Вы находитесь на странице: 1из 11

TURBULENCE MODELING

Tips & Tricks: Turbulence Part 1 - Introduction to Turbulence Modelling


We will now focus on Turbulence Modelling, which is a critical area for any engineer involved with
industrial CFD. There are a number of different approaches so it is important that you have solid
grounding in this area to enable you to choose the appropriate model for your simulation
requirements. It is worth noting that in August 2012, LEAP will be hosting Dr. Florian Menter to run a
series of Advanced Turbulence Training courses in Melbourne, Sydney and Perth. Dr. Menter is a
world recognised expert in turbulence modelling, and more information on his visit to Australia can
be found here.

The ANSYS CFD Solvers solve the Navier Stokes and conservation equations, but as direct solutions
are not possible to resolve for any flows of an industrial Reynolds number then we need to do some
modelling, as opposed to resolving the values directly.

The equations that we used are not closed and so we need to use Turbulence Modelling to close the
equation set and then iterate towards a solution. We can use what is called a Reynolds Averaged
Navier Stokes (RANS) approach, or we can use an Eddy Simulation technique which resolves the
larger eddies in the flow and is only really required when you have separation or large recirculating
regions.

The most commonly used models are the RANS models due to their low cost in terms of compute
power and run times. The Eddy Simulation methods can be quite mesh sensitive but will yield much
better results for separated and recirculating flow, albeit over much longer run times.

The RANS models apply a Reynolds decomposition technique to the Navier Stokes equations which
breaks the velocity down into its mean and fluctuating components. This decomposition leaves us
with one unknown value, which is termed the Reynolds Stress. We use Turbulence Models to resolve
the Reynold’s Stress and close the equation set. There are two ways we can go about resolving this,
the first (and most commonly used approach) is to use an isotropic value for the turbulent viscosity
value which is called the an Eddy Viscosity Model, the other way is to solve using the Reynolds Stress
Model (RSM) for the 6 separate Reynolds Stresses, which results in an anisotropic solution.

EDDY VISCOSITY MODELS

The limitation with Eddy Viscosity models is that they use an isotropic value which may not be
appropriate and hence can increase the diffusion in your result. Obviously solving for the 6 Reynolds
Stresses and dissipation will be more accurate, but you are then solving extra equations which will
increase your run time considerably. There are further modifications to the 2 equation Eddy Viscosity
models that yield similar results to RSM which I will elaborate on soon.
I have listed below the most commonly used Eddy Viscosity turbulence models over the past ten or
so years and the intended use behind their development.

The common options are:

 Spalart-Allmaras – One equation model for attached aerodynamic analysis.

 k-epsilon – Two equation model for free shear and non-wall bounded flow behaviour. Was
the previous industrial standard.

 k-omega – Two equation model for wall bounded flows, not commonly used.

 SST (Shear Stress Transport) - Two equation model blending the freestream advantages of the
k-epsilon model with the wall bounded advantages of the k-omega model. This is the new
industrial standard and should be the default choice for most applications.

Note: Modifications for Curvature Correction to the SST model give comparable results to RSM and
the SST model also works very well with the Transition Model now available in CFX and FLUENT.

Options like Curvature Correction are very useful in tackling problems where previously you would
have required an RSM model. The Transition Models can be of benefit also, depending on your
application and give much better drag prediction as they will maintain laminar flow along a body and
develop natural transition points, as well as calculate regions of bypass transition. I won’t discuss
these here, but there is extensive information in the ANSYS User Guides. These additional options
are available in ANSYS CFX and ANSYS FLUENT.

REYNOLDS STRESS MODELS

The Reynolds Stress models were used widely for high swirling flows, for example, the flow in a
cyclone separator. In the past we would use the RSM models for this type of flow, but nowadays you
should use the SST model with the Curvature Correction model enabled.

