Lecture 52 : Boundary Conditions II

IIT Kanpur July 201
20 followers
2 Video Views·Oct 5, 2024
Hello everyone, welcome back. In the last couple of classes, we discussed about the
turbulence models or the k-omega turbulence model, k-epsilon turbulence model and the
blending of these two models, which is the SSTQ-omega model. And we saw that these models
work well for certain problems. SSTQ-omega particularly is very useful if your flow has
got separation. And again, all of these models also have many different versions, which were
basically proposed at different points of time. So, when you are using it, you have
to make sure that which version you are using. And you should also read the relevant paper
to basically find out what kind of coefficient that you require to simulate a particular
flow. So, I mean, it is very important to know about your flow. If you are expecting
your flow, I mean, that you will have separation or soft or expansion. Accordingly, you have
to basically use a model, which is best for those kind of flows. Also, you have to make
sure that your coefficients, the model coefficients that you are using are correct. Sometimes,
you have to tune the model coefficient also. Like I talked about this structure parameter
or the basically viscosity limiter, that you may have to change in certain flows to get
a better behavior compared to the external. That is why I always insist that when you
start with the turbulence modeling, you have to make sure that the model that you are going
to use is validated for the class of flow that you want to simulate.
Let us say, you want to simulate a flow, which is having separation. You should not
be using, for example, a model like k-epsilon model. Now, you can validate the k-epsilon
model for a attached flow and then continue to use that model for a separated flow. That
is not going to give you correct result. A very common example is, let us say, you want
to predict the lift on a wing at different angles of attack. So, your C L versus alpha
curve, if you see, it will go something like this. At certain point, this will be the point
of C L max. After that, it will start dropping. So, this is the stall. This is the stall angle.
For example, if you want to use the k-epsilon model, what will you see typically? I think
I have seen this kind of diagram earlier that it is going to work very well. Then, there
will be abrupt stall, may be early or may be late. Your C L max is also not predicted.
Why? Because as you increase the angle of attack, this is your alpha and this is your
C L. As you increase the angle of attack, the flow is going to be separated. So, you
had this initial flow. Let us say, this is the wing or an aerofoil. The flow is attached
nice. Any model will give you the good result. As soon as you go to a very high angle of
attack like this, may be this may make the same aerofoil. As soon as you went like this,
your flow is going to be separated here. Now, where is the onset of separation? So,
that will basically dictate what is the lift function that you want to get. How big is
the separation? Whether it is closed separation or open separation? So, all these questions
that you need to basically discuss or understand. Then, only you will be able to use CFD very
effectively. So, knowing about flow is very important. As you go to high angle of attack,
you know that flow is going to separate. So, you should be very alert and very careful.
Let me change the model and find the model which can make the prediction better. Then
also it may not be very successful. Sometimes, the flow become inherently unsteady. So,
I mean there are some scales which are unsteady in nature. So, that time if you are using
RANS and no matter what model you are using, it is not going to give you correct result
because the RANS is the average steady state flow. So, in that case, either you have to
switch to U-RANS which is very expensive. So, U-RANS as we had discussed is unsteady
RANS or you have to do to more expensive approaches like LES if possible or in between
a RANS and LES which is the eta g dissimulation. Now, we are going to look at a very interesting
model which is called the transitional model. So, the flow may be in some region, it may
be laminar. In some region, it may be turbulent and in between it may be transitional. So,
you know that when the flow is laminar, the skin friction is low. As soon as you go to
turbulent, the skin friction becomes very high. In the transitional region, it starts
rising and transitional region is very high and then the turbulent region is basically
higher than the laminar flow. So, if you are using a model, a turbulent model in a laminar
flow for example, there are certain region where it is laminar. It is going to produce
the viscosity, eddy viscosity which is not there in the flow. It is giving you an artificial
mixing which is not physical. So, it is not going to give you the correct behavior. So,
with that basically philosophy that we need to predict the behavior where the flow is
changing from laminar to turbulent, the transitions model have been developed.
There are two, three transition models in the literature. I am going to discuss one
model, but there are other models also. For example, one model is basically four equation
SST K omega model. It is one model. It is also called transitional SST model. Other
name is transitional SST. So, you should be very careful about the nomenclature. K omega
is standard K omega. SST K omega is a blending of K epsilon and K omega. Transitional SST
model is basically SST model, but some changes in the K omega equations along with that
you have two more equations. I will discuss, I mean come to those two equations very soon.
Then, you have basically K, K L, K T equation. We also call the laminar kinetic energy model.
