Linear and Nonlinear FEM Airil Et Al

LINEAR AND
NONLINEAR FINITE
ELEMENT ANALYSIS
FOR SOLIDS AND with
FLUIDS (UNEDITED MATLAB
1ST DRAFT)
AIRIL YASREEN MOHD YASSIN, MOHD
AL AKHBAR MOHD NOOR,ERWAN
HAFIZI KASIMAN, MOHD ZHAFRI
JAMIL
LINEAR AND NONLINEAR FINITE ELEMENT ANALYSIS
FOR SOLIDS AND FLUIDS (unedited 1st draft)
2013
CONTENTS
PART 1-LINEAR FORMULATION (A: 1D)
CHAPTER 1: BASIC CONCEPT OF NUMERICAL TECHNIQUES 2-16
CHAPTER 2: GALERKIN FORMULATION: BAR ELEMENT 17-45
CHAPTER 3: GALERKIN FORMULATION: BEAM ELEMENT 46-69
CHAPTER 4: PLANE STRUCTURES: TRUSS AND FRAME 70-101
CHAPTER 5: EIGENPROBLEMS: FREE VIBRATION AND BUCKLING 102-139
CHAPTER 6: DYNAMICS: FORCED VIBRATION 140-164
PART 1-LINEAR FORMULATION (B: 1D)
CHAPTER 7: SCALAR ELEMENT (HEAT TRANSFER) 165-188
CHAPTER 8: PLANE ELASTICITY (PLANE ELASTICITY) 189-227
PART 2-NONLINEAR FORMULATION (A: 1D)
CHAPTER 10: NONLINEAR BAR ELEMENT 228-312
PART 2-NONLINEAR FORMULATION (B: 2D)
CHAPTER 12: FLUID DYNAMICS (NAVIER-STOKE EQUATIONS) 313-363
© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

ERWAN HAFIZI KASIMAN, MEng, MOHD ZHAFRI JAMIIL, MEng Page 1
2013
CHAPTER 1
BASIC CONCEPT OF NUMERICAL TECHNIQUES
1.1 INTRODUCTION: WHAT IS FINITE ELEMENT METHOD (FEM)?
Finite Element Method or FEM can be both “everything” and “nothing”. At one end, FEM is
everything when it allows engineers to get information (i.e. stresses, displacement, forces)
of complex physical phenomenon for design purposes. At the other hand, FEM is nothing
because the information obtained is actually nothing but solution of partial differential
equation (PDE) or ordinary differential equations (ODE). In this context, FEM is nothing but
another numerical technique to solve PDE or ODE.
Realizing how FEM can be “everything’ is important because it can motivate its study but
realizing at the same time how FEM can be “nothing” guides the proper learning of FEM that
is, any discussion must begin from the first principle (i.e. PDE, ODE) if strong understanding
is desired.
1.2 BASIC CONCEPT OF NUMERICAL TECHNIQUES
Having said how FEM is just another numerical technique, indeed, there are various
numerical techniques available nowadays such as;
i. Finite Element Method (FEM)

ii. Finite Difference Method (FDM)
iii. Boundary Element Method (BEM)
iv. Meshless or Meshfree Method (MESHFREE)
However, despite their variations, all the methods share similar concept that is;
“to covert the continuous nature of PDE (or ODE) into ‘equivalent’ simultaneous algebraic
equations”.
In elaborating the concept, we discuss herein the solution of the simplest forms of ODE that
is of a bar element. By leaving the derivation for later, the ODE of a bar element (as shown
in Figure 1.1) can be given as:

2013
Figure 1.1 Bar element / structure
Domain equation
(1.1)
where u and q are the axial displacement of the bar and the external force acting on the bar,
respectively. and are the Young’s modulus and the cross-sectional area of the bar
respectively which both are constant for a linear problem. L is the length of the bar.
Complementing the domain ODE are the boundary conditions given as follows (which
detailed derivation and discussion are delayed until Chapter 2)
Boundary conditions (equations)
(1.2)
|
| (1.3)
The ODE is considered solved when a solution, is found which satisfies all the
equations above (i.e. Eqns. 1.1-1.3).
In fact, the exact (closed-formed) solution of the problem can already be obtained by direct
integration, thus;
(1.4)
( ) ( )
However, despite the availability of Eq. (1.4), the ODEs of the problem (Eqns. 1.1-1.3) are
still descritized numerically herein so as to demonstrate the basic concept of numerical
techniques. The problem is chosen due to its simplicity allowing for easy tracing of the
discussion.
To convert the ODEs (i.e. Eqns. 1.1-1.3) to its ‘equivalent’ simultaneous algebraic equations,
we start by assuming a guessed solution in the forms of polynomials. In our case we guess

2013
(1.5)
Then, we satisfy Eqn. (1.3) by inserting Eqn. (1.5) into the equation to give;
| (1.6a)
which gives
(1.6b)
Next we satisfy Eqn. (1.2) by inserting Eqn. (1.5) into the equation to obtain
|
(1.7)
Finally, by inserting Eqn. (1.5) into Eqn. (1.1) the following is obtained.
(1.8)
Observing Eqn. (1.8), it must be noted that, whilst each of Eqns. (1.6-1.7) is an act of forcing
the equation to a certain values (i.e 0 and P, respectively) hence the “satisfaction” of the
equations, the insertion of the guess function (Eq. 1.5) into the domain equation (Eqn. 1.1)
is yet a satisfaction of the original equation hence the use of the inequality symbol ( ). As
we are going to see, the forcing of Eqn. (1.8) at several locations within the domain to a null
value is what satisfies the equation and what creates sufficient number of equations.
By grouping Eqns. (1.6 - 1.8) together, we can see that, so far, we have established three
simultaneous equations as follows
(1.9 a)
(1.9 b)
(1.9 c)
However, observing Eqn. (1.9) we can notice that:
- We have five (5) unknown constant, a1, a2, a3, a4, and a5 but with only three (3)
simultaneous equations.
- The last equation that is Eqn. (1.9 c) (previously Eqn. (1.8)) is still not algebraic but
continuous in x. Also, the left hand side of the equation is not equal to the right hand
side because the guessed function is yet the solution of the ODE, as mentioned
previously.