EDDY SIMULATION

The more advanced options for resolving turbulence are the simulation based models. These
generally require much better mesh resolution and involve much longer run times as they can only be
run as an unsteady simulation. These are used primarily for resolving large scale separations and
recirculating regions. The options in order of least expensive, to most expensive are:

- Scale Adaptive Simulation

- Detached Eddy Simulation

- Large Eddy Simulation

The Scale Adaptive Simulation is the best point to start when stepping up from a RANS modelling
analysis. The Scale Adaptive Simulation is based on an unsteady SST RANS model but calculates the
local length scales and resolves accordingly. The advantage of this is that it can be run on a good
quality RANS mesh without any additional meshing requirements. For some flow behaviours this has
been shown to give very similar results to Large Eddy Simulation (LES) and Detached Eddy Simulation
(DES). This is why we would recommend this as a starting point for unsteady simulations with
separated or recirculating flows, and as a good alternative to an unsteady RANS SST simulation.
SAS solution in a Swirl Combustor

The Detached Eddy Simulation resolves the boundary layer and the smaller eddies by using a RANS
solution in those regions. This can produce quite good results for a fraction of the cost of LES, but is
very sensitive to the mesh resolution. You need to ensure a very good quality mesh in the boundary
layer and a good transition region from the boundary layer to the freestream to help the solver
switch from the RANS solution in the boundary layer to an LES solution for the larger eddies. This
adds complexity and has been known to cause erroneous results.

DES solution in a Swirl Combustor

Large Eddy Simulation is the most expensive of the three simulation options and requires a very fine
grid to accurately resolve the eddies, especially in the boundary layer. Once the eddies are too small
to be resolved by the grid, they fall back into a sub-grid model. As this sub-grid model is used only to
resolve the smallest eddies which tend to have more universal properties, this yields very accurate
results, but because of the large mesh requirements this is not a commonly used approach in
industrial settings. We can now use LES in specific regions of the mesh only using Embedded LES. This
means that we can run an unsteady RANS in the bulk of the flow and a full LES calculation in only the
required separated regions hence making it more practical for industrial use.

Each of these simulation methods when used correctly will yield a better result than the steady state
or unsteady RANS models for separated flow or recirculating flow behaviour but it is more expensive
to do so.

As far as Turbulence Models are concerned, the SST is a good default choice for most flow problems.

Tips & Tricks: Turbulence Part 2 - Wall Functions and Y+ requirements


Previously we have discussed the importance of an inflation layer mesh and how to implement one
easily in ANSYS Meshing. We also touched upon the concept of mesh y+ values and how we can
estimate them during the inflation meshing process. In other posts, we also discuss the different
turbulence models and eddy simulation methods available to ANSYS CFD users. In today's post, we'll
talk in more detail about y+ values apply to the most commonly used turbulence models.

From our earlier discussions, we now understand that the placement of the first node in our near-
wall inflation mesh is very important. The y+ value is a non-dimensional distance (based on local cell
fluid velocity) from the wall to the first mesh node, as you can see in the image below. To use a wall
function approach for a particular turbulence model with confidence, we need to ensure that our y+
values are within a certain range.

y+ definition

Looking at the image above, we need to be careful to ensure that our y+ values are not so large that
the first node falls outside the boundary layer region. If this happens, then the Wall Functions used
by our turbulence model may incorrectly calculate the flow properties at this first calculation point
which will introduce errors into our pressure drop and velocity results. The upper range of
applicability will vary depending on the flow physics and the extent of the boundary layer profile.

For instance, flows with very high Reynolds numbers (typically aircraft, ships, etc) will experience a
logarithmic boundary layer that extends to several thousand y+ units, whereas low Reynolds number
flows such as turbine blades may have an upper limit as little as 100 y+ units. In practice, this means
that the use of wall functions for these class of flows should be avoided as their use will limit the
overall number of mesh nodes that can be sensibly placed within the boundary layer. In general, it is
recommended that you endeavour to place sufficient inflation layer cells within the boundary layer,
rather than simply focusing on achieving any particular y+ value. This will be covered in detail in a
future post

In addition to the concern about having a mesh with y+ values that are too large, you need to be
aware that if the y+ value is too low then the first calculation point will be placed in the viscous
sublayer (logarithmic) flow region and the Wall Functions will also be outside their validity (below
about y+ < 11). You can imagine that this would become an issue if a mesh intended to be used with
wall functions is then refined near the wall. Fortunately, the use of scalable wall functions in ANSYS
CFD products now takes care of these problems and produces consistent results for grids of varying
y+. Without any further user involvement, the scalable wall functions activate the local usage of the
log law in regions where the y+ is sufficiently small, in conjunction with the standard wall function
approach in coarser y+ regions.

