Lecture 50 : Advanced Turbulence Models II

IIT Kanpur July 201
20 followers
Sep 26, 2024
Hello everyone, in the last class we started discussing about the eddy viscosity model
and we looked at the zero equation models like the Prandtl-Milchner model, Von-Riess-Milchner
model, C. B. K. Smith model and then we started talking about the two equation models and
we derived the equation for the kinetic energy from the first principle and we found different
terms in that the temporal term, advection term, production term, diffusion term.
And then we saw that we do not have any way to basically derive the equation for the basically
the dissipation rate. So, what we did we followed the basically different terms I mean most
of the equation that we see whether it is a mass conservation, momentum conservation,
energy conservation you have these terms available the temporal term, advection term and then
you have the production term and the diffusion term.
So, we use those terms and derive the I mean equation the dynamical equation the transport
equation for the epsilon which is the dissipation rate. So, I have basically written those equations
here you can see this is my temporal term, advection term, dissipation term and this
is the source plus sink. This is again I am giving you for the standard K epsilon and
as I told that K epsilon model is a class of model I have been telling that the equation
for the epsilon may differ depending on which model we are using. We are using a standard
K epsilon or R and G or basically realizable. And you can see there are several coefficients
here which is sigma K, sigma epsilon and then sigma C epsilon 1, C epsilon 2. So, the values
of these also may vary depending on the model. Even for the same model depending on when
it was proposed you have basically we have seen that this value is evolved. And how this
value is evolved basically you get the more and more accurate experimental data or the
D N S data. You use these models and plot it and see when basically it matches perfectly
with your experimental data or the D N S data those values you use.
In fact, if you are using these models in a solver you can change these values to suit
your requirement the values of basically C epsilon 1 and C epsilon 2. So, now we have
two things K which is kinetic energy unit is meter square per second square. And then
you have epsilon which is unit is meter square per second cube. We are looking for nu t which
is eddy viscosity whose unit is meter square per second. Again we will go back to the dimensional
argument we want to use this K and epsilon and get this nu t. So, how do you do that
you can use this basically by the dimensional argument. So, this will if you take K square
by epsilon. So, this is this will become meter. So, this will become m to the power 4 by second
to the power 4 meter square by second cube. So, that will become meter square per second.
So, your nu t must be proportional to K square by epsilon. We will also look at like how
it is related to the argument that you have been making about the velocity scale and the
length scale. But for now you can see by the dimensional argument nu t proportional to
K square by epsilon. And then we use a proportionality constant which is called C mu. So, this will
become C mu K square by epsilon. And if you use the definition of the eddy viscosity which
is nu t that will be this is the kinetic eddy viscosity this is the eddy viscosity this
will become C mu rho K square by epsilon. This is epsilon. So, once you know the K and
epsilon which will you can get by solving these dynamical equations along with your
mean flow equations. You can find eddy viscosity use this eddy viscosity in the equation that
you have the momentum equation as well as your energy equation. And you can solve the
problem for the mean flow. One more advantage of this eddy viscosity model compared to let
us say Reynolds model is that it provides more stability to your numerical system. Why
because as you have been talking that if a problem has got higher numerical viscosity
it is likely to be more stable. In fact, it is more stable. Remember the argument
that you had made about this Lax-Friedrich scheme and we said that it is highly dissipative
because it has got a higher numerical viscosity. So, in effect if you talk about the eddy viscosity
we are multiplying. So, we earlier let us say you had mu del u bar by del y. But now
by using the argument of eddy viscosity your effective viscosity became mu plus mu t del
u bar by del y. So, your effective viscosity increased. So, that means numerically it
has become more and more stable. So, that is another argument in favor of the eddy viscosity
model compared to trying to get a dynamical equation for the Reynolds stress itself.
So, this is again I have put all the equations together for the k and for the epsilon and
I have basically indicated different terms. So, this is the production of kinetic energy
and dissipation and based on the model we have this the production of the dissipation
rate if I can call that and this is the dissipation. So, these are all model terms. So, we saw
that our mu t is given by C mu tau k square by epsilon, but how does it fit to our definition
that my viscosity should involve a length scale and a velocity scale. So, you can compare
let us say if I compare it. So, your mu t is equal to C mu k square by epsilon, but
your mu t is also equal to basically k to the power half is a velocity scale multiplied
by some length scale which I am writing as l m.