2013
So to get the sufficient number of equations (and to convert Eqn. (1.9 c) into algebraic) we
argue that, since Eqn. (1.9 c or 1.8) is obtained from domain equation, the equation must
hold (must be true) throughout the domain thus we can evaluate Eqn. (1.9 c) everywhere in
the domain as much as we need. In our case, to complement Eqn. (1.9 a) and Eqn. (1.9 b),
we evaluate Eqn. (1.9 c) at three (3) location in the bar says at x=L/3, L/2, 2L/3 to obtain
(1.10 a)
( ) ( )
(1.10 b)
( ) ( )
(1.10 c)
( ) ( )
Eqns. (1.10) are the results of forcing Eqn. (1.8) to a null value at several locations within the
domain as mentioned previously. Also, as can be seen, such an act not only represents the
satisfaction of the original ODE but also convert the ODE into a set of algebraic equations.
Now, by re-grouping Eqns. 1.9 (a – b) and Eqns. (1.10) we then have a sufficient number of
algebraic equations, given as;
(1.11 a)
(1.11 b)
(1.11 c)
( ) ( )
(1.11 d)
( ) ( )
(1.11 e)
( ) ( )
Eqns. (1.11) is thus the ‘equivalent’ simultaneous algebraic equations of the ODE of the
problem which are given originally as Eqn. (1.1 to 1.3). In other words, we can say that;
“Eqns. (1.11) are the ‘equivalent’ algebraic forms of Eqn. (1.1 - 1.3)”.
So this is basically the main concept shared by all numerical techniques such as FEM, FDM,
BEM and MESHFREE. But it is also the character of a numerical technique to treat the

2013
equations in matrix forms as this what suits computer programming. In this context, Eqns.
(1.11) can be arranged in matrix forms as
( ) ( )
( ) ( ) (1.12)
{ } { }
( ) ( ) ]
[
or
[ ]{ } { } (1.13)
where
( ) ( )
[ ]
( ) ( )
( ) ( ) ]
[
{ }
{ }
{ }
{ }

2013
Note: , , and are the alternative notations known as indicial/tensorial notations. We

are going to use this notation system and the matrix forms interchangeably.
By solving Eqn. (1.11) or (1.12), we would then obtain the numerical values of , , ,
and and by inserting these values into the original guessed function, Eqn. (1.5) we thus
obtain the numerical solution to the problem.
EXAMPLE 1.1
Let’s put what we have learned so far into practice. We are going to compare the result
obtained from Eqn. (1.12) with the closed-form solution of Eqn. (1.4) for the particular
problem below (Figure 1.2).
Figure 1.2 Cantilever bar with uniformly distributed load and a single point load
For this problem, the equation in the form of Eqn. (1.12) can be given as
E 1.1
[ ]{ }
{ }
By solving Eqn. (E1.1) using Matlab command “\”, (source code 1.1), the values of ai are thus
as given in Table 1.1

2013
Table 1.1: Values of guessed solution’s constant
Value
And by inserting the values into Eqn. (1.5) we thus obtained the numerical solution of the
problem as
The accuracy of this numerical solution can be assessed by comparing its value with the
exact/closed-form solution which is obtained from Eqn. (1.4) as
( ) ( )
The plot of unumerical and uexact throughout the length, L of the bar and their respective values
at several locations are given in Fig. E1.1 and Table 1.1. Based on the plot and the table, it
can be concluded that the basic concept of the numerical technique able to provide similar
results hence solution as compared to the ones provided the closed-form solution.
However, it must be noted that this is generally not the case. Generally a numerical solution
would not be accurate but converge to the “accurate” solution with the increase in the
order of polynomial of the guessed functions (or degree of freedom and mesh density later
on).
Figure 1.3 Comparison between unumerical and uexact

2013
Table 1.2: Comparison of values between unumerical and uexact
x=1m x=2m x=3m x=4m x=5m

uexact (m)
unumerical (m)
1.3 BASIC CONCEPT OF FEM: WEIGHTED RESIDUAL METHOD
In previous section, we have discussed the basic concept shared by all numerical techniques
that is, to convert the PDE or ODE of a problem into its equivalent algebraic equations.
Herein, we are going to focus our discussion on Finite Element Method (FEM). FEM is a
numerical method that widely used in engineering field of today. The theoretical argument
for FEM can be done in three different approaches which are;
i. Direct Method
ii. Variational Method
iii. Weighted Residual Method (WRM)
In our present discussion, our main interest is on WRM where we are going to leave the
discussion on Variational Method for later. Due to its simplicity, we are not going to discuss
Direct Method.
Actually, the specific form of WRM that is employed in present day FEM is known as
Galerkin WRM formulations. However, before we get into that, we are going to discuss first
the basic idea (or concept) of WRM, as given next.
1.3.1 BASIC CONCEPT OF WRM
The basic concept of WRM can be better argued by observing the evolution of the
discussion from the previous basic concept of numerical technique. We start by re-
observing Eqns. (1.1-1.3). Since we are going to discuss the basic concept of WRM by using
these equations as our starting point, we re-write the equations herein as
(1.14 a)
(1.14 b)
|
| (1.14 c)
The treatment on Eqn. (1.14a) is what differentiates WRM from other numerical techniques.
Whilst in previous discussion, Eqn. (1.14a) is evaluated at three different locations in the
domain (bar) so as to yield the necessary three algebraic equations. In WRM, we have quite
a different argument for this.

