Computational Fluid Dynamics

Professor. Suman Chakraborty

Department of Mechanical Engineering
Indian Institute of Technology, Kharagpur
Lecture No. 01
Introduction to CDF

Let me welcome you all to this introductory course on CFD or Computational Fluid
Dynamics. In this particular course, we will emphasize on the fundamental principles that
govern the implementation of CFD in practical applications. Now, the first question that
should come is what is CFD, why are we doing CFD and when should we do CFD? We will
try to answer these one by one.

So, what is CFD? CFD has a full form of Computational Fluid Dynamics, but its scope is not
limited to only fluid dynamics. So, when we say computational fluid dynamics, we
essentially mean computational transport phenomena, which involve computational fluid
dynamics, heat transfer, mass transfer or any process which involves transport phenomena
with it. We will come into that in more detail subsequently.

Now, next we will try to see that “when CFD”; that is, what are the important and
challenging applications that we have for CFD these days. So, let us look into some of the
interesting applications of CFD.

(Refer Slide Time: 1:30)

The first application which has been the most common one is the aerospace application. Let
us consider an aircraft as an example. So, if you consider an aircraft, it has interior and an
exterior. So, when we consider the design of an aircraft, you have to design the interior, i.e.,
the working of the air conditioning and ventilation system. And when you have an exterior,
you are basically considering an airfoil section and there is a flow past an airfoil section
which you are intending to model. Also, you have the combustion chamber within the aircraft
and you need to know about the transport phenomenon within the combustion chamber. One
can have CFD for spacecraft applications also.

(Refer Slide Time: 2:58)

Next, let us consider the automobile applications. So, if you look into automobile
applications, what you see is that again the basic philosophy is very, very similar; you want to
see the fluid flow behavior inside the car as an example. The whole idea is to make a good
design so that the passenger feels comfortable inside the car. However, the other important
consideration is the external flow, i.e. how fluid flow takes place across and outside the car
because that determines the drag force on the car and that is where most of the clever design
goes into. But one should not undermine the importance of design that goes for designing the
ventilation system inside the car. Then, there is the combustion chamber, and combustion
chamber design is also very, very important from the transport phenomena viewpoint.
(Refer Slide Time: 3:59)

Automobiles and aircraft applications are very traditional applications of CFD and that is
why we have begun with those applications. But there are many applications which are not so
traditional, but which are very importantly used in modern day applications like the
biomedical applications. You can have basically interesting flow patterns in bronchial tubes.
And these pictures/videos show that how the flow patterns are like in the human heart which
is being modeled as sort of a pump in CFD applications. Another interesting case where fluid
flow analysis is very, very important is the flow through arteries. Apart from these, there are
examples where it is not the fluid flow, but heat transfer is important. For example, if one has
a tumor and somebody is trying to have a laser treatment to destroy the tumor. So, how the
temperature is distributed across the tumor on the laser treatment must be visualized. So, for a
medical practitioner, this is very, very important because medical practitioner can get an a
priori idea regarding the possible extent of destruction of the cells with the laser treatment.
So, accordingly, the parameters of the laser treatment may be designed so that healthy cells
will not be damaged. Of course, there may be some damage, but it will not be damaged
beyond a critical extent. And only the tumor cells which are designed to be destroyed, will be
destroyed by the laser treatment.

(Refer Slide Time: 5:53)

CFD has lots of applications in chemical industries. A nice example is where you have
mixing of two different chemical streams and the interface emits some bubbles. It is a very
complicated interfacial phenomenon and CFD sort of tries to capture these type of
phenomena by involved computational modeling. You can also have injection of streams,
separation, mixing; these are all important chemical engineering applications and CFD has a
big role to play in many of these applications.

