Basics of Finite Element Analysis – Part II

Prof. Nachiketa Tiwari

Lecture -03
Steps followed in FEA

Welcome to Basic of Finite Element Analysis course. This is the new course part 2, is
the third day of the first week of this course. And in last lecture we reviewed the basic
notion as to how the basic philosophy and the line finite element analysis, where
complex problem is analysed, by breaking the domain into smaller elements and then we
find solution for each specific element. And in the last class we did not do this exercises
in context to solving ordinary or partial differential equations, but whatever we learnt in
the last class or reviewed, we found that the accuracy of this kind of an approach
increases if we increase the number of elements, if we increase the nature or order of
polynomial functions which are used as interpolation functions, and also once we
increase the computational accuracy.

So, what we will do in today’s module is actually again review what are the different
steps when we do finite element analysis. So most are time, when we talk about finite
element analysis, that is in context of solving partial or ordinary differential equations so
it is when we have to solve these equations and if we have hard time finding a closed
form solution, then we use finite element method to find approximate or numerical
solutions for the same equation. And do today is first get an over view as to how we go
in this direction in on a step by step basis, and then maybe in the next couple of classes
we will actually do some examples.
(Refer Slide Time: 02:26)

So, the first step when, so the steps in which are used as in the finite element analysis are
the first step is that we discretize the domain. So we break the overall region in which we
want to find the solution of the domain and we discretize the domain into sub domains.
And these sub domains are called elements, so this is the first step. Second and we will
learn the details of it maybe later today, is that once we broken the overall domain into
sun domains, then over each element or each domain we assume how unknown variable
changes over the element.

For example, suppose I have a rod, and am pulling this rod like this and I am also
applying some traction on the rod in some other region. And what I am interested in
finding out is the displacement in the rod at all these individual points. So my unknown
is u which is the displacement in this direction in the direction of pulling of the rod. And
I am interested in finding the displacement at all internal points in the rod. So the first
step is the overall length of this rod is let say l, and I will break this overall domain into
smaller sections. So in this case I broken it into 4 different elements. Element number 1,
element number 2, element number 3 and element number 4. This is the first thing then
the second step is that we assume how u varies over each element. We do not know the
value of u at each point, but we assume how u varies over each element.

So, I can assume that u is constant, over element one it could be also constant. Over
element 2 it could be constant, over element 3 on so on and so forth. Or I can assume that
u varies linearly over element 1, 2, 3 and 4. Or I can assume that u varies quadratically
over individual elements. I can also assume that u varies linearly an element 1 and 2 and
it varies quadratically on element 3 and 4. That is also fine, but we assume that over each
element how this unknown which is u it varies. And when we make that assumption
basically we are assuming that if it is a linear thing then we assume that u is equal to a
plus b x, over the length of an element right were a and b are unknown. Or if it is a
quadratic variation then we assume that u is equal to a plus, b x plus c x square over each
element and a b c are again unknown and they may change from one element to other
element.

So, this is the second thing. The third step is that we find out the error, so in the
differential equation once we make this assumption. So and we multiply this error by
weight function and integrated over the domain. So and I will explain that later first let
me just write it. So we multiply error by weight function and integrated over element or
sub domain. What does that mean so suppose this problem is governed by an equation
minus d by d x, a x, d u over d x is equal to q, both that is problem. Or to make it a little
more general I will say plus c u is equal to q.

So, this is the governing differential equation and if we can solve this equation in context
of this particular problem when we will have the solution, but suppose the solution is not
easy or it getting a close from solution is not easy, then what do we do first step is that
we broke this entire domain into 4 elements. Second is we assume that in first element let
us say assume, to that first element u is equal to a plus b x, here also a 1 plus b 1 x in this
case u is equal to a 2 plus b 2 x and so on and so forth.

Then the third step is we plug this, in this equation. And once we plug it if the solution is
really true exact the left hand and the right hand side will be same, but because this is an
approximate solution the left hand in the right side will not be same which means that
there will be a difference in the values of lefts side and right side and that difference is
called error. That difference is called the error, so that is the error we calculate and we
multiply it by some weight function.