2013
In WRM, Eqn. (1.14 a) is written as
(1.15)
Residual error is a function emerges due to the fact that the guessed solution is not the
actual solution of the domain ODE. For the guessed solution of Eqn. (1.5), the residual error
function, R is as given by Eqn. (1.9c), re-given herein for easy viewing as
(1.16)
Plot of Eqn. (1.16) would give us the graphical view of the residual error function which a
typical plot is as shown in Figure 1.4.
Figure 1.4: Typical plot of residual error function, R
Since this error function should not exist, integrating the function and setting it equal to
zero will also mean forcing the area beneath the curve to become zero, as shown in Figure
1.5.

2013
Figure 1.5: Forcing of the area of the residual error function to zero
This action of forcing the area under the residual curve would give us one algebraic equation
in a similar way if we evaluate Eqn. (1.9c) at one location within the domain as previously
discussed. However, our present approach is more effective because the integration
somehow disperses the error in a more smooth and ‘overall’ manner.
But so far, this only gives us one equation, while the fact is, we need as much equations as
necessary. In previous discussion, we evaluate at three locations to provide the three
required algebraic equations, remember?
Herein, to provide the sufficient number of algebraic equations, what we do is to firstly

create three independent residual error functions and to integrate these resulting functions
over the domain before setting them to zero. In accordance, to create the three
independent residual functions, we multiply Eqn. (1.16) separately with independent
function, i.e. 1, x, x2 to give
(1.17a)
(1.17b)
(1.17c)
where R1(x), R2(x), and R3(x) are the three ‘new’ residual error functions which all can be
typically plotted as in Figure 1.6

2013
Figure 1.6: Plots of residual error functions, R1(x), R2(x), and R3(x)
Now by integrating separately each of the residual functions and setting it to zero, the
action would force the area under each curve to ‘disappear’ thus results in the three
required algebraic equations, given as
(1.18a)
∫ ∫
(1.18b)
∫ ∫
(1.18c)
∫ ∫
By conducting the integration we obtain:
(1.19a)
(1.19b)
(1.19c)

2013
Now, as we did previously, by grouping Eqns. (1.9a-1.9b) which actually the satisfaction of
Eqns (1.14b-c) and Eqns. (1.19a-b) together, we get:
(1.20a)
(1.20b)
(1.20c)
(1.20d)
(1.20e)
In matrix forms, Eqns. (1.20) can be given as
(1.21)
[ ]{ }
{ }
Equation (1.21) is the discretised equation of the bar ODE obtained from the basic WRM.
EXAMPLE 1.2
Let’s solve the previous problem of EXAMPLE 1.1 this time by employing WRM. We are
going to compare the result obtained from this approach with those obtained using basic
numerical technique given by Eqn. (E 1.1) and the closed-form solution of Eqn. (1.4).
For the given problem problem, the WRM discretised equation in the form of Eqn. (1.21)
can be given as
E 1.2
[ ]{ } { }

2013
By solving Eqn. (E1.2) using Matlab command “\”, (source code 1.2), the values of ai are thus
as given in Table 1.3
Table 1.3: Values of guessed solution’s constant
Value
And by inserting the values into Eqn. (1.5) we thus obtained the numerical solution of the
problem as
The accuracy of WRM can be assessed by comparing its value with those previously
obtained in EXAMPLE 1.1. The plot of WRM results together with unumerical and uexact are
given in Figure 1.7 their numerical values at several locations along the bar are given in
Table 1.4.
Figure 1.7 Comparison between uWRM , unumerical and uexact

2013
Table 1.4: Comparison of values between uWRM , unumerical and uexact
x=1m x=2m x=3m x=4m x=5m

uWRM 1.25 x 10-7 5 x 10-7 1.125 x 10-6 2 x 10-6 3.125 x 10-6
uexact (m) 1.25 x 10-7 5 x 10-7 1.125 x 10-6 2 x 10-6 3.125 x 10-6
unumerical (m) 1.25 x 10-7 5 x 10-7 1.125 x 10-6 2 x 10-6 3.125 x 10-6
Based on the plot and the table, it can be concluded that WRM gives similar results to
previous approaches hence validates its concepts. However, as mentioned previously, such
accurate result obtained herein is generally not the case. Generally, like any other numerical
techniques, WRM converges to the “accurate” solution with the increase in the order of
polynomial of the guessed functions (or degree of freedom and mesh density later on).
1.4 CLOSING
So far we have discussed about the basic concept shared by all numerical techniques. Then
we discussed about the basic concept of weighted residual method (WRM). However, it has
also been mentioned that the specific form of WRM employed in present day FEM is
Galerkin WRM. Since this is the actual discussion on FEM, we are going to discuss this in the
next chapter.
1.5 SOURCE CODE

%--------------------------------------------------------------------------
% Clear previous data and graph
%--------------------------------------------------------------------------
clc;clear all;
cla
%--------------------------------------------------------------------------
% Declare symbol
%--------------------------------------------------------------------------
syms x
%--------------------------------------------------------------------------
% Input
%--------------------------------------------------------------------------
E=200e6; % Young's modulus (kN/m)
A=0.04; % Area (m^2)
L=5; % Length (m)
P=0; % Point Load (kN)
q=2; % Uniform load (kN/m)
%--------------------------------------------------------------------------
% Calculate matrix k and f for Ritz Method and wrm
%--------------------------------------------------------------------------
kz=[1 0 0 0 0;
0 E*A 2*E*A*L 3*E*A*L^2 4*E*A*L^3;
0 0 2*E*A 6*E*A*(L/3) 12*E*A*(L/3)^2;
2013
0 0 2*E*A 6*E*A*(L/2) 12*E*A*(L/2)^2;