(Refer Slide Time: 6:36)

While chemical engineering application is a sort of a traditional application, you can also
have not so traditional application just like the biomedical one. Another example is
electronics. These days, with the advent of new generation electronic devices, the sizes of
electronic devices are getting more and more reduced, but if you are still having the same
power rating, then what happens? Then basically what you have to do is you have to dissipate
a much greater amount of heat per unit volume than what you had to dissipate for larger
devices. That means, you must have more efficient electronics cooling strategies. So, no
matter whether it is your laptop or desktop or any other electronic device, you require
efficient cooling strategies. If you are cooling it by different means, for example, if you have
a fan cooling or like cooling through different other technologies (we will not go into such
technologies, but there are interesting technologies these days) one has to assess the fluid
flow and heat transfer around the flow passages in electronic devices. It may not always be a
forced convection type of cooling; you can also have cooling because of phase change in a
loop system which is called as a heat pipe. You can have, for example, a system where you
have a hot spot in the device and there is a fluid which gets evaporated by taking heat and
then if the fluid gets transported to a different place which is colder, it gets condensed. So,
you have a cyclic evaporation and condensation process in this way, the entire energy is sort
of handled in the form of latent heat. So, there is no sensible change in temperature and there
is sort of a cyclic phenomenon that goes on occurring.

(Refer Slide Time: 8:34)

There are lots of other energy related applications of CFD, like the emission characteristics
from a coal fired boiler, flow in fluidized bed systems and separators, wind turbines and so
on. So, you can have very interesting energy related applications, and fluid flow and heat
transfer and mass transfer in many of these applications turns out to be very, very critical and
CFD plays a big role towards understanding that.

(Refer Slide Time: 9:15)

Now, when you have fluids, many times we forget that fluids interact with their boundaries
and that gives rise to fluid structure interactions. So, when we have fluid structure
interactions, it is important to model the CFD with a structural model. So, CFD is basically a
fluid dynamics model, but you should also have a model for solid mechanics or structural
mechanics and there must be an interaction between these models, which can give rise to
interesting fluid structure interactions.

Let us consider an example. So, let there be a balloon, which sort of interacts with a surface
and tends to get deformed as it interacts with the surface. So, there is a model of the surface
because it has its own elasticity and stiffness. And the fluid filled balloon has its own
modeling; the balloon has a membrane, so, it has also its own flexibility. So, all in all,
apparently, it might appear to be very, very simple problem, but it is actually a very complex
problem in fluid mechanics.

Let us suppose there are dolphins which are sort of interacting with a pool of fluid. So,
dolphins are approximately modelled. CFD is all about modeling; it is not that somebody is
representing the real dolphin, but representing it may be like a flexible membrane. So, you
have a flexible membrane that represents or that models the dolphin artificially and that sort
of interacts with the fluid and creates a deformation in the fluid itself as well as there is a
deformation in the body of the dolphins. These types of interesting problems can also be
addressed through CFD. In general fluid structure interaction problems talk about some
structures, which have some vibration or motion, and they interact with an incipient fluid
which is there in the surroundings, and how they interact eventually with each other. So, the
fluid interacts with the structure, the structure interacts with the fluid, it is not a one way
coupling, but it is a two way coupling that is very, very important. Otherwise, one can have a
separate model for fluid and a separate model for solid, but whatever the disturbance that the
fluid gives to the structure, the structure has a deformation on the basis of that, that
deformation is fed back to the fluid and that modifies the velocity field within the fluid. So, it
is a two-way coupling and a dynamic coupling. And the great challenge is to actually take
into account all the intricacies of these dynamics, while modeling fluid structure interactions.

(Refer Slide Time: 12:01)

CFD also has lots of applications in larger scales like marine applications. So, you can have a
flow past ships and boats and even submarines, and though these are of much larger length
scales than many of the examples that we have seen till now, these are still very, very
important. Huge component of the marine design nowadays relies on the use of CFD.

(Refer Slide Time: 12:32)

CFD has lots of applications in materials processing or manufacturing. One may try to
simulate the grain growth in a material, the nature of which gives rise to the material
properties. So, a CFD simulation may be used to determine how the grain grows. Of course,
this always need not be a deterministic simulation; one can also use stochastic simulations
like Monte Carlo simulations and so on.

