fluid we want to model. In order to run the simulation and check
our boundary conditions for accuracy, we will need to set fluid properties. FLOW-3D includes databases
of solid and fluid properties for a number of materials.

1. Select the Model Setup > Fluids tab.

 2. Select Materials > Fluids Database... from the menus at the very top of the interface.

3. Choose Water_at_20_C, and select Load Fluid 1.

4. You have already set the unit system, so the only units available are SI.

5. Select OK, then OK again, then Close the database. Save again.

Edit Fluid Parameters and Add Limited Compressibility

1. Expand the Properties > Fluid 1 tree and the sub-branches.

2. Note that fluid properties are grayed out if they require inactive physics models (e.g., thermal
properties).
3
3. Check that the values for fluid Density ρf = 1000 kg/m and dynamic Viscosity μf = 0.001 kg/m/s.

3
4. The values are approximate. Change Density = 998.21 kg/m . This is true for clear water at 20°C
and standard pressure (101,325 Pa).
-1
5. Locate Compressibility and set it to 4.55e-10 Pa . This is so-called limited compressibility. It is
used in Incompressible flow simulations (like this one) to allow the cell pressure to vary as if
density was changing due to compression. Volume stays constant. This makes it easier for the
solver to converge (faster runtime), and allows pressure waves (e.g., water hammer) to
propagate. The value entered is correct for clear water at 20°C and standard pressure. For your
 2
future reference, the equation is: Limited Compressibility RCSQL = 1/K = 1/(ρf W ). K is the fluid
bulk modulus, ρf is the fluid density, and W is the adiabatic speed of sound.

6. Save your simulation!

Initialize Fluid Regions & Hydrostatic Pressure

If you run your simulation right now it will take a long time to reach steady-state because it will fill the
empty domain before topping the weir. You should add initial fluid to the simulation to speed up the
time to steady state. The upstream and downstream initial depths should match the free surface
elevation at the boundaries. You will also specify a hydrostatic pressure distribution for fluid in the
domain at t = 0 and at the pressure boundaries. Hydrostatic pressure is specified as an initial condition,
but it also applies at all vertical pressure boundaries for the entire run.

1. Navigate to the Model Setup > Meshing & Geometry tab.

2. Open the Initial Conditions Window and then the Initial > Global > Pressure branches.

3. Change the Pressure from Uniform Pressure to Hydrostatic Pressure. The surface pressure will
be that of the initial void pressure, which defaults to zero, and will be maintained by the Z-max
pressure boundary (where F = 0). The boundary pressures and initial void pressures match at 0
Pa. Matching the void, initial, and boundary pressures can be very important when the void
pressure varies via bubble physics.

4. Open the Global > Fluid Initialization branch and set Fluid Initialization = Use Fluid Elevation
and Initial Fluid Elevation = 1.5 m. This will fill the entire domain to the level of the downstream
boundary condition.
5. Close the Global tree branch, then right-click Fluid Regions and select Add a Fluid Region.

6. Name the new fluid region Upstream. Fluid regions overwrite global areas that they overlap.

7. Note that the region will Add Fluid, and that the Fluid Fraction = 1 (100% Fluid 1 = water).

8. Open the Limiters branch of the tree fluid region tree.

a. Specify X High = 0 m to fill the upstream channel to the weir.

b. Specify Z High = 4.2 (to match the upstream boundary condition free surface elevation).
c. The unspecified limits default to the domain extents.

9. Note that initial Velocities, Pressure, and Turbulent Kinetic Energy can also be specified, but do
not need to be here.

10. Save your work.

 11. View the FAVORized geometry using the computational mesh. You should already know how to
do this from previous exercises. This time select Fluid Surface = Fluid 1.

Add Mass/Momentum Sources to the Pipes

Mass/momentum sources and sinks are flat planes that add or remove fluid. They are completely
porous to flow approaching from behind them, and are used to represent nozzles, pipes, intakes, vents,
and other devices. Mass/momentum sources are ideal when fluid must be added somewhere inside the
domain. They can also be used to remove fluid if they are placed so they are always underwater.

1. Navigate to the Model Setup > Physics tab, select Fluid Sources, and check the box to activate
the physics. Select OK.
2. Go back to the Model Setup > Meshing & Geometry tab and open the Mass/Momentum

Source Window .

3. Right-click Mass Momentum Sources and select Add a Mass Momentum Source.

4. Click to check the Enabled box to activate the first source.

5. Enter the values shown below. The Volumetric Flow Rate gives 2 m/s velocity: QV = U x A.

1. Adjust Transparency so you can see the circular source at the origin.

2. Right-click Mass Momentum Source 1 and select Copy. In the new source, change only
Geometries > Translation > Y = -1.

3. Check that your model appears as shown, with flow vector pointing outward, and Save.
Add a Flux Surface Baffle to Measure Weir Flow
Baffles are flat planes along grid lines that block or measure flow. They can be solid or porous. Flux
Surfaces are a special kind of baffle which measure flow parameters and can be completely open to flow
(100% porous).

1. Still on the Meshing & Geometry tab, open the Baffles window and select Add .

2. Name the new baffle weir flux. Output will be titled with this name.

3. Check the Define as Flux Surface option. Now the baffle will measure flow rates, depth, and
hydraulic energy and head.

4. Define the baffle location and parameters:

a. Open Baffle Region 1 > Definitions. Only one definition should be specified per baffle.
b. Specify X Coordinate = 0. The baffle now defines the plane at x = 0.
c. Open Baffle Region 1 > Limiters. These are used to bound the defined plane.
d. Specify Y Low = -1.7, Y High = 1.7, and Z Low = 3. This makes the baffle slightly larger than
the opening, which may be important when the baffle moves to the nearest grid lines.
e. Open Porosity Properties and set Porosity = 1. This is not strictly required, since flux
surfaces cannot be completely solid, and will default to porosity = 1 if the default 0 is left.

5. Save your work.

Add a History Probe to Measure the Approach Depth
History Probes are single points which measure the cell-centered flow parameters. They can be made to
move with solid objects or float freely in the flow. Here you will create a fixed point probe that will act
as a staff gauge for measuring the flow depth. Like flux surfaces, the history probe output is labeled
with it’s name.

1. Still on the Meshing & Geometry tab, open the History Probes Window .
2. Select Add and enter the values shown below.

3. Save again.
 Water & Environment Training on FLOW-3D v11
Exercise 4: Fluid, Physics, and Initial Conditions

Set the Finish Time

The default runtime for a new simulation is 1 second, and this is not long enough to reach steady state.

Rule of Thumb
A conservative measure of the time for a free-surface fluid system to reach steady
state is 10 to 20 times that required for a fluid packet to pass through the system. A
characteristic velocity in the domain needs to be estimated to determine the
characteristic (packet-transit) time scale.

Without experimental or simulation data we can only estimate the characteristic velocity. The velocity of
1/2
the water falling under gravity though a distance H of 2.7 m (the drop height) is V = (2 g H) = (2*9.81
2 1/2
m/s *2.7 m) = 7.3 m/s. The total length of the mesh domain is about 12 m, so the packet-transit time
is approximately 12/7.3 ≈ 1.6 seconds. Real-world time to steady state is then about 16 to 32 seconds.
Model time to steady state also depends on the mesh resolution: for example, meshes that are too
coarse or have very poor aspect ratios may not converge to steady state at all.

1. Navigate to the Model Setup > General tab.

2. In the upper left corner, enter Finish Time = 35 seconds.

Set the Number of Processors

It’s good practice to leave at least one processor available for the OS and other programs to use.

1. Specify Number of Processors = 4.

2. Save your work!

Select Output Options

The solver will automatically write 10 data sets (e.g., every 10 seconds) that can be used to generate 2-D
and 3-D plots. These data sets are called Restart data because they include all of the information
necessary to describe the flow field and can be used as the initial and/or boundary condition
configuration of a new simulation called a Restart simulation. More frequent output can be added as
Selected data; the default frequency adds 100 data sets (e.g., every 1 seconds) containing only the
variables you select. These Selected Data sets are written in addition to the Restart data.

1. Navigate to the Model Setup > Output tab.

2. Find Additional Output on the left. Selections made here will be added to Restart Data and
History Data (time-dependent data recorded for the flux surface and history probe).
 3. Check the box for Hydraulic Data under Additional Output. The parameter is also added to the
Selected Data list in the center of the screen.

4. Under Selected Data, check the boxes for:

a. Fluid fraction,
b. Fluid velocities,
c. Hydraulic Data, and
d. Pressure

5. Note that the Restart Data Interval, Selected Data Interval, and History Data Interval can be
th th th
forced to other than the default 1/10 , 1/100 , amd 1/100 of the finish time, respectively.
Intervals are specified in seconds and can usually be left to the default.

6. Save your work!

Set Higher-Order Momentum Advection

The default momentum advection is quite robust and usually very accurate for linear flow. It tends to
diffuse momentum (and kinetic energy) when the flow is curving. There will likely be some flow
curvature near the weir corners, so although it may not help (and will definitely slow down the run) we
will add a higher order momentum advection method.

1. Go to the Model Setup > Numerics tab. Locate the Momentum Advection options on the right-
hand side.

nd
2. Select Second Order Monotonicity-Preserving momentum advection. This option is 2 -order
st
accurate in space and 1 -order accurate in time, and is stable in cells that have a free surface.

3. Save your work! Have you figured out that this should happen with every change?

Run the Simulation

You already pre-processed the simulation in Exercise 3. A careful modeler would preprocess the
simulation again now, but throw caution to the wind and run the model anyway!

1. Select Simulate > Run Simulation > Local from the menu at the top.

2. The run will usually take between 8 and 10 hours, depending on your computer’s power.

Your instructor is available to answer any questions you’d like to discuss!

Exercise 5: Post-Processing & Analysis
 Water & Environment Training on FLOW-3D v11
Exercise 5: Analysis

Hydraulics Exercise 5: Analysis

Problem Description & Objectives

The results from the simulation you built in the previous four exercises will be analyzed. This exercise
applies and expands the techniques introduced in Exercise 1.

FLOW-3D Learning Objectives

In this exercise you will learn techniques to:
1. Use 3-D plots to check the results qualitatively.
2. Review the Runtime Plots to check for steady state.
3. Review Solver Output text and files to check for convective flux errors.
4. Review the meaning and significance of convective flux errors.
5. Use 2-D plots to visualize the location of the minimum stability criteria.
6. Use Probe plots to quantify flow rate and depths.
7. Discuss model comparisons and possible sources of error.

Load the Results File

The starting point for this exercise is the output generated at the end of Exercise 4. The flsgrf file is 4.9
GB, so it is not offered for download or copy. If you have not completed Exercise 4 yet, please ask your
instructor for a copy of the results file.

1. Launch FLOW-3D®.
2. Select the Simulation Manager tab.
3. Click to highlight the Hydraulics Examples > Weir Finish simulation that you ran.
4. Select the Analyze tab.
5. If prompted to select a file, continue with step 7 below.
6. If not, select the Open Results File button at the bottom left of the Analyze tab.
7. Select Custom to display flsgrf and prpgrf files.
8. Select flsgrf.Weir_Finish.
9. Select OK. The file loads and the GUI switches to the Analyze > 3-D tab with all options reset.

Select 3-D Plotting Options

1. Go to the 3-D sub-tab.

2. Select Data Source = Selected.

3. Check that Iso-Surface = Fraction of Fluid so you plot the water/air and water/solid interfaces.

4. Select Color Variable = Velocity Magnitude.

Page 1 of 13
 5. Click Additional Variables below the Color Variable selection.

6. Select Flow Depth, Free Surface Elevation, Depth-Averaged Velocity, and Froude Number, and
click OK to accept the variables for plotting.

7. Open the Time Frame sliders all the way: Min = 0 and Max = 35 seconds.

8. Open the X, Y, and Z Limits sliders all the way.

9. Select Component Iso-Surface Overlay = Solid Volume. This plots the solids as well.

10. Select Options = Activate Plain Streamline Seeding.

View and Interpret the 3-D Results

1. Click the Render button at the lower right. The display switches automatically to the Display tab,
and a new Streamline Edit dialog appears. Find a good setting for the Transparency slider.

2. Use the Next and Previous buttons, and/or double-click on the Available Time Frames listed, to
see the results at different times. Note that the baffle is displayed as a gray solid plane, even
though it’s completely porous. Hide it on the View menu > Baffle Options dialog.

3. Adjust the Transparency slider. Observe that the colored results are only the values at the
surface; changing transparency does not allow you to see the velocity profile inside the flow. 2-D
plots or 3-D Mesh Slices can be used to visualize internal (submerged) flow properties.
4. Select the other Color Variables from the dropdown in the upper left corner of the interface. Do
the results make sense?

5. Set Color Variable = Froude Number, and open Tools > Options from the menu above the
display window. Uncheck the default values option and enter the values shown below. Note that
the nappe appears subcritical. This is because Hydraulic Data is depth-averaged along z, so the
Froude number includes the depth and velocity of the pool beneath the nappe.
 6. Re-activate the Default Values option and click Apply to return to the normal color scale.

Note how a supercritical jet expands away from the wake, while the rest of the pool is subcritical. This
makes the appropriateness of the downstream boundary unclear. Pressure boundaries are best for
subcritical tailwater, while Outflow boundaries are best for supercritical flow. The X-max and Y-max
boundaries should be placed farther from the nappe to avoid this difficulty.

Also note the contraction of the jet as it falls; the degree of contraction in the CFD model depends on
the momentum advection method, volume-of-fluid (VOF) method for free-surface tracking, and the
geometric resolution of the weir crest.

Finally, note that the upstream depth (as shown for Fluid Depth and Free Surface Elevation) does not
vary much. This indicates very minor head loss due to friction. The degree of friction appearing in the
model depends on the water velocity, the roughness height of the component, and the mesh resolution
near the bed.

Add Streamlines
1. Access the Streamline Edit dialog or select Tools > Streamline Edit. If the dialog doesn’t
appear, return to the Analyze tab, activate Plain Streamline Seeding, and Render again.

2. Set a single streamline starting point and edit it’s appearance:

a. Select the X direction radio button for the Seeding Plane.
b. Slide the Seeding Plane slider back and forth and watch the green plane move in the display.
c. Adjust Transparency so you can see the entire plane.
d. Hold Shift and left-click on the plane at a point near the bed and approximately on the
centerline of the flow.
e. Select Render.
f. A line should appear showing the streamline from t = 0 to the selected time in the Available
Time Frames list.
g. If no line appears, troubleshoot by:
i. adjusting transparency,
ii. selecting a later time frame (for a time when fluid is moving),
iii. making sure that the point you added isn’t inside the bed (no flow there).
iv. Use Clear and try again, if necessary. Remember to Render again after adding a point.
h. Increase the number of Integration Steps by a factor of 10 (add a zero) and Render again.
The number of steps controls the resolution of the line, as well as the length. Usually 10,000
or 100,000 integration steps are sufficient to complete the line. Repeat the increase until
the line reaches the end or no longer changes. If the line terminates at a free surface, try a
different starting location.
i. Locate the Object List pane in the lower left. Right-click the Streamlines entity there to get
display options.
j. Try changing the display shape to Ribbons, adjusting the transparency, and changing the
color and weight (ribbon scale percentage) of the streamline. Also try adjusting the
transparency of the other entities (solid and fluid surfaces),
3. Make many streamlines:
a. Select Clear. The seeding points are erased, but the lines remain.
b. Move the slider so that it is located at the minimum X value -6.0655 m.
c. Activate the Range checkbox and set the number of points to 10.
d. Hold Shift and click on the seeding plane in a corner near the bed, and then in the opposite
corner near the surface. A line of 10 points appears between the points you clicked, for a
total of 12.
e. Repeat the Shift + click process for the other two corners, to make an X with 24 points.
f. Render the streamlines and adjust the display (global and streamline-entity) as before. You
may need to add more integration steps (100,000 shown below) to complete the lines.
g. Select Clear to remove the seed points and Close to remove the plane. The lines remain.
h. Step through the Available Time Frames and observe the early oscillations of the
streamlines and how they eventually become steady.
4. If you want to change the streamlines, select Tools > Streamline Edit from the menu above the
display to reopen the dialog and seeding plane.
Review the Runtime Plots
1. Select the Simulation Manager tab.

2. Select the following plots from the list on the left of the screen:

a. Stability Limit & Time Step Size: the time step size is very close to the stability limit,
matching it in most cases. There are two types of solvers: implicit and explicit. Implicit
solvers are iterative, while explicit solvers are direct (one step). The stability limit is the
smallest time step required for any explicit solver (e.g., convection). The time step dt may be
less than the stability limit when an implicit solver (like pressure) does not converge to the
error tolerance: then it will restart the calculation with a smaller dt. The only implicit solver
in our model is the pressure solver, so this suggests that the pressure solver is working well.
b. Time Step Size: Repeats part of the information from the last plot. Move the mouse over
the plot to find that the time step size oscillates about 0.005 seconds per step.
c. Epsi & Maximum Pressure Residual: The maximum residual is always below epsi, so the
pressure solver converged to a solution in every time step. See Exercise 1 for more
explanation.
d. Pressure Iteration Count: the default pressure solver is the Generalized Minimum Residual
method (GMRES), which should usually operate between 1 and 10 iterations per cycle.
e. Fill Fraction: the simulation is about 53.5% full of fluid for the second half of the simulation.
The relatively flat shape of this plot suggests that fluid volume in the simulation is near-
constant, and therefore outflow = inflow. This is one criterion for steady-state.
f. Conv. Volume Error (% Lost): the total volume lost due to numerical mass calculation error
is less than 0.1%. This is acceptable. Typically less than 1% to 3% loss is acceptable.
g. Interblock Boundary Volume Error (% Lost): the total volume lost due to interpolation at
inter-block boundaries is about 2.5%, and follows the same criteria above. The two values
should be considered additive. 2.5% is on the high side. It suggests that one of the nested
blocks has its boundary in a region of high gradients. Later you will locate where this occurs.
h. Volume of Fluid 1: the flow volume appears to reach quasi-steady state by about 60
3
seconds, with an average near 163 m in the simulation.
i. Fluid 1 Surface Area: the water surface area comes to quasi-steady state around 10
seconds.
j. Mass-Averaged Kinetic Energy: MKE here is the sum of total kinetic energy in all cells
divided by the mass of fluid in the simulation. It reaches quasi-steady state around 20
seconds.
k. Mass-Averaged Turbulent Energy: This integral is the sum of time-mean TKE in all cells,
weighted in each cell by the fluid mass. TKE appears to be quasi-steady by about 10 seconds,
but there are too many oscillations to be certain. You could launch a Restart Simulation
from the end of this one to see if TKE levels off later, or maybe it’s just because of the
splashing on the topography under the nappe.
l. Particle Count: The number of particles in the simulation stays constant at 1,330 particles.
These particles are used to define the Mass/Momentum sources in the pipes (as point fluid
sources) and do not move with the flow.
 Water & Environment Training on FLOW-3D v11
Exercise 5: Analysis

Use Solver Text Output to Diagnose Errors

1. Locate the Warnings & Errors button above the runtime plots. It means warnings are
present.

2. After the meshblock mismatch errors discussed in There is one warning, which is displayed
below.

Solver Messages:
convective flux exceeded stability limit
at t= 3.0009E-02 cycle= 1 iter= 51 delt= 3.0009E-02 mesh block 4
restarting cycle with smaller time step
maximum failure ratio = 2.26944E+00 is in x-coordinate direction
at cell ( 82, 34, 17) mesh block 4

3. The meaning is as follows:

a. At t = 0.03 seconds, a packet of fluid tried to move more than one cell at a time. The packet
probably accelerated unexpectedly, which is why the solver didn’t predict the velocity of the
packet. The stability limit is the requirement that fluid packets only move one cell per time
step. The solver takes care of the problem by restarting the time step with a smaller dt.
b. The error occurred in Mesh Block 4, the conforming block.
c. The offending cell number is i, j, k = 82, 34, 17. Remember that i,j,k values are block-specific,
while x,y,z values are global.
d. The problematic fluid packet was moving in the x-direction. In the next step you will find the
problem cell and identify why the problem occured.