So, here you can see that there is another term. So, there is a turbulence kinetic energy
K T, but there is another term which is K L. So, I think this is not K. This is K L
and K T. Then, there is another term. So, this is laminar kinetic energy model. So,
you not only solve for the turbulence kinetic energy, but also solve for laminar kinetic
energy. Then, there are some other, I mean transition
models which are coming because as we become more and more ambitious, initially when the
CFD was started, it was like just laminar flow we wanted to study. Then, we went to
fully turbulent flow and we found some success with the different turbulence models. Now,
we have become more ambitious and we want to do transitional flow also and we want to
basically capture full regime. The flow can be laminar in one region, turbulent in one
region. So, if you want to just see this boundary layer where the flow is laminar or turbulent,
I also discussed in the last class. So, before that I think this is K L, K T and omega.
So, you had K omega, but now you have the scale which is the laminar and turbulent kinetic
energy. They are separated. The laminar kinetic energy will be lower than the turbulent kinetic
energy. So, if you want to do this may be on a boundary layer, so this is the laminar
flow, initially laminar. This is the edge of the boundary layer. Then, it becomes transitional
little bit and then it becomes fully turbulent. So, you have region, this is I call this laminar
which is transitional and this is turbulent. RANS works here or the most of the turbulence
model, fully turbulent flow model works here. The laminar you do not require eddy viscosity.
So, mu t equal to 0 and in between you can have. So, this will have mu t and this is
a value. Let us call it mu transitional which is basically in between mu and 0 and mu t.
So, can we have a model which predicts all the regions? So, one approach is look at a
term which is called intermittency. So, what is intermittency? It basically defines how
much the flow is intermittent. So, if you look at the laminar flow, I mean
it is a very nice smooth flow and basically particles are not jumping here and there.
So, it is not intermittent. You can predict the behavior here. So, in this case you will
say intermittency is equal to 0. This is gamma equal to 0 here. On the other hand, the other
side of the spectrum, you have the turbulent flow where there is a very high intermittency.
You have eddies of all sizes, particle jumping from one layer to another layer and so on
and so forth. So, here we will call it gamma is equal to 1.
In between you can have transitional region where if you look at the theory of turbulence,
you will see that first there will be 2D stability. Then, there will be 3D stability from the
laminar flow and then you will have the formation of this small islands which are kind of looks
like eddies. Then, they become more and more violent and finally flow becomes turbulent.
So, here the intermittency can be between 0 to 1. So, intermittency you can think of
it as an indicator whether the flow is laminar or the turbulent. So, we will be using this
intermittency in our equation k and omega equation. For example, we will be discussing
these 4 equations in this STK omega model. So, basically find out where there will be
production of kinetic energy and how much the production will be. So, if the flow is
not turbulent, there is no production of kinetic energy, turbulence kinetic energy and how
much the dissipation will be and so on and so forth.
But, then there is another question, how do you find like in the flow where this gamma
should be 0 or where the gamma should be 1 or where basically it will be between 0 to
1. So, that is another problem that we have that means that how do you find out the onset
of the transition. So, this point is called the onset. So, you can call this as onset.
This is where it onset started and this is basically the your length of the transition
region, this is the length. So, where it is starting and where it is ending, this is the
onset of a transition and end of the transition. So, that is also I mean something that needs
to be predicted by the model. In fact, in the earlier models when if you are using a
transitional model, you had to give this is the onset, this is the point I want the flow
to become transitional and then it automatically start changing the value of mu t from that
region, otherwise the mu t will be 0. But, now models are smart, they can predict the
onset of transition and also they can predict basically how much will the size of the transition
when it will become turbulent, whether it will become turbulent and those kind of things.
Further, the transition can be of many types. There can be natural transition that means
you do not perturb the flow anywhere, the flow will naturally become transitional. It
starts with laminar flow and it will slowly become transitional. Then, you can have a
transition where basically you are trying to put some perturbations, may be the free
stream intensity is high, there is a turbulence in the background and because of that your
transition will not be natural, but it will be kind of accelerated.
It may so accelerated that it may go to stage, it is called the bypass transition. That means
this entire region that you are saying may very small or may not exist. I mean may not
exist to your eye that it become transitional, it become turbulent in a very small region.
So, that is called the bypass transition. You can also have a transition which is a
separation induced transition. So, for example, I was talking about the example of this flow
separation on an aerofoil or a wing. So, let us say flow is coming from this side and
then when it goes here, it becomes separated here. So, when it becomes separated, your
flow may become transitional within the separation. So, this is called the separation induced
transition. The flow may be laminar upstream, but because of the separation some activity
started happening in the boundary layer and that may made the flow transitional.