So, CFD need not always be a deterministic solution, it need not always give a deterministic
set of approach, but there are stochastic approaches also in CFD. One such approach is used
where one is interested about the transport phenomena during a butt-welding process. If one
is interested to create a joint by bringing two materials from the two sides, CFD may be used
to study the transport phenomena involved in the formation of that butt welded joint. The
CFD results may be used to visualize the growth of dendritic crystal, and its solidification
from a melt, and you can see interesting patterns as the material is processed. You can also
have interesting applications of CFD in mold filling. So, if you are filling a mold with a
molten metal, it is very important to know a priori how well the mold will be filled up,
because of a particular filling process. If the filling process is not appropriate, it can give rise
to casting defects and such kinds of defects are not wanted and one cannot do lots of trials
and experiments, because that will be hugely expensive. So, if CFD gives a sort of a
guideline that these are the processing parameters that are required to have a very efficient
mold filling process, then that can optimize the process and make sure that defects are
minimized.

In processing of materials, we often require that the product should be as homogeneous as

possible. Now, materials usually have impurities and when there is a fluid flow, these
impurities are getting transported from one point to another with the flow and many of the
materials actually have to undergo this process, because many of the processing of materials
involve the transport of molten material. So, if you have a solidification process that means,
prior to the solidification the material was there in a molten state, just like a casting process.
Now, if you have impurities, the impurities will be transported by convection from one point
to another point with the molten material. Once it solidifies, whatever was the impurity at that
particular location it will be frozen at that location and it will give rise to a distribution of
species across the material. And if this distribution of the species is undesirable, then that will
give rise to undesirable mechanical properties. So, to know how to impose a control over the
distribution of species that are being added to a particular system, one has to assess the
velocity and the temperature field and the species concentration field within the domain
which constitutes the domain of the materials processing application.
(Refer Slide Time: 16:20)

Now, if you consider applications in small scales, there are interesting applications of CFD in
microfluidics or microscale fluid flows where one of the primary purposes is to mix two
different fluids. Now, microfluidics deals with the studies on fluid flows at small scales in
channels of micron or submicron dimensions. In such small length scales, usually it is not
possible to take the advantage of turbulence because Reynolds number is very, very small.
So, if you want to mix two streams, there are different ways in which you can do that: one is
by clever designing of the channel or you can also try to have a good mixing by using a
pulsating flow and so, with a pulsatile flow input, the objective is to create good mixing
between different fluid layers.

In microfluidics, one also has important studies considering or concerning droplet dynamics.
As an example, in a T type of junction droplets are produced because of shear, and then
droplets are elongated and sort of compressed and so on. So, they are deformed and it is very
important to see that how these droplets evolve, because many times you put reactants with
these droplets and allow reactions to get performed in small scale (micro reactors). Therefore,
it is very important to understand that how the droplet dynamics evolves in small scales.

Now, in micro fluidics also fluid structure interactions are important. Let us consider an
example where you have an oscillating flap located within a micro channel or a small
channel. So, what this flap tries to do is it tries to oscillate with a particular frequency and
amplitude and there is some material which wants to undergo a reaction at the wall. So, this
flap is trying to force the material on the wall at certain locations where the reactants are kept.
And it is not only at one single spot where the reactants are kept. There are different locations
where these are kept. So, one needs to optimize the movement of these flaps, so that all the
reactant locations get the sample for reaction. And these flaps essentially act like a
combination of a mixer and a pump. Since mixing is not so easy in small scales, this is one of
the strategies by which one can have a better mixing by using mechanical flaps in small
scales.

Of course, there are many, many other applications. But our objective is not to go deep into
any application, but to appreciate that different fields have different challenging applications
of CFD.

(Refer Slide Time: 19:35)

Now, if you come to some field of sports, there is a huge application of CFD in sports. For
example, if there is there is somebody who is driving a racing car, then how the fluid flows
past it is very important because again in such applications, it gives a very important idea
how to minimize the drag. And to do that one needs to design the car and to design the car
one needs to test it. And of course, fabricating it in the real size all the time and doing the
testing may get over expensive. So, before the actual fabrication, during the design stage,
CFD is very commonly used. Even when human beings are running, how the fluid flow takes
place around human beings dictates what is the total drag force and helps determine the
running posture that interacts very nicely with the fluid flow in a way that one can have a
technique by which one can minimize the drag force under certain conditions.