4. Use the small X in the upper right corner of the dialog to close the Warnings & Errors.

5. Scroll up through the text output displayed below the plots and find the same warning. Notice
that the cell location and mesh block are not given in this summary.

6. Select Diagnostics > Solver Errors from the menu at the top of the screen: the same information
is displayed. Some errors report additional information to the Solver Errors file, so it should
always be checked.

The only run-time error reported in this simulation is a single convective flux error which does not affect
solver accuracy. Except for the purposes of this exercise, you can safely ignore isolated errors. If
convective flux errors recur frequently, they will at best slow down the simulation, and at worst cause it
to terminate before completion. Errors should be addressed when they are repetitive.

Use 2-D Plots to Diagnose Errors

1. Go to the Analyze > 2-D tab.

2. Click the Mesh Block button at the lower right and select Mesh Block 4 only. When you click OK,
the Limits sliders will reset to show the I J K cell number in the selected block at their sides.
3. Set plotting extents for Block 4:
a. Set Contour Variable = X-Velocity and Vector Type = Plain Velocity Vectors.
b. Select Plane = XZ.
c. Select the Mesh checkbox to plot the grid lines.
d. Select Data Source = Selected.
e. Slide the Max Time Frame slider (or use the associated arrow) so that t = 0 to 0.7 seconds.
f. Slide the X Limits sliders so that I = 78 to 86. The problem cell (I = 61) will be in the center.
g. Move the Y Limits sliders so that both Min and Max J = 34. This will display one planar slice
along the cell centers of Block 4. The slice will contain the problem cell.
h. Slide the Z Limits sliders so that K = 2 to 32.

4. Select contour plotting options:

a. Click the Contour Options button in the bottom central portion of the tab.
b. Activate the No Contour Smoothing checkbox to turn off the aesthetic blurring of cell-
centered values.
c. Select Blanking Variable = None. This clears the Blanking Variable Value entry. All cell
values will be displayed, whether they contain fluid or not.
d. Select OK to close the dialog.

5. Check that you’ve made the selections above (marked below for reference) and click Render.
The two plots closest in time to the error should be similar to the ones on the next
page.
6. Locate the cell at I = 82, J = 34, K = 17:
a. Select multi as the plot type and then select the time frames for t = 0.35 and t = 0.73
seconds from the list at the right. The problem occurred before these plots, but they still
show why.
b. Check that the text below the plot shows the range ix = 78 to 86 and kz = 2 to 32.
c. Start with the bottom-right cell at K = 2 and count up to 17. From the left, start at I = 78 and
count up to 82. This is the cell specified by the warning message.
d. Induce an explanation (or see the next page).
The problem cell is just inside the top of the pipe. Notice the recirculation zone to the right above the
pipe discharge jet. Convective flux errors are associated with sudden accelerations. The flow is making a
corner here, so it can be induced that the discharge, forming in quiet water, interacted with the pool
and experienced an unexpected acceleration at t = 0.03 seconds. Even at t = 0.35 and 0.7 seconds you
can still see where the recirculating pool flow accelerates from the downward direction to the horizontal
direction as it’s caught by the pipe discharge. By the visible time steps, however, the flow is predictable
enough that the dynamically chosen time step accounts for the change in direction.

Identifying the problem cell is useful when there are many errors in a row. In this case the failure is of a
type that can usually be ignored, especially since it occurs only once. Common causes of serious (that is,
numerous) errors include:

 bad geometry files: fix with pyAdmesh and netfabb or re-export from CAD,
 unresolved or poorly resolved gaps between geometry: use finer cells or different geometry,
 incorrect boundaries or sudden changes in boundary conditions or mass/momentum sources,
 incorrect initial conditions, and/or
 flying droplets/splashing of fluid.

Use Probe > General History Data for Measurement Location Output
1. Go to the Analyze > Probe tab and select Data Source = General History.

2. Display units:
a. Click Units at the bottom right of the screen.
b. Activate Show Units on Plots.
c. Select Units System = SI.
d. Select OK to close the dialog.

3. Scroll down in the Data Variables list and select the following:
a. staff gauge: flow depth (around #14). This is output from the History Probe you placed
upstream.
b. fluid 1 volume flow rate @ weir flux (around #59). This is output from the Flux Surface you
put at the weir.

4. Render the plots graphically, using multi plots to look at the data side-by-side.
 5. Get measurement station output as text:
a. Return to the Analyze > Probe tab and select Output Form = Text (in the lower left).
b. The options for Integrate, Differentiate, and Moving Average become inactive, as they are
only available for graphical output.
c. Check that the Time Frame Min and Max cover the entire duration of the run.
d. Click Render.
3
e. The text output is given with the variable titles: TIME (s), HYFD (m), and FFLUX1 (m /s). It
appears that the flow rate reaches approximate steady state around 20 seconds.
f. The time-averaged flow rate can be found by copying and pasting the data into a
spreadsheet. If you were to do this now, you would use a text import function to separate
the space-delimited values and some method of averaging the later quasi-steady results.
3
g. The average flow rate Q over the weir after 20 seconds is approximately 2.42 m /s.
h. The average upstream centerline depth h1+P after 50 seconds is 4.20 m.

6. Select Continue to close the text output.

View the Mesh and Find The Multi-Block Volume Error

Now you will practice what you know.
1. Make a 2-D plot with the following settings:
a. Plot all 4 mesh blocks.
b. Plot the flow’s center line as an XZ plane.
c. Plot Fraction of Fluid.
d. Turn off contour smoothing.
e. Turn off cell blanking. All cells should show values with no white space.
f. No Velocity Vectors.
g. Plot only the last output time, which is when the multi block error was at maximum.
h. Plot the Mesh.
2. Do you see any multi-block problems? Answers on the next page…
Luckily for you, the diffusion and volume loss appear to occur after the critical control at the weir crest. If
you are interested in flow rate over the weir, the multi-block error won’t affect your results. If, however,
you’re interested in the nappe impingement and its effect on the downstream topography, then the
model has a problem and should be re-meshed to enclose the nappe in a single block.

Compare Results to Physical or Empirical Data

At this point, you should:

1. Compare the model results to the physical or empirical results.

2. Identify the relative error as e = (QCFD – Qphysical)/Qphysical.
3. Is the relative error < 5%? If not, try to explain why or make the model better. Sources of error
may include:
a. Inter-block gridline mis-match may cause momentum diffusion before or at the weir.
b. Non-cubic cell aspect ratios may diffuse momentum.
c. Poor resolution of the crest may cause too little contraction or nappe spring.
d. Mesh-resolution studies address many potential sources of error. Perform one by reducing
all cell sizes by 2x, running, and repeating until Q varies less than 2 – 3%.
e. Momentum advection and/or VOF methods may diffuse momentum and/or free surfaces.
f. Turbulence model assumptions may affect results.
g. Empirical equations have limits of validity that affect their accuracy: does the modeled
situation match the experiments that provided the data?
h. Physical test data may be scaled incorrectly, especially for turbulent effects.
i. The CFD model may not match the physical model.
j. The physical model data includes measurement uncertainty. Are error brackets known?
 3
This model predicts a flow rate of about 2.4 m /s, while the Kindsvater and Carter (1957) equations
3
predict about 2.1 m /s. The disagreement is about 13%. A careful check of the limits of applicability of
their equation (found in Bos 1989) show that the model is within the region of applicability for approach
head depth, crest height above the channel, channel size, and drop length of the nappe. Feel free to
suggest how this model could be improved to your instructor. For one successful example, see Abd El-
Hady Rady (2011) “2D-3D Modeling of Flow Over Sharp Crested Weirs”, Journal of Applied Sciences
Research.

Congratulations on completing your first FLOW-3D simulation from start to finish!

Exercise 6: Intro to Air Entrainment
 Water & Environment Training on FLOW-3D v11
Exercise 6: Air Entrainment

Hydraulics Exercise: Air Entrainment

Problem Description & Objectives

Air entrainment downstream of a hydraulic jump will be simulated in three ways. Entrained air will be
modeled as a massless scalar, as a miscible scalar with mass, and as a drifting dispersed phase with both
density and bubble-drag effects. The student will set up the first model from a set of specifications as an
exercise in critical thinking. Hints and suggestions will be included after the specifications.

The experiment models a bent pipe that discharges below the surface of a standing pool. A hydraulic
jump forms in the downward bend of the pipe, as shown in Figure 1, below. To reduce the simulation
time, only the downstream portion of the pipe will be modeled, using 2-D CFD, and this domain is
illustrated in Figure 1. The 2-D assumption will limit the validity of the results; the exercise is for
illustration of the various approaches rather than a formal validation.

X-min
Z-max

X-max

Z-min

Figure 1: Schematic of Experimental Apparatus and CFD domain

FLOW-3D Learning Objectives

In this exercise you will practice techniques to:

1. Create 2-D geometry and mesh,

2. Use component gravity to create an angled (with respect to gravity) domain,
3. Set boundary conditions (including turbulent energy) based on Froude number and depth,
4. Use the Air Entrainment model, and
5. Use the Density Evaluation and Drift-flux models.

Page 1 of 13
Determine Component Gravity Parameters
We want to model the downstream portion of the pipe in such a way that the x-axis is parallel to the
pipe walls, and the z-axis is normal to the flow direction, as illustrated in Figure 2. The downstream leg
of the pipe is 15 cm diameter and angled at a 10% slope (i.e., the slope of the pipe is 10 cm/100 cm).
We can use this to determine component gravity:
z

15 cm
θ
gz = - g cos θ 2
g = 980 cm/s

10
gx = + g sin θ θ x
100
Figure 2: Determining component gravity from pipe slope
-1
The angle of the pipe θ as measured from horizontal is tan (10/100) = 5.71059°. It is easy to show that
θ is also the angle between gravity and the vector normal to the flow direction. Therefore:

2
gx = 980 sin 5.71059° = + 97.5136 cm/s
2
gz = - 980 cos 5.71059° = - 975.136 cm/s

2 2 2
Another way to solve for gz would be to start with slope: gx = gz/10, so [gz + (gz/10) ] = 980 .

1. Add a new simulation called Air Entrainment Scalar.

2. Set the Simulation Units = CGS.
3. Set Gravity component gx and gz given above on the Physics panel.

Determine Upstream Velocity from Froude Number and Depth

The flow accelerates and things as it crests the bend of the pipe. Most air entrainment experiments
characterize the flow by a depth and Froude number at the control point at the pipe bend. In this case,
let the Froude number be 3.0 and the depth be 5 cm.

√
where:
Fr = Froude number, given as 3.0,
U = average velocity of the fluid,
2
g = gravity, 980 cm/s ,
h = mean depth of the flow, given as 5 cm.
Technically, h is the hydraulic mean depth, equal to the wetted area Aw divided by top surface width T.
Because this is a 2-D model that ignores the curvature of the pipe, the true meaning of h will also be
ignored (again contributing to the results being “for illustration only”).

1. Add a new mesh block, ignoring it’s dimensions for now.

2. Rearrange the Froude equation and solve to find the mean upstream velocity U = 210 cm/s.
3. Make the upstream (x-min) boundary velocity-type:
a. Set X-Velocity = 210 cm/s.
b. Set Free Surface Elevation = 5 cm.

Compare Estimates of Upstream Turbulent Kinetic Energy and Dissipation

Boundary conditions have zero turbulent kinetic energy (TKE) by default. This is often correct if the fluid
enters from a resting state like a reservoir. An easy but time-consuming way to model upstream
turbulence is to model the upstream pipe flow explicitly, and then make a restart simulation with an
upstream grid-overlay (GO) boundary to copy the calculated TKE to a new boundary in a smaller domain.

A less certain but much faster approach is to use empirical estimates of turbulent parameters. There are
several ways to approach this. Ideally the approach is consistent in all respects with the turbulence
model and flow geometry. In this case, although you will use the RNG turbulence model and a partially
full pipe modeled as a 2-D plane, the empirical estimates below are derived from the standard k-e
model and empirical approximations for full round pipes. The discrepancy between approaches is a
source of uncertainty.

where:
Tu = initial intensity of turbulence (dimensionless, equation is empiric for fully-developed pipe flow),
Re = pipe Reynolds number (dimensionless),
2
Aw = wetted area (cm ),
DH = characteristic hydraulic length, taken to be the pipe hydraulic diameter (cm),
h = depth of 2-D flow (cm),
2 2
kT = mean turbulent kinetic energy per unit mass (cm /s ),
2 3
εT = rate of dissipation of turbulent kinetic energy per unit mass (cm /s ),
LT = turbulent length scale, aka turbulent mixing length, about 7% of pipe diameter (cm),
Pw = wetted perimeter (cm),
U = mean fluid velocity (cm/s),
μ = fluid dynamic viscosity (g/cm/s),
3
ρ = fluid density (g/cm ),
 3
The upstream depth of flow is 5 cm, velocity is 210 cm/s, density is 1 g/cm , and viscosity is 0.01
g/cm/s. The Reynolds number is therefore 4.2E5, so the flow is fully turbulent. Assume the pipe is
hydraulically smooth. The following empirical upstream boundary turbulence values apply:

 Tu ≈ 0.0156 (1.56%)
2 2
 kT ≈ 16 cm /s
 LT ≈ 0.35 cm (potential range 0.25 – 0.65 cm)
2 3
 εT ≈ 30 cm /s .

A second way to approach the problem is to put brackets on the initial turbulent intensity. Most pipe
flows exhibit Tu between 1% and 5%. In rivers Tu often exceeds 10%! The brackets give us two more
possible values:
2 2 2 3
Estimation Method kT (cm /s ) εT (cm /s )
Full-pipe Tu ≈ 1% 7 8
-1/8
Full-pipe Tu ≈ 0.16Re ≈ 1.6% 17 33
Full-pipe Tu ≈ 5% 165 998

***** Note that there is no widely agreed-on method of deducing upstream turbulent parameters, and,
as shown in Figure 3 below, the effect of an arbitrary value can be significant! For this reason, some CFD
modelers prefer to model upstream turbulence explicitly, even if it requires more than one simulation.
Figure 3: Volumetric fraction of entrained air and jump shape at 3 seconds as a function of upstream
2 2 2 3
boundary turbulence (top to bottom: kT = 80, 660, 4400 cm /s , εT = 140, 3600, 61500 cm /s )
Problem Specification (Run 1: Scalar Air Entrainment)
General, Physical, and Numeric Specifications
o Simulation name: Air Entrainment Scalar
o Simulation units = CGS
o Simulation time = 5.0 seconds
o Number of processors = 2
o Initial time step = none needed

Physics
o A ir
Entrainment
 rate coefficient = 0.5,
 surface tension coefficient = 73,
3
 air density = 1.2E-3 g/cm ,
o
Gravity
2
 gx = 97.5136 cm/s ,
2
 gz = - 975.136 cm/s ,

o Viscosity &
Turbulence
 viscous flow,
 renormalized group (RNG) turbulence model,
 dynamically-computed TLEN

Fluid Properties (generic properties for water)

3
o Density ρ = 1.0 g/cm
o Viscosity μ = 0.01 poise (i.e. g/cm/s)
2 2
o Compressibility = 1/(ρ w ) = 4.653E-11 cm-s /gm

Output
o Additional Output = Hydraulic Data
o Selected data at 1% time
intervals
 dynamic viscosity
 hydraulic data (add to Additional Output)
 volume fraction of entrained air

Geometry, Initial, and Boundary Specifications

o No geometry. Walls will be defined as boundary conditions.
o Boundaries and Initial Regions as shown in Figure 4, below.
o Initial Pressure = Hydrostatic

Measurement Stations
o Flux Surface Baffle w/ Porosity = 1 at X = 95 cm
o History Probe at X = 60 cm, Y = 0.5 cm, Z = 7.5 cm
o Sampling Volume from X = 60 to X = 95, Y = 0 to 1, Z = 0 to 15 cm

Set up the model and check each bullet off as you go.
 X-min = 0 cm X-Direction Total Cells = 200
z Y-Direction Total Cells = 1 Z-max = 15 cm
Boundary = Z-Direction Total Cells = 30 X-max = 100 cm
Velocity Y-min = 0, Y-max = 1 Boundary = Wall Boundary =
U = 210 cm/s X-Low2 = 40 cm FLUID REGION 2 Stagna. Pressure
FSE = 5 cm Z-Low2 = 5 cm
2 2
TKE = 16 cm /s U2 = 0 cm/s P = 6000 g/cm/s2
DTKE = 30 FLUID REGION 1: Z-High1 = 5 cm, U1 = 210 cm/s Fluid Fraction = 1

Z-min = 0 cm, Boundary = Wall x

Figure 4: 2-D definition sketch w/ initial fluid regions. Measurement locations not shown.

Check the Set Up

Use the problem specification statement and the sketches above and below to set up the simulation. If
you are unsure of a value or selection, ask the instructor. Check that you’ve added all the definitions.
You may need to right-click the History Probe and Baffle to adjust their transparency and the probe’s
size. Note that there is no solid geometry in this simulation.