If you compare these two, you can write the length scale as C mu k to the power three
half by. So, this is your length scale based on this length scale you can remember this
C mu is just a constant it is a proportionality constant. So, if you have know this length
scale you can define a number which is called the turbulent Reynolds number. You have seen
the Reynolds number. Now, we are talking about turbulent Reynolds number. You might
be seeing the different Reynolds different turbulence model like low Re turbulence model
high Re turbulence model. So, that Re there basically indicates the turbulent Reynolds
number. What is Re turbulent Reynolds number? So, like the way you have the diffusion of
Reynolds number right how the Reynolds number is defined let me write it here. So, this
is Reynolds number is initial forces by viscous forces right. So, this is basically rho u
square initial forces this is mu u by l right that becomes the rho u l by mu right this
is your Reynolds number. In the same speed we will define the turbulent Reynolds number
which we call as the turbulent forces basically the forces due to the turbulence stress and
divided by the viscous forces. So, the region where the turbulent force is more dominant
that will be your high turbulent Reynolds number where the viscous force is dominant
that will be your low turbulent Reynolds number. As you might guess that where is the viscous
force is dominant very close to the wall right in the viscous sub layer right. So, if you
can elaborate it. So, we will be writing it. So, this will be rho u square this u square
so we get to see what is the u square definition and divided by mu u by l right. So, for the
turbulent forces the u square definition will be turbulent kinetic energy because u is k
to the power half. So, u square will be rho k mu and this will be k to the power half
just see. So, the k is k to the power half is our u. So, let me just work this out u
and then. So, length if you look at the length scale this is k to the power 3 by 2 by epsilon.
So, let me write that this is k to the power half divided by k to the power 3 by 2 by epsilon
right. So, if you work this out this will become rho k square by mu epsilon which is
given here. So, remember this indicates the ratio of the turbulent forces which is like
given by this k square and by the viscous forces which is dominated by the viscosity
multiplied by your dissipation right. So, we will come back to this turbulent Reynolds
number later when we talk about the low Reynolds number model.
Now, the question is how do we determine the coefficients right like say you have the what
are the coefficients you have mainly you have c mu which is basically the multiplying constant
when you try to get this mu t from k and epsilon and then you have c epsilon 1 and c epsilon
2. We also had like mu alpha earlier or I think there was another mu in the dissipation
equation right. So, let us try to get this c mu again these are all empirical models
but still there is I mean lot of experimental insights into deriving these values otherwise
these values can be anywhere and you will not have a way to basically find out the value
of this. So, if you talk about how do we get the determination of c mu what we are going
to do we are going to talk about something called local equilibrium.
Most of the models that you see are the adivisory models which are popular they actually follow
the local equilibrium what it says is that your production is equal to dissipation locally.
Now, if you talk about the production, so the production is given by minus u prime e
prime bar del u by del y right and we can equate it to basically your dissipation but
again this is stress this is a stress and if you go by the argument the constant stress
layer this is equal to your u tau square and in the log layer you have this u tau by k
y kappa y right. So, epsilon is nothing but u tau cube by kappa y that is one definition
of your dissipation rate by going the equilibrium this is a very important assumption and actually
it fails in many regions and that is why you get that your model is not behaving correctly.
So, we are assuming that your dissipation rate is equal to the production of the kinetic
energy right. Now, if you go look at the turbulent boundary layer, so you can also write it as
the turbulent viscosity is equal to tau by del u by del y right and tau is nothing but
your rho 2 rho u tau square by the definition of the friction velocity u tau and also you
have del u by del y is equal to u tau by k y this is the log layer.
So, if you combine these two this definition tau turbulent and del u by del y. So, you
will get this value of rho u, so this is basically rho u tau square which is your tau turbulent
by u tau by kappa y right that is a mu t. So, and you can write as mu t by rho is equal
to u tau by k into kappa y right. So, this is basically kappa u, so your mu t is equal
to u tau kappa y right. So, remember this mu t what is your mu t and what is your dissipation
rate and you can combine that. So, let me write it here mu t equal to u tau kappa y
and earlier we had epsilon equal to u tau by kappa y epsilon equal to tau by kappa y
right and you can use this definition of kappa y here u tau q by kappa y. So, you can combine
these two, so that will become mu t u tau into u tau q by epsilon this will become u
tau 4 by epsilon right. So, this is your mu t by going by the argument
of the log layer and the constant stress layer, but also the mu t is c mu k epsilon. So, if
you compare these two you will get c mu equal to u tau square by k square right and u tau
is nothing but your turbulent stress square. So, minus u prime v prime square by kappa
square right. So, this is your the value that we are getting by comparing in the turbulent
boundary layer right. So, we have seen in the experiment that the ratio I mean typically
the ratio of the basically the shear stress by the turbulent kinetic energy turbulent
shear stress by turbulent kinetic energy is around 0.3, where this is seen in most of
the turbulent flows. So, then epsilon will be basically minus u prime v prime bar by
k square which will become 0.09 c mu. So, this how you get the value of c mu and
this is typically I mean very robust and does not change too much on the other hand you
can also use the epsilon equation and by comparing with the experimental data we are going to
use the derivation. You get the value of this c epsilon 1 and c epsilon 2 which is
around 1.44 and basically 1.92 and then these are the values which are used again the from
the modeling sigma k equal to 1 and sigma epsilon equal to 1.3. These values have evolved
over time you can look at different models and this value will be very different in different
models. Similarly, your epsilon transport equation is also heavily modeled and varies
among different models. So, epsilon is a class of models.