The dynamics of sports balls is also very, very important like golf ball, cricket ball, etc. How
fluid flow takes place around these is important because it is the extent of turbulence or the
nature of the flow whether it is laminar or turbulent, that usually dictates the movement of the
ball, such that you can have the swing of the ball and you can have other sorts of other types
of lateral movement. So, in sports CFD is very extensively used for scientific purposes.

(Refer Slide Time: 21:26)

In other applications of mechanical engineering, CFD is used in one of the traditional

applications viz. turbo machines. In turbo machines, CFD is very, very commonly used for
design of blades or flow passages. For example, in a Pelton wheel turbine, you have a wheel
and then on the periphery of that you have buckets and fluid jet is interacting with the bucket
and this entire process can be simulated by CFD. So, you can have CFD applications in turbo
machines where you can sort of assess in great details of how fluid flow takes place, for
example, across the impeller of a pump. Now, this is very, very important because it is
virtually impossible to analyze such flows analytically, because the blade passages are very,
very complicated. It is not as simple as like flow over a flat plate or flow in a parallel plate
channel or flow in a circular pipe. Because of such complicated geometry across which the
fluid flow is taking place, it is almost inevitable that one has to go through computational
roots for solving such problems.

So, we have seen some applications of CFD. And we have now developed an idea that in
these applications, why CFD is important, but the bigger issue or the bigger question is, is
CFD inevitable? That means is it that we require CFD in all cases?

(Refer Slide Time: 23:10)

That is one interesting question that we would like to answer because many times just
because this is a CFD course, we will be having a bias towards CFD. So, we will say that yes,
CFD is necessary. But to be honest, let us try to answer this question that “Is CFD necessary
in all cases?” So, the second question that we wanted to answer is “when CFD” and “why
CFD”.

So, we have understood by this time that CFD is about numerical simulation of some
governing equations for transport phenomena. That is enough in a summarized form, but, to
solve the transport phenomena problem, you could also have an analytical approach and you
could also have an experimental approach. So, these are alternative approaches to CFD. So,
question is that when should we prefer other approaches or when should we prefer CFD? Let
us try to look into that.

First of all let us consider experimental investigation. We should keep in mind that there is no
substitute for experimentation because seeing is believing. So, there are many idealizations
that we can sort of do while we are doing a numerical modeling or analytical modeling, but
when we do experiment and see something is happening, that is the reality. Having said that,
there are certain important limitations or restrictions that can go with modeling. What are
experimental investigation or experimental modeling? So, when we say modeling we can be
both experimental modeling and numerical modeling.

Experimental modeling in full scale is expensive and often impossible. For example, suppose
you are interested to study fluid flow in a steel making vessel. In a steel plant, in the steel
making vessel there is a very interesting flow pattern because you may have jets of say
oxygen being input to the fluid. You can have the jets coming from both the top and from the
bottom and so, there can be multiphase flows in the system. So, it can be a very complicated
flow and eventually the final property of the steel is a strong function of how the flow is
taking place in the steel making vessel. But, imagine that you have a molten steel at such a
high temperature plus the medium is also not transparent. So, you cannot have a full-scale
experimental model with the real fluid under those applications. So, even if it could be
possible in this case, in the real case, it is not possible to study the fluid flow exactly in the
system. There are certain examples where it is possible, but it is very, very expensive.

So, in general full-scale modeling may be expensive, often impossible. Suppose you are
designing an aircraft. So, before designing or before coming up with the final dimensions if
you want to make a lot of trial by experimenting with the real sized aircraft, then that is very,
very expensive and one would not, or one should not go for such a design, because it will
involve a lot of expense. Additionally, just like any tool has its own errors, experiment also
has important errors, which are related to measurements. So, it is not that seeing is believing.
Although one may argue that seeing is indeed believing, measuring is not always believing.
So, whatever you are measuring may have lots of error associated with that. Now, many
times because of limitations with the full-scale modeling, one goes for a scaled model, the
scaled model may be smaller than the actual model, it may also be larger than the actual
model depending on whatever is convenient. The scaled model should satisfy certain
important objectives, that once you have a scaled model, you should not scale it up or scale it
down in such a way that the flow physics has changed. For example, say you are interested to
study flow in a micro capillary. Now, you say that, well, it is difficult for me to do
experiments in microcapillary. So, I will do same experiment in a pipe, which is sort of
geometrically similar version of the capillary but hugely expanded in dimensions. So,
geometrically you can maintain the similarity, but the physics of the problem might change,
because in the microcapillary surface tension force is very, very important, but when you go
to a large pipe, surface tension force may not be that important. So, when you are considering
a scaled model, one has to keep in mind that at least the physics does not change, but many
times one has to make a compromise. And once that compromise is made, it becomes a
simplified version of the reality, but not the reality itself. Sometimes, it is very difficult to
extrapolate results from the model to what will happen for a prototype because of incomplete
similarities. Since many times exact similarity, i.e., physical similarity, geometrical
similarity, kinematic similarity and dynamic similarity may not be properly maintained, it
may be difficult to extrapolate the results, plus measurement errors are there even in the
scaled model.