Figure 5: Finished mesh

Run #1: Scalar Air Entrainment

1. Run the simulation: it should take minutes or less.
2. Check if the flow has reached steady state by looking at the runtime plots. It would be difficult to
claim that flow is at steady state.
3. Check your work by 2-D plotting Dynamic Viscosity and Volume Fraction of Entrained Air as
contour variables using Selected data. See that the boundary layer develops rapidly and
observe the pattern and magnitude of scalar air entrainment.
4. Check the final time step output against Figure 6 below before proceeding. If your results don’t
match, try to determine why, and then ask an instructor for help.
 Figure 6: Check your scalar results visually at the final time-step

Use Analyze > Probe tab > General History data to check the results for the History Probe, Flux Surface,
Sampling Volume, and Total Domain. These three measuring techniques are the built-in methods of
tracking entrained air. Activate the following check-boxes (units are given following the variable):

3 3
 fluid probe #1: volume fraction of entrained air: cm air/cm air-water mix
3
 fluid 1 volume flow rate @ flux surface 1: cm /s air-water mix x +/- unit vector
3
 entrained air flux @ flux surface 1: cm /s air, may be negative x +/- unit vector
3
 sampling volume 1: entrained air volume: cm air at time t
3
 entrained air volume: cm air at time t in the entire domain including ghost cells

Note that the outputs above count only entrained air scalar and do not include resolved void bubbles.
Do not use Mesh Dependent History data, as it will only report the specified air volume fraction in the
ghost cells (zero), and not the calculated volume crossing it.

Example Calculation for Flux Plane:

1. Select Units > CGS to avoid converting the output units.
2. Render the Probe > General History data variables above as Text. The variables are written left-
to-right in the same order that they are listed (top-to-bottom) in the selection list.
3. Find TIME = 5 seconds (the last output time).
3
4. Find FFLUX1 (the Fluid 1 Flow Rate at the Flux Plane) = 1050 (cm air-water mix)/s. Since Density
Evaluation physics are not activated, the density of the mixture is the density of pure water. No
density variation due to entrained air is included. Therefore the liquid flow rate through the
3
plane QV,WATER = 1055 cm water/s.
5. Find FPSCL1 (the volumetric air concentration at the plane) = 96.1 “amount”/s. The units for
FPSCL1 are tricky: they represent a scalar, so FLOW-3D reports them as g/s. In actuality, scalars
can be defined with any consistent measure of amount unless Density Evaluation physics are
active: then density scalars must be defined as mass/volume. So what is the measure of
“amount” for the scalar Air Entrainment model? It’s a dimensionless volumetric fraction: the
volume of air per volume of liquid. When the fraction and velocity are integrated over the area
of the flux plane, the result is a volumetric flow rate (despite what FLOW-3D reports). So FPSCL1
3
= QV,AIR = 96.1 cm air/s.
6. % concentration of air in the mixture = QV,A/QV,MIX = 96.1/(1055 + 96.1) = 8.3% by volume
crossing the flux plane at t = 5 seconds. This is an instantaneous value in an oscillating flow; the
time-average is better for reporting and is 6.6% by volume over the final 2.5 seconds.
The Scalar model should only be used without Density Evaluation physics when the volume of air is less
than 1% or 2%. When the fraction is higher the mixture should include bulking and density effects.

Run #2: Add Density Evaluation Physics

When the entrained air exceeds some volumetric concentration, bulking begins to increase the mixture
volume and decrease the mixture density. Experiments suggest that bulking begins to affect the flow
when the entrained air is between 1% and 3% by volume. Modeling the bulking mixture effects requires
adding Density Evaluation physics.

1. Select Simulation Manager > File > Add Simulation Copy…

2. Name the copy Air Entrainment Density.
3. Activate Model Setup > Physics > Density Evaluation > First Order Approximation to Density
Transport Equation. Entrained air transport models must use the transport equation option.
4. Open the Air Entrainment physics dialog. Notice how Activate Bulking is now active.
5. No other changes are required. Save and Run Simulation.

The simulation will run more slowly with the added physics, but will still finish quickly due to the small
number of cells (6000 total).

1. Open the results file. Make sure you’re not using the old results from the last simulation!
2. Check Solver Warnings: there are many F-packing errors at the toe of the jump. Ignore them.
3. 2-D plot Volume Fraction of Entrained Air, using Contour Limits = Global, and Vector Type = No
Velocity Vectors.
4. Look for the extent and characteristics of the entrained air. The air now rises to the surface, but
cannot leave the water, even at free surfaces (a limitation that will be addressed in the next
run).
5. Compare your last output to Figure 7, below. Did you set up the model correctly?

Figure 7: Check your density-evaluation results visually

The Flux Plane is visible in the plot as a dotted line. Calculating as before, we find at t = 5 seconds:
3
1. QV,MIX = FFLUX1 = 1236 cm air-water mixture/s: 25% higher than in the scalar run! Explain why.
3
2. QV,AIR = FPSCL1 = 78 cm air/s. Note that the units are still incorrectly reported as g/s: when
3
baffle scalar flux FPSCL1 is entrained air its units are cm /s.
 3. % concentration of air in the mixture = QV,AIR/QV,MIX = 6.3% by volume at t = 5 seconds. The
time average is 6.7% by volume over the final 2.5 seconds, similar to the scalar-only physics
results. The magnitude of oscillation is much smaller with density physics than with scalars only.

Run #3: Add Drift-Flux Physics (Two-Phase Flow)

The results of the model with Density Evaluation physics look more physical than with scalar-only
entrainment. This indicates that the bubble concentration cannot be modeled as if it were molecularly
dissolved in the water. Now consider the possibility that the air is entrained as a cloud of tiny bubbles,
which not only bulk the fluid (volume and buoyancy effects) but interact with the flow in three
additional ways: by drifting at a different velocity than the water, by exerting form and wake drag on the
water, and by escaping to the atmosphere through the free surface. These extra three effects are
modeled by Drift-Flux physics. With the drift-flux model, you define the properties both of a continuous
Phase #1 (in this case, water), and a dispersed Phase #2 (air bubbles).

1. Make a copy of Air Entrainment Density and name it Air Entrainment DriftFlux.
2. Activate Drift-flux on the Physics tab and apply water air bubble values per Figure 8, below.
3. Save and Run Simulation.

Figure 8: Parameters for air bubbles and water for drift-flux model
A brief description of the parameters in Figure 8 follows. Values for air bubbles can be found in the
literature, and a description of the drift-flux model equations that can be used to fit parameters to
experimental data can be found in the User Manual and in Technical Note 83 at users.flow3d.com. Note
st nd
that the Drift Flux model requires 1 or 2 order Density Evaluation physics.

 Drag Coefficient is the fully-turbulent limit of the drag coefficient CD. FLOW-3D uses a smooth
function that gives Stokes settling in the laminar regime and drag defined by CD in the fully
turbulent regime. The default value of CD is 0.5, which is correct for dense spheres. Values of
0.75 to 1.5 are typical for sand and gravel. 0.95 has been demonstrated to be accurate for
3
buoyant spheres (ρS < 900 kg/m ), where the rising sphere exhibits a corkscrew motion due to
pressure/buoyancy effects that causes more drag; see Karamenev (1992).

 Average Particle Radius: typically taken to be between 0.5 mm and 1 mm for air bubbles, but
the data scatter is significant. Impinging jets often exhibit bidisperse-normal distribution of
bubble radius, which complicates the estimate of a single mean radius.

 Richardson-Zaki Coefficient Multiplier: this should be 1.0 for the standard Richardson-Zaki
correlation, which describes the additional effect of particle-particle (bubble-bubble) wake
interaction in a concentrated cloud of the dispersed phase. The multiplier can be used to curve-
fit existing data, if needed. See the Tech Note for the full equation.

 Viscosity of Two-Phase Mixture: the viscosity used in the Navier-Stokes cell-to-cell momentum
equations, regardless of phase concentrations. Typically assumed to be the viscosity of the
continuous phase, so here it is the viscosity of water. Values entered here will change the value
of Fluid 1 Viscosity on the Fluids tab. If the cell-to-cell viscosity varies, consider borrowing the
solidification variables as discussed in the lecture on Multiphase Flow. An example of this
situation would be two liquids (water and oil, e.g.) that flow as unmixed phases in different parts
of the domain. Here, the continuous phase will always be water.

 Viscosity of Phase #1 is the viscosity used in the in-cell evaluation of relative (separation)
velocity between the continuous and dispersed phases. Set this to the viscosity of water also.

 Viscosity of Phase #2 is the viscosity of the dispersed phase if it were to become so

concentrated as to be continuous. This only has meaning for dispersed phases that can flow,
such as oil droplets. Leave it blank, and it will default to Fluid 1 Viscosity, that of water. This is
the best representation of the system, because no matter how many air bubbles are present,
they are still separated by thin liquid films.

 Density of Phase #1 is the density of the continuous phase, pure water. Values entered here will
change Fluid 1 Density on the Fluids tab.

 Density of Phase #2 is the density of the dispersed phase, pure air. This value was entered on
the Air Entrainment physics dialog, and changes made here will change the value there.

 Minimum Volume Fraction of Phase 1 is the irreducible fraction of water allowed in a cell. Unity
minus this value is the maximum fraction of air allowed in a cell. The correct value should be
 some small number that reflects the thin film of water between bubbles, but this value is not
known, so use 0 as a reasonable approximation.

 Maximum Volume Fraction of Phase 1 is the upper limit fraction of water permitted in a cell. 1–
this value is the irreducible minimum air permitted in a cell. Water can be present without air in
it, so 1 is appropriate.

 Volume Fraction of Phase #2 at Inversion Point is the maximum volumetric fraction of air in a
cell before the continuous and dispersed phases switch so that air is considered continuous. If
modeling another liquid, 0.5 would make sense. Because we assume air bubbles will always be
discontinuous (separated by films), 1 is appropriate. The equations of the Drift Flux model do
not make physical sense for any other value when the two phases are different states (liquid vs.
solid or gas).

 Allow Gas to Escape at Free Surface allows phase #2 to disappear at free surfaces and should be
activated for the best physical representation of bubbles.

The model will take even longer to run than the last one, due again to the added physics. Review the
results as before, but use Auto contour limits this time. Compare your results to Figure 9.

Figure 9: Final time step (t = 5.0 seconds) for drift-flux model version of problem

Calculate as before to find the instant and time-averaged air flow:

3
1. QV,MIX = FFLUX1 = 1141 cm mixture/second at t = 5 seconds.
3
2. QV,AIR = FPSCL1 = 82 cm air/s, 5% greater than in the density-evaluation only model.
3. % concentration of air in the mixture = QV,AIR/QV,MIX = 7.2% by volume at t = 5 and 6.6% by
volume averaged over the final 2.5 seconds, about the same as the density-only model.

One last calculation:

1. View the results in 2-D using Restart Data and Contour Variable = Macroscopic Density.
2. Now use Probe > General History Data to plot text for fluid 1 volume flow rate @ flux surface 1
(FFLUX1) and mass flow rate @ flux surface 1 (MFLUX).
3
3. Divide the mass flow rate (g/s) by the volume flow rate (cm /s) at the last time step to find the
spatially-averaged mixture density ρMIX at the flux surface. Use the equation:
The percentage of air calculated from the density is the same as was determined by dividing entrained
air flux @ flux surface 1 (FPSCL1) by fluid 1 volume flow rate @ flux surface 1 (FFLUX1). This verifies
that FPSCL1 is the volume fraction of air in the mixture, not g/s. You can use this approach for any
density scalar to determine if the reported units are correct.

Intro to Testing for Uncertainty

One interesting test is to examine the model sensitivity to the air bubble size. If the mean radius is
increased from 0.5 mm to 1 mm, the entrained air flow rate increases by a factor of 3.75! The sensitivity
of this input increases uncertainty in the model. It’s an example of input uncertainty.

Another important test is the mesh dependency of the results. Turbulence is known to be quite mesh
dependent, and since the physics model uses TKE to calculate air entrainment, the volume of entrained
air is also likely to be mesh dependent. In this case, halving the cell sizes in each direction twice for a
total of three runs on three meshes changes the time-average air concentration from 6.6% to 9.8% and
then to 4.0%. Note that the variation is mostly due to the effect of the cell size on the hydraulic jump
itself: the jump location experiences significant boundary effects from the nearby downstream pressure
boundary, and so it’s a poorly-posed problem to begin with.

One way to use the air entrainment model is to find a mesh that you can live with and then use it to
model a known air entrainment experiment with similar Reynolds and Froude numbers as the target
model. Calibrate the air entrainment rate coefficient and/or bubble size until you match the experiment,
and then model the target case.

If no experimental data is available with similar Fr and Re, then successively vary the mesh resolution, air
entrainment rate coefficient, and bubble size. Use probability distributions that describe the likely
possibilities of the “real” value of the input variables to reduce the number of models, if possible. Find
the range of CFD output and develop uncertainty bands around the mean results.

The above two options are time-consuming and are appropriate when quantitative results are needed,
like the exact volume of entrained air. It’s much easier to use air entrainment physics when all you need
are qualitative results. For example, if one design option causes much less air entrainment than the
other, you can pick the better design without worrying much about the input and numeric uncertainty!

References

Brethour, J.M. and C.W. Hirt (2009). Drift Model for Two-Component Flows. FSI-09-TN83Rev.
Flow Science Inc.: Santa Fe, NM.

Hirt, C.W. (2003, 2011). Modeling Turbulent Entrainment of Air at a Free Surface. FSI-03-TN61R.
Flow Science Inc.: Santa Fe, NM.
Exercise 7: Intro to General Moving Objects
 Water & Environment Training on FLOW-3D v11
Exercise 7: General Moving Objects

Hydraulics Exercise: General Moving Objects

This is a simple example to illustrate the general moving object model in FLOW-3D. Surface tension and
viscosity are also used. By now you should be familiar with the basic set up methods in FLOW-3D.

Problem Description and Objectives

The model represents a simple piston-driven inkjet print head with a contaminant (particle of dirt) in the
ink. There is air above the print head, liquid ink inside it, and a solid object in the ink.

Select General Settings

Begin on the Model Setup > General tab. The motion and compressibility of the air can be neglected in
this case, so use the one fluid method. The remaining parameters are selected based on knowledge of
the physical case.

1. Number of Fluids = One

2. Interface Tracking = Free surface or sharp interface
3. Flow Mode = Incompressible
4. Units = CGS / Kelvin
5. Finish time = 1E-4 seconds
6. Number of Processors = 2

Select Physics Options

1. Moving and Simple Deforming Objects

a. Activate general moving objects (GMO) model
b. Select Implicit for the Moving object / fluid coupling. The Implicit option is usually better
when a moving object will have coupled (as opposed to prescribed) motion. The dirt particle
will move freely with the fluid.

2. Viscosity and turbulence

a. Activate viscous flow
b. Select Laminar as the turbulence option because the flow is expected to have a low
Reynolds number.

3. Surface tension
a. Activate surface tension model
b. Choose Explicit approximation
c. The remaining parameters will be entered on the Model Setup > Fluids tab.

Page 1 of 10
Define Fluid Parameters

Only the fluid density, viscosity, and surface tension properties are needed for this simulation.

1. Fluid 1
a. Name = Ink
3
b. Density = 1 g/cm
c. Viscosity = 0.06 Poise (g/cm/s)
d. Surface tension:
2
i. Surface tension coefficient = 73 g/s
ii. Contact angle = 15 degrees. This is a global parameter, and applies to wall boundaries
and subcomponents that do not have a different contact angle specified as a surface
property.

Define Geometry

The general process for the Model Setup > Meshing & Geometry tab will be to define the geometry,
assign properties to the components, define the mesh and boundary conditions, and then set the
initial state of the fluid and void. Start by defining the components. The tank will primarily be
defined using the mesh boundaries, but the filling tube and valve will be components.

1. Geometry: defined in the Geometry Window .

a. Nozzle
i. Create a solid cylinder subcomponent with the following dimensions. Name the new
Component 1: Nozzle.
1. Z low = undefined (blank, not zero), Z high = 0.15 cm
2. Cylinder outer radius = undefined (make it blank after you build the cylinder)
3. Cylinder inner radius = 0.0105 cm. The inner radius is like pipe inner diameter; it
must set it in the tree under Limiters and with the undefined outer radius will make
everything outside the inner radius solid.
ii. Create a solid torus subcomponent with the following dimensions:
1. Component = Nozzle
2. Torus Minor radius = 0.0075 cm
3. Torus Major radius = 0.0105 cm
4. Z low = undefined, Z high = 0 cm. This must be set in the subcomponent tree
Limiters. It cuts the torus across the centerline so only the lower half remains.
5. Cylinder outer radius = 0.0105 cm. Also set in the subcomponent Limiters; this
bevels the outer surface of the torus so it’s straight perpendicular in z.
iii. Move the torus subcomponent into position by translating it 0.015 cm in Z.
 iv. Check the effect of the additional limiters by FAVORizing the geometry: use a
Renderspace with extents -0.02 to 0.02 in all directions and 0.0005-cm “cells”. Because
the part is small relative to the unit system, the camera may not be zoomed in enough

when you Render. Use to reset the view.

b. Piston
i. Create a cylinder subcomponent and assign it to a new Component 2: Piston.
1. Z low = undefined, Z high = 0.002 cm
2. Cylinder inner radius = undefined, Cylinder outer radius = 0.0085 cm. This makes
the piston solid.
3. FAVORize again and check your work against the picture below. You may need to
adjust transparency and/or rotate the view to see the piston inside the cavity.
c. Dirt Particle
i. Create a sphere subcomponent and assign it to a new Component 3: Dirt.
1. Sphere outer radius = 0.001
2. Translate it 0.012 cm in Z.
Now add Component Properties to the Piston and Dirt components:

2. Use the tabs at the bottom of the Component Properties Window to toggle between the
components. The Type of Moving Object setting must be set in the Geometry Window under
the Component itself.

a. Component 2: Piston: the piston will move in the z-direction to eject an ink droplet.
i. Component 2 > Type of moving object = Prescribed
ii. Component 2 > Component Properties > Type of Moving Object > Edit Moving Object
Properties > Initial/Prescribed Velocities tab > Tabular Z Translational Velocity
Components per the table below (see next page).

b. Component 3: Dirt: the particle will move with the fluid.

i. Component 3 > Type of moving object = Coupled
ii. Component 2 > Component Properties > Type of Moving Object > Edit Moving Object
Properties > Motion Constraints tab: all translations coupled, all rotations prescribed
(defaults to zero, no rotation permitted).
3
iii. Same dialog > Mass Properties tab: Mass Density = 7.6 g/cm

Check your Moving Object Setup dialogs against the images on the next pages.
Define Mesh, Boundary, and Initial Conditions

Now add a 2-D cylindrical mesh around the geometry.

