Finite Element Analysis

Prof. Dr.B.N.Rao
Department of Civil Engineering
Indian Institute of Technology, Madras

Lecture - 33

Let us continue with what we left behind in the last lecture. So, we will see some more
examples on applications based on general two-dimensional boundary value problems.
And as a part of that, in todays lecture, we are going to see two-dimensional ideal,
irrotational, incompressible fluid flows around an object. So, basically if you recall, what
we did in the last class, we looked at torsion problem.

Basically, once we get the governing differential equation and associated boundary
conditions, we compare the differential equations, and the boundary conditions with the
corresponding equations ofgeneral two-dimensional boundary value problems.And once
we identify the corresponding coefficients, we can easily write the finite element
equations. So, that is what basically we did. So, similar manner let us solve some
problems in todays class using orthe finite element equations; that we already developed
for general two-dimensional boundary value problems.

(Refer Slide Time: 01:50)

So, ideal fluid flows around that irregular object. The problem of two-dimensional ideal,
irrotational, incompressible fluid flow around an object is solved using either stream line
formulation or potential formulation. So, basically there are two approaches through
which, we can solve this problem, and both formulations are presented here that means,
we are going to look at the both the formulations.So, first let us start with stream line
formulation.

(Refer Slide Time: 02:24)

The basic parameters in the stream line formulations are shown in figure below. The
field variable is here for stream line formulation. The field variable is the stream
function, which is denoted with psiand it is related to the fluid velocities in x and y
directions. And also, the solution domain is extended far enough from the obstruction.
So, that there is no effect of obstruction on the flow characteristics.
(Refer Slide Time: 03:19)

So, the fluid velocities in x and y directions are related to stream function psi, which is
function of x and y through these equations. That is, velocity in the x direction is partial
derivative of psi with respect to y. Fluid velocity in y direction is equal to minus of
partial derivative of stream function with respect to x. The governing differential
equation is as follows. So, basically we need to solve this differential equation subjected
to someboundary conditions imposed on psi.

(Refer Slide Time: 04:21)

The fluid rate between any pair of stream lines is given by this equation Q i j is equal to
psi i minus psi j. That is, i and j are any pair of stream lines and also the solution domain
is extended far enough from the obstruction. So, there is no effect of obstruction on the
flow characteristics.And for the figure given here, we can easily identify the boundary
conditions.

(Refer Slide Time: 05:06)

The boundary conditions are as shown. That is, at the far left end, V x is specified; V y is
equal to 0 and at the far right end, V y is equal to 0 again. And stream function at the top
is denoted with psi subscript t and stream function at the bottom is denoted with psi
subscript b. The constant values of stream lines at the top and bottom are determined
from the flow rate. So, what is the difference between psi t and psi b? That is given by
the flow rate.
(Refer Slide Time: 06:20)

Now, let us consider an example. Consider flow around cylinder of diameter 4.5
centimeters. Away from the cylinder, the fluid velocity in the horizontal direction is V x
is equal to 5 centimeter per second. So, this is the problem. V x is specified; V y is equal
to 0. The diameter of a cylinder is given as 4.5 centimeters and total width of flow is
given as 12 centimeters. So, the constant psi values; that is, psi t and psi b as I just
mentioned, it can be determined from the flow rate per unit thickness. Here thickness is
assumed to be 1 unit.

(Refer Slide Time: 07:32)

So, flow rate or unit thickness is given by 12 times 1 times V x; that is, 5 centimeter per
second and it comes out to be 60 centimeter cube per second. So, the difference that is
psi t minus psi b is 60 centimeter cube per second.

(Refer Slide Time: 08:09)

So, the flow rate between any two stream lines is the different between them. So, and
because of symmetry, the two stream lines are equal and actually because of symmetry,
we can model only a quarter of this domain.In that case, psi t minus or psi t is going to be
minus ofpsi b is equal to 30, which is half of 60.

(Refer Slide Time: 08:59)

A distance of 8.25 in front of cylinder is assumed to be enough for constant velocity
condition; because we need to select solution domain far enough from the
obstruction.So, that no effect of obstruction on the flow characteristics.Using symmetry,
only a quarter of solution domain is to be modeled and the quarter model is shown are
along with all the boundary conditions. Since V y is equal to 0, partial derivative of
stream function with respect to x is equal to 0 on the extreme left edge.

