Why Are My Results Inaccurate?

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 5

No. The error message means your regions are not connected.

Check the
interfaces/mesh which connect these regions. Do not disable the expert setting.
Why are my results inaccurate?
1. Firstly, is the simulation correctly set up? Does your model include all relevant
physics?
2. Has your solution converged to a reasonable value (see above)? For a simple
analysis, an RMS residual of 1E-5 should be sufficient. Keep lowering your
residual value until the solution no longer converges monotonically.
3. Perform a sensitivity analysis on the relevant features of your simulation. In its
simplest form, a sensitivity analysis involves altering one parameter (e.g., grid
refinement) while keeping the other parameters constant and observing the
effect this parameter has upon the simulation accuracy (e.g., by monitoring
skin friction). The purpose of a sensitivity analysis is to obtain a simulation
that is both accurate and cost effective. For just about every simulation a
sensitivity analysis on mesh size and convergence tolerance is required.
Depending on what you are modelling other sensitivity analyses may be
required, covering things such as:
o Domain size (how close can the boundaries be to the region of
interest?)
o Grid density
o Grid quality
o Grid type (e.g., structured versus unstructured),
o Spatial discretization (or advection) scheme
o Temporal discretisation scheme
o Turbulence model,
o Turbulence numerics,
o Turbulence intensity,
o Boundary condition type,
o Any empirical models used, such as heat/mass transfer coefficients,
drag laws, free surface model,
o Two- versus three-dimensional simulations,

o and timestep size.


4. Perform an error analysis:
o Calculate your discretization errors. A calculation should be performed
on each quantitative parameter of interest (e.g., lift). The Journal of
Fluids Engineering provides an explanation of predicting discretization
errors here: Calculating Discretization Errors. In terms of errors,
discretization errors are the greatest concern to a CFD modeler.
Discretization errors are those that occur as a result of modeling the
governing flow equations as algebraic expressions in a discrete spacetime domain. Discretization errors should approach zero as the grid is
resolved (i.e., as more grid points are generated and the solution
becomes grid-independent). However, discretization errors also depend
on grid quality (e.g., aspect ratio, orthogonality, skew). A well-refined
grid is not necessarily a high quality grid and, thus, grid resolution and
quality sensitivity analyses are recommended.
o Physical approximation errors can cause significant accuracy issues.
This type of error is caused by physically approximating something to
simplify the simulation (e.g., modeling a real fluid as an ideal gas).
o Computer round-off errors are caused when a computer stores floating
point data. Depending on the accuracy your computer stores the data
at, some round-off may occur. This type of error is usually negligible
compared to the others and can be easily decreased by running the
simulation in Double Precision mode.
o Iterative convergence errors are due to the resolution of the RANS
equations. Since the Navier-Stokes equations are not solved directly,
the solution iterates until a specified convergence criterion is reached
(e.g., a residual value of a certain size is reached). This type of error
can be decreased by allowing convergence to an acceptable level (see
above).
o Usage errors are caused by user specifications. For example, the user
may run a turbulent simulation as laminar or an unsteady simulation as
steady. These errors can be monitored via a convergence study.
2. And finally - consider the results are you comparing against. Are they
accurate? How do you know? Are you modelling EXACTLY the same
conditions as the results you are comparing against?

In my experience:
Simple simulations with straight forward flows can converge in under 100
iterations.

Most flows with moderate complexity converge in 100-200 iterations.


Complex flows (eg shock waves, multiphase, radiation, chemistry,
transitional flows) will take longer than 200 iterations to converge.
This is assuming a high quality mesh and correct convergence settings. If
you cannot do a high quality mesh or don't have good convergence
settings then things will take a LOT longer.

I can't understand you. You should define an interface, not a boundary condition. In
the interface definition, you should set your fluid domains(you should have at least 2
pieces), and the regions, where they connect. It is very simple, and clearly discussed
in the help.

If the interface connection isn't automatically generated, you should click


on the Interface icon on the top, and set what do you need. Here Interface
means a defined connection betweed fluid domains, which need to be
automatically or manually adjust.

If you have separate bodies, they will each meet separately. What you need to do is go
into DM, select all the parts and right click to create a multi-body part.
This will mesh the assembly with conformal mesh. Then Make sure to put all the
bodies in the same named selection so that they appear as one cell zone in FLUENT.

My steady state solution converges for a while but stops converging before
reaching my convergence criteria
Failure to obtain full convergence is a common issue for steady state simulations. It
often occurs when doing mesh refinement studies, where coarse mesh simulations
converge fully and quickly but as the mesh is refined the simulation fails to achieve
the same residuals convergence. Frequently the cause for this behaviour is by the
coarse mesh not resolving small shedding features, but as the mesh is refined smaller
flow features are resolved and unless they are properly handled by the simulation
setup it can lead to convergence problems.
To resolve the issue work through the following points:
1. Read the CFX documentation. Specifically section 15 ("Advice on Flow
Modeling"), section 15.10.6 ("Problems with Convergence"), and and any best
practises guide applicable to your simulation.

2. The first thing to consider is whether your simulation is sufficiently converged


even though your specified criteria has not been met. To check whether your
simulation is sufficiently converged, output the parameters important to your
simulation to monitor points and display them in the solver manager.
Appropriate parameters could be lift, drag, pressure loss, flow rate - what ever
is an important parameter to your simulation. If these parameters are not
changing to an accuracy tolerance suitable for your simulation then your
simulation is almost certainly OK as it is and no further work needs be done.
3. If important parameters have not converged so a tighter convergence is
required, the next thing to try is trying to get the simulation as it currently is to
converge. Tips here are:
o Use a larger physical time step. A time step approximately equal to the
average residence time in the simulation domain is a good guide for
most simulations. If it is a recirculating system without an inlet or
outlet then use the turn over time of the largest flow feature. You can
get the residence time in CFX-Post by placing a streamline and looking
at the "Time" variable on it. The maximum value of time is the
residence time.
o Use Local Timescale Factor. A factor of about 5 is a good guess to start
with. If this is successful you should run the final few iterations to
convergence with a physical timescale (not local timescale all the way
to convergence).
o For some simulations using double precision can help, particularly if
there is a large range between maximum and minimum values of
dimensions or flow parameters (velocity, pressure etc). It can also
useful for buoyancy driven simulations.
o If using the hybrid differencing scheme you can consider reducing the
blend factor. Don't reduce it below 0.75 without showing it does not
harm accuracy by a sensitivity check.
4. If you cannot get it to fully converge then you should look at why it has not
converged and try to fix the problem. Check:
o The physics of the simulation is correct.
o Do a test run with the residuals included in the result file. It is likely a
small region of the flow has high residuals while the rest is converging.
Consider why are the residuals high in that region - Is it:
1. Poor quality mesh - the fix is obviously do a better quality
mesh
2. A physical instability, such as vortex shedding - the fix here is
to use a larger timescale, a coarser mesh in the vortex shedding

region, decrease the blend factor (if using hybrid differencing)


or use a lower order turbulence model. The first option is
preferred as the latter options can have accuracy implications.
5. If you still cannot get the simulation to converge then try running it as a
transient simulation. Adaptive timestepping can be useful here to quickly find
the appropriate timescale. Transient simulations are much slower than steady
state simulations so be aware that you will need extra patience.
6. If the transient simulation shows the results to not be steady state then give up
on the steady state model as the flow is transient and needs a transient solution
to properly capture it.

You might also like