1. Mesh: defined in the Mesh Window . Right-click Mesh – Cartesian and select Change to
Cylindrical Coordinates. Add a mesh block and define the 2-D cylindrical mesh to enclose the
geometry and align the grid lines (using X Mesh Planes) with the outside edges of the piston and
nozzle as shown on the next page.
2. Open the Boundaries branch of the Mesh – Cylindrical tree.
a. Z-min boundary is Wall-type.
b. Z-Max boundary is an Outflow boundary to allow fluid to exit the domain smoothly.
c. All other boundaries are or Symmetry-type.

The initial conditions for this simulation are an initial depth and pressures such that output will be in
gauge pressure (local ambient atmospheric pressure = 0).

3. Open the Initial Conditions Window .

a. Global > Pressure = Uniform Pressure
2
b. Uniform Pressure = 0 Barye (g/cm/s )
c. Global > Fluid Initialization = Use Initial Fluid Elevation
d. Initial Fluid Elevation = 0.0149 cm above Z = 0.
2
e. Global > Void Initial State > Pressure = 0 g/cm/s
Set Output Options

The default output rates are acceptable. Add Fluid Velocities and Pressure as Selected Data. The GMO
location is always output in any time step, so adding any Selected Data will give 101 GMO visualizations.
If you wanted more frequent numeric GMO output (e.g., mass-center coordinates, forces acting on the
th
object, etc.), you could set History Data Interval < 1E-6 seconds (the default 1/100 of the simulation).

Set Numerics Options

There are some numerical options that will help make this simulation run well.
1. Initial Time Step DELT = 1E-9 seconds
2. Momentum advection = Second order monotonicity preserving
3. Viscous stress solver = Successive under-relaxation
4. Multiplier for Convergence Criteria EPSVIS = 0.9
5. Over-Relaxation Factor OMEGVS = 0.8
6. Maximum Number of Iterations ITVSMX = 50

Run and Troubleshoot the Simulation

1. Run the simulation. If it’s set up as defined above, there will be dozens of consecutive pressure
iteration failures.
2. Terminate the simulation: select it in the Queue on the Simulation Manager tab and click .
3. Look at Diagnostics > Solver Errors to find the IJK locations of the failures. At first most of them
occur at i = 46, j = 2, k = 3. This is just above the bottom of the domain. Later iteration failures
occur at different locations. This doesn’t look like a problem with a boundary condition, because
the Z-min boundary is Wall type.
4. Try giving the solver more overhead to converge: go to Numerics > Pressure Solver
Convergence Controls and set Minimum Number of Iterations ITMIN = 2, Maximum Number of
Iterations ITMAX = 500, and GMRES Subspace Size MRSTRT = 50. These options will only slow
down the solver, and will not relax the convergence criteria. If they work, the solver will run
more slowly, but accuracy will either be improved or will not be affected.
5. Run the simulation again. This time there should be a few (around 3) pressure iteration failures
at the very start of the simulation, and after that there should be none.
6. Check the Solver Text for the implicit solver convergence performance: pressure res/epsi
should be small (around 1E-7 to 1E-2) and viscous stress res/epsi should be on the order of E-1.

Check for instances where the viscous solver did not converge. The default max iterations for the
Successive Under-Relaxation Viscous Solver is iter = 20, which you raised to 50. Are there any lines
where res/epsi > 1, and an asterisk * is written next to res/epsi to mark a failure? Why not?
There are a few viscous stress iteration failures, but they are not consecutive and, unlike pressure
iteration failures, they do not affect the time step dt.

Postprocessing Visualization

1. Plot Selected Data for Pressure, and then Velocity Magnitude, in 2-D with No Velocity Vectors.
2. Activate the Horizontal Symmetry checkbox and Render.
3. Compare your output to that shown below at t = 3.9E-5 seconds.
4. Plot Probe > General History data > Component 3 GMO Shear Force Magnitude in Space
System to see the particle drag magnitude.

The dirt gets trapped in the ink droplet and drag makes the droplet move more slowly and elongate
the droplet. This would cause poor printer performance.
Exercise 8: Intro to Saturated Porous Media
 Water & Environment Training on FLOW-3D v11
Exercise 8: Saturated Porous Media

Hydraulics Exercise: Saturated Porous Media

Problem Description & Objectives

The flow of water through a crushed coal media filter will be simulated using the Saturated Porous
Media physics model. Drag coefficients will be determined from experimental data given by Abood
(2009). The model results will be compared to the experimental results. The purpose of this exercise is
to practice setting up and parameterizing a saturated porous media problem.

FLOW-3D Learning Objectives

In this exercise you will practice techniques to:

1. Select a saturated porous drag model,

2. Compare empirical coefficients to project-specific coefficients,
3. Calculate FLOW-3D drag coefficients from more common coefficients,
4. Set up a simulation from a list of specifications,
5. Specify a boundary hydrograph so boundary velocity varies with time, and
6. Analyze simulation results and compare them to experimental data.

Review the Physical Experiment

The physical experiment is described in An Experimental Model for Flow Through Porous Media Using
Water Filter, by D. W. Abbood of Al-mustansiryiah University, Baghdad, Iraq (published in the
proceedings of the Thirteenth International Water Technology Conference, 2009, Hurghada, Egypt). The
experiment consisted of recirculating clean water through parallel columns of packed media and
measuring head loss vs. velocity. Multiple experiments were conducted with varying flow rates (0.001
to 0.020 m/s), varying temperatures (9 °C to 57 °C), varying porosities (0.36 – 0.65), and varying
characteristic grain size (0.5 mm to 2.0 mm). Five different media types were studied: crushed sand,
crushed coal, glass beads, crushed garnet, and crushed porcilinaite. Measurements were taken from
three flow meters and four differential pressure gauges. The media was packed as homogenously as
possible using a constant-height-of-fall pluviation method. Porosity for each media was determined
from the packed column volume and mass and the known microscopic media density. Several tests were
performed to ensure the packed bulk density of the media was repeatable. The experimental setup is
shown in Figure 1, below.

Page 1 of 19
 Figure 1: Experimental apparatus from Abood (2009)

Select Model Approach

The media tubes are straight cylinders without bends, so the Cylindrical coordinate system can be used.
Select the number of dimensions of the model:

 1-D
o Pros: fastest run-time, easiest analysis.
o Cons: velocity and pressure profile due to inlet/outlet effects and wall effects not modeled.
 2-D
o Pros: fast run-time, easy analysis, inlet/outlet/wall effects can be approximated.
o Cons: Must assume axially and radially symmetric flow about z-axis.
 3-D
o Pros: inlet/outlet/wall effects and non-axial/radial symmetric flow can be modeled.
o Cons: slowest run-time, most difficult analysis (requires spatial and temporal averaging).

Based on the above, a 2-D run in cylindrical coordinates is suggested to model axial and radial flow
symmetrically around the bed tube axis.
Select Physics Models

Physics models should be considered carefully for all simulations. Here we will check whether gravity,
surface tension, viscous flow, turbulence, and porous media physics are needed.

GRAVITY
Because the media vessel is arranged vertically, we might assume that gravity must be included.
However, looking at the results indicates the opposite: even at very low flows, there is still a head loss
across the media. We can demonstrate this using the Extended Bernoulli Equation: if the fluid is
incompressible, this relation holds true for any two points 1 and 2 along a flow path:

Here E represents energy. Subscript Z is the potential (elevation) energy, P is the pressure energy, and V
is the kinetic (velocity) energy. Subscripts A and E are energy added or removed by hydraulic machinery,
f is the energy lost due to friction, and 1 and 2 are two points along a streamline. In this case because
the tubes are uniform diameter, the velocity is constant (conservation of mass), and the velocity and
hydraulic machinery terms can be omitted. Converting the remaining terms from energy to hydraulic
head (in units of feet) gives:

where P and z are pressure and elevation at point 1 or 2, ρ is the fluid density, g is gravitational
acceleration, and hf is head loss due to friction. Let z2 be the outlet reference elevation and z1 be the
inlet elevation: they will be 0.4 m apart. Frictional losses will be zero when there is no flow, and the
above equation then indicates that the static pressure head at the outlet will be greater than the static
pressure head at the inlet by 0.4 m. This implies that at flow rates near zero, the pressure will increase
from the inlet to the outlet.

The graphical results reported in the paper show a head loss at all flow rates. We can (correctly) infer
that the researcher has already accounted for static head before reporting the results. The experimental
results reported are purely a function of the inlet and outlet pressures in the absence of gravity. They
would be equally valid for horizontal flow through the same media. Therefore do not use Gravity and
Non-Inertial Reference Frame physics.

SURFACE TENSION

The Surface Tension physics model is never active inside porous media components, since surface
tension effects are included empirically in the Porous Media physics, including capillary action. Surface
Tension physics can be used outside of porous media where there are free-surfaces. Since there is no
free surface in the model, do not use Surface Tension physics.
VISCOSITY AND TURBULENCE

Walls are free-slip surfaces when flow is modeled as inviscid. We should include viscosity at the least so
that wall effects are not neglected. Viscous Flow physics should be used. FLOW-3D always assumes that
flow in porous media is Laminar, but turbulence may be included for free-water regions above and
below the media.

Check the assumption that flow in the media will be laminar: as will be calculated below, the pore
Reynolds numbers (Reynolds number within a pore of the media) will be less than 40. True turbulence
(eddies) in porous media begins at approximately ReP = 100 to 500. This is a different threshold than
form drag, the effects of which begin to occur around Rep = 1. FLOW-3D’s assumption of laminar flow in
porous media is valid for this case.

Check whether turbulence occurs in free-water regions: the media-filled pipe is 0.24 m in diameter, and
the maximum free-stream velocity is 0.02 m/s. Hand calculation shows that ReD outside the media =
4,800. Turbulent flow in circular pipes begins at approximately ReD = 4,000. The flow may be turbulent
at the maximum velocity, but will be laminar at lower velocities. Since the flow is delivered to the tubes
via small-diameter pipes, it is probably entering the tubes with some turbulence for most flow rates.
Activate viscous flow and RNG Turbulence physics.

Select a Porosity Model

Porous media flow exhibits viscous and inertial drag and may be enhanced or retarded by capillary
effects. There are many empirical approaches that predict the velocity, pressure, and fluid fraction of
flow within the media. Several of these approaches are incorporated into FLOW-3D as Porous Media
physics options. Each approach uses different coefficients and is valid for a specific set of conditions.
Because fitting experimental data to empirical equations is a detailed process, it is important to first
identify the model which best represents the physical case.

Given that the media is saturated at all times during the experiment, we will disregard the Unsaturated
Flow model, and the three drag models associated with it.

There are three choices left: the D’Arcian Saturated Drag model, Permeability Dependent Saturated
Drag, or the Forcheimer Saturated Drag model. In general, the D’Arcian model applies when pore size is
small and flow is relatively slow: ReP < 1. The permeability and Forcheimer models are different
formulations of the same quadratic drag curve. They are valid for all ReP up to the onset of true
turbulence, and fit many experimental data sets. You should decide D’Arcian model and the two
quadratic models on the basis of the pore Reynolds number, ReP:

where ρ is the fluid density, μ is the fluid dynamic viscosity, Dp is the characteristic pore diameter, and
Up is the mean microscopic (or pore) velocity of the fluid. Microscopic velocity is almost never measured
directly due to the difficulty of inserting probes between grains. Instead, macroscopic (or superficial)
velocity Umacro is measured as the transit time across the media, neglecting tortuosity. Macroscopic and
microscopic velocities are theoretically related by the media effective porosity:

where ηeff is the average effective porosity. Effective porosity is the porosity that is available for flow,
and excludes pores that cannot be filled or drained. In a completely saturated loose media ηeff can be
assumed to be the same as bulk porosity η (0.50 in this case).

The experimental velocity given in the paper is superficial velocity Umacro (range 0.001 – 0.020 m/s). Use
porosity to convert to macroscopic velocity to microscopic velocity, and then find the pore Reynolds
number for each velocity assuming the pore diameter Dp is equal to the average grain diameter (1 mm):

EXPERIMENTAL CALCULATED CALCULATED

SUPERFICIAL MICROSCOPIC PORE
VELOCITY VELOCITY REYNOLDS #
Umacro Up = Umacro/ηeff ReP (dimless)
(m/s) (m/s)
0.001 0.002 2.0
0.002 0.004 4.0
0.003 0.006 5.9
0.004 0.008 7.9
0.005 0.010 9.9
0.006 0.012 11.9
0.007 0.014 13.9
0.008 0.016 15.8
0.009 0.018 17.8
0.010 0.020 19.8
0.011 0.022 21.8
0.012 0.024 23.8
0.013 0.026 25.7
0.014 0.028 27.7
0.015 0.030 29.7
0.016 0.032 31.7
0.017 0.034 33.7
0.018 0.036 35.7
0.019 0.038 37.6
0.020 0.040 39.6
Table 1: Calculating microscopic velocity and pore Reynolds number

The pore Reynolds number ranges from 2.0 to 39.6. The D’Arcian flow regime is strictly speaking only
valid to ReP = 1, but is generally assumed accurate up to ReP = 10. Therefore select the Permeability
Dependent Saturated Drag model, which is accurate for the D’Arcian region and outside of it. The
particle size (1 mm) and porosity (0.50) are larger than in the soils D’Arcy studied, so it makes sense that
his model does not apply here.
Review the Permeablity-Dependent Saturated Drag Equation

You need to determine two coefficients: D’Arcian Permeability (PERM) and Non-D’Arcian Permeability
(PERM2) for the porous media component. The equations to derive these values from known data are
given below as a formal exercise, based on the equations in the User Manual.

where K1 is the media’s experimental D’Arcian permeability and K2 is the coefficient of non-D’Arcian
permeability. ∆P/∆x is the pressure loss per unit distance traveled superficially, ρ is fluid density, and μ is
the fluid dynamic viscosity. Of the three saturated porous drag models, only Permeability-Dependent
Saturated Drag is written in terms of macroscopic velocity Umacro; the rest use pore velocity UP.

To use the relationship above, the experimental head loss ∆H needs to be converted to pressure loss ∆P.
They are related by density ρ and gravity g:

Figure 2 and Table 2, below, show the experimental head loss data (∆H/∆x), digitized using EnGauge
freeware, and the calculation of unit pressure loss (∆P/∆x) for use in the coefficient fitting equations.

Figure 2: Experimental head loss vs. superficial velocity

 SUPER. MICRO. PORE HEAD PRESS.
VELOCITY VELOCITY REYNOLDS LOSS LOSS
Umacro Up = Umacro/ηeff ReP (∆H/∆x) (∆P/∆x)
(m/s) (m/s) (dimless) (m/m) (Pa/m)
0.000 0.000 0.0 0.000 0
0.001 0.002 2.0 0.048 472
0.002 0.004 4.0 0.089 872
0.003 0.006 5.9 0.189 1,853
0.004 0.008 7.9 0.215 2,108
0.005 0.010 9.9 0.267 2,617
0.006 0.012 11.9 0.315 3,089
0.007 0.014 13.9 0.390 3,816
0.008 0.016 15.8 0.482 4,725
0.009 0.018 17.8 0.564 5,524
0.010 0.020 19.8 0.631 6,178
0.011 0.022 21.8 0.720 7,050
0.012 0.024 23.8 0.809 7,923
0.013 0.026 25.7 0.902 8,831
0.014 0.028 27.7 0.991 9,704
0.015 0.030 29.7 0.991 9,704
0.016 0.032 31.7 1.191 11,666
0.017 0.034 33.7 1.314 12,865
0.018 0.036 35.7 1.496 14,646
0.019 0.038 37.6 1.588 15,555
0.020 0.040 39.6 1.607 15,736
Table 2: Experimental Head Loss Converted to Pressure Loss

The pore velocity UP and the experimental pressure loss (∆P/∆x) is now known and the permeability
coefficients can be found.

Find Drag Coefficients

With the correlated variables known, there are two ways to find the drag coefficients:

 Fit literature drag coefficients from empirical relations to the superficial equation, or
 Fit case-specific experimental data to the superficial equation.

The second option is always a better choice when case-specific data exists. Compare the coefficients
from both options:

Empirical Correlations Without Experimental Data

The Ergun (1952) equation for packed spheres and a relation for crushed angular materials (sand) by
Idelchik (1986) are plotted below along with the experimental data.
 Figure 3: Empirical correlations are not recommended

The chart shows neither the Idelchik correlation nor the Ergun equation is a good fit. An historic
overview of empirical correlations with examples are given in the source paper (Abood 2009). The
purpose of this step was to demonstrate the poor fit of most empirical correlations. Now you will fit the
experimental data directly.

Curve-Fitting Experimental Data

The drag model equation is given below in simplified form for fitting experimental data:

( )
A and B are coefficients fitted to the experimental data. Use a statistics package or a spreadsheet
nd
program to fit a 2 order polynomial ‘trend line’ to the experimental data. This gives a well-fit
correlation as shown below.
 nd
Figure 4: Fit a 2 -order polynomial to the experimental data

This is obviously a better match than generalized correlations. From the fitted equation, we see that

A = 421,674

B = 19,540,327

Compare the fitting equation to the full equation with K1 (PERM) and K2 (PERM2):

( )

And then solve for K1 and K2 using the known fluid density and viscosity:

K1 = μ/A = 2.3715E-9

K2 = ρ/B = 5.1084E-5
Problem Specification

Set up the problem following the model specifications below:

General and Numeric Tab Options

o Simulation name: Saturated Porous
o Simulation units UNITS = SI
o Simulation time TWFIN = 3598 seconds
o Number of processors NPROC = 2
o Initial time step DELT = 0.001 sec
o Minimum number of pressure iterations ITMIN = 2
o Maximum number of pressure iterations ITMAX = 500
o GMRES subspace size MRSTRT = 25

Physics
o Porous Media - Permeability Dependent Saturated Drag
o Viscosity & Turbulence
 Viscous Flow
 RNG Turbulence with Dynamically-Computed TLEN

Fluid Properties (Water, measured experimentally @ 24 – 26 °C)

3
o Dens i ty ρ = 998.2 kg/m
o Vi s c osi ty μ = 0.001008 Pa-s (i.e. kg/m/s)
2
o Compressibility = 1/(ρ W ) = 4.55E-10 (assume speed of sound W ≈ 1483 m/s)

Output
o Selected data at default 1% time intervals
 Drag Function
 Fluid Velocities
 Pressure

Geometry
o 1 Subcomponent
 Shape = Cylinder
 Name = Packed Bed
 Radius = 0.12 m
 Z Low = 0.0 m
 Z High = 0.3 m
 Axis along origin
o 1 Component
 Name = Packed Bed
 Component Type = Porous
o Properties (Component Properties > Porous Properties)
 Porosity = 0.5
 Darcian Permeability = 2.3715E-9
 Non-Darcian Permeability = 5.1084E-5
Problem specification is continued on the next page:
Mesh Specifications
o Cylindrical (right-click on Mesh-Cartesian and select Change to Cylindrical Coordinates)
o 2-D Mesh Block
o Domain
 X (r) = 0 m to 0.12 m ----------------- # Total Cells = 5
 Y(θ) = 0 degrees to 30 degrees --- # Total Cells = 1
 Z (z) = -0.012 m to 0.312 m -------- # Total Cells = 20

Boundary Conditions
o X Min: Symmetry (S) (r = 0 is a discontinuity, and should always be symmetry)
o X Max: Wall (W)
o Y Min & Y Max: Periodic (Pd) (flow can rotate out one side and back in the other)
o Z Min: Gauge Stagnation Pressure (P)
 F=1
 P = 0 Pa
o Z Max: Velocity (V)
 varies with time: import velocities & times from a .csv file (described below)

Initial Conditions
o Global > Fluid Initialization = Use Fluid Elevation.
o Initial Fluid Elevation = 0.312 m.