And because of symmetry, partial derivative of stream function with respect to y, which
is indirectly related to the V y; fluid velocity along y directionis also 0 along the line of
symmetry on the right hand side. And again, since the model is symmetry; because of
symmetry even psi is equal to 0 on the bottom edge of this quarter model. Similarly, the
edge which is coinciding with the cylinder psi is equal to 0 and psi is equal to 30 and that
is coming from flow rate, which we just calculated. So, this is the quarter model of the
solution domain. So, we need to solve the second order differential equation; that we
have seen earlier in terms of psi subjected to these boundary conditions.

(Refer Slide Time: 11:40)

So, the boundary condition on the left vertical face is partial derivative of psi with
respect to x is equal to 0; because of assumption of uniform flow in the x direction there.
The boundary condition on the right vertical face is partial derivative of stream function
with respect to s is equal to 0, because of symmetry.And the boundary condition along
the top face of the model is psi is equal to 30. Lower side of the model corresponds to the
middle of entire solution domain. So, stream function is equal to 0 there. So, these are
the boundary conditions.

(Refer Slide Time: 12:32)

So, we need to solve this second order differential equations subjected to these boundary
conditions. So, that is the problem statement. Now, everything is given to us. Now, our
job is to compare this governing differential equation with general two dimensional
boundary value problems; the differential equation, that we have taken for general two
dimensional boundary value problems. And the corresponding boundary conditions,we
need to compare with the boundary conditions of this particular problem.Identify the
coefficients and we can actually write the finite element equations.
(Refer Slide Time: 13:13)

So, now let us look once again general two dimensional boundary value problems. This
is a statement, that we started out with and this needs to be satisfied over the domain A
subjected to any of these boundary conditions; essential boundary condition and natural
boundary condition.

(Refer Slide Time: 13:39)

And using Galerkin criteria, what we did is, after substituting finite element
approximations, a complete equations for general two dimensional boundary value
problems reduces to this; where, each of these k x, k y, k p,k alpha, r beta,and r q are
defined here.

(Refer Slide Time: 14:09)

And for linear three node triangle element, since the shear functions are very simple for
linear triangle, it is possible to carry out all integrations in closed form. Assuming k x, k
y, P, Q are constant over element to get element equations in explicit form.So,for three
node triangle element, k x turns out to be this; after multiplication of two vectors, we get
this.

(Refer Slide Time: 14:50)

Similarly, k y and k p and here, since the components of P are not going to be constant,
we need to perform some numerical integration.Terms in k p matrix are not constant.
Fortunately, simple formula is available for integrating shape functions over triangle.

(Refer Slide Time: 15:14)

This is the formula. So, applying this formula, all theparameters in that formula are
defined; alpha, beta, gamma. N 1, N 2, N 3 are the three shape functions of the linear
triangular element. A is the area of triangle and symbol exclamation is for factorial and
using integration formula, the terms in k p matrix can be evaluated like this.

(Refer Slide Time: 15:50)

And then, we required to evaluate r q. Once we identify q, what is q? We can evaluate r
q.For evaluating r q also, we can use this formulaand one component of that r q is
simplified and the details are shown. And carrying out similar kind of integrations, we
can get all components of r q. So,therefore r q is given by one third Q times A and3 by 1
column vector consisting of all 1s.

(Refer Slide Time: 16:26)

And now comingto the line integrals,once we identify once we carefully comparethe
given boundary conditions with the corresponding boundary conditions of general two
dimensional boundary value problems,we can identify what arethe corresponding
parameters?And we can assemblethe boundary integrals, k alpha.For illustration, k alpha
is given here; it isk alpha evaluated alongside 1 2 details are given here. The shape
functions of nodes 1 and 2 can be obtained using one dimensional Lagrange interpolation
formula.And simplification of this leads to the similar integrations can be carried out
along the other two edges.
(Refer Slide Time: 17:18)

Similarly, r beta alongside 1 2:So, all these procedures, which we have seen earlier needs
to be repeated.