So, where should you start? We have learnt that the wall function approach and y+ value required is
determined by the flow behaviour and the turbulence model being used. If you have an attached
flow, then generally you can use a Wall Function approach, which means a larger initial y+ value,
smaller overall mesh count and faster run times. If you expect flow separation and the accurate
prediction of the separation point will have an impact your result, such as the drag or lift forces
experienced by the ellipse below, then you would be advised to resolve the boundary layer all the
way to the wall with a finer mesh. Please refer to this post for a more detailed explanation of
appropriate turbulent wall function and modelling approaches.

Wall Function applicability

Once we know our preferred approach, we can estimate the thickness for our first inflation layer cell
using the equation below, which can be used to calculate the distance value for a specific velocity
fluid and the required y+ value (based on the flow over a flat plate). This is usually a good initial
estimate and the y+ value we aim for will depend on our turbulence model selection.

Note that Δy is the distance of the first node from the wall, L is the flow characteristic length
scale, y+ is the desired y+ value, Re_L is the Reynolds Number based on your problem's characteristic
length scale.

Unfortunately, as the y+ value is dependent on the local fluid velocity which varies across the wall
significantly for most industrial flow applications, it is not possible to know your exact y+ prior to
running an initial simulation. For this reason, it is important that you get into the habit of checking
your y+ values as part of your normal post-processing in ANSYS CFD-Post so that you can make sure
you are in the valid range for your flow physics and turbulence model selection.

Our next post in this series concentrates on the feasibility and selection of different wall functions,
based on the applied turbulence modelling strategy.

This is still an area of active research and is a hot topic for many of our CFD users. If you have any
questions or comments, please leave a message below or contact our CFD Technical Support team for
more detailed technical information on these topics.

Turbulence Part 3 - Selection of wall functions and Y+ to best capture the


Turbulent Boundary Layer
In recent posts in our series of Turbulence Modelling posts, we have covered boundary layer theory
and touched on some useful meshing and post-processing guidelines to check you are appropriately
resolving the boundary layer profile. Today we will consider three critical questions that are often
asked by CFD engineers when developing or refining a CFD simulation:

- Am I using the correct turbulence model for the type of results I am looking for?

- Do I have an appropriate Y+ value and a sufficient number of inflation layers?

- Am I using the right wall function for my problem?

This topic is so important because we know that in turbulent flows the velocity fluctuations
within the turbulent boundary layer can be a significant percentage of the mean flow velocity, so it
is critical that we capture these effects with accuracy. A Reynolds averaging approach using
turbulence models will provides us with an estimate of the increased levels of stress within the
boundary layer, termed the Reynolds stresses. In order to appreciate the use of wall functions and
the influence of walls on the turbulent flowfield, we should first gain familiarity with the composite
regions of the turbulent boundary layer:

Composite regions of the turbulent boundary layer

In the laminar sub-layer region (Y+ < 5) inertial forces are less domineering and the flow exhibits
laminar characteristics, which is why this is known as the low-Re region. Low-Re turbulent models
(e.g. the SST model) aim to resolve this area and therefore require an appropriate mesh resolution to
do this with accuracy. This is most critical for flows with a changing pressure gradient where we
expect to see separation, as observed below.

Predicting separation in a diffuser-type geometry

In the law of the wall region, inertial forces strongly dominate over viscous forces and we have a high
presence of turbulent stresses (this is known as the high-Re composite region). If using a low-Re
model, the whole turbulent boundary layer will be resolved including the log-law region. However, it
possible to use semi-empirical expressions known as wall functions to bridge the viscosity-affected
region between the wall and the fully-turbulent region.
Contours of the eddy viscosity ratio on a low-Re grid illustrating high turbulent viscosity in the log-law
region as opposed to the laminar sub-layer

The main benefit of this wall function approach lies in the significant reduction in mesh resolution
and thus reduction in simulation time. However, the shortcoming lies in numerical results
deteriorating under subsequent refinement of the grid in wall normal direction (thus reducing the Y+
value into the buffer layer zone). Continued reduction of Y+ to below 15 can gradually result in
unbounded errors in wall shear stress and wall heat transfer (due to the damping functions inherent
within the wall function approach).

Bearing all of the above in mind, and keeping our eye on finding the right balance between accuracy,
stability and speed, we can tackle a wide variety of CFD problems using the following guidelines:

What results am I interested in and am I using the right turbulence model?