Figure 5: Check the mesh, geometry, and boundaries

Specify a Time-Dependent Velocity Boundary
The time for a fluid packet to pass through the simulation can be found from the domain length and the
superficial velocity of the fluid (t = L/Umacro). This is the front time of the velocity profile. The pressure
field also needs time to come to a steady-state solution, so the front time should be multiplied by a
safety factor. Assuming a 2x safety factor for pressure, the table of time and velocity values looks like
this:

POROUS MEDIA LENGTH: 0.300 m

MODEL DOMAIN LENGTH: 0.324 m
STEADY STATE TIME MULTIPLE: 2
SUPER. FRONT EST. STDY TOTAL
VELOCITY TRANSIT STATE TIME
Umacro tfront tsteady,est. ttotal
(m/s) (s) (s) (s)
0.020 0
0.020 25 50 50
0.019 26 53 103
0.018 28 56 158
0.017 29 59 217
0.016 31 63 280
0.015 33 67 346
0.014 36 71 418
0.013 38 77 495
0.012 42 83 578
0.011 45 91 669
0.010 50 100 769
0.009 56 111 880
0.008 63 125 1,005
0.007 71 143 1,148
0.006 83 167 1,314
0.005 100 200 1,514
0.004 125 250 1,764
0.003 167 333 2,098
0.002 250 500 2,598
0.001 500 1,000 3,598
Table 3: Determining runtime of each velocity for the boundary profile

This is the basis for the Finish Time that was given above. FLOW-3D interpolates linearly between
tabular data points. Given the above table, FLOW-3D will generate an inflow velocity profile that looks
like Figure 6 below:
 Figure 6: Incorrect interpolation of velocity profile

This is not desired! The model should match the experiment. To do this, each velocity must be held
constant long enough for the results to become steady. Adding two points per velocity gives us the
desired profile, which looks like this:

Figure 7: Desired velocity profile

At each step in the velocity profile, the elapsed time between the lower step and the upper step is
arbitrarily selected to be one second, since FLOW-3D requires that each boundary condition time step
have only one associated value. The steps will cause rapid change at the inlet boundary condition, a
known cause of pressure problems! This is why increased pressure solver overhead was specified above.

Generate a table in a spreadsheet program, starting with the fastest velocity and working down, with a
header row, and save it as a .csv file. Note the negative values mean the velocity vector is pointing
toward the z-min boundary.
TIME W
0 -0.020
49 -0.020
50 -0.019
102 -0.019
103 -0.018
157 -0.018
158 -0.017
216 -0.017
217 -0.016
279 -0.016
280 -0.015
345 -0.015
346 -0.014
417 -0.014
418 -0.013
494 -0.013
495 -0.012
577 -0.012
578 -0.011
668 -0.011
669 -0.010
768 -0.010
769 -0.009
879 -0.009
880 -0.008
1,004 -0.008
1,005 -0.007
1,147 -0.007
1,148 -0.006
1,313 -0.006
1,314 -0.005
1,513 -0.005
1,514 -0.004
1,763 -0.004
1,764 -0.003
2,097 -0.003
2,098 -0.002
2,597 -0.002
2,598 -0.001
3,597 -0.001
Table 4: Time-step/velocity entries saved as .csv file

Import the table as follows:

 1. Copy BoundaryVelocity.csv to the project folder (Saturated_Porous).
2. Open the Z-Max boundary condition.
3. Select Specified Velocity as the Boundary Type.
4. Specify Fluid Fraction = 1.
5. Click on the Velocity button above the X, Y, and Z velocity components.
6. Select Import to Column = W in the lower right of the dialog.
7. Check Import from Simulation Directory to tell the interface where to start browsing.
8. Click Import Values.
9. Browse if necessary and open the BoundaryVelocities.csv file.
10. Click the header W to select the row to import (there is only one row in this file), and then OK.
11. You have the options of Import Tables to Prepin or Link External CSV Data to Prepin. The first
option will copy the data into the prepin file so the .csv file is no longer needed, but limits the
number of points per table to 500. The second option adds a link path to the prepin file to tell it
where to look: the .csv file is required to stay with the prepin file, but can have more than 500
points. For this case, select Import Tables to Prepin.
12. Check that the W column is populated, and that the chart appears as shown below.
13. Click OK to get out of the Boundary Dialog and Save your work.

Figure 8: Importing velocities from a .csv file

You could also enter all these values manually in the table, but importing from a .csv file is much easier.
Run the Simulation and Review Runtime Output

The model is completely set up, so go ahead and run it.

1. Review Solver Warnings . There are a few messages that the simulation is nearly steady.
They occur between velocity changes at the boundary, implying that the duration of the steps is
sufficient to get steady flow between velocity changes.

2. Select the Mass-Averaged Mean Kinetic Energy runtime plot and note how the kinetic energy
relates to the magnitude of the velocity profile.

Figure 9: MKE relates to velocity

3. Now take a look at the Pressure Iteration Count plot. The most iterations reported is 2 per
cycle: increasing the maximum number of pressure iterations permitted ITMAX = 500 was
unnecessary. Notice that even though the minimum number of iterations was specified ITMIN =
2, the solver does not always iterate twice.
 Water & Environment Training on FLOW-3D v11
Exercise 8: Saturated Porous Media

Analyze 2-D Results

1. Using what you already know, Open the results file and analyze the flow in 2-D.
2. Plot Pressure and Plain Velocity Vectors.
3. Turn off Contour Smoothing.
4. Make sure you’re plotting with SI units.
5. Check your work against the final time step shown below.

Figure 10: Pressure drop (Pa) across porous media at t = 3598 sec
 Water & Environment Training on FLOW-3D v11
Exercise 8: Saturated Porous Media

Analyze Results using Probe

1. Using what you know, plot Probe > Selected data for Pressure and Z-Velocity.
2. Plot results at the center of the top (inlet) of the cylinder (I = 2, K = 22).

The rest of the analysis is illustrated below. The steps are:

1. The pressure loss ∆P across the media is determined by subtracting the outlet pressure (I = 2, K =
2) from the inlet pressure (I = 2, K = 22) at each output time. A more thorough approach would
spatially average the inlet and output pressures, but this step is skipped here.
2. Divide ∆P by the length of the porous media (0.3 m) to find the unit loss ∆P/∆x.
3. Let Umacro = the average of the inlet and outlet velocities at each output time.
4. Convert pressure loss ∆P/∆x to head loss ∆H/∆x.
5. Plot the unit head losses against their matching average velocities.
nd
6. Calculate and plot 2 -order polynomial fitting equations for both sets of data.
7. Compare the results to the physical experiment, as shown below.

The model output matches the experimental data with a high degree of accuracy. Further statistical
error correlations can be developed from the point data and from the polynomial fitting curves. The
exercise is now complete. If you have any questions, please ask one of your instructors.

Figure 11: Agreement between experiment and CFD

 Water & Environment Training on FLOW-3D v11
Exercise 8: Saturated Porous Media

References
 Abbood, D.W. 2009. An Experimental Model for Flow Through Porous Media Using Water Filter.
th
Proc. 13 International Water Technology Conference: Hurghada, Egypt.

 Bejan, A. & D. Nield. 2006. Convection in Porous Media, 3rd Ed. New York, NY: Springer
Science+Business Media.

 Ergun, S. 1952. Fluid flow through packed columns. Chem Engrg. Prog. 48, 89-94.

 Flow Science. 2009. FLOW-3D v9.4 User Manual. Santa Fe, NM: Flow Science Inc.

 Freeze, R.A. & J.A. Cherry. 1979. Groundwater. Englewood Cliffs, NJ: Prentice-Hall.

 Idelchik, I.E. 1986. Handbook of Hydraulic Resistance. New York, NY: Hemisphere Publishing Corp.

 Irmay, S. 1958. On the theoretical derivation of Darcy and Forchheimer formulas. Eos, Trans. AGU 39,
702-707.

 Masch, F.D. & K.J. Denny. 1966. Grain size distribution and its effects on the permeability of
unconsolidated sands. Water Resour. Res. 2, 665-677.
Exercise 9: STL Topography from LiDAR
 Water & Environment Training on FLOW-3D v11
Exercise 9: Converting Topography to .stl

CREATING STL FILES FROM

TOPOGRAPHIC POINT CLOUDS USING
TOPO2STL AND FREEWARE
INTRODUCTION

Topography can be defined as “the configuration of a surface including its relief and the
position of its natural and man-made features”. It can also be defined as “the art or practice
of graphic delineation in detail … of natural and man-made features of a place” (Merriam-
Webster 2013). Digitizing topographic data for use in FLOW-3D combines both definitions:
the process of recreating accurate, detailed surface features requires both art and practice.

Point clouds are one of the most common ways that topographic data is recorded. Each point is a
measurement of surface location, defined by three coordinates in some reference system:
latitude/longitude/elevation, for example. The raw point data may be digitized from contour maps,
exported from GIS (geographical information system) software, or may be the output of LiDAR (Light
Detection and Ranging), TLS (terrestrial laser scanning), or satellite profiling equipment.

Stereolithography file format (.stl) is a non-proprietary system for recording 3-D surfaces.
The format is raw and unstructured, meaning that the elements of the surface can be
recorded in any order. The surface elements are triangles (called facets) composed of
three points and a normal vector pointing away from the surface. The points and vectors
are recorded in Cartesian coordinates. The .stl file format may be either binary (smaller file
size) or ASCII (readable in text editors); see Appendix A for the format definitions. FLOW-
3D uses .stl files as inputs to define complex geometry and as outputs to describe
resolved fluid and solid surfaces.

This exercise introduces FLOW-3D’s TOPO2STL utility for creating .stl parts from large
point clouds. No single program is ideal for converting large surface point sets to 3-D
geometry, so a method is presented here that combines freeware utilities with
TOPO2STL to get excellent results. Freeware tools introduced in this exercise include
VIM, MeshLab, CloudCompare, Admesh, and netfabb Basic.

OVERVIEW OF THE BASIC PROCESS

The steps to convert a point cloud of XYZ data to .stl are as follows:

1) Precondition the data for use with TOPO2STL:

Page 1 of 26
 a. Create a space-delimited ASCII text file with one coordinate per line and no headers.
Each line must contain three numbers, separated by spaces, so that the first number is
the X coordinate, the second number is the Y coordinate, and the third number is the Z
(elevation) coordinate. The coordinate units must all be the same (e.g., feet or meters).

b. Visualize and check the data. The first test of the raw data is visual examination. Look
for outlying points, strange stretching (indicating that units are not the same), noise,
and other problems. Locate the minimum and maximum extents of the raw data.

c. Clean the data by manually removing artifacts. The most common artifacts are people
and equipment that are picked up by the recorder and outlying points that are added
inadvertently (usually when filtering the data with software).

d. Fill in surface elevations for absent data. Many point sets only include data in a region of
interest (like a river bed). The areas outside of this region are “blank”: there is no data
there. These outlying regions should be filled in with a uniform surface elevation, usually
one that is higher than the any of the known data. This optional step is highly
recommended because it speeds up the .stl conversion by a factor of 100-plus, and
because it produces better output with fewer artifacts (Figure 1 below).

e. Pre-decimate (subsample) the point data. This is an optional step that reduces the
number of points in the data set. The benefits of pre-decimating include more accurate
.stl surfaces, faster conversion times, and more reliable results when using software
post-condition the data. TOPO2STL also subsamples the data during the conversion, so
the purpose of this step is to prepare the data by making it more uniform. It’s important
to use subsampling algorithms that preserve the data integrity (i.e., that don’t create
new points) and that evenly distribute points (e.g., that don’t cause higher point density
near contour lines).

Figure 1. 1-inch resolution .stls with (right) and without (left) data pre-conditioning
2) Use TOPO2STL to convert the point surface to a 3-D .stl file:

a. Select spatial resolution. TOPO2STL generates a rectangular sampling grid around the
minimum and maximum horizontal extents of the point cloud. The grid is made of
square stencils of the specified side length. Each grid cell retains a single coordinate
point out of all the original points in that cell; the points that are not retained are
thrown away and do not appear in the .stl output. This is one method of sampling
(decimation). Cells that do not have any original points in them are assigned a
coordinate based on the elevations of the points in nearby cells. The end result is a
more-or-less rectangular grid of coordinate points that are called vertices.

b. Select z-minimum elevation. TOPO2STL generates a network of triangular facets using

the vertices as corners. This creates the upper surface of the .stl part. The lower surface
of the .stl part is created as a flat plane at the specified z-minimum elevation. The .stl is
closed by stitching the upper and lower surfaces together with more triangular facets.

3) Post-condition the .stl file:

a. Simplify unnecessary regions to reduce file size. TOPO2STL creates .stl facets that are
about the same size, regardless of location. In regions where there wasn’t any data in
the original point cloud, this adds unnecessary size to the file. Recombining facets into
larger ones only where no data originally existed is the recommended way to reduce the
final .stl size.

b. Convert the .stl to binary format. Binary format .stl file size is about 20% of ASCII format.
Converting the ASCII TOPO2STL output to binary format is recommended.

c. Repair the part. TOPO2STL and other converters create artifacts in the output. Some of
the most common artifacts are small holes, disconnected edges, triangles with bad
aspect ratios, and inverted facets. These must be fixed before using the .stl file in
FLOW-
3D or unrealistic results might occur near the bad surfaces.

4) Quality control: Compare the .stl file surface to the original point cloud data to verify the
accuracy and precision of the conversion. Statistical measures that will be demonstrated in this
exercise include the mean error (ME), mean absolute error (MAE), signed and unsigned
minimum and maximum errors (EMIN, EMAX, |EMIN|, |EMAX|), standard deviations (σE and σ|E|), and
2 2
chi-squared tests of continuous probability functions (χ NORMAL and χ WEIBULL).
EXAMPLE DATA: RIO GRANDE RIVER SCALE MODEL

The example point cloud file (MiddleRioGrande.txt, see Figure 2 below) can be downloaded from the
FLOW-3D Users Site at users.flow3d.com. The data is from a joint project of the U.S. Bureau of
Reclamation and Colorado State University, and is provided as a courtesy by Dr. Amanda Cox.
Measurements were taken via ground-based LiDAR of a scale model of the Middle Rio Grande River in
New Mexico. The physical model was built at the Colorado State University Engineering Research
Center. The measurement units are feet.

Figure 2. Scale model LiDAR data

DETAILED EXAMPLE OF THE BASIC PROCESS

We’ll follow the basic process outlined above. Several ways of accomplishing each step will be given. The
workflows given in this note are only one way of approaching the solution. At a number of points in the
process, you’ll be asked to save your work with a different file name. Saving your work with different
file names is important. It helps you keep track of your progress, revert to earlier data states, and
compare different stages and approaches.

1) Format the text file. The original file won’t load in TOPO2STL because it is not formatted
correctly. The file size is 445 MB: too large to work with in most text editors. Two ways to clean
and check the data are given here.

a. Use CloudCompare to import and re-export the data: CloudCompare is a useful free tool
for working with large point cloud files. It does not yet include as many features as
MeshLab, but the features it includes are well-implemented and easy to use. It can be
downloaded from www.danielgm.net/cc/.

Open CloudCompare and select File > Open. Set File Type = ASCII and select
MiddleRioGrande.txt. Set Skip Lines = 2 to ignore the headers, and then set the three
column labels at the top to coord. X, coord. Y, and coord. Z. The column labels are a
useful feature when the raw data is not in the order TOPO2STL expects. Select OK to
begin the import. When the file is done loading, select MiddleRioGrande – Cloud in the
DB Tree pane at the upper left: this displays some data about the cloud in the Properties
 panel. Note there that the file has 15,010,488 points. Save the highlighted cloud as
MiddleRioGrande_CCtxt.xyz.

b. Use MeshLab to import and re-export the data: MeshLab is a powerful and free open-
source tool for working with meshes and point clouds. There are many ways to use
MeshLab with point clouds and .stl files, a few of which will be detailed below. MeshLab
can be downloaded from meshlab.sourceforge.net. It is not as user-friendly as
CloudCompare, and “Mr. P’s” tutorials (www.youtube.com/user/mrpmeshlabtutorials)
are highly recommended when you are ready to learn more.

First make a copy of MiddleRioGrande.txt and rename it MiddleRioGrande.asc. Open

MeshLab and select File > Import Mesh. Set the file format type to ASC (ascii triplets of
points) and open MiddleRioGrande.asc. Set Header Row to be Skipped = 2 (per what
you saw in CloudCompare) and deselect Grid Triangulation, because this import filter
won’t work with non-uniform point distributions. Select OK to begin importing the data.
The progress bar at the bottom will fill several times as various importation steps are
completed. Note at the bottom of the screen that the number of vertices (points) is
15,010,488, as before.

Importing the data as file type .asc works correctly but importing as .xyz type produces
an incorrect extra point at coordinates 0, 0, 0 at the beginning of the file. This is because
the .xyz format cannot contain headers and there is no option to skip initial lines.
MeshLab has a lot of little quirks like this.

Select File > Export Mesh As, set Files of Type = XYZ Point Cloud, and name the file
MiddleRioGrande_MLASCtxt.xyz. Click Save and on the next dialog deselect Normal so
that the file saves with only three coordinate values per line (instead of six).