(Refer Slide Time: 17:28)

Once we identify the corresponding parameters for thisideal fluid flow around an
irregular object.So, for side 2-3,3-1 integrals can be evaluated in similar manner and only
difference is going to bethe placement of zeros in the matrices consisting of shape
functions.
(Refer Slide Time: 18:03)

And for side 23 and 31,k alpha and r beta are looks like this.So, once we identify the
corresponding coefficients, we can get the element equations. And once we get element
equations, we can get the global equation system based on nodal connectivity. And after
imposing the essential boundary conditions, we can solve for the nodal values.

And once we solve for the nodal values here, for this particular problem of fluid flow
around an irregular object, around an irregular object the nodal parameters are going to
be the stream function values. Once we obtain stream function values, we can findpartial
derivatives of stream functions; stream function with respect to x and with respect to y
and which are related to the velocity components or fluidvelocities along x and y
directions.
(Refer Slide Time: 19:05)

So, stream function calculations:Once the nodal values psi are known, stream function
can be interpolated using element shape functions like this.

(Refer Slide Time: 19:24)

And also, sometimes we require this. What is the integral psi dx dy over the element or
entire domain? Here, the formula is written for one particular element. So, we know psi
is equal to N 1psi 1 plus N 2 psi 2 plus N 3 psi 3,which can be written in matrix and
vector form.As it is shown on the right hand side of the equation and substituting, since
psi 1,psi 2, and psi 3, the stream function nodal corresponding to the nodes.Nodal stream
function values are constant.They can be taken out of the integral.

And once we simplify this integralwith N 1, N 2, and N 3 integrated over dx dy or the

entire area of the element for linear triangular element, these integrals turns out to bearea
of triangle divided by 3. So, substituting this integral, that is psi integrated over the entire
element can be approximated as area of the triangle element divided by 3 or area of
triangle element multiplied by average value of a stream function at all the nodes. That
is, A over 3 times are in brackets psi 1 plus psi 2 plus 3 as shown in the right hand side of
the last equation.

(Refer Slide Time: 21:07)

Now, we require to solve this differential equations subjected to these boundary

conditions. So, proceeding in the way that is explained, we can get the element
equations.And once the element equations are obtained, we can assemble the global
equation system; impose the essential boundary conditions. Here, essential boundary
conditions are psi is equal to 30 on the top phase; psi is equal to 0 on the bottom face. So,
these are the essential boundary conditions. After imposing these essential boundary
conditions, we canor we need to impose these essential boundary conditions; the
nodeswhich are along these boundaries.
Once we impose these essential boundary conditions, we can solve for the nodal values
and do all kinds of post processing.So, now comparing with general two dimensional
boundary value problems, see that, it is mentioned as torsion problem;it should be ideal
fluid flow problem.So, for this particular problem,when we compare withgeneral two
dimensional boundary value problem, it turns out that k x is equal to k y is equal to 1;P is
equal to 0 and Q is equal to 0.So, once again there is a mistake. It should be fluid flow
problem, instead of torsion problem.

(Refer Slide Time: 22:33)

So, now let us seethe solution, how it looks or how it converges, when we take different
meshes? Here coarse mesh solution, by coarse mesh I mean the mesh; the quarter model
consists of 24 triangular elements. So, this is the quarter model and it consists of 20
nodes.The corresponding elements are here 24 triangle elements. So, using this
discretization, we can solve this problem by imposing essential boundary conditions at
nodes 9, 30, 70, 1, 2, 3, 4, 8, 12, 16, and 20. By imposing essential boundary conditions
over these nodesand we can solve for the stream function value at other nodes. And here
at each node, there is only one degree of freedom, which is going to be the stream
function.
(Refer Slide Time: 23:45)

So, the nodal solution, the details are given here at all nodes 1 to 10 in this table and next
table shows for the rest of the nodes.

(Refer Slide Time: 24:00)

(No audio from 23:59 to 24:10)

(Refer Slide Time: 24:12)

And solution derivatives and integral over each element, they are shown in this table.
Partial derivative of stream function with respect to x, partial derivative of stream
function with respect to y and integral of stream function over the element domain 1 to 9
elements are shown here.

(Refer Slide Time: 24:39)

Andrests of the elements are shown. The other slide 10 to 19, details are given here.