0 0 2*E*A 6*E*A*(2*L/3) 12*E*A*(2*L/3)^2]; % Ritz
kwrm=[1 0 0 0 0;
0 E*A 2*E*A*L 3*E*A*L^2 4*E*A*L^3;
0 0 2*E*A*L 3*E*A*L^2 4*E*A*L^3;
0 0 E*A*L^2 2*E*A*L^3 3*E*A*L^4;
0 0 (2/3)*E*A*L^3 (3/2)*E*A*L^4 (12/5)*E*A*L^5]; % WRM
fz=[ 0; P ;-q;-q;-q]; % Ritz

fwrm=[ 0; P ;-q*L;-(1/2)*q*L^2; -(1/3)*q*L^3]; % WRM
%--------------------------------------------------------------------------
% solve for unknown
%--------------------------------------------------------------------------
dz=kz\fz; % Ritz
dwrm=kwrm\fwrm; % WRM
% Ritz coeff.
az1=dz(1,1);
az2=dz(2,1);
az3=dz(3,1);
az4=dz(4,1);
az5=dz(5,1);
% WRM coeff.
aw1=dwrm(1,1);
aw2=dwrm(2,1);
aw3=dwrm(3,1);
aw4=dwrm(4,1);
aw5=dwrm(5,1);
%--------------------------------------------------------------------------
% Plot Result
%--------------------------------------------------------------------------
% Exact
Uexact=@(x)-q / E / A * x ^ 2 / 0.2e1 + (P + q * L) / E / A * x;
% Ritz
Unum=@(x) (az1+(az2*x)+(az3*x^2)+(az4*x^3)+(az5*x^4));
% WRM
Uwrm=@(x) (aw1+(aw2*x)+(aw3*x^2)+(aw4*x^3)+(aw5*x^4));
fplot(Uwrm,[0 L],'-*r')
hold on
fplot(Unum,[0 L],'-.sk')
hold on
fplot(Uexact,[0 L],'-ob')
hold on
xlabel('x');
ylabel('u (x) ');
legend('UWRM ','Unumeric','Uexact')

LINEAR AND NONLINEAR FINITE ELEMENT 2013
ANALYSIS FOR SKELETON STRUCTURES
CHAPTER 2
GALERKIN FORMULATION: BAR ELEMENT
2.1 INTRODUCTION
In the previous section, we have discussed the basic concept of numerical technique and the
basic concept of WRM. In this chapter we are going to discuss on the specific form of WRM
that is employed in present day FEM that is Galerkin WRM. However, since it has also been
mentioned that FEM is nothing but a numerical solution to PDE (or ODE), it is important in
any FEM endeavour, for the analyst to be familiar with the relevant PDE (and ODE) to
include its derivation and its closed-form solution (if available). This way, next time the
analyst is required to embark on a project or study, he or she is then trained to look at the
problem from the first principle and identify all the relevant aspects and concerns before he
or she employs FEM in getting the solution to the problem.
2.2 ORDINARY DIFFERENTIAL EQUATION (ODE) OF BAR PROBLEM
As mentioned, it is vital for an analyst to work into the problem from first principle and in
many cases; this would mean from the derivation of the relevant PDE (or ODE). Since in the
previous chapter, we have been introduced to the ODE of bar problem, herein we are going
to show the derivation of the ODE. Figure 2.1 shows a bar element and its differential
element.
Figure 2.1: Bar structure and its differential element
© AIRIL YASREEN YASSIN, PhD AND MOHD AL-AKHBAR NOOR, MEng Page 17
It can be argued that, although we have an axial force, F at the left side of the differential
element, due to the ‘disturbance’ along the differential length, dx (due to external load, for
example) the magnitude of the axial force at the right side of the differential element must
change. However we don’t know the exact change of this force, else we would not have this
problem in the first place would we? But by assuming the change is continuous, we can say
that the force at the right-side of the differential element can be represented by Taylor
series, as shown in Figure 2.1b).
But by assuming higher order terms as insignificant and since this is a 1D problem
(i.e. ( ) ( )), a state as shown in Figure 2.2c) is considered. This is an important
argument thus must be grasped by the readers because as we going to see later, the
derivation of PDE (or ODE) for other problems will be based on the same argument as well.
Having established the differential element and the corresponding forces acting on it, we
are in the position to derive the ODE for the bar problem. Since the bar only deform in axial
direction, only equilibrium in x-direction need to be considered thus;
(2.1)
By rearranging gives:
(2.2)
From Hooks Law, we know that
(2.3)
where σ is the axial stress, E is the Modulus Young and ε is the axial strain. Since
(2.4)
and
(2.5)
By inserting Eqns. (2.4) and (2.5) into Eqn. (2.3) gives

(2.6)
( )
By differentiating Eqn. (2.6) once gives
(2.7)
( )
By inserting Eqns. (2.7) into (2.2) would then give the ODE which we have previously
encountered in Chapter 1 (Eqn. (1.1)), that is
(2.8)
( )
For every ODE (or PDE for that matter), the domain equation/s must be accompanied by
boundary equations and for this particular case, the equations are given as below:
a) Natural/force boundary conditions
( ) (2.9a)
|
( )
| (2.9b)
b) Essential/displacement boundary conditions
| (2.9c)
| (2.9d)
Note that the sign convention for the natural (force) boundary conditions above is based on
the assumption that the boundary conditions are the reactions at support hence the
opposite directions to the internal forces.Since Eqn. (2.8) is a 2nd order ODE, two boundary
conditions out of the four given above must be known in prior so as to have a well-posed
problem.
Having established the bar ODE, we are now all set to discuss the basic concept of Galerkin
WRM. This is given next.
2.2.1 FUNDAMENTAL OF GALERKIN WRM
We have discussed the basic concept of WRM that is, the integration of a function consists
of a product of the residual function and a weight function and forcing this integrated value
to zero in getting an algebraic function.
The Galerkin WRM differs in the following aspects:
1. The guessed solution is expressed in terms of shape of functions, and degree of