So, then again the prediction of separation bubble is a task. What kind of bubble it should
be? It is an open bubble or closed bubble. Flow may be transitional before this. Then
you have to find out how this bubble is changing the behavior of the boundary layer there or
flow may be laminar upstream. So, then it is a case of laminar separation bubble. So,
all those questions come into picture and that is why it becomes a very tricky business
for the model to predict this kind of flows. So, the true transition models like KKL omega
or KLKT omega, whatever you want to say or this 4 equation SST omega, they have certain
merits and demerits in the sense that they can predict some of these transitions. Maybe
one model can predict the natural transition well, but not the bypass transition. The other
model can predict the separation induced transition better. So, you have to basically
look at the behavior of this model. As I have been telling the RANS model is there is a
lot of experience needed to use these models in a difficult flow, in a complex flow. I
mean for simple flow like flow in a pipe or flow in a channel, it is very easy to use,
but when you have all these kind of complexity comes in and they are there in your engineering
flows. You want to use this turbulence model for engineering flows. So, there you have
to be very careful and you need to do a good study before you start using this model.
So, let us look at this 4 equation K omega SST model. So, the base line equations are
basically still your K omega, K n optional equations. So, this is your K equation and
this is your omega equation, but you see that there are certain changes in the equations.
So, this is your production of kinetic energy. So, earlier it was P K, now we have made it
P K tilde and then this is epsilon term which is the dissipation rate and this term also
has been modified. Finally, you have a blending term which was F 1 that one also has been
have been modified slightly here. So, other terms remain the same, but there are certain
terms which have been modified. So, look at let us look at the modifications of these
terms. So, the P K hat is nothing but P K multiplied by something called gamma effective
or in other words you can also take it as gamma. So, gamma effective is basically gamma
is the intermittency as we discussed and then this intermittency I mean you can see if the
flow is becoming transitional naturally. Then of course, you can look at the picture of
this boundary layer and see how the gamma is changing, but if the flow is having separation
then you have to also consider the gamma due to the separation. So, it basically takes
the maximum of gamma and gamma due to the separation intermittency which is developing
due to the separated boundary layer. So, this what we are seeing here is that now
my production of kinetic energy is no longer just basically dictated by your turbulence
model, but there is a term which is multiplies to this which basically controls the production
of the turbulence kinetic energy. So, let us say we have gamma effective equal to 1
equal to 0. So, that is a laminar flow. So, in this case your production of the kinetic
energy completely stops there is no production of the turbulence kinetic energy if there
is no turbulence. On the other hand if gamma effective is equal to 1 this is a fully turbulent
flow. So, in this case it basically goes back to the original S S T Q omega model.
Similarly, you can see here the epsilon also be changed. So, here you can see that it takes
a value I mean forget about the gamma separation. Separation transition will not cover in the
class, but you can go back and look at it here you can write it here as minimum of.
So, this is 0.1 or 1.0 epsilon. So, what it does basically this is incorrect it cannot
be minimum 0.11. So, there has to be gamma somewhere. So, what it says is that may be
it will be some factor some factor of gamma. So, the dissipation is also controlled by
gamma. So, but what this modification says is that
it if the flow is laminar when the gamma is equal to 0. So, again I am saying that
this is not gamma, but some combination of gamma which makes sure that it is 1 in the
fully turbulent flow and the value greater than may be 0.1 in other regions. So, if the
flow is laminar the function works in such a way that you have gamma is equal to. So,
the dissipation is equal to I mean epsilon hat is equal to 0.1 into epsilon. So, I figure
it out. So, this should be you can call it as minimum of maximum of gamma and 0.1 and
then 1.0 multiply by epsilon. Let us say in the flow is laminar then gamma is equal to
0, but you do not want to take 0 because if there is no dissipation also then also
you will have there is no turbulence also then also you will have dissipation of the
turbulence kinetic energy that turbulence kinetic energy may come from the free stream.
So, you have to dissipate that part also. So, then also you will have epsilon hat will
be equal to 0.1 of this if in other cases it is by default 1.0 that means it recovers
to the original k omega equation and there is some modification in the f 1 term also
as we saw here. So, because of this changes that now the p k and epsilon depends on the
gamma it may be possible that the idea that we had originally that it should go back to
the k omega equation near the wall that may not work with the original blending function.
So, that blending function is also modified to ensure that we have k omega very close
to the wall even when the gamma is very small or gamma is equal to 0.