The last thing I want to discuss is the low Re formulation. So, like if you recall the
Prandtl model one distinct model what we said that the model which basically works in the
fully turbulent region away from the wall may not work very close to the wall. The reason
for that is basically the viscous layer dampens the eddies. So, you have to make some argument
that how the effect of turbulent forces go down very close to the wall because we have
this viscous sub layer very close to the wall. Let us say here where your value of the turbulent
forces are not that dominant. So, like the one this change the Prandtl model by adding
some damping the same thing is done in the low Re formulation of the co epsilon model.
So, that the co epsilon model works not only away from the wall, but also very close to
the wall. So, what they have done basically they have used different damping for different
coefficient. So, this is f 1 f 2 and f mu which is for c epsilon 1 c epsilon 2 and c
epsilon c mu. So, these terms basically multiply wherever the c epsilon 1 you multiply with
the f 1 c epsilon 2 multiply by f 2 and divided by c mu the definition of basically mu t you
multiply by f mu. You can see that these are the numbers which are functions of Re t and
if you go very close to the wall the value of Re t decreases because the turbulent forces
decrease and that has been accounted by using these kind of functions here in f 2 and f
mu. They also tried using the damping for f 1, but they found that it basically does
not affect the model that much. So, they have kept the f 1 equal to 1, but they have used
the damping based on the turbulent Reynolds number in f 2 and f mu and that gave us the
formulation of the low Re formulation. So, that it can work all the way to the wall and
you get the essential damping due to the viscosity incorporated in the k epsilon model.
So, this is how you see. So, this is basically if you just look at the f mu which multiplies
with your c mu. So, your mu t is basically f mu c mu and k square epsilon that is what
we had here earlier. I am observing the c earlier. So, you can see that very close to
the wall f mu is much less than 1. So, that means your eddy viscosity is very small, but
it will be smoothly if you go away from the wall. It is going to increase and if you go
to the fully turbulent region, it becomes 1 and that your definition refers to mu t
equal to c mu k square by epsilon. One more topic that I want to briefly mention
is the wall function. So, earlier like when the computers were not very powerful, what
you could do like you know that how the flow vary very close to the wall. You know there
is a region where u plus equal to y plus and there is a region which is typically y plus
less than 5 and y plus greater than 30. We have u plus equal to 1 by kappa l n y plus
and plus c 1 or b the intercept because this is log low and this is the log in the viscous
sub layer. So, what they used to do that you can actually provide, you can assume a profile
here very close to the wall. This is the node very close to the wall and if you can assume
or you give that profile to the model and ask it that you do not have to solve it, but
you can use this profile directly. Of course, this is valid only when the boundary layer
is in equilibrium and when the it is the fully turbulent velocity profile. If there is a
separation or there is a reattachment, these things will not work because they will not
follow this law. Let us say there is a separation, your velocity profile will become like this
in the case of separation, but for other attached flows, you can use these wall functions.
Then the advantage here is that you do not have to resolve all the way to the wall. If
you want to let us say resolve this y plus equal to u plus, you need at least 5 grid
points. So, that will be minimum point should be y plus equal to 1 and then 2, 3, 4, 5,
so that you can resolve this region, the linear law.
Similarly, here also we have to resolve in between you will have buffer zone that also
you have to resolve. So, you could use this wall function, so that you can just model
this entire thing using assumed velocity profile for a turbulent boundary layer and just solve
it. Of course, it will give you a poor prediction for separation and reattachment. The other
thing is that you should not put grid points between 5 and 30. Why? This is the region
of the buffer zone where none of these laws are defined. Generally, you are asked to basically
put the y plus less than 1 or around 1. If you are solving to the wall, if you are using
wall function, then put it the value above 30, so that you put the wall function up to
30, which are following this law. After that, your model is resolving through the simulation.
So, these are the wall functions. So, these finally, we have the limitations of the k-epsilon
turbulence model. It is valid only for the fully turbulent flows. It will give poor predictions
as we expect for the rotating flows and swelling flows. Despite of the fact that we have incorporated
the low Re model, it is not very good for flows with separation or even with shocks.
If there is a lot of swirl or rotation, the assumption of the local equilibrium or the
boundary equilibrium goes breaks down. This does not give you the good prediction. So,
this is usually preferred for high Re applications where there is no separation or reattachment.
If you want to let us say solve, so we use this model for a complicated turbine blade
where there is a lot of separation. For example, the flow is coming like this and this is separating.
This model is likely to not give you the good prediction. The k-epsilon model is not good.
You have to have like separation and all like in turbo machine. Similarly, flow with strong
shocks, this is not good. So, what we will see in the next class is that we will look
up for a look a model which works very well, very close to the wall. That model is called
the k-omega model, but then there are issues with this model also. We will try to blend
the k-epsilon and k-omega model and that will give you a model called SST k-omega model.
So, this is what we are going to discuss in the next class and that will be our final
class on this RANS modeling. I hope it gave you a good picture of how these models work,
what are the eddy viscosity model and how things work. After that, we will look at these
models. I mean it will give you an idea that how to use this model, how does this coefficient
work. So, with that we will stop the class today. We will meet in the next class.