freedoms, instead of interpolation functions and its constant. In other words and in bar
problem in particular, whilst the basic WRM uses the following guessed function
in Galerkin WRM, the following guessed function are used instead
where and are the shape functions and and are the degree of freedoms (dofs).
We are going to discuss in detail on shape functions and degree of freedoms in the next sub-
section.
2. Galerkin WRM involves integration by part (IBP) which purposes are to:
i. reduce the order of the original derivative terms so as to relax the continuity requirement
ii. induce the natural (force) boundary conditions explicitly into the formulation
Point i) and ii) deserve further elaboration. For point i), as we going to see later, the IBP will
reduce the order of derivation of bar problem from to . Such a reduction would mean
that a guessed function, say a linear polynomial, ( ) once inserted into the former
would vanish but a constant if inserted into the latter. This is what we mean by “to relax the
continuity requirement”. Due to this, the final equation of Galerkin WRM is known as weak
statement because it has been “weakened” from the original PDE (or ODE). Accordingly, the
original form of the PDE (or ODE) is known as the strong statement of the problem.
For point ii), as we are going to see, the IBP will induce boundary terms which exactly in the
forms of Eqn. (1.2) (or Eqn. (1.14b)). Having these terms induced rather naturally, these
terms are called natural boundary conditions hence the interchangeable use of the term
with force boundary conditions. Due to this natural induction of the force (or natural)
boundary condition, we are not going to see the separate satisfaction of the force boundary
conditions as we saw previously in the basic WRM (i.e Eqn. (1.20b)).
3. Due to the use of dofs (hence the advantage of Galerkin WRM), we are also not going to
see the separate satisfaction of the displacement boundary conditions as previously seen in
Eqn. (1.20a). Instead, the displacement boundary conditions will be imposed only after the
complete establishment of the algebraic equation and prior to the solution of the equation.
Due to this, what is known as displacement is also interchangeably known as essential
boundary condition in Galerkin WRM and FEM for that matter.
Actually, there are two types of Galerkin WRM which difference depends on the type of
weight function. If the shape functions are used as the weight functions, the method is
called Bubnov–Galerkin but if other functions is used (for example, interpolation function as
in the basic WRM) the method is called Petrov–Galerkin. However, herein, our main interest
will be on the former, that is Bubnov–Galerkin method.
So the above is the description about Galerkin WRM (on how it differs from the basic WRM,
discussed previously in Chapter 1). Observing such an evolution is important as it can allow
for easier but deeper understanding of FEM.
Having established all these, we are now all set to discuss in detail the Galerkin WRM and
FEM for that matter and we are going to do this by solving the same bar ODE which we have
previously discussed in Chapter 1 and which derivation have been made earlier in this
chapter. To do this, we start by discussing the derivation of shape functions and the
definition of dofs for a bar problem. Before that, let’s note that Galerkin WRM is referred to
FEM from now on.
2.2.2 DOF, AND SHAPE FUNCTIONS, FOR BAR ELEMENT
FEM bar element can be usually of two types; linear element and quadratic element. These
elements are shown in Figure 2.2; the former has two nodes and the latter has three nodes
per element.
Figure 2.2: Degree of freedoms of bar elements
Herein, we are going to derive the shape functions for both elements and we start by
assuming an interpolation function, ( ) in the forms of polynomial, thus
a) For linear bar element
( ) (2.10a)
b) For quadratic bar element
( ) (2.10b)
Despite the familiar use of polynomial functions, it is preferable to deal with a special set of
functions, known as shape functions, and a special sets of coefficient, known as degree
of freedoms (dof), . This means, Eqns. (2.10) must be equivalently expressed as:
( ) (2.11a)
( ) (2.11b)
The equations can be expressed compactly in components forms as:
( ) for (2.12a)
( ) for ,3 (2.12b)
Being referred as the degree of freedoms or dofs, , , and in Eqns. (2.11) (or in
Eqns. (2.12), are actually the nodal values of the axial displacement, ( ) These dofs are
graphically described in Figure 2.2.
The conversion from Eqns. (2.10) to Eqns. (2.11) or (2.12), hence the derivation of the shape
functions, can be done as follows.
For linear element
i. Evaluate Eqn. (2.10a) at the location of the nodes (i.e. at both ends, ) and
equate them according to the dofs will give:
( )| ( )
(2.13a)
( )| ( )
ii. Solving for the values of then gives:
(2.13b)
iii. Inserting Eqn. (2.13b) into Eqn. (2.10a) and by factorizing give:
(2.13c)
( )
By comparing Eqn. (2.13c) and Eqn. (2.11a), it can be concluded that:
(2.13d)
Both and take the shapes as shown in Figure 2.3. Take note of the shapes as they will
be referred later in the discussion of integration by parts and the resulting boundary terms.
Figure 2.3: Linear shape functions
For quadratic element
i. Evaluate Eqn. (2.10b) at the location of the nodes (i.e. at both ends,
) and equate them according to the dof give:
( )| ( ) ( )
( )| ( ) ( ) (2.14a)
( )| ( ) ( )
ii. Solving for the values of then gives:
(2.14b)
iii. Inserting Eqn. (2.14b) into Eqn. (2.10b) and by factorizing give:
(2.14c)
( ) ( )( ) ( ) ( )
By comparing Eqn. (2.14c) and Eqn. (2.11b), it can be concluded that:
( )( )
( )
(2.14d)
( )
, and take the shapes as shown in Figure 2.4. Take note of the shapes as they will
Figure 2.4: Quadratic shape functions
2.2.3 DISCRETIZATION BY GALERKIN METHOD
Having established the shape functions and the degree of freedoms of the bar element, we
are all set to proceed with the discretization of the ODE.
For linear element, inserting Eqn. (2.13c) into Eqn. (2.8) gives:
( )
(2.15a)
whilst for quadratic element, inserting Eqn. (2.14c) into Eqn. (2.8) gives:
( ) (2.15b)
Take note that, from now on, we are going to deal frequently with the component forms as
this would eliminate the necessity to distinguish between linear and quadratic elements.
With this, it should also demonstrate the advantage of such notation system. The
component forms of Eqns. (2.15) is given as
( ) (2.16)
Since we have number of unknowns (because we want to determine the values of the
dofs, ) we thus need number of simultaneous equation. These required equations can be
obtained by multiplying Eqn. (2.16) subsequently with a set of weight functions. This act is