All algorithms have the potential to produce noise or error in the results. Converting the data
with several methods and comparing the results is recommended. Comparing the files:

CLEANING METHOD NUMBER OF POINTS DIFFERENCE IN COORDINATE

(OUTPUT IS .XYZ FORMAT) (15,010,488 IN ORIGINAL) VALUES FROM ORIGINALS
CloudCompare .txt import 15,010,488 +/- 3E-6 feet (< 1 micrometer, μm)
MeshLab .asc import 15,010,488
+/- 2E-6 feet (< 1 micrometer, μm)
15,010,489
MeshLab .xyz import (excluding the extra point)
(adds point at 0,0,0)
Table 1. Comparison of cleaning methods for MiddleRioGrande.txt

The precision loss is due to rounding of large numbers. Few data sets require micrometer
precision, so the formatting method will usually depend on factors like file size, data column
order, and type(s) of bad values that must be removed. This particular data set contains a highly
precise measurement record. Though it will have negligible effects on modeling results, the
 cleaned file with the least introduced error should be used. The remainder of the exercise will
use MiddleRioGrande_MLASCtxt.xyz.

2) Visualize and check the data. The next step is checking that the topography looks right, locating
any bad data points, and making note of the minimum and maximum data extents. Two
methods are introduced.

a. Open and view the point cloud in MeshLab. Select File > Import Mesh and load
MiddleRioGrande_MLASCtxt.xyz. Select Render > Render Mode > Points from the

menus or on the Render toolbar to display the points.

b. Color the points by elevation. Select Filters > Quality Measure and Computations > Per
Vertex Quality Function. Set func q = -z (you may have to use the Tab key to gain access
to the input box), and activate map into color. Click Apply. Select Render > Color > Per
Vertex from the menu bar to display the colored results.

Figure 3. Points colored by elevation from blue (low) to red (high)

c. Find approximate min/max coordinates. Select Filters > Color Creation and Processing >
Colorize by Vertex Quality. The min and max values reported are the inverse of the min
and max z-elevations (97.2 and 99.3 feet) because the quality was set to the negative of
z-elevation. Changing the min and max values manually equalizes the color, which will
be useful later. Click Cancel.

d. Find precise min/max coordinates. Open the data in CloudCompare. Select Tools >
Projection > Export Coordinate(s) to Scalar Field(s) and export X, Y, and Z. Select the
coordinate to view under Properties > Scalar Fields in the lower left: the numeric
coordinate values are given as the slider limits of the displayed SF Scale (see Figure 4
below). Check again for outlying points and record the precise min/max values (Table 2).
 Figure 4. Min/max coordinates (mapped into scalar fields in CloudCompare)

LIMIT X (feet) Y (feet) Z (feet)

Minimum coord. 13.74089909 7.39987087 97.17911530
Maximum coord. 65.13487244 183.29547119 99.29799652
Table 2. Precise minimum and maximum coordinates of original data from CloudCompare

3) Clean the data by manually removing artifacts. In the image below, you can see that there
are some obvious artifacts: many look like people or equipment. They should be removed
before triangulation, and since they are clearly visible at this stage now is the best time to
remove them. The next exercise in this series describes how to remove large sections at a
time with CloudCompare, which is more useful for records from large digital elevation
databases.

a. Turn on point selection. Back in MeshLab, select the Edit > Select Vertexes tool from

the menu or from the Edit toolbar above the display. Hold the left mouse button
and drag around the points to select them. Appendix C introduces useful MeshLab
keyboard and mouse commands for making additive and subtractive selections.
b. Delete a set of points. Press Ctrl+Del to delete vertexes (or select Filters > Selection >

Delete Selected Vertices from the menu or from the Action toolbar).
c. Save your work after each operation: there is no Undo command in MeshLab! Select File
> Export Mesh and save the cleaned data in .xyz format with no normals as
MiddleRioGrande_MLASCtxt_clean.xyz. For this exercise, just remove a few artifacts.
 Figure 5. Removing some point artifacts

4) Fill in empty regions with a flat plane. This step produces more accurate and attractive .stl files.
Pre-filling also speeds up the .stl conversion time by a factor of 100 or more!

a. Create the plane. Select Filters > Create New Mesh Layer > Mesh Generator. Set X scale
= 60 and Y scale = 180 (the absolute length of the grid sides) per Table 2 above. Set Num
vertices = 1443 and 4327 points in X and Y, respectively, for ½-inch spacing. Activate
Centered On Origin, and click Apply. The plane will be built with both points and facets
on a new layer named Grid Generator (see Figure 6 below).

b. View layers. Select View > Show Layer Dialog from the menus or on the Standard
toolbar. Click on the icons to hide and compare the original point cloud and new
plane layers. If you need to reset the view, you can select Windows > View From >
Front.

Figure 6. New plane added

c. Translate the grid to position. Select the new layer, then Filters > Normals, Curvatures,
and Orientation > Transform: Move, Translate, Center. Set Z-Axis = 99.3 and Freeze
 Matrix to actually change the point coordinates. Apply the filter, and then set the Z-Axis
= 0 again so the grid doesn’t rise another 99.3 feet next time you click Apply. Enter X-
axis = 1 for translation and repeatedly click Apply until the grid has shifted so that it just
encloses the point cloud extents. Set X-axis = 0 and repeat the process for Y-axis. The
plane should overlap the original data with a small margin on all sides.

d. Color the layer. Assign a per-vertex attribute = –z on Filters > Quality Measure and
Computations > Per Vertex Quality Function and manually equalize the colors with
Filters > Color Creation and Processing > Colorize by Vertex Quality to Min = -99.3 and
Max = -97.2. Check that the new grid color is red, meaning it is higher than the
maximum elevation of the original data.

e. Remove the plane vertices that overlap the point cloud edges. Select the original point
cloud topography layer and apply Filters > Selection > Select All. Then select Filters >
Sampling > Vertex Attribute Transfer. Make sure the source mesh is the point cloud
and the target mesh is the new plane. Deselect all default attributes and select Transfer
Selection. Set Max Search Distance = 0.5 (absolute world units). Apply then Close the
filter. Hide the point cloud layer, select the plane layer, and delete the vertices that
have been selected with the Delete Current Set of Selected Vertices tool (see
Figure 7 below).

Figure 7. Delete points in the new plane that are within 0.5 feet of the river data

f. Remove the plane vertices that overlap the point cloud center. Keep the original data
hidden and turn off the coloring by selecting Render > Color > None. Activate the Select

Vertexes on a Plane tool and roll the mouse-wheel to set the HopThru size to less
than 0.5. Hold Control and click-drag the mouse to add and size selections. Select and
delete all of the points inside the originally-scanned data area. Use the Select Vertexes
tool where necessary to make sharp edges in the selection and manually clean up
any points that escape the first attempt (see Figure 8 below). Check your work by
showing both layers together and individually (see Figure 9 below). Export the final
plane in .xyz format with no normals in case MeshLab crashes during the next step.
 Figure 8. Select and delete remaining plane points that overlap river data

Figure 9. Final configuration of plane and river layers, colored by elevation

g. Merge the plane and river data. Show both layers, then right-click one of the layers in
the Layer Dialog and select Flatten Visible Layers. Select Keep Unreferenced Vertices or
all original points will be deleted! Don’t delete the existing layers yet. Export the
flattened layer with no normals as MiddleRioGrande_MLASCtxt_clean_fill.xyz.

h. Reload the data as points only. Now Delete all the layers (on the Layer Dialog) and
Import the point cloud you just saved (which has been stripped of all facet data and will
now be easier to work with). Remember how to colorize? Repeat the colorization
process (per vertex quality function = -z). Your point data should appear similar to
Figure 9 above.

5) Pre-subsample the data. There are many reasons for decimating a large point cloud (stability
and run-time of various meshing algorithms, for example), and there are usually ways to do it
that retain a high degree of fidelity. Two common methods are Clustered Vertex subsampling,
which reduces the data via a one-per-gridded-cell approach (either averaging the point
coordinates in a cell or taking the single point closest to the center), and Poisson-disk
subsampling, which uses a stochastic method to retain points. The Clustered Vertex cubic grid
method tends to concentrate points near contour lines (see Figure 10 below). Poisson-disk is
therefore recommended for topographic applications.
 a. Apply Filters > Sampling > Poisson-disk Sampling. Activate Base Mesh Subsampling
(very important) and select Explicit Radius (absolute world unit) = 0.0833 for 1-inch
samples. Export the subsampled mesh with no normals as
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833.xyz.

Figure 10. Subsampling. Left: original point cloud. Middle: 1-inch clustered vertex subsampling
(emphasizes 1-inch contour lines). Right: recommended 1-inch Poisson-disk subsampling

6) Use the TOPO2STL utility to convert to.stl. Download TOPO2STL from the Utilities section at
users.flow3d.com. Uncompress it and run Topo2STL.exe to launch the file selector. Browse and
pick MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833.xyz. The file will take several minutes to
load. Check that the Data Size (point count) reported is about 870,000 points.

a. Set point-sampling stencil. Set Spatial Resolution = 0.0833 feet. Very large .stl files that
exceed the available graphics card memory will not load in the FLOW-3D interface,
although they can be added manually to prepin files that are run from the command
line. The spatial resolution selection directly affects the file size. Post-conditioning can
reduce the output file size.

b. Estimate the total number of surface points. Find the number of points TOPO2STL will
2
either retain or create by multiplying the data extents in X and Y (60 x 180 = 1800 ft )
 2
and dividing by the spatial resolution squared (0.0833 ) to get the approximate total
number of points (1.6 million). A good rule of thumb is that .stl output files are about 1
GB for every 11 million points (including converting the output to binary .stl format).

c. Other conversion settings. Set STL Minimum Z-Coordinate = 90 ft. Set the name of the
output file as MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833.stl.

d. Generate the .stl. Click Convert to begin. Table 3 shows the conversion time for various
spatial resolutions, with and without pre-conditioning the data set. Adding planar data
to blank regions speeds up the conversion by a factor of 50 to 100. Pre-decimating the
data set after filling in the empty regions gives additional speed improvements.

POINT PRE- TOPO2STL CONVERSION STATISTICS

SPACING FOR ESTIMATE OF RAW DATA SET: CLEANED DATA SET CLEANED DATA SET
MESHLAB TOTAL NOT CLEANED, NOT WITH EMPTY REGIONS WITH EMPTY REGIONS
PRE- SURFACE PRE-FILLED, NOT PRE-FILLED, NOT PRE- PRE-FILLED &
DECIMATION POINTS TO BE DECIMATED DECIMATED POISSON-DISK PRE-
& TOPO2STL GENERATED DECIMATION TO
SAMPLING BY TOPO2STL TARGET TOPO2STL
SPACING
¼-inch 24.96 million TOPO2STL did not 18.6 million pts in/ 6.94 million pts in/
(0.0208 feet) points complete the 53 minutes/ 45 minutes/
conversion within 4 21.40 million pts out/ 21.36 million pts out/
days. Process was 10.7 GB ASCII .stl 10.7 GB ASCII .stl
manually terminated.
½-inch 6.24 million 15.0 million pts in/ 18.6 million pts in/ 4.43 million pts in/
(0.0416 feet) points 18 hours/ 20 minutes/ 10 minutes/
4.88 million pts out/ 5.83 million pts out/ 5.85 million pts out/
2.45 GB ASCII .stl 2.93 GB ASCII .stl 2.95 GB ASCII .stl
1-inch 1.56 million 15.0 million pts in/ 18.6 million pts in/ 0.87 million pts in/
(0.0833 feet) points 3.5 hours/ 10 minutes/ 2 minutes/
1.22 million pts out/ 1.39 million pts out/ 1.37 million pts out/
629 MB ASCII .stl 714 MB ASCII .stl 707 MB ASCII .stl
2-inch 0.389 million 15.0 million pts in/ 18.6 million pts in/ 0.26 million pts in/
(0.1666 feet) points 10 minutes/ 8 minutes/ <1 minute/
0.308 million pts out/ 0.356 million pts out/ 0.349 million pts out/
158 MB ASCII .stl 183 MB ASCII .stl 179 MB ASCII .stl
Table 3. Conversion times and other data for four TOPO2STL resolutions
(tested on Intel Core-i7 3930K CPU with 16.0 GB of RAM & Windows 7 Professional 64-bit)

7) Fix the output and convert it to binary format. All automated triangulation algorithms create
minor errors: disconnected edges between triangles, missing triangles (holes), and reversed or
incorrectly-computed facet normal vectors. These errors can cause bad pressure and velocity
solutions in FLOW-3D. Additionally, TOPO2STL‘s ASCII output is five times larger than binary
format. Fixing and converting the format of the .stl is always required. Three methods of fixing
and converting the part are demonstrated here; the third also introduces post-conversion
simplification of unnecessary surfaces. The three methods use the free programs qAdmesh,
netfabb Basic, and MeshLab.

a. Use qAdmesh for repair & format conversion. This software is a user-interface for the
free command-line program Admesh. It comes bundled with FLOW-3D. Source code and
theory for Admesh are available at
https://sites.google.com/a/varlog.com/www/admesh-htm. Launch qAdmesh from
FLOW-3D by navigating to the Model Setup > Meshing & Geometry tab and selecting
Tools > qAdmesh from the menu above the viewing pane. Select Browse under Input
and load MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833.stl. Set the
Output type = Binary STL, and name the output file
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833_admesh.stl. Select
Default check, and click Apply.

The results are given as text output: qAdmesh identifies about 5,400 facets with
disconnected edges, about 2.73 million incorrect normal vectors, and two identified
surface shells (which are called parts in qAdmesh).

Run qAdmesh again on the fixed output until no more changes are made to make sure
that the part is repairable. Notice that the repaired part still has 2 shells. This is not
ideal, but since the parts are fully closed and have no inverted normals it should not
affect the FLOW-3D model.

b. Use netfabb Basic for repair & format conversion. This free software can be
downloaded from http://www.netfabb.com/basic.php. It includes visualization, manual
and automatic analysis and repair, and other options.

Select Project > Add Part and load the ASCII TOPO2STL output file
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833.stl. The entire part is
selected so all faces are green. Clicking anywhere except on the part will deselect it. Re-
select the part by clicking on it or selecting it in the Parts tree at upper right, which
works much like the layers in MeshLab. Use the eye logo to toggle part visibility.
Notice the warning label in the lower-right corner. It indicates that a loaded part has
detectable problems.

Select Extras > New Analysis > Standard Analysis . A new Part Analysis branch is
added to the tree: selecting the Standard Analysis sub-branch displays data about the
part: it has two shells, two holes, an unclosed surface, and about 5,400 unjoined edges
(called boundary edges in netfabb): very nearly the same as the number of facets with
 unclosed edges reported by qAdmesh. Note that netfabb does not report incorrect
normal vectors like qAdmesh, but it does correct them automatically.

Figure 11. TOPO2STL output (1-inch-spacing, pre-decimated-and-filled) in netfabb Basic

Select Extras > Repair Part to create a new Part Repair branch in the tree. When
this branch is selected, individual facets and points can be added and deleted and
automatic repairs can be applied. Select the Part Repair branch and click Automatic
Repair at the bottom of the information display. Choose Default Repair and Execute.
Accept the changes by clicking Apply Repair to generate a new fixed part. You can
choose to leave the original part in the tree or delete it.

Select the newly repaired part in the tree and perform a new Standard Analysis on
it. Two holes have been filled, all facets have been connected, and unlike the qAdmesh
results, there is only one shell. Repair algorithms are implemented differently in
software packages, and almost always produce slightly different results. In this case the
Netfabb Basic output is slightly preferable to that of qAdmesh.
 Choose Part > Export Part > As STL to save the repaired geometry in binary format.
Name the part
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833_netfabb.stl. The binary
file size will always be about 81.5% smaller than the ASCII file size. Re-check it with
qAdmesh and verify there are no artifacts.

c. Use MeshLab for simplification, repair, and format conversion. MeshLab can be used to
reduce the size of .stl files, which is particularly important for those created with a fine
resolution. The goal is to reduce the file size without simplifying the data in the region of
interest. The way to do this is to recombine and simplify only the facets of the
surrounding box.

Import MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833.stl into MeshLab.

Choose Unify Duplicated Vertices during import.

Select only the non-river plane faces using Filters > Selection > Conditional Vertex
Selection, with Boolean function = (z > 99.298) or (z < 97) (see Table 2). Note that or
must be in lowercase. Activate Strict Face Selection to avoid picking up any river data.

Once the non-river surfaces are selected, apply Filters > Remeshing, Simplification and
Reconstruction > Quadric Edge Collapse Decimation with the following options: Target
Number of Faces = 100000, Quality Threshold = 0.3, Preserve Boundaries of the Mesh
(Boundary Preserving Weight = 10), Preserve Normal, Preserve Topology, Planar
Simplification, Post-Simplification Cleaning, and Simplify Only Selected Faces which is
very important to avoid simplifying the river! Check that the non-river facets have been
simplified as in Figure 12 below. Export the simplified part in Binary Format as
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833_MLquad.stl.

Figure 12. Simplification of the non-river surrounding box in MeshLab

Apply Filters > Selection > Select None to deselect the faces. Select Filters > Quality
Measure and Computations > Compute Topological Measures. The Layers dialog lists
the text results. There are a number of boundary edges (meaning the part is not closed),
non-manifold vertices, and incident faces.

Activate Render > Show Non Manif Edges and Render > Show Non Manif Vertices to
visualize where the problems are. They are at the four lower corners of the box.

Figure 13. Locating non-manifold vertices with MeshLab

Apply Filters > Selection > Select non Manifold Vertices and then delete them with the

tool. The bottom of the box will be deleted as well, since all facets in it share the
same four corner vertices.

Start the Edit > Fill Holes tool . No holes should be marked as non-manifold (thanks
to the previous step). If there are non-manifold holes, first close and re-launch the Fill
Holes tool: sometimes it’s wrong. Mark the Select check-boxes for holes with only three
edges (if any exist) and activate the Trivial Holes fill option. Select Fill and wait for the
Accept check-boxes to appear: mark all the boxes and click Accept at the bottom of the
 tool dialog. If there were many trivial holes, re-export the mesh in case MeshLab
crashes later.

Now select the big hole, this time using the Self Intersection repair option. Click Fill,
wait, and then click Accept. MeshLab will start to close the hole and incorrectly report
that there are no more holes to fill. Ignore this message (click OK to close it) and repeat

the process by re-launching the tool. Keep selecting, filling, accepting and re-
launching the tool until no more holes are reported when the tool starts. The bottom of
the part should now be closed (Figure 14 below). Export the part, and check Compute
Topological Measures again: there should be no boundary or non-manifold edges. The
part should be two-manifold and composed of a single component.