(No audio from 24:48 to 25:01)

(Refer Slide Time: 25:03)

20 to 24 details are given here. We can also plot, what are called stream lines.
Corresponding to the lines along which, stream function value is constant.

(Refer Slide Time: 25:30)

So, stream lines are shown in this figure. Also we can plot, derivatives of stream
functions with respect to x and y.
(Refer Slide Time: 25:50)

Derivative of stream function, x derivative of stream function, y derivative of stream

function; because x derivative of stream function is related to velocity in the y
directionand y derivative of stream function is related to velocity in the x direction. And
they are shown here, contours of derivatives of stream function with respect to x and y.
So, whatever finite element solution that is shown is using 24 triangular elements. The
problem can be repeated by using finer mesh.

(Refer Slide Time: 26:36)

Solution domain is divided into 96 triangular elements. So, the discretization is shown
hereand everything remains same, except that size of the problem increases; because we
require to solve for more number of nodal values.

(Refer Slide Time: 26:57)

So, let us look at, how the solution looks? Stream lines, psi is equal to constant; those
lines are these.

(Refer Slide Time: 27:12)

Andthe derivatives of stream function, psi with respect to x and with respect to y.So,we
solve this problem using 24 elements and 96 elements.

(Refer Slide Time: 27:32)

Let us see, how the solution matches? Both coarse and fine mesh gives a reasonable
stream line that is expected. Because the solution itself may converge very fast; whereas,
derivatives of solution require finer mesh. Fine mesh stream lines are smoother, which
indicates a better solution. Derivative contours for coarse mesh show reasonable general
trend. Fine mesh solution results show better velocity contours. And solution near left
boundary is uniform; thus confirming our choice of constant velocity boundary at a
distance of 8.25 centimeters away from cylinder.
(Refer Slide Time: 28:37)

So, now let us look at, solving this ideal fluid flow around an irregular object through the
different formulation; that is potential formulation. The velocity components are related
to a potential phi by these relations. The fluid velocity along x direction, y direction are
related to the potential phi, where this equations. The governing differential equation,
derived from Eulers momentum equation and conservation of mass is this one. So,
basically we require to solve this second order differential equation; solve for phi
subjected to some boundary conditions.

(Refer Slide Time: 29:32)

Appropriate boundary conditions as given here are as shown in the figure. That is,
essential boundary condition at the top and bottom are partial derivative of potential with
respect to y is equal to 0. And V x is specified on the left vertical side and V y is equal to
0.So, we can solve or we can redo the problem that we already did, using potential
function for potential formulation.

(Refer Slide Time: 30:15)

Andwe let us look at the problem statement. For the problem that we already looked at,
consider fluid flow around cylinder of 4.5 centimeters and V x is 5 centimeters per
second. Again, only a quarter of solution domain needs to modeled, because of
symmetry.
(Refer Slide Time: 30:48)

And the boundary condition on the left vertical phase ispartial derivative of potential
with respect x is equal to phi; becauseV x, fluid velocity in the x direction at that location
is already given in the problem statement. This is because of the assumption of uniform
flow in the x direction. Boundary condition on the right vertical phase is potential is
equal to constant; because of symmetry. For the numerical solutions a constant a value of
50 is assigned to thepotential phi along this phase.

This is arbitrarily assigned. Arbitrarily, since phi is equal to potential is equal to

constantand that constant value is taken as 50 just arbitrarily. The boundary condition
along the top and bottom phase of this model are partial derivative of potential with
respect to y is equal to 0, which are indicated in the figure and all the details of geometry
are shown in the figure. So, now with this information, we can solve the second order
differential equation with respect to the potential; for this particular problem, over this
domain subjected to these boundary conditions.

By comparing this with the problem statement of general two dimensional boundary
value problem and then, we can actually get the finite element equations. And then,
asusual the procedure is same; assembling and applying the essential boundary
conditions solving for the nodal values. Here the nodal values, turns out to the potential
value. And again this problem, at each node you have only one degree of freedom, which
is going to be the potential phi value.
(Refer Slide Time: 33:15)

So, these are the boundary conditions that we already looked at.