Experimental investigation has many advantages as well. Here we have only itemized the
critical concerns. It is not that we have considered just the positive points, but we have
considered it very, very critically where the constraints are also taken into account.

If you come to theoretical calculations, theoretical calculations for the same problem will
have two sorts of approaches one is an analytical solution. Analytical solution, if it exists,
there is nothing like it because it is exact and it gives exact point to point variation without
any approximation, but the question is, does the analytical solution exist for the case that you
are interested for? Well, for most of the practical cases, analytical solution does not exist. So,
it exists only for a few cases. Those few cases are very, very important. We will see that CFD
cannot stand alone by itself, it requires a good help from experimental as well as analytical
solutions because you are required to benchmark your solutions. Say you are solving a very
complex problem of a fluid flow and you have written a code or you have developed a code,
how do you validate that? You can validate that against some problems, which are either
standardized in their solution by analytical means, or there are benchmark experiments with
which you can compare. So, you cannot really discuss CFD devoid of the considerations of
analytical solutions and experimental solution. That is why we are bringing these into
perspective. Analytical solutions exist only for a few cases and in cases when they exist,
sometimes they become very, very complex. So, I have seen students who by looking into
analytical solutions for a problem, easily get driven away from the problem and they never
come back to solving the problem. And the reason is that sometimes the expressions are so
big and so cumbersome, and wherever inside those expressions, some physics is hidden, it is
very difficult for a beginner to appreciate that in many cases. But if analytical solution exists,
there lies its own elegance and its own beauty and it needs to be appreciated and it can be
used as a benchmark for more complicated numerical solutions.

Numerical solutions exist for almost every problem. Of course, whenever you gain
something, you lose something also, when you say that numerical solution exists for almost
every problem, what we mean by numerical solution is that you are getting solutions at
discrete points in the domain, and not continuously at all locations. So, the continuous nature
of the solution gets sacrificed, but at the expense of that, you at least get a solution for even
complex problems.
(Refer Slide Time: 31:55)

So, if you now compare modeling with experimentation, what are the advantages of
modeling? Advantages of modeling are as follows; they are cheaper than experimentation in
many cases. For example, if you have established a particular code or developed a particular
code or you have got a code which can be used for a particular problem, then you can
simulate it for different cases and for that, expenses do not mount up. It is the same code that
you are running just with different parameters. Of course, you are incurring computational
time and if it is a highly sensitive code to computational efforts, that is, you are for example,
doing a direct numerical simulation for a turbulent flow problem or solving a problem in a
highly parallelized computing environment, then, it involves a lot of cost. It does not mean
that computation is free, but only thing is that many times it may still be less expensive than
the corresponding trials which one could have done with experimental investigation.

With modeling, you get more complete information. Why? Say with the experiment, you
really do not have a scope of getting all the details of all the variables that you are looking
for. Because, for example, you are interested for some flow measurement, temperature
measurement, velocity measurement. Of course, you can do that within a complete flow field
by numerical simulation, you can do it also by experimental simulation, but if you are
interested to do that, it will involve a huge expense. Not only that, in many cases it is ruled
out as we said, for example, if you are interested for a flow visualization in a molten steel,
say if you want to get a velocity profile within that, or velocity vectors within that,
numerically you can generate, experimentally it is not possible. One can handle any degree of
complexity, i.e., it can be a very complex problem and one can handle it, but there are
important buts.