So that error so mathematically we will say that that error then we multiply by weight
function and we will discuss what is the weight function later, so we multiply that by
weight function and then we multiply and then we integrate it over the domain of the
element. So this let say this here x 1, this is x 2 so I am integrating it over x 1 to x 2, and
this I equated to 0. And the reason why we do it in all that stuff we have explained it in
basics of finite element part one. So if you have doubt I will definitely encourage you to
go back and review what are we talk about at the last class.

So, this is the third step and this step gives us and we will see that later it gives us good
large number of equations at each for each element. So we will have this equation for
element one will have us different equation of same nature for element 2 and so on so
forth. And for each element we get several equations, we get several equations. So we
get element level equations. So this third step it gives us element level equations.

(Refer Slide Time: 11:01)

Then the fourth step is assemble element level, we assemble element level equations.
And then these then what we get is assembly level equations.

Then the next step is that I apply boundary conditions. And we will again do all this
through an example so we apply boundary conditions. So for instance in this problem if
the boundary condition would be that at x equal to 0, there is the force on the left side
this is the one boundary condition and at x is equal to l there is the force tensile force
which is being applied at the and which is getting inserted in the right side, or in the
positive x side those are the 2 boundary conditions. We can apply those boundary
conditions and once we have applied the boundary condition then we solve for
unknowns, in the equations in assembly level equations and this is how we get the
solution, there is how we get the solution for the overall thing.

(Refer Slide Time: 13:24)

So, what we will do is, we will again the next 10 15 minutes, will illustrate this
methodology, will go through an example. So suppose I have a hollow cylinder and so
this is all material and suppose since this hollow cylinder heat is flowing. How could
heat be flowing? Suppose, this is the metallic hollow cylinder and I am sending current
though this, and is a long cylinder so this current times i square time the resistance of the
cylinder will generate heat at every point in the cylinder so heat is getting generative.

And what I am doing is for using some method at, so this is outside radius, this is inside
radius r i and at some other location, which on the radius is r. Which varies r, this is an
access symmetric problem. And what I am also saying is that on the outside radius t at r
0 is equal to t nought. So suppose I am immersing the cylinder in say ice. So the outside
temperature is maintain that 0 degree, centigrade or I am immersing it in boiling water
outside temperature is maintained at 100 degrees centigrade, this is one thing. And the
inside radius what I do is I put some thick layer of an insulator some perfect insulator.
Which means so what are the boundary conditions in this case? Boundary conditions that
t at r o is equal to t nought and no heat transfer at r i.

Now, the governing differential equation for this kind of a system is something like this.
Minus 1 over r, d by d r, k r, d t over d r is equal to q r. Here k represents the thermal
conductivity of the material, t is the temperature at any point in the medium, r represents
radius and because it accesses symmetric problem for same value of r temperature will
be same and q is the amount of heat which is getting generated, so rate of energetic
radiant in the medium that is q.

So to find r, so our aim what is the aim, aim is to find r for different values of r between
so, this is the entire range of r, r is starting from r i to r o, at different values of r, what is
the value of temperature so we want to find temperature as a function of r, this is our
goal and we can solve this problem by solving this equation and applying boundary
conditions. Now what we will do is we will generate a finite element formulation for this
method. So this is an access symmetric problem, so what I will do is, I will look at only
radial line. So the first step is what we discuss was that we break the entire domain into
small elements.

So, step one, discretisation of domain. So my domain is this r i r o. Let us say this is my
centre origin; so the overall length of the domain is r o minus r i and I can break it up
into smaller elements. So this first element is h 1 long, second element is h 2 long, third
element is h 3, long h 4 long. And if I want I can break it up into whatever number of
elements I want, so this is the first step. The second step, is step 2, assume interpolation
functions or approximation functions for the variable for each element.

So, what is the variable in this unknown variable in this problem, t. So for each element
we assume how t is going to vary. So we will say that t, it is the function of r and for the
eth element I will call it t e, for the eth element I will call it actually subscript t. So
temperature is varying as a function of r for eth element, to the value of e will be 1 for
first element, it will be 2 for second element, it will be 3 for third element; so t e is equal
to now am assuming a function interpolation function. So I am saying this is equal to t j,
psi j, r e and j is equal to 1 to n. Now here psi j e is a known function. It is a known
function. An example of this function could be 1. It is a constant right psi j is the known
function, another example of this function could be a plus b x. It is a known function, but
here a and b they are unknown constants.