Figure 14. .stl bottom before (left) and after (right) removing non-manifold vertices and filling holes

Finally, apply Filters > Normals, Curvatures and Orientation > Recompute Face Normals
and then Filters > Normals, Curvatures and Orientation > Normalize Face Normals.
Export the part and check it with netfabb Basic (no errors detected) and qAdmesh
(about 5,400 bad normals detected). Fix the normals with qAdmesh and save the
output. Re-check the output with netfabb Basic and qAdmesh: both programs should
agree that the part is fixed.

The three repair methods introduced above are compared in Tables 4 and 5 below. The
MeshLab simplification and repair procedure with qAdmesh finishing is recommended for
general use because it can be performed on parts that are too detailed to load with the other
methods, and it further reduces the final .stl file size.
 SPATIAL qAdmesh repair netfabb Basic repair MeshLab simplification &
RESOLUTION repair w/ qAdmesh finish
¼-inch 1E5 target facets/
(0.0208 feet) Fixed: 4 vertices, 2 holes,
21599 normals/
1 shell output/
Too many facets to fix in qAdmesh and
949 MB binary .stl
netfabb Basic without prior simplification.
½-inch 1E5 target facets/
(0.0416 feet) Fixed: 4 vertices, 4 holes,
11079 normals/
1 shell output/
243 MB binary .stl
1-inch Fixed: 9509 facets, Fixed: 2 holes, 5440 1E5 target facets/
(0.0833 feet) 2733148 normals/ borders, 2 shells/ Fixed: 4 vertices, 8 holes,
2 shells output/ 1 shell output/ 5400 normals/
130 MB binary .stl 130 MB binary .stl 1 shell output/
64 MB binary .stl
2-inch Fixed: 4765 facets, Fixed: 2 holes, 2726 1E5 target box facets/
(0.1666 feet) 692898 normals/ borders, 2 shells/ Fixed: 4 vertices, 6 holes,
2 shells output/ 1 shell output/ 2722 normals/
33 MB binary .stl 33 MB binary .stl 1 shell output/
20 MB binary .stl
Table 4. Repair data for TOPO2STL output of pre-cleaned, pre-filled, and pre-decimated point sets

SPATIAL qAdmesh repair netfabb Basic repair MeshLab simplification &

RESOLUTION repair w/ qAdmesh finish
¼-inch 1E5 target facets/
(0.0208 feet) Fixed: 4 vertices, 1 hole,
21618 normals/
1 shell output/
Too many facets to fix in qAdmesh and
951 MB binary .stl
netfabb Basic without prior simplification.
½-inch 1E5 target facets/
(0.0416 feet) Fixed: 4 vertices, 3 holes,
11054 normals/
1 shell output/
242 MB binary .stl
1-inch Fixed: 9587 facets, Fixed: 2 holes, 5480 1E5 target facets/ Fixed: 4
(0.0833 feet) 2759748 normals/ borders, 2 shells/ vertices, 14 holes,
2 shells/ 1 shell output/ 5446 normals/
132 MB binary .stl 132 MB binary .stl 1 shell output/
64 MB binary .stl
2-inch Fixed: 4824 facets, Fixed: 2 holes, 2760 1E5 target facets/
(0.1666 feet) 707198 normals/ borders, 2 shells/ Fixed: 4 vertices, 11 holes,
2 shells/ 1 shell output/ 2722 normals/
34 MB binary .stl 34 MB binary .stl 1 shell output/
20 MB binary .stl
Table 5. Repair data for TOPO2STL output of pre-cleaned & pre-filled point sets without pre-decimation
8) Check the output (quality control). The final step is to visually and numerically compare the
repaired .stl to the original artifact-clean point cloud to determine if the conversion process was
accurate.

a. Load the comparison data. Open CloudCompare and load

MiddleRioGrande_MLASCtxt_clean.xyz (set file type to ASCII) and
MiddleRioGrande_MLASCtxt_clean_fill_PDSub0833_T2STL0833_netfabb.stl (file type
STL).

b. Get preliminary comparison statistics. Hold Control + left-click to select both Cloud and

Mesh entities in the DB Tree. Select Tools > Distances > Cloud/Mesh Dist . The
dialog that opens lists Approximate Results for the distance magnitudes between
original points and the nearest .stl facets. It will shortly be demonstrated that these
values are incorrect. The estimate of Max Relative Error gives the likely error of the
distance results themselves as a function of the distance d between any point and the
.stl. The Approximate Results are chamfer distances and not reliable for small d
(Girardeau-Montaut et al. 2012).

c. Get precise magnitude statistics and map them into color. Locate the Precise Results
section. For best accuracy, set Octree Depth = 9 or 10. Deactivate Signed Distances and
click Compute. After the calculation is complete click OK to exit the dialog and map the
point results into a scalar field.

Find the text Console panel at the bottom of the screen (press F8 to toggle it). The text
output gives a new estimate of the mean and standard deviation σ: about 0.01 feet (3
mm) for both. The mean unsigned error is referred to as the Mean Absolute Error (MAE).

d. Find the maximum error magnitude. The min and max SF scale slider values are the
minimum and maximum error magnitudes. In this case |E|MIN = 0 and |E|MAX = 0.242
feet ≈ 7 cm. The real maximum error is much smaller than was reported by the
Approximate Results.

In the DB Tree, un-check the Mesh part to hide it and then select the Cloud entity. Check
that Current Scalar Field = C2M Distances in the Properties panel. Activate Display Color
Scale under SF Scale. Under SF Scale, slide the displayed minimum slider all the way to
the right (or set the minimum value box to the same as the maximum). This makes all
values less than the maximum render as gray, and values equal to the maximum will be
displayed as red. Deselect NaN in grey to hide the non-maximum points. Move the
mouse near the upper-left of the display window to display Point Size options. Toggle
the Mesh display under DB Tree, rotate, pan, and zoom the view, and increase Point
 Size (+) until you locate the point of greatest error. Reactivate the Mesh entity to see
the .stl.

Figure 15. Maximum error between .stl and point cloud is high on a steep bank slope

e. Visualize the error magnitude. Reset the displayed sliders to the minimum and
maximum permissible (check that they match the color saturation slider values). Set the
color saturation maximum slider value box = 0.06 feet. Now all error magnitudes larger
than about 1.8 cm appear in red, illustrating the regions of greater error. As visible in
Figure 16 below, there are also some red areas that appear to be electrical cable for the
ground-based LiDAR camera and that were not included in the point cloud. From this
visualization it is clear that the majority of the error magnitudes are much less than 0.03
feet (less than 1 cm).

Figure 16. Visualization of error magnitude with blue = 0 and red = 0.06 feet.

f. Visualize the error magnitude distribution. Select Tools > Statistics > Compute Stat.
Params, and set Distribution = Gauss to compare the actual error magnitudes to a
normal distribution. The resulting histogram contains all the information that has
already been determined: the maximum and minimum errors are the ends of the
abscissa, and the mean and standard deviation are summarized (and rounded) above
the chart. More significant figures can be displayed in the Settings options.
 The white line represents a normal distribution (based on the mean and standard
deviation). The normal distribution is not a good fit! Repeat the process and select
Weibull as the distribution type. The Weibull distribution is a good fit to the error
magnitudes. The curve parameters are given above the chart. Results should look like
Figure 17. The error mean and deviation should be reported in terms of these unsigned
2
values. The Console (F8 to toggle) gives a Chi-squared (χ ) test for each fit.

Figure 17. Statistical fitting of the error magnitude distribution: normal (left) and Weibull (right)

g. Visualize the signed error distribution. Select both the cloud and .stl layers, and re-run
the Tools > Distances > Cloud/Mesh Dist tool. Activate Signed Distances under Precise
Results and Compute the error to map it into the scalar field of the point cloud. Select
the point cloud layer and run the Tools > Statistics > Compute Stat. Params tool first for
Distribution = Gauss and then for Weibull. The results are shown in Figure 18. It’s clear
2
from these and the χ tests that the signed error is uniformly distributed about the
mean and that the .stl surface error (relative to the point cloud) is nearly unbiased
(neither higher nor lower on average).

Figure 18. Statistical fitting of the signed error distribution: normal (left) and Weibull (right)
 Table 6 below shows comparative mean error magnitude (MAE) results for various sampling
spacing and pre-conditioning approaches. The 3 repair methods described in Tables 4 & 5 gave
identical statistics (within 0.3 μm) for each method and spacing, so the results shown in Table 6
are true for all repair methods, regardless of how many shells the method produced.

POINT POINT CLOUD TO .STL DISTANCE MAGNITUDE (FEET)

SPACING FOR RAW DATA SET: CLEANED DATA CLEANED DATA SET
MESHLAB NOT CLEANED, NOT SET WITH EMPTY WITH EMPTY
PRE- PRE-FILLED, NOT REGIONS PRE- REGIONS PRE-FILLED
DECIMATION DECIMATED FILLED, NOT PRE- & POISSON-DISK PRE-
& TOPO2STL DECIMATED DECIMATION TO
SAMPLING TARGET TOPO2STL
SPACING
¼-inch TOPO2STL could not Max: 0.247521 Max: 0.146637
(0.0208 feet) convert the raw data Mean: 0.004533 Mean: 0.004532
within 4 days σ: 0.004472 σ: 0.004175
½-inch MeshLab could not Max: 0.307642 Max: 0.200105
(0.0416 feet) simplify non-river Mean: 0.006653 Mean: 0.006409
facets in 12 hours σ: 0.006960 σ: 0.006309
1-inch Max: 0.33902451 Max: 0.324913 Max: 0.242234
(0.0833 feet) Mean: 0.010307 Mean: 0.010395 Mean: 0.009717
σ: 0.012193 σ: 0.012463 σ: 0.010644
2-inch Max: 0.383123 Max: 0.379844 Max: 0.316623
(0.1666 feet) Mean: 0.016889 Mean: 0.017011 Mean: 0.016472
σ: 0.023922 σ: 0.023929 σ: 0.019153
Table 6. Quality Control statistics for different conversion methods and spacing

CONCLUSIONS AND SUMMARY

An approach for converting large topographic point clouds to 3-D .stl parts has been recommended.
Tables 3 and 6 indicate that better conversion speed and better accuracy can be achieved by pre-filling
empty regions and pre-decimating the data set. Tables 4 and 5 illustrate the file-size reduction that can
be achieved through post-conditioning the .stl part: simplifying the non-important surfaces and
converting it to binary format. Recommended filters include Poisson-disk subsampling for pre-
decimation and quadric edge collapse decimation for post-conditioning simplification.

The method outlined in this exercise has been tested on a 15-million point river (Middle Rio Grande
scale model), a 61-million point watershed (Dalalven river), and several other sets. Comments and
experiences should be addressed to the author ([email protected]) or Flow Science support
staff ([email protected]).
REFERENCES
1. Girardeau-Montaut, D., Bey, A., and Marc, R. (2012). “CloudCompare User’s Manual for Version
2.1.” Le Faive, R., translator. http://www.danielgm.net/cc/, retrieved 3/7/2013.
2. Merriam-Webster (2013). “topography.” http://www.merriam-webster.com, retrieved
3/7/2013.
3. Willmott, C.J., Ackleson, S.G., Davis, R.E., Feddema, J.J., Klink, K.M., Legates, D.R., O’Donnell, J.,
and Rowe, C.M. (1985). “Statistics for the Evaluation and Comparison of Models.” Journal of
Geophysical Research 90(C5): 8995-9005.
 APPENDIX A – STL FORMAT

.stl ASCII format

solid <partname>
facet normal <ni> <nj> <nk>
outer loop
vertex <v1X> <v1Y> <v1Z>
vertex <v2X> <v2Y> <v2Z>
vertex <v3X> <v3Y> <v3Z>
endloop
endfacet
………..
<repeat beginning with “facet normal” through “endfacet” for every facet in the part>
………..
endsolid <partname>

indentation must be with spaces (no tabs)

<nI> <nJ> <nK> are facet normal vector components in single precision float and may be negative.
<v1X> <v1Y> <v1Z> are the vertex Cartesian coordinates (single precision float) and may not be negative.

.stl binary format

by te s data type data content

80-byte ASCII <partname>
4-byte unsigned long integer <total number of facets>
4-byte floating point <ni>
4-byte floating point <nj>
4-byte floating point <nk>
4-byte floating point < v1X >
4-byte floating point < v1Y >
4-byte floating point < v1Z >
4-byte floating point < v2X >
4-byte floating point < v2Y >
4-byte floating point < v2Z >
4-byte floating point < v3X >
4-byte floating point < v3Y >
4-byte floating point < v3Z >
2-byte unsigned integer <attribute byte count>
………..
<repeat beginning with the “ni” line through the “attribute byte count” line for every facet in the part>
………..
 APPENDIX B - COMPARISON OF TEXT EDITORS FOR LARGE POINT CLOUDS

It is helpful to have a text editor which includes the following capabilities:

 Opens large (up to 2 GB or larger) ASCII files

 Finds and converts tabs, commas, and spaces using regular expressions,
 Finds and converts DOS, Unix/Linux, Macintosh, and Netscape line-end characters,
 Includes a column-edit method for block editing.

PROGRAM CAN OPEN & CAN OPEN & CONVERTS TAB INCLUDES FREEWARE
VIEW 445 MB VIEW 953 MB TO SPACE IN COLUMN ?
1,3 1,2
POINT TEXT FILE 445 MB FILE EDIT MODE
1,2
CLOUD
Sublime Text 2 YES. Loads entire file into NO. Uses 4 GB LIMITED. NO.
memory. memory. Does Large column
not complete operations
within 1 hour. fail.
TextPad 6 YES. Loads YES. Converts NO. Unreliable. YES. NO.
whole file into some Requires
memory. characters. multiple tries.
gVIM 7.3 32-bit with YES. Becomes briefly YES. Total time LIMITED. YES.
LargeFile plugin unresponsive during scrolling about 3 Requires text
operations due to piecewise minutes. commands.
loading method.
Crimson Editor 3.70 NO. Out-of-memory error. YES. YES.
EditPad Lite 7 YES. NO. Location is NO. Uses all NO. Available MAYBE.
loaded available in EditPad Free for
unreliably: memory. Does Pro only. non-
final line # not complete commercia
changes on within ½ hour. l use only.
separate tests.
ConTEXT 0.98.6 NO. Out-of-memory error. YES. YES.
PilotEdit Lite YES. Loads entire file into NO. Does not YES. YES.
memory. complete
YES (loads into memory first). within ½ hour.
Notepad++ 5.9.8 YES. Loads NO. Uses 1 GB of memory, then YES. YES.
entire file into crashes.
memory.
Table B-1. Comparison of eight text editors
Notes:
1) Tested on Intel Core i7-3930K 3.20 GHz CPU w/ 16.0 GB RAM, Windows 7 Professional 64-bit OS
2) MiddleRioGrande.txt, courtesy of Colorado State University and US Bureau of Reclamation.
3) “Enwiki9.txt”, courtesy of Matt Mahoney and the Large Text Compression Benchmark project
(http://mattmahoney.net/dc/text.html, accessed Nov 19, 2012).
 APPENDIX C – A BRIEF GUIDE TO KEYBOARD AND MOUSE COMMANDS
IN MESHLAB

Navigation commands

 Left click and drag to rotate the part and coordinate system around the trackball center.
 Mouse wheel to move the part and coordinate system along line of sight relative to the trackball.
 Center mouse button and drag to pan the part and coordinate system.
 Shift+mouse wheel to change camera field-of-view.
 Control + mouse wheel to move “near ” c li ppi ng plane.
 Control + Shift + mouse wheel to move “far ” c li ppi ng plane.
 Double-left-click to place the trackball center at the clicked-on point.
 Alt + Enter to enter full screen mode.
 Control+Shift+left mouse button and drag to change lighting direction (only if there are normals).

Selection commands (when a selection tool is active)

 Left-click and drag to draw a box and select all faces or vertices in it (even those behind surfaces).
 Alt+left click and drag to select only visible faces or vertices in the box.
 Ctrl+left click and drag to add the faces or vertices in the box to the already-selected group.
 Shift+left click and drag to remove the faces or vertices in the box from the already-selected group.
 Ctrl+Alt+left click and drag to add only visible faces in the box to the group.
 Shift+Alt+left click and drag to remove only visible faces in the box from the group.
 Esc to toggle between navigation mode and the selection tool.

Important considerations

 The trackball is the point about which the coordinate system (and part) rotates and moves.
 There is no undo function: after every major processing step, export the mesh with a new filename.
 Some filters are less stable: most do better with a few million points or less. Consider subsampling.
 The MeshLab project file (.mlp) file records mesh file path, transformation, & camera/lighting data.
 The .mlp file stores no information about the points or faces. That data is stored in the mesh file.
 The transformation matrix is a 4x4 record of movement (scaling, rotation, translation) of the mesh.
 The Freeze Matrix option means that when the mesh is exported it will include the transformations.
Exercise 10: STL Topography from Google Earth
 Water & Environment Training on FLOW-3D v11
Exercise 10: Making STL Topography from Google Earth

MAKING STL FILES FROM GOOGLE

EARTH USING SKETCHUP AND CAD2STL
INSTALL NECESSARY SOFTWARE
You will need to install the following
programs:

 SketchUp Make 2014 (http://www.sketchup.com/ )

 TIG Extrude Tool 2.5 (http://sketchucation.com/resources/pluginstore?pln=ExtrudeTools)
 Cad2Stl v1.4 (users.flow3d.com)

The TIG Extrude Tool plugin will come in handy later so you won’t have to manually draw each .stl
face. Follow the steps below to install it.

1. Launch SketchUp Make 2014.

2. Select Window > Preferences from the top menu.
3. Choose Extensions from the list on the left.
4. Choose the Install Extension... button at the bottom.
5. Specify the ExtrudeTools_v2.5.rbz file.
6. Make sure the Extrude Tools plugin is active in System Preferences and click OK.
7. If the Extrude Tools toolbar doesn’t appear, right-click on the toolbar and select
Extrusion Tools from the list.

Figure 1. Install the TIG Extrude Tools v2.5 extension in SketchUp Make 2014

Page 1 of 11
 Water & Environment Training on FLOW-3D v11
Exercise 10: Making STL Topography from Google Earth

GET TOPOGRAPHY FROM GOOGLE EARTH USING SKETCHUP

Getting topography is simple in SketchUp. The resolution may not be as fine as that from national or regional
DEM databases, but it is often adequate for general modeling.