What are those buts? Like when you are handling a very complex problem many times what
is not known is a correct description of the boundary condition. So, you see, a numerical
solution is as good as the input that is going in. So, if you are solving a problem numerically
and you are getting a solution, whether you need to believe the solution or not, it is up to the
level of the input. So, what do we mean by the level of the input? The input mainly deals with
the input of thermo physical properties. For example, input of properties as a function of
temperature. So, if you are giving the properties as a function of temperature and composition
very accurately then it is a very good thing, but from where will you get that? So, you must
have reliable sources of property data and correct boundary conditions. Many times we give
boundary conditions in numerical simulation, which are not correct representatives of
physical reality like for example, we say isothermal boundary conditions. In reality whoever
has done experiment will understand that it is one of the toughest things to maintain an
isothermal boundary even in a very simple case. So, whatever we are sort of planning to be
isothermal in the real experimental case may not be maintained as isothermal. So, the
understanding is that there can be lots and lots of uncertainties with regard to the boundary
conditions and one has to consider these uncertainties very seriously.

Sometimes the problem may be so nonlinear and then, with a slight change in the initial
condition, with a slight change in boundary condition, the solution may change, the solution
may bifurcate to an entirely different one from what you are interested to get. And that is
where it can be very, very critical. So, we should not basically be so emphatic and say it is
disadvantage of modeling, but may be one can say that a limitation of modeling is that it
deals with a mathematical description and not with the reality. So, when you see a boundary
condition, it is a mathematical description which may or may not correspond to the reality.
So, one has to keep in mind that it is your responsibility as an analyzer to make sure that it
deals with a reality as much as possible to whatever extent possible.

But mathematical description can be inadequate. So, how it can be inadequate? Suppose, we
have considered the property data and the boundary condition. What about the governing
equations themselves? So, are the governing equations representing the correct physics that
you are looking for? It may be or it may not be. It may be that the correct physics that you are
trying to sort of capture cannot be captured by the equations that you are employing. So, then
no matter whether you are using correct properties or correct boundary conditions, your
solutions will not be the desired ones, simply because you are not representing the correct
physics.

Sometimes modeling, particularly for nonlinear problems, can be tricky because nonlinear
problems may have multiple solutions. And by modeling you might get each of these
solutions by having different initial guesses. For example, if you are using iterative schemes
for solution. So, the question is what would be your correct solution? Because may be when
you are doing experiments, you will get only one solution during one experiment. So, how do
you relate to your experiments with the modeling, that is a very important concern, so to say.

To summarize this discussion on modeling versus experimentation we can say that, there is
actually no substitute for experimentation, but, there are limitations associated with
experimentation- may be because of expenses or sometimes it is because of the inability to
capture many things during experiments like for example, capture the velocity field in a non-
transparent material flow as an example. So, because of these limitations, one cannot do
experiments on a trial basis. So, if one is interested to go for a design, then the best route for
solving a problem may be like this: first, make a model of the problem and then try to
approach the problem mathematically, then come up with some solutions, either numerical or
analytical depending on whatever is more convenient. Of course, if analytical solution exists
that will remain to be more convenient, but if it does not exist, then one has to go for
numerical solutions. And from those solutions, reduce your number of trials, so, that you
come up with only a few sets of data with which you do final set of limited and very
controlled experiments. So, study of CFD is not to make us go away from experiments, but
actually to help us in doing better experiments, so, that we can get a domain where we have
very restricted and very highly refined sets of data towards a good experiment design. And
then on the basis of that design, if we do experiments that will give us a good feel of what is
happening in reality. So, then with that experiment, we can compare some of the CFD output.
However, we cannot compare some of the CFD output with the experiment, because such
details you may not get from experiment, but if other details are verified, then of course, we
can say that the CFD model is representing the reality in a very nice way.

And then you can do many sets of parametric studies and come up with nice conclusions
which may not be possible with just experimentation. So many times in old days,
experimentation was like a hit and miss type of trial, just like an Alchemist, where one is
mixing A with B to see that whether C is becoming gold. But those approaches towards
experiments were very ancient. Nowadays approaches towards experimentation is very, very
sophisticated and in fact CFD plays a big role towards good experimentation.