So, this a known function and t j e these are unknown constants. So suppose the overall
known function is a plus b x. Then in this case it is some of 2 is smaller functions. First
function is a and the second function is b x, right. In a is unknown, that is same as t j and
a times 1, so 1 is the known function which is psi j. The second function is b x, there b is
unknown and so that is t j, that represents t j and psi j is x which is the known function.
So that is there, so in that in this case it would not be a plus b expect maybe a plus b r,
because the variable is r.

So, I am saying that temperature over element e is a sum of products of t js, and psi js,
psi js are known function, t j is an unknown constant. And I will have, I can some several
of these, so n is what is the number of interpolation functions. So let us called this
governing equation as equation a and let us call this equation as equation b. So the third
step is step 3, step 3, is plug b into a. So plug this equation b into a, and what will be get
because this is an approximation of the thing, so we will get an error.

Plug b into a to get error and then the other part of the step 3 is multiply error I call it e,
by some weight function. Now how do we choose weight function, we will see that later
and we have also explain that in f a 1, so we multiplied by weight function and here we
will just take weight function as, another functions psi I and again it is a function of r,
weight function r. And the value of I could vary from one to n. So multiply error by
weight functions psi i and integrate the product over element and equate it to 0.

So, what are we doing? We are what essentially in this step we are doing is, we are
calculating the error, assigning a weight to the error because we are multiplying it by a
weight function, and integrating this overall multiple of weight function over an error
over the domain and equated to 0. So in this sense the weighted integral of the error is 0.
So that is why this is the weighted integral that is dual approach, so weighted that is dual
approach, in finite element analysis. So error is not 0 at ever point, but over the entire
domain if I integrate the error, then it is in a 0 in a weighted sense, into 0 in a weighted
sense, the size of the element essentially as I keep my element size smaller and smaller.
(Refer Slide Time: 26:10)

Essentially when I do this step, I go towards a point that error is approaching 0 at every
point, because over the entire domain it is between 0. And as I in string the size of the
element that element is increasingly becoming more and more like a point.

So, the error is between 0 at every point over, but again in an approximate sense. So
when I do this what is the equation I get, so this is what I get, minus 1 by r, d by d r, k r d
minus; so this is the error, right this is the error. And what am I doing? I am multiplying
this error; see here what have I done. I have substituted u by t j psi j, some of t j psi j, so
this is the error. And I am multiplying this error by your weight functions psi i for the eth
element and this is also a function of r. And this entire thing is getting integrated over the
equating it to 0, and what are the limits of integration, so this this equation is for eth
element.

So, for eth element the first, starting of eth element is, h 1 e and the ending of eth
element is h 2 e. These are the coordinates, so for each the eth element this is there, now
we said that i is equal to 1 to n and n is the number of interpolation functions. So if I
have so just look at this thing here, n is the number of interpolation functions right, how
many unknowns we have integer in in this functions. So I all s I js are known or n known
there known, t js are n known, how many t js are there n. So total number of unknowns is
total number of unknowns is n. Because t js are n in number so total number of n known
is n, now in this equation i can assume how many values i is equal to 1 to n. So first I
multiply the error with psi 1, I get one equation. Then I multiply this error with psi 2 an
integrated I get second equation.

So, total number of equations I will get at element level will be n because i is equal to 1
to n and total number of t js which are unknown a n. So the total number of t js is n and
total number of equations which we generate at element level is n. So this is not one
equation, this represents n equations, because i equal 1 to n. This is not one equation this
represents n equations, because i equal 1 to n. Not because j is going from 1 to n, because
j is being summed. So this term does not change, it is the psi I it changes as I changes. So
this represents n equations and there are n variables at element level. So at least at the
element level we have the number of equation and number of unknowns is same.

So, this is step three, so what we will do is we have done these 3 steps, what we will do
is next is we the way we have explained is that step 4, is we assemble all these equations.
And in the step following that is, that once we have done the assembly. Then we apply
the boundary conditions, and once we have applied the boundary condition, then we
solve for these unknowns which are t js and that is how we get the solution.

So this completes our lecture for today and in the next lecture will develop this
discussion even further. And we will also review this whole notion of week formulation
in that context. So with this we will close for today and will continue this discussion
tomorrow.

Thank you.