(No audio from 33:18 to 33:32)

(Refer Slide Time: 33:35)

So, now let us look at, again for different mesh solution, how it looks coarse mesh
solution? Solution domain is divided into same number ofelements as we did for the
previous formulation; that is 24 triangle elements. The discretization is shown here.
(Refer Slide Time: 33:59)

And the potential lines that is, phi is equal to constant are shown in this figure. Equi-
potential lines; potential line with phi is equal to constant are called equi-potential lines.

(Refer Slide Time: 34:22)

Andx derivative of potential and y derivative of potential, which are related to the
velocity in the x and y directions respectively are shown here.The contours and these
solutions correspond to 24 triangle elements, as we did earlier; we can actually go for
finer mesh.
(Refer Slide Time: 34:51)

Solution domain is divided into 96 triangle elements.The mesh is shown here.

(Refer Slide Time: 34:59)

And the potentiallines or equipotential lines obtained, using 96 triangle elements. The
solution is shown here.
(Refer Slide Time: 35:19)

And contours of x derivative of potential, y derivative of potential; they are shown here.

(Refer Slide Time: 35:40)

And let us see, let us make some observations.The potential lines are perpendicular to the
stream lines.The stream lines are the lines that we obtained using the previous
formulation. Stream function formulation and the potential linesare the lines that we just
obtained using potential. Potential based solutions are similar to the stream line based
solutions. Fine mesh results show better velocity contours than the coarse mesh.
(No audio from 36:15 to 36:25)

So, this is one application of general two dimensional boundary value problems.So, now
let us look at other application, that is two dimensional steady state heat flow.

As usual only thing is, we require to know, what is the governing differential equation
and what are the boundary conditions? So that, we can make a comparison with general
two dimensional boundary value problems; identify the coefficients and get the finite
element equations.So, consider the problem of finding temperature distribution through
long chimney like structures. A slice of t units thick of such body is shown in the figure.
Assuming no temperature gradient through thickness, the problem can be modeled as
two dimensional problems. So, we are assuming, no gradient of temperature through the
thickness.

(No audio from 37:38 to 37:47)

The governing differential equations can easily be derived using conservation of energy.
And to do that, consider a differential element as shown here. So, the following notation
is used.Temperature at any point is denoted with dx y; k x, k y are the coefficients of
thermal conductivity along x and y directions. The corresponding units are given and q x
is the heat flux in the x direction and q y is the heatflux in the y direction and Q is heat
generated for unit volume.So, using this notation and taking a differential element,we
can derive the governing differential equation.So, let us consider this differential
element, in which all quantities are indicated.
(Refer Slide Time: 39:13)

Heat flux along x direction,y direction and the heat generated andthe dimensions of this
differential elements are dx, dy and the coordinate system is also defined x y in the
figure. So, using the convention that heat entering the control volume is positive and
leaving and that is leaving is negative.

(Refer Slide Time: 39:43)

The conservation of energy dictates that, writing the equilibrium equation, we get this.
And cancelling terms, which are higher order or cancelling terms, which are same in
values, we get this. And substituting the heat flux, that is q x, q y in terms of thermal
conductivity coefficients along x and y directions.

(Refer Slide Time: 40:27)

We know that, q x and q y are related to thermal conductivity coefficients through these
equations. So, substituting q x, q y we get the following governing differential equations
for steady state heat flow in atwo dimensional domain.Andclearly, this is a second order
differential equation and it is similar towhat we already looked for general two
dimensional boundary value problem.So, the governing differential equation is a second
order partial differential equation.
(Refer Slide Time: 41:14)

And the boundary conditions are as follows. Temperature specified on part of boundary,
insulated boundary, no heat flow across the boundary. In that case q n, n is nothing but
outward normal to the surface; n x, n y are the components of the out word normal to the
surface. Since this is insulated, so q n is equal to 0. And we can also consider the other
case, heat loss due to convection on the boundary.In that case, q n is going to be a
nonzero value, which is going to be the functionof convection coefficient and
temperature of the ambience.

(Refer Slide Time: 42:09)

Heat loss due to convection on a boundary, in that case q n is given by this. So, we have
two kinds of boundary conditions. Irrespective the boundary condition, what we need to
do is, we need to compare the corresponding equation with the equation of general two
dimensional boundary value problemand identify, what are corresponding coefficients,
or corresponding parameters?