If our aim is to accurately predict the boundary layer velocity or thermal profile, or if the developing
boundary layer will tend to separate (due to a changing pressure gradient – and not because of sharp
edges or discontinuities in the geometry), then we recommend the use of a low-Re model. Low-Re
models are also required for accurate pressure-drop or drag calculations. We highly recommend the
use of the Shear Stress Transport (SST) model, but all ω-based models or ε-based models with
enhanced wall treatment may be used. For high speed external aerodynamic flows, the one-equation
Spalart-Allmaras model (with Y+ < 2) may also be considered to reduce the computational time.
Alternatively, for flows where wall-bounded effects are not a priority, or if separation is expected to
occur only due to sharp changes in the geometry, an ε-based wall function approach is more than
sufficient. In ANSYS CFD, all ω-based models and the SST model are capable of resorting to a wall
function formulation (automatic wall treatment) in the presence of coarse mesh resolutions near the
wall without any further user input. Wall function models are also useful for calibrating our CFD
models, due to the decreased simulation time.

Flow pattern with separation and reattachment on a rotor blade


What is my Y+ value and do I have a sufficient number of prism layers?

When using low-Re models or any models with enhanced wall treatment, the average Y+ value should
be on the order of ~1 to ensure we are capturing the laminar sub-layer. When using wall function
models, the Y+ value should ideally be above 15 to avoid erroneous modelling in the buffer layer and
the laminar sub-layer. High quality numerical results for the boundary layer will only be obtained if
the overall resolution of the boundary layer is sufficient. This requirement in some cases is more
important than achieving certain Y+ values. The minimum number of cells to cover a boundary layer
accurately is around 10, but values of 20 are desirable. The total thickness of the prisms should be
implemented such that around 15 or more nodes are actually covering the boundary layer. Our next
post in this series on the turbulent boundary layer will cover a very useful and practical technique to
post-process the resolution of the boundary layer, and offer insight into modifications required to
improve accuracy.

Boundary layer velocity profile modeled with standard k-e for three different mesh densities using
Enhanced Wall Treatment

Am I using the right wall function?

In ANSYS CFD, all turbulence models are y-plus independent. However selecting the most appropriate
wall function is dependent on level of refinement of our wall adjacent mesh, or the relative scales in
our flow. Use of the standard wall function (ε-based models) implies that our boundary layer mesh
lies entirely within the log-law region of the boundary layer. For industrial applications, this in
fact might be difficult to achieve due to varying geometrical and velocity scales associated with our
model – and therefore grids inherently designed with arbitrary refinement. We highly recommend
the use of the scalable wall function, which offers an elegant solution to this ambiguity often
encountered. This wall function virtually displaces the mesh to a Y+ ~ 11.225 (transition to the log-law
composite layer) irrespective of the level of refinement, thereby avoiding the erroneous modelling of
the laminar sub-layer and buffer region. It is also important to note that for grids designed with a Y+ >
11.225, the scalable wall function will provide identical results to the standard wall function.
Enhanced wall treatment may further be selected for ε-based models on refined low-Re grids, and is
also formulated such that it can perform well for meshes of intermediate resolution. However the use
of enhanced (or non-equilibrium) wall treatment for low-Re modelling of the turbulent boundary
layer is generally not recommended and more confidence in our solution can be obtained by
selecting a suitable ω-based formulation, such as the SST model.
Turbulence Part 4 - Reviewing how well you have resolved the Boundary Layer
In recent posts we have comprehensively discussed inflation meshing requirements for resolving or
modeling wall-bounded flow effects due to the turbulent boundary layer. We have identified the y-
plus value as the critical parameter for inflation meshing requirements, since it allows us to
determine whether our first cell resides within the laminar sub-layer, or the logarithmic region. We
can then select the most suitable turbulence model based on this value. Whilst this theoretical
knowledge is important regarding composite regions of the turbulent boundary layer and how it
relates to y-plus values, it is also useful to conduct a final check during post-processing to ensure we
have an adequate number of prism layers to fully capture the turbulent boundary layer profile, based
on the turbulence model used (or more precisely, whether we aim to resolve the boundary layer
profile, or utilize a wall function approach). In certain cases, slightly larger y-plus values can be
tolerated if the boundary layer resolution is sufficient.

How can I check in CFD-Post that I have adequately resolved the boundary layer?