So, let us look at the gamma equation. So, this is the gamma equation here. So, the four
equation that we have. So, k equation omega equation gamma equation which is the intermittency
and then you have i theta equation or i theta t you can call it which is the momentum thickness
based Reynolds number. I suggest that you read about this momentum thickness I mean
you can have if you have a boundary layer like this you can define a thickness in many
different ways one is of course, your boundary layer thickness with delta where you say that
your u is equal to 0.99 or 0.9 of u infinity 9 type of u infinity or you can have something
called theta which is given here momentum thickness which is theta or you can have a
delta star which is basically your the displacement thickness.
So, without going into the details of this, this is a fluid mechanics thing and I expect
that you go and read it. So, the momentum thickness is used as an indicator to find
out where the flow may become transitional. So, the purpose of momentum equation because
as I told you that omega can be become 0 and 1 or in between 0 and 1, but it needs a location
where it should be switched on. So, that location will be given using this momentum thickness
based Reynolds number and of course, you need some experimental data where you say that
if this is the momentum thickness then you the flow becomes basically transitional.
So, you will be solving these equations compare with the experimental data or computational
data like came from d n s and make sure that if my r e theta t found from this equation
is basically let us say r e theta critical or then you basically start turn on this gamma
equal to 1 here, gamma greater than 0 here and then you keep on calculating this r e
theta t and keep on comparing with the experimental data till you basically see that this value
is in the region of the transitional flow or in the turbulent flow. So, this is used
as an indicator and this I mean the nature of these equations are again the same you
have the temporal term you have the term which is advective then you have the production
of this intermittency production of the thickness Reynolds thickness based Reynolds number and
then basically this will be d gamma this is dissipation of gamma and then you will have
this transport which is there in both the terms. These equations are also heavily modeled
and this is d here and you can see that the production of this gamma depends on two parameter
I mean the many parameter, but there are many coefficients model coefficients, but there
are two parameters which for which I want to attention is the f length and f onset.
So, the f onset what it does is basically it switches the intermittency production before
that there is no intermittency, but it has to be determined by something. So, that is
determined by this R e theta t equation and f length basically controls the strength.
So, if the production rate is very high it may become turbulent early. So, going back
to the same picture when it becomes I mean when it becomes transitional and when it becomes
fully turbulent. So, this onset and length this has to be determined and they are done
by basically comparing your R e theta t which is also a transport equation. So, this R e
theta t is also being evaluated everywhere at every time and is compared against experimental
data for which you know that it is a robust parameter which has got distinct value in
the laminar flow and in the turbulent flow and in the transitional flow.
By keep on comparing this R e theta t with those experimental data we switch on this
gamma, gamma will work then gamma will be used in your k and omega equation. So, in
the k equation it will change your production dissipation of this turbulent kinetic energy
and omega equation it will change the blending rate as we saw earlier. So, this is what about
the transitional model again you can see that many model coefficients there in this transitional
model also it is not a within this scope of this course that I cover all the models, but
let me just briefly give you some idea about this some of the other models which are used
in the industry.
So, right now we have been using this linear eddy viscosity model. So, basically your stress
Reynolds stress is proportional linearly to your ballistic radial, but in the literature
you have non-linear eddy viscosity model also which basically takes care of your anisotropy.
What do you mean by anisotropy that you have this u prime and v prime and w prime. So,
what we do in this linear eddy viscosity model we assume that they are of 1 is to 1 is to
1 in all the direction of fluctuations are basically kind of equal that is an isotropy,
but that is not true in the practical flow and it can lead to some wrong predictions.
So, what this non-linear eddy viscosity models does I mean I have written only for the simple
shear flow is that it takes Reynolds stress depending not only on the depending not only
on the ballistic radial linearly, but also in a quadratic way or may be a cubic way and
so on. So, forth right as you would I mean understand
that this makes this model more expensive, but then these models work in certain flows
for example, this cubic epsilon or there is a models like E R S M, E S M etcetera.
They are they come in this class of non-linear eddy viscosity model they are more expensive
than the eddy viscosity model, but not much. Again here also coefficients can be tuned
to produce the correct anisotropy and it can basically produce better prevalent behavior
in certain flows, but there is of course, no theoretical foundation behind this. It
is just that you are trying to make this model more and more complex if the flow becomes
complex that is how it is, but there is no underlying theoretical foundation that why
it should depend quadratically or cubically or on higher powers.