(Refer Slide Time: 42:54)

And here one more thing, care should be taken in assigning signs to q n term. And the
sign convention adopted isheat flowing into the boundary is positive and heat flowing
out of the boundary is negative.
(Refer Slide Time: 43:13)

This is the governing differential equations. So, this problem is clearly of general two
dimensional boundary value problem form considered in the previous chapter or
previous lectures. And if we compare, only thing that we observe is, when compare with
the general two dimensional boundary value problem, P term is missing. So, P does not
exist here or P is equal to 0; so KP is going to 0. Convection boundary condition, if it is
insulated boundary, alpha and beta are going to be 0.If it is convection boundary
condition, then it is handledby taking alpha is equal to minus h; beta is equal to h times T
infinity. T infinity is nothing but, temperature of the surround.So,with this
understanding, we can solve a problem.
(Refer Slide Time: 44:17)

So, now let us take an example. Determine temperature distribution through a square
medium with which is a 6 inches in diameter. And here the problem is given in f p s units
butdoes not matter as long as we stick with the consistent units. And pipe carrying hot
liquid at 500 Fahrenheit is placed at the center as shown in the figure below. Outside
temperature is at 100 Fahrenheit and k x, k y thermal conductivity coefficients along x
and y directions are given here. And you can see here, this problem we can take
symmetry into account and because of symmetry, only one eighth of the solution domain
needs to be modeled.

(No audio from 45:11 to 45:19)

So, nowlet us look at, one eighth model.

(Refer Slide Time: 45:25)

Note because of symmetry only, one eighth of the solution domain needs to be modeled.
So, this is the one eighth model. And once we take symmetry into account, the associated
boundary conditions along the line of symmetry are shown there,that is partial derivative
of T. T here is temperature; partial derivative of T with respect to normal is equal to 0
along their lines of symmetry.

So, as usual we need to solve the second order differential equation, partial differential
equations subjected to these boundary conditions,that isT.T value at the inner diameteris
500 degrees Fahrenheit or T value at the center is 500 degrees Fahrenheitor T value at
the inner diameter and T value at the outer diameter is 100 degrees Fahrenheit. Subjected
to those conditions and subjected to the natural boundary conditions along the line of
symmetry, we need to solve this problem.
(Refer Slide Time: 46:58)

And as usual we can proceed with finite element discretization and here eight element
solution is shown. So, finite element mesh with eight triangle elements is shown here.

(No audio from 47:11 to 47:21)

(Refer Slide Time: 47:22)

Element numbers are shown in bold. A 500 degree Fahrenheit temperature is specified at
nodes 3, 6, 9. A 100 degree Fahrenheit temperature is specified at nodes 1, 4, 7. And zero
flux is specified along 7-8-9 and 1-2-3 edges. So, let us see. So, 500 degrees Fahrenheit
temperature is specified on the nodes 3, 6, 9; because those are the nodes, which are
along the inner diameter of the model. And temperature of 100 degrees Fahrenheit is
specified along 1, 4, 7; because that is the edge, which is in contact with outside of the
model. And zero flux is specified along 7, 8, 9; which is one of the line of symmetry and
1, 2, 3 which is the other line of symmetry and the nodal so with that boundary
conditions.

(Refer Slide Time: 48:45)

Nodal temperature values are computed, the model is solved. After applying the essential
boundary conditions and the nodal temperature contours are shown here.
(Refer Slide Time: 49:09)

And numerical values at all the nodes are given in this table. Nodes 1 to 9.

(Refer Slide Time: 49:21)

And elements 1 to 8, heat flux along x direction, y direction; they are given here.
(Refer Slide Time: 49:42)

And now, the model is as we did for fluid flow case, a finite discretization is taken. So,
32 element solution is shown here. For that, the finite element mesh looks like this.

(Refer Slide Time: 50:06)

And the nodal temperature contoursand still model is refined further. Using 72 elements;
it should be 72; instead of that, it is typed as 32; and 72 element mesh is shown here.
And the nodal temperature contours, for 72 element solution is given in this figure. If we
compare this solution with 8 elements, 32 elements and 72 elements, we can clearly see
convergence of solution both for temperature and heat flux values.