For the majority of industrial cases, it is recommended to use the two-equation turbulence models,
or models which utilize the turbulent viscosity concept and the turbulent viscosity ratio (i.e. the
turbulent viscosity over the molecular viscosity). We can make use of this concept to visualize the
composite regions of the turbulent boundary layer, and ultimately visualize how well we are resolving
the boundary layer profile. Consider the conceptual case-study of the turbulent flow over an
arbitrarily curved wall. Prism layers are used for inflation, and tetra elements in the free-stream. Once
we have calculated the solution, within CFD-Post we can create an additional variable for the eddy
viscosity ratio. Then by plotting this variable on a suitable plane, and superimposing our mesh in the
near-wall region, we can visualize the boundary layer resolution.

Figure 1: RKE with standard wall function – y-plus of 75 with 10 prism layers

Figure 1 provides an example of a reasonable wall function mesh. There is a good cell transition from
the prisms to the free stream tetra elements. The y-plus we have prescribed at the first cell indicates
we are in the logarithmic composite region of the turbulent boundary region, which is the region
largely dominated by inertial forces and thus we have high levels of turbulence. The turbulence
gradually dissipates as we approach free stream conditions (where the levels of turbulence are
governed by inlet conditions), which is expected. At this stage, we could even reduce the number of
cells in the inflation layer as we are clearly capturing the logarithmic region layer before approaching
the free stream. Correspondingly, we could aim to reduce the y-plus value (y-plus ~ 20) to better
capture the increase in turbulent viscosity as we move from the inner layer to the outer layer of the
logarithmic region.
Figure 2: SST low-Re model – y-plus of 1 with 20 prism layers

Figure 2 provides a good mesh for a low-Re turbulence model. We observe that the transition in size
from the final prism layer to the free stream tetra elements has been regulated well. Since we have
prescribed a y-plus value of 1 we are within the laminar sub-layer, which exhibits laminar flow
characteristics (thereby resulting in no turbulent viscosity). As we gradually move through the buffer
region and into the logarithmic region we see a large rise in the viscosity ratio before it dissipates into
the free stream. This maximum value will generally occur near the middle of the boundary layer,
which also gives us an indication of the physical boundary layer thickness (twice the location of the
maximum eddy viscosity ratio gives the boundary layer edge). As per the example given in Figure 2, it
is essential that the prism layer is thicker than the boundary layer as otherwise there is a danger that
the prism layer confines the growth of the boundary layer.

Figure 3: SST low-Re model – y-plus of 0.25 with 20 prism layers

Figure 3 ia an example of a poor quality mesh for a low-Re turbulence model (such as SST k-omega).
In this case we have unnecessarily prescribed a very low y-plus value yet we have not compensated
by appropriately allowing for more prism layers in the inflation layer. Therefore, we are capturing the
laminar sub-layer to an excessive detail, and the boundary layer does not transition to the logarithmic
region until we are well inside the free stream. Consequently there are cells which are not aligned
with the direction of the flow and thus our boundary layer profile will not be well resolved (its growth
may be confined by the extent of the prism layers), hence affecting our drag or pressure-drop
calculations.
Figure 4: RKE with standard wall function – y-plus of 1 with 20 prism layers

Figure 4 is an example of a poorly defined mesh for a standard wall function turbulence model. The
accuracy of wall function or high-Re turbulence models (e.g. the k-epsilon variants) cannot be
confirmed modeling the laminar sub-layer and thus should be avoided. In ANSYS Fluent, the laminar
stress-strain relationship is employed when the mesh is below a y-plus of 11.225 (noted as the
transition to the logarithmic region). After which, the logarithmic wall functions are employed. This is
an example where a low-Re turbulence model should be used (c.f. Figure 2), or alternatively we could
aim to increase our y-plus value such that it resides in the logarithmic region (c.f. Figure 1).

Figure 5: RKE with scalable wall function – y-plus of 1 with 20 prism layers

The problems in Figure 4 can be overcome using scalable wall functions, as shown in Figure 5 using
the same mesh. The purpose of scalable wall functions is to force the usage of the logarithmic law.
Here we can see a turbulent viscosity distribution which is analogous to the case presented in Figure
1 (with the exception that we are now capturing the increase in turbulent viscosity as we move from
the inner layer to the outer layer of the logarithmic region). For this simple case, we could potentially
save simulation time by coarsening the mesh immediately adjacent to the wall, or alternatively we
could opt for a low-Re turbulence model. The real advantages of the scalable wall functions arise for
complex flows on grids of arbitrary refinement (or correspondingly flows with various boundary layer
scales) since it will provide consistent modelling.

Вам также может понравиться