So, this one class of model which is non-linear eddy viscosity model and finally, this model
which is Reynolds stress transport is an expensive model. So, here we say so in the earlier slide
it is not u prime v prime, but you can call it u prime square v prime square and w prime
square basically the energy or the stress in all the direction normal stresses. So,
this Reynolds stress transport model so it basically takes care of this anisotropy much
better because now we are not reducing all these normal stress component or the shear
stress component into one single parameter which is mu t. That is what we have been doing
in the eddy viscosity model, but here we will be solving equations for u prime u prime
bar u prime v prime bar u prime w prime bar v prime v prime bar v prime w prime bar and
then u prime v prime bar we have already done. So, there are six equations w prime w prime
bar. So, there are three normal stresses and then
we have got three shear stresses. So, we will have six equations in your RSTM model
Reynolds stress transport model and you have to basically derive these equations the same
way as we derive the turbulence kinetic energy equation. Separately for each of these there
will be six transport equations and the nature of the equation will be very close to what
we have been seeing will be a temporal term, you will have a term represent the direction
there will be production term, there will be dissipation term and there will be diffusion
term. So, all these terms will be there in these
equations also for each shear stresses and there will be lot of model coefficients and
the one thing you will see that let us say if you want to derive the equation for u prime
u prime bar you this is a double correlation you get a triple correlation as an unknown
it may be u prime u prime v prime or may be u prime v prime v prime bar something like
this. So, that you will figure out as you try to derive these equations for each of
these stresses. So, we wanted to close this equation, but the closure problem remains.
Now, the closure problem has shifted to the triple correlation that is what this model
is done, but the idea is that as you go to higher and higher correlation may be the effect
of modeling these terms may be lesser than the modeling it right here itself.
Secondly, the anisotropy also because you are doing this stress term separately. So,
we are basically taking care of these terms separately which is typically in a flow is
like if the flow is in x direction u prime square will be 1.0 v prime square will be
0.4 and w prime square bar will be 0.6. So, this is how the anisotropy is. So, this stress
term takes care of that the production adhesion term as you told that like the turbulent kinetic
energy you want to derive those terms you will get these terms accurately, but the diffusion
terms are usually modeled as in the k epsilon equation.
It has got 6 equation, but this is less stable and then the EVM equation and the reason for
that is that in the EVM equation we get an extra viscosity which is mu t. So, this mu
t basically makes this scheme more stable because your effective viscosity is mu plus
mu t that advantage we do not have in this particular equation. So, this is less stable
in the viscosity model, but this has got in terms of like physical picture it is looks
more accurate although there is modeling at the higher level, but it is definitely more
expensive and in practical situations it does not offer much advantage to like the
most recent EVM based models like the, but in some flows it may be a useful model.
So, this is how the cost is. So, of course, the DNS is the most expensive approach after
that there is LES and now these three that you see 1, 2, 3 these are basically your WS
model these are the RANS approaches. So, the most expensive is the Reynolds transfer model
because you are solving 6 equations and then you have the non-linear viscosity model because
your mu t is proportional not only to the del u by del t, but also del u by del t square
and so on and so forth. And the least expensive is basically mu t proportional to del u by
del t which is the linear adiabatic viscosity model you want to call it this is a linear
adiabatic viscosity model which we have in this case you have basically k epsilon, k
omega, k SST, k omega and also the transition models in this class you have this cubic k
epsilon and ERSM this is RSTM model. It has also got some versions, but I am not
covering those things and then you have LES which will be covering towards the end of
this course and DNS is anyway solving the Navier Stokes equation as it is without any
modeling, but the requirement of the grid size is very, very high. So, high that it
is not even affordable with the kind of computers super computers that we have for the practical
problem. So, this ends the discussion about this Reynolds
average Navier Stokes equations or RANS equations turbulence modeling and this I took some time
to discuss about this RANS because this is the model that we will be using the most unless
you are in academia the chances of using DNS is not very high. LES is now becoming a tool
for industries for some problems, but still it is not mainstream you will not be using
LES for your regular design tool or let us say you have gone to the parametric study
you will not be using LES. So, in most cases you will be using your RANS model despite
all the advantages disadvantages that we talked about what limitations of this model.
So, that will stop here we will next discuss about the boundary conditions I has basically
delayed the discussion of boundary conditions because we also need to discuss about the
boundary conditions for the turbulent parameters. So, that is why we will next take the boundary
conditions and we will also see the how to set up the boundary conditions for the turbulence
parameters like k and I mean like turbulence intensity or length scale and so on. So, we
will meet in the next class. Thank you very much.