similar to the one conducted to Eqns. (1.17) during the discussion of the basic concept of
WRM.
Also, as mentioned earlier in this chapter, for Bubnov-Galerkin formulation, the weight
functions are also the shape functions. For example, for a linear bar element, the two
simultaneous equations can be obtained as follows:
( )
( )
(2.17a)
( )
( )
So, in component forms, the required simultaneous equations can be given as:
( )
( ) (2.17b)
The continuous nature of Eqns. (2.17) (since it is still in ODE forms) can be converted into
the equivalent algebraic forms by integrating over the length of the bar. This action is
referred to as ‘weighting the residual’; an action from which an approximate solution can be
obtained by forcing the residual to zero. The algebraic forms of Eq. (2.17b) are thus given as:
( )
∫ ( ) (2.18)
It can be seen that Eqn. (2.18) involves the integration of a product of functions, thus
integration by parts (IBP) can be employed. The reasons for employing the IBP, as
mentioned earlier in this chapter, are:
i. to reduce the continuity requirement since only first and lower derivatives will be dealt
with
ii. to induce (make explicit) the natural boundary conditions as boundary terms resulting
from the IBP
Employing IBP to Eqn. (2.18) gives:
( )
∫ ( ) ∫ | (2.19)
In Eqn. (2.19) above, the dofs, are taken out from the integral sign because they are
constants thus do not involve in both differentiation and integration. The last terms on the
right hand side of Eq. (2.19), are called boundary terms resulting from the IBP.
By inserting Eqn. (2.12), the boundary terms can be given as:
( )
| | (2.20)
And further comparison with Eqns. (2.9a) and (2.9b), one can identify that Eqn. (2.20) is
actually the natural boundary conditions of the bar; hence it can be given that:
( )
| | | | (2.21)
Eqn. (2.21) deserves further explanation. The prevalence of the natural/force boundary
conditions, i.e. and within the equation depends on the accompanying value of the
shape functions, being evaluated at the location of the respective node i.e. | and
| . As can be seen, say from Figure 2.3 (linear element), the value of the shape
functions at the location can be given as:
at | |
(20)
at | |
As a result, for linear bar element, the boundary terms of Eqn. (2.21) can be specifically
given as:
| | ( ) ( )
(2.23)
| | ( ) ( )
But, for quadratic bar element, there is one particular feature worth of attention, that is, the
evaluated value of always null at the location of the edge nodes (refer Figure 2.4). By
taking this into consideration, the boundary terms of Eqn. (2.29) for a quadratic bar element
are thus given as:
| | ( ) ( )
(2.24)
| | ( ) ( )
| | ( ) ( )
Due to these various conditions and combination, it is common in the discussion of finite
element formulation to write Eqn. (2.19) simply as:
∫ ( ) ∫ (2.25)
where refers to boundary terms which can be nodal loads or reaction forces at support.
Eq. (2.25) completes the discretization of the bar’s ODE by Galerkin Method.
2.2.4 LOCAL LOAD VECTOR, { } AND STIFFNESS MATRIX, [ ]
Now, let’s discuss the other term on the right hand side of Eqn. (2.25), that is ∫ .
Furthermore, let’s call this term as equivalent nodal loads and denote it as , thus:
∫ (2.26)
It is termed as equivalent nodal loads because these loads are the equivalent effects of a
load, that is acting on the domain; effects as perceived at the nodes. To better describe
such equivalence, let’s refer to Figure 2.5. As shown in the figure, the distributed loading,
( ) which is acting on the domain of the bar would deform the bar in the same manner as
if the bar is subjected to the two nodal loads (or three for quadratic bar) which values are
calculated by Eqn. (2.26).
Figure 2.5: Equivalent nodal loads
Having defined Eqn. (2.26), Eqn. (2.25) can be rewritten as:
∫ ( ) (2.27)
Eqn. (2.27) has been derived for one particular bar element, thus it is called local
equilibrium equation. For a more compact statement, Eqn. (2.27) is usually written in
component forms as:
(2.28)
where is termed as the local stiffness matrix of the bar element. and , on the other
hand, are the local vector of dofs and the load vector of the bar element, respectively. Thus,
the stiffness matrix and the load vector of the bar element can be given as:
∫ ( ) (2.29)
Alternatively, in matrix forms, the local equilibrium can also be represented as:
[ ]{ } { } (2.30)
In an expanded matrix forms, the stiffness matrix of linear and quadratic elements can be
respectively given as:
(2.31a)
∫ ( ) ∫ ( )
[ ] [ ]
∫ ( ) ∫ ( )
[ ]
∫ ( )
[ ] [ ]
∫ ( ) (2.31b)
[ ]
(Note: due to limited spacing, only elements and are expanded).
Stiffness matrices which are given in Eqns. (2.31) are still in integral forms. Generally, to
carry out the integration, numerical integration is used due to the complexity. However,
since those in Eqns. (2.31) are simple, direct integration is used herein. The main purpose is
that, with direct (analytical) integration, the evolution of the procedure can easier be
followed since all variables remained to be seen. For demonstration purposes, the
integration of are shown below for both linear and quadratic elements. The fully
integrated stiffness matrix of both elements is then given.
(2.32a)
∫ ( )
∫ (( ) ( ))
(2.32b)
∫ ( )
∫ ( ) ( )
The fully (or analytically) integrated stiffness matrix for both elements can be given as:
(2.33a)
[ ] [ ]
[ ]
[ ]
(2.33b)
The load vectors are respectively given as:
∫
{ } { } { }
∫
{ }
(2.34a)
for a specific case of uniform distributed load, it can be given that
{ } { }
{ } { } { } ∫
∫
{ } (2.34b)
for a specific case of uniform distributed load, it can be given that
{ }
{ }
Note that, Component forms and matrix forms will be used interchangeably in our
forthcoming discussion.
2.2.5 ASSEMBLY OF ELEMENTS: GLOBAL STIFFNESS MATRIX, [ ] AND LOAD VECTOR, { }
A body or a structure is meshed by many elements in a finite element analysis. The meshing
requires the assembly of local stiffness matrix, [ ] and local load vector, { } of individual
element to form the global stiffness matrix, [ ] and load vector, { }. Although in practice,
hundreds and thousands of elements might be assembled, herein, only the assembly of two
bar elements is demonstrated for easy tracing of the procedure. Figure 2.6 shows the two
assembled bar elements together with the local and global numbering of nodes and dof.
Figure 2.6a: Local and global numbering of linear bar element
Figure 2.6b: Local and global numbering of quadratic bar element
Based on the above, it is obvious that we need to have extra notation to distinguish
different elements. In doing this, we introduce the use of superscript as follows:
i) The local stiffness matrix of element e is denoted as or [ ]