1. Select File > Geo-Location > Add Location... or select the icon in the toolbar.
2. Search and/or Navigate to find your location of interest.
3. Click the Select Region button and drag the corner pins to enclose the area of interest.
4. Click the Grab button to lay the selected topography image on the SketchUp map.

Figure 2. Locate your region of interest in SketchUp.

MAKE A WATERTIGHT SURFACE FROM THE TOPOGRAPHY USING SKETCHUP

1. Select File > Geo-location > Show Terrain to three-dimensionalize the terrain map.
2. Select the model by clicking it, then right-click and select Unlock from the context menu.
3. Right-click the model again and select Explode.
 Figure 3. Three-dimensionalize the terrain.

Figure 4. Unlock the terrain model in SketchUp.

Figure 5. Explode the terrain to create a triangulated surface from the terrain model in SketchUp.
4. Click the background to deselect the triangulated surface. Then double-click the object to select it
with edges.
5. Select Extrude Edges by Vector from the Extrusion Tools toolbar.
6. Extrude in -z axis. You may need to move the mouse somewhat until it extrudes by the blue axis.
7. Click the Select arrow icon from the main toolbar and a dialog will appear.
8. The tool sometimes extrudes reversed faces. If the faces are reversed they will appear gray, as in
Figure 7. If they are reversed, select Yes. Otherwise select No.
9. Another dialog will appear asking whether to explode extruded faces. Select Yes.

Figure 6. Reselect the triangulated surface as edges in SketchUp.

Figure 7. Extrude the surface downward following the blue (z) axis and check and fix reversed faces.
10. Select Shapes > Rectangle and draw a large rectangle on the canvas to completely surround the
topography.
11. Click the Move tool icon and lower the plane so it cuts the extruded faces without cutting the surface.

Figure 8. Draw a rectangle to surround the topgraphy.

Figure 9. Place the plane so it cuts through the extruded faces.
12. Adjust the camera and make a selection that includes all the extruded faces and the rectangle.
13. Right-click the selection and select Intersect Faces > With Selection.
14. Draw a selection box to select the surfaces below the rectangle and Delete the extended
surfaces, including the rest of the rectangle that extends past the topography.
15. Select All ( Ctrl+A) and then right-click the selection and select Make Group from the context menu.
16. Select File > Export > 3D Model... to export the geometry file. Save the file as .obj or .3ds format.

Figure 10. Select extruded faces and the rectangular plane together.

Figure 11. Intersect the faces and then delete the unneeded bottom faces and rectangle sides.
 Figure 12. Export the completed model.

CONVERT THE SURFACE USING CAD2STL

1. Open Cad2Stl by running run_cad2stl.bat.

2. Select Add and browse to select your .3ds or .obj file.
3. Select the file from the list and click convert.
4. Cad2Stl will save the .stl file in same folder and same name as the original file by default.
5. Check and fix any errors in the .stl using tools like pyAdmesh or netfabb Studio.

Figure 13. Convert the surface to .stl format.

Figure 14. Repair the .stl until it’s watertight.
Exercise 11: STL Topography from USGS Database
 Water & Environment Training on FLOW-3D v11
Making .STL Files from USGS Data

Making .STL Files from USGS Data

TABLE OF CONTENTS
INTRODUCTION ............................................................................................................................................. 1
STEP 1 – DOWNLOAD USGS DATA FOR THE REGION OF INTEREST .............................................................. 1
STEP 2 – USE MicroDEM TO CONVERT DATA TO POINT-CLOUD FORMAT ................................................... 2
STEP 3 – USE CloudCompare TO COMBINE AND TRIM TOPOGRAPHY ......................................................... 4
STEP 4 – USE MeshLab TO PREPARE THE TOP SURFACE............................................................................... 6
STEP 5 – ESTIMATE THE MINIMUM STL RESOLUTION .................................................................................. 9
STEP 6 – CONVERT THE POINT CLOUD TO .STL WITH TOPO2STL ................................................................. 9
STEP 7 – SIMPLIFY/RECOMBINE FACETS ..................................................................................................... 11
STEP 8 – DELETE NON-MANIFOLD VERTICES .............................................................................................. 12
STEP 9 – CLOSE TRIVIAL HOLES ................................................................................................................... 12
STEP 10 – REPAIR THE BACK FACE .............................................................................................................. 13
STEP 11 – FIX FACET NORMALS .................................................................................................................. 14
STEP 12 – CHECK ACCURACY....................................................................................................................... 15

INTRODUCTION
This note outlines a method for converting large topographic data sets to .stl format for use with FLOW-
3D. This note builds on the exercise Creating .STL Files from Topography, available on the FLOW-
3D Users Site. Many of the methods in this note are described in more detail in that exercise. This
note
adds instructions for data sets from the U.S. Geological Survey (USGS) National Elevation Dataset (NED).
The example describes converting an area of interest from 1/3-arc-second resolution USGS data in
ArcGrid format to DEM, point cloud, and finally .stl formats. The method can also be used to convert
GeoTIFF formats and floating point formats to .stl. This exercise uses open source freeware and FLOW-
rd
3D’s TOPO2STL utility. There are many other possible workflows that involve 3 -party commercial
software that may be faster or more reliable. The basic processes and principles will be the same
regardless of software.

STEP 1 – DOWNLOAD USGS DATA FOR THE REGION OF INTEREST

The region of interest is the inundation map of the St. Francis dam break flood of 1928, from the
California coastline near Ventura up to and including the St. Francis dam and reservoir (Figure 1).
Topographic data is downloaded for free from the USGS National Elevation Dataset (NED) at
http://viewer.nationalmap.gov/viewer/. The download will contain all 1-degree x 1-degree tiles that
Page 1 of 16
contain the region of interest. For this case, the region of interest results in two compressed files, each a
1 degree-square (35 N and 119W through 120W), in 1/3 arc-second ArcGrid format. Unzip each file into
a new temporary working folder located on the root. Refer to http://ned.usgs.gov/Ned/faq.asp for
details on NED data, including formats, accuracy, and methods of collection.

Figure 1. Approximate region of interest, viewed at http://earthexplorer.usgs.gov/

STEP 2 – USE MicroDEM TO CONVERT DATA TO POINT-CLOUD FORMAT

Download MicroDEM from Professor Peter Guth’s page at the U.S. Naval Academy website:
http://www.usna.edu/Users/oceano/pguth/website/microdem/microdem.htm. Install the software
with full installation and default directory options. Update the program (.exe) and help file (.chm) builds
from File > Tools, or download them from the MicroDEM website. Right-click the .chm file and select
Properties > Unblock to allow it to work in Windows 7.

FWTools provides the functionality to read the USGS formats. Download and install the latest version
from http://fwtools.maptools.org/. In MicroDEM, Set Options > Directories > FW Tools to the bin folder
of the installation (e.g., C:\Program Files (x86)\FWTools2.4.7\bin). Note that MicroDEM is under heavy
development as of Fall 2013, so refer to its Help file for more current instructions.

Select File > Data Manipulation > Resample > DEM formats > ADF Directories. Select the folder that
contains the ArcGrid data (e.g., n35w119_13sec_ArcGrid_a). When prompted if the output should be
GeoTiff format, select No. A MicroDEM-format DEM file (.dem) will be created in the data subdirectory
of the ArcGrid folder. Repeat for all ArcGrid sets. Select File > Open > Open DEM to open and view the
new files. Arrange them as shown in Figure 2.
 Figure 2. Two USGS 1-degree tiles, converted to DEM with MicroDEM and FWTOOLS

MicroDEM has the option to simultaneously open & merge, but memory limitations in the executable
prevent this from being useful for large data sets. Select File > Save DEM > ASCII > ASCII XYZ (.xyz) and
the following options for each of the DEM files.

Figure 3. Recommended MicroDEM Save options

If you are computer-savvy you can choose to download .bat scripts from USGS instead of using
MicroDEM. The scripts use the GDAL library to convert the data without a software user-interface. The
scripts are available as of November 2013 at
http://imsdemo.cr.usgs.gov/Website/Ned_Conversion_Scripts.zip. The GDAL library can be downloaded
from http://www.gdal.org/, and must be installed and included in the PATH environment variable
before running the scripts. The same GDAL library is included in FWTOOLS for MicroDEM.

STEP 3 – USE CloudCompare TO COMBINE AND TRIM TOPOGRAPHY

Open the new files in CloudCompare. You’ll be shown a column list for each cloud where you can check
the precision of the MicroDEM output. When prompted, shift each cloud by -300,000.00 m in x, -
3,000,000.00 m in y, and 0.00 m in z. This prevents loss of precision due to machine rounding. Select
both clouds in the DB Tree pane and use the Fuse tool on the menu bar to combine them into a
single object. Save the fused cloud as ASCII .xyz format with the options shown below. The new file will
be about 197 million points (6.3 GB) that will be saved with the original coordinates. Color the points by
elevation: Tools > Projection > Export Coordinate(s) to Scalar Field(s) > Z.

Figure 4. Recommended CloudCompare Save options

Figure 5. Fused UTM projections in CloudCompare

Use Edit > Scalar Fields > Filter by Value to create a new cloud with elevations above 570 m removed.
This is because the initial water level behind the St. Francis Dam was 559.4 m (1,835 ft) per Outland
(1963), as cited in Sanders (2007). Select the Segment tool from the menu bar to cut out the region
of interest. The segmentation tool opens with a new toolbar. Use left-click to add a point to the
selection polygon and right-click to close the polygon. Select the action to make the contents of the
selection polygon a separate point cloud. Edit the cutoff values in the Properties pane > SF Display
Params section to help visualize the elevation gradients. Select to accept the new point cloud. Repeat
as necessary. The goal is to minimize the X and Y extents of the topography. Save the final point cloud.

Fig 6. Final point cloud, filtered to below 570 m and segmented

Reload the point cloud with the new recommended shift, rounded by order of magnitude. For example:

Figure 7. CloudCompare temporary shift options for segmented topography

Use Tools > Projection > Export Coordinate(s) to Scalar Field(s) > X,Y,Z to export the coordinates. Check
the extents in the Properties pane > SF Display Params section. Toggle between coordinates by changing
the Active dropdown selection in the Properties pane > Scalar Fields section. Calculate the required shift
values so that the number of significant digits is minimized, and all coordinates are positive.

Use Edit > Apply Transformation and enter the calculated values for the Tx and Ty entries in the
transformation matrix. This step permanently shifts the saved cloud coordinates. The FLOW-3D output
will need to be shifted back by the opposite sign of these values to get UTM coordinates. This approach
minimizes the effect of rounding error in the CFD calculations.

Figure 8. CloudCompare permanent shift options for shortest positive coordinates

Select Tools > Scalar Fields > Delete All(!) to eliminate the scalar fields from the saved output so that it
can be opened in MeshLab. Save the shifted cloud (now with no scalar fields) and re-open it. Repeat the
coordinate checking process to verify and record the new extents. For example:

LIMIT X (m) Y (m) Z (m)

Original max/min 288,889.7 to 3787191.3 to -0.167 to 570.000
coordinates 365,560.8 3826153.5
Translate -286,000.00 -3,780,000.00 0.00
New max/min 2889.7 to 79560.8 7191.3 to 46153.5 -0.167 to 570.000
coordinates for
modeling
Coordinate Extents 76,671.1 38,962.2 570.167
Table 1. Translated Area of Interest

STEP 4 – USE MeshLab TO PREPARE THE TOP SURFACE

Skip the pre-decimation step described in the Creating .STL Files from LiDAR exercise. The goal here is
to maintain the 1/3-arcsec resolution of the source data. Use MeshLab’s Grid Generator filter to create
a plane with extents slightly larger than those in Table 1: use 78,240 m x 39,600 m with 15-m spacing.
 Figure 9. MeshLab grid generation options

Use the Translate filter to move the grid so it overlaps the topography extents and is at elevation
284.915 m, midway between the maximum 570 and minimum -0.17 m. Select All the points in the
topography layer and use the Vertex Attribute Transfer filter, with Transfer Radius as close to 290 m as
possible. Delete the overlapping points on the plane grid, and then raise the plane back to 571 m.

Figure 10. Delete vertices that overlap the topography

Duplicate the plane layer. Select Vertices with the and tools to select those points that
comprise the ocean. Invert the selection and Delete the land points. Use the Geometry Function filter
to set the elevation of all points in the layer to Z = -2 m.
 Figure 11. Lower the sea floor to -2 m in MeshLab

On the original plane layer, Select and Delete only the ocean region, and set the elevations of all
remaining points in that layer to Z = 571 m (above the flow region). There are now three layers: two
horizontal planes (one at -2 m and one at 571 m), and one layer of USGS topography. Flatten all three
layers in MeshLab to combine them. Save the final combined layer with no normals.

Figure 12. The final surface combines ocean, bounding box, and topography layers
STEP 5 – ESTIMATE THE MINIMUM STL RESOLUTION
Use rules of thumb to find the minimum resolution. The number of surface points in TOPO2STL output is
a function of the area of the rectangular bounding plane and the square resolution stencil. On average,
the size of the TOPO2STL ASCII-format .stl output in megabytes (MB) can be estimated as:

Binary-format .stl files have the identical data but are 5.42 times smaller than ASCII format:

The above estimates are accurate within 15% of the actual size. The file can be made smaller by
recombining facets to simplify the unused (planar) surfaces. The ratio between high-resolution
topography and the entire surface can be estimated by checking the number of points in each layer
during Step 4 above:

The file size reduction is a function of the area of the upper surface, the area of the bounding plane, and
the aggressiveness of the simplification algorithm. The simplified file size can be estimated from the
averages of past projects:

The estimate is usually within 20% of the actual size of the final, simplified, binary-format file.

Using these rules of thumb, the simplified size of the output file will be 141 MB with 30-m resolution
and 565 MB with 15-m resolution. 420 MB is the approximate maximum binary file size that can be
repaired with 32-bit STL tools like qAdmesh and netfabb Basic, so the 15-m resolution file will need to
be repaired with a 64-bit tool like MeshLab or a professional version of netfabb.

STEP 6 – CONVERT THE POINT CLOUD TO .STL WITH TOPO2STL

Launch TOPO2STL from FLOW-3D > Utilities menu > Topo2STL. Set Minimum Z= -250 m and Spatial
Resolution = 30 m. Repeat the process with 15 m resolution.
 Figure 13. Convert the point cloud to .stl with TOPO2STL

Figure 14. TOPO2STL output (30-m spacing), viewed in MeshLab

STEP 7 – SIMPLIFY/RECOMBINE FACETS
Import the ASCII .stl output into MeshLab. Use the Conditional Vertex Selection filter to pick the facets
that have all three vertices above 570.9 m. Use the Quadric Edge Collapse filter to reduce the bounding
box facet count with a target of 90% fewer facets and the other options shown in Figure 16 below.

Figure 15. Select the bounding .stl surface in MeshLab

Figure 16. Recombine the bounding surface facets to reduce them by 90%
STEP 8 – DELETE NON-MANIFOLD VERTICES
Select Render > Show Non-Manifold Vertices to mark vertices with connecting facets that do not adjoin
perfectly. Use the Select Non-Manifold Vertices filter to select and then delete these. Repeat as
necessary to eliminate all problem vertices, which should all be in the simplified area as shown below.

Figure 17. Render, select, and delete non-manifold vertices

STEP 9 – CLOSE TRIVIAL HOLES

Use the Select Borders filter to identify holes generated from the first simplification. Use the Close
Holes filter to auto-close all holes with fewer than 10 edge facets.

Figure 18. Select and close holes with fewer than 10 edge facets
 Figure 19. Close small holes

STEP 10 – REPAIR THE BACK FACE

Delete the four non-manifold vertices on the back corners and the facets that share them, if you didn’t
already in Step 8. Use MeshLab’s Fill Hole tool to repair any trivial holes that remain from simplification,
and then iteratively use the Self-Intersection repair option on the single large hole on the back. You will
have to re-launch the Fill Hole tool after each operation to slowly fill in the back side.

Figure 20. Locate and delete non-manifold vertices with MeshLab

 Figure 21. Fill holes and repair the .stl bottom with MeshLab

If a smaller .stl file size is desired, repeat Steps 7 through 9 until the desired size is achieved. Note that
too much simplification can make a watertight surface impossible.

STEP 11 – FIX FACET NORMALS

Figure 22. Normals before (left) and after (right)

Use the Recompute Face Normals and Normalize Face Normals filters. Save the output. Launch
qAdmesh from the FLOW-3D user interface (Model Setup > Meshing & Geometry > Tools menu). Check
and repair the .stl, including the normals, which will be slightly re-adjusted from MeshLab’s orientation.
If the file is small enough (< 420 MB), re-check the saved final .stl with the netfabb Basic Standard
Analysis.
STEP 12 – CHECK ACCURACY
Use CloudCompare’s Compute Cloud/Mesh Distances tool with Octree Level = 10 and compare the .stl
file to the original trimmed point cloud. Do not include the generated 2-D grid that was added to make
the upper surface of the .stl.

SIGNED ERROR ERROR MAGNITUDE

STANDARD STANDARD
MEAN DEVIATION MEAN ± DEVIATION
NED (USGS 2003) 2.44 m ~2.3 m
-0.32 m 2.42 m th
(NED vs. 13,305 geodetic controls) (RMSE) (1/2 95 percentile)
15-m TOPO2STL spacing
6 -0.10 m 3.6 m 2.4 m 2.6 m
(.stl surface vs. 20.3 x 10 points)
30-m TOPO2STL spacing
6 -0.28 m 7.9 m 5.5 m 5.7 m
(.stl surface vs. 20.3 x 10 points)
Table 2. Accuracy statistics for 1/3 arc-second data: original and converted to .stl

As of June 2003, the USGS NED data exhibits vertical uncertainty vs. the real world that is similar to the
uncertainty generated by converting the NED data to .stl format with 15-m spacing. For details of the
NED error estimation, see USGS (2006) “Vertical Accuracy of the National Elevation Dataset”, online as
of 04/29/2014 at http://ned.usgs.gov/documents/NED_Accuracy.pdf. The area of interest (the valleys
where the St. Francis flood occurred) has a very low error, as shown in Figure 23. The error magnitude
distribution is shown in Figure 24: it is best described as a Weibull distribution. The 15-m .stl file (Figure
25) shows the least error, and will be used in subsequent FLOW-3D simulations.

Figure 23. Original point cloud, colored by distance from 15-m resolution .stl surface
(blue = 0 m, red = ±7.6 m i.e. mean + 2 standard deviations)
Figure 24. Histogram of point-cloud to 15-m .stl facet error. Gray region is outside 2 standard deviations.

Figure 25. Close-up of the final 15-m .stl.