ii) The local load vector of element e is denoted as or { }
iii) The vector of dof of element e is denoted as or { }
Having introduced the new superscript system, the local equilibrium for each element can
thus be given as:
[ ]{ } { } [ ]{ } { }
(2.35a)
Element 1 Element 2
[ ]{ } { } [ ]{ } { } (2.35b)
Element 1 Element 2
Now the assembly of the elements can be done as follows.
[ ]{ } { }
or in global numbering as:

(2.36a)
[ ]{ } { }
which can be simplified as
[ ]{ } { }
[ ]{ } { }
(2.36b)
[ ]{ } { }
[ ]{ } { }
Based on Eqns. (2.36), it can be deduced that[ ], { } and { } are the global stiffness
matrix, global vector of dof and global load vector, respectively.
2.2.6 IMPOSITION OF ESSENTIAL BOUNDARY CONDITIONS
As mentioned earlier, to have a well-posed problem, essential (or displacement) boundary

conditions must be imposed prior to the solution of the equilibrium equation. There are
various ways to impose boundary conditions such as row-column condensation, penalty
method, Lagrange multiplier method etc. Herein, row-column condensation is employed.
The basic concept of the row-column condensation can be explained as follows. Assume
that we have a set of three simultaneous equations given as:
(2.37a)
Now, let’s assume that is known (the boundary conditions) i.e. ̅ . By inserting this
known value into Eqn. (2.37a) gives:
̅
̅ (2.37b)
Since ̅ is known, terms associated with it can be moved to the right hand side of Eq. (54b),
thus:
̅
̅ (2.37c)
By examining Eq. (54c), one can realize that we have two unknowns with three equations.
Since we need only two equations for the solution, we can just use say, the first and the
third equations. (We can also decide to use the second equation together with either first or
third equation), thus
̅
̅ (2.37d)
Now, the condensation can be made more obvious if we discuss the above in matrix forms.
Eqn. (2.37a) can be given in matrix forms as:
[ ]{ } { } (2.37e)
If ̅ is known, Eqn. (2.37e) can be condensed (thus the employment of the boundary
conditions) as follows:
̅
[ ]{ } { } (2.37f)
̅
The above explains the basic concept of row-column condensation where it can be
concluded that, an imposition of number of boundary conditions to a stiffness matrix of
size and to a load vector of , would condense the former to a size of
( ) ( ) and the latter to as size of ( ) . Once condensed, the solution
of the matrix system can proceed.
Now, let’s employ this method directly on our global equilibrium equation as given by Eqns.
(2.36). For the purpose of discussion, let’s assume that the right end of the bar is having a
specified displacement. Such a condition leads to the following essential boundary
conditions:
For linear element: ̅
For quadratic element: ̅
The imposition of the above boundary conditions to the global equilibrium equation can be
done as follows.
[ ]{ } { }
̅
which by condensation gives
̅
[ ]{ } { ̅}
(2.38a)
(2.38b)
[ ]{ ̅ } { }
̅
̅
̅
̅
[ ]{ } { } (2.38c)
Eqns. (2.38) are now ready to be solved. There are various ways to solve a matrix equation
such as direct inversion, Gauss Elimination, Cholesky Decomposition, etc. In the
accompanying MATLAB source codes, Gauss Elimination is adopted. Symbolically, the
solution of the matrix equation is represented as
{ } [ ] { } (2.39d)
EXAMPLE 2.1
Let’s solve the previous problem of EXAMPLE 1.1 this time by employing FEM. The problem
is solved by the assembly of two bar elements. Both linear and quadratic bars are
considered. The results are then verified against the closed-form solution of Eqn. (1.4).
a) Linear bar element
Due to the symmetrically, element 1 and element 2 would have a similar local stiffness
matrix and load vector, thus,
[ ] (E2.1)
and
{ }
(E2.2a)
{ }
(E2.2b)
Note that is the local reaction at support of element 1 as can be seen in Figure 1.2.
The assembled global stiffness matrix, [ ] is given as
[ ] [ ] (E2.3)
and the assembled load vector, { } is given as
{ } { } { } (E2.4)
To emphasize the global reactions forces, is expressed as . This is unknown variable

because its corresponding dof is the essential boundary condition which in turn, is a known
value. The whole equilibrium equation can thus be given as
[ ]{ } { } (E2.5)
By imposing the essential boundary conditions ( ), the assembled global equilibrium

equation is reduced to
(E2.6)
[ ]{ } { }
By solving Eqn. (E2.6) using Matlab command “\”, (source code 2.1), the values of the
assembled global dof are thus obtained as:
{ } { } (E2.7a)
Eqn. (E2.7a) is the assembled global solution. By inserting these values accordingly into
Eqns. (2.30) the local forces can then be determined. However, before let’s determine first
the global reaction forces, which can be done by inserting Eqn. (E2.7a) into Eqn. (E2.5)
which yields
{ } [ ]{ } { } (E2.7b)
The global reaction force can thus be given as
{ } { } (E2.7c)
Now, let’s proceed with determination of the local values. The elemental local solution for
each element is thus
{ } { } { } { }
(E2.8)
{ } { } { } { }
(E2.9)
Eqns. (E2.8) and (E2.9) are the elemental local solution.
Element 1
By inserting Eqn. (E2.8) into Eq. (2.30), the local force (or reaction if at support) can be
determined as follows.
{ } [ ]{ } { }
[ ]{ } { } (E2.10)
{ } { }
{ }
Element 2
{ } [ ]{ } { }
[ ]{ } { }
(E2.11)
{ } { }
{ }
Table 2.1: Values of linear bar global degree of freedom, { },
Linear bar
b) Quadratic bar element
matrix and load vector, thus,
(E2.12)
[ ]
and
{ } (E2.13a)
{ } (E2.13b)
Note that is the local reaction at support of element 1 as can be seen in Figure 1.2.The
assembled global stiffness matrix, [ ] is given as
(E2.14)
[ ]
[ ]
[ ]
and the assembled load vector, { } is given as
{ } (E2.15)
{ } { }
To emphasize the global reactions forces, is expressed as . This is unknown variable

because its corresponding dof is the essential boundary condition which in turn, is a known
value. Having established [ ] and { }, the whole equilibrium equation in the form of Eqn.
(2.36c) can thus be given as
[ ]{ }
{ }
(E2.16)
By imposing the essential boundary conditions ( ), the assembled global equilibrium

equation is reduced to
[ ]{ } { } (E2.17)
(E2.18a)
{ }
{ }
Eqn. (E2.18a) is the assembled global solution. By inserting these values accordingly into
Eqns. (2.30) the local forces can then be determined. However, before let’s determine first
the global reaction forces, which can be done by inserting Eqn. (E2.18a) into Eqn. (E2.16)
which yields
{ } [ ]{ }
{ }
(E2.18b)
The global reaction force can thus be given as
{ } { } (E2.18c)
{ } { } { } { } (E2.19)
{ } { } { } { } (E2.20)
Eqns. (E2.19) and (E2.20) are the elemental local solutions.
Element 1
{ } [ ]{ } { }
[ ]{ } { } (E2.21)
{ } { }
{ }
Element 2
{ } [ ]{ } { }
[ ]{ } { } (E2.22)
{ } { }
{ }
Table 2.2: Values of quadratic bar global degree of freedom, { },
Quadratic bar
The validity of both results thus formulation can be assessed by comparing their values with
those previously obtained in EXAMPLE 1.1. The plot of FEM results against the closed-form
solution (Eqn. (1.4) are shown in Figure 2.7 which numerical values at several locations
along the bar are given in Table 2.3.
Figure 2.7 Comparison between ulinear, uquadratic and uexact
Table 2.3: Comparison of values between ulinear, uquadratic and uexact
x=0m x = 1.25 m x = 2.5 m x = 3.75 m x=5m

ulinear - -
uquadratic
uexact
Based on the plot and the table, for the first time the convergence nature of a numerical
analysis becomes obvious. As can be seen, the assembled two linear elements provide quite
a poor approximation except at the location of the nodes where the results agree with the
closed-form solution. A better hence a converged solution is provided by the quadratic
elements thus an immediate demonstration of the beneficial effect of the use of higher
order elements. This is an immediate demonstration to the statement “FEM converges to
the “accurate” solution with the increase in the order of polynomial of the guessed functions
(or degree of freedom and mesh density later on)” which was stated in previous chapter.
2.4 SOURCE CODE (LINEAR BAR)
clc;clear all; %Clear previous data
%Input
A=0.04; % Area (m^2)
L1=2.5; % Length 1st Element (m)
%Create local stifness n forces

klocal1=[(A*E)/L1 -(A*E)/L1;
-(A*E)/L1 (A*E)/L1];
-(A*E)/L2 (A*E)/L2];
flocal1=[q*L1/2;q*L1/2];
flocal2=[q*L2/2;q*L2/2];
%Assemble
Kglobal=zeros(3,3);
Kglobal(1:2,1:2)=Kglobal(1:2,1:2)+klocal1;
Fglobal=zeros(3,1);
Fglobal(1:2,1)=Fglobal(1:2,1)+flocal1;
FG=Fglobal;
Fglobal(3,1) =Fglobal(3,1)+P; % Point Load
KG=Kglobal;
%Impose BC
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Fglobal(1,:)=[];
%Solve for global displacement

U=Kglobal\Fglobal
% local result Element 1

U1=[0;U(1,1)];
f1=(klocal1*U1)-flocal1

U2=[U(1,1);U(2,1)];
% Reacttion
UU=[0;U(1,1);U(2,1)];
R=KG*UU-FG
%--------------------------------------------------------------------------
% Plot Result
%--------------------------------------------------------------------------
L=5;
Uexact=@(x) -q / E / A * x ^ 2 / 0.2e1 + (P + q * L) / E / A * x;
FEM2Nodex=[0;2.5;5];
FEM2Nodey=[0;U(1,1);U(2,1)];
plot(FEM2Nodex,FEM2Nodey,'--sk')
hold on
fplot(Uexact,[0 L],'-.*b')
hold on
xlabel('x');
ylabel('u (x) ');
legend('Ulinear ','Uexact')
2.5 SOURCE CODE (QUADRATIC BAR)

Source Code Quadratic Bar
%Input
A=0.04; % Area (m^2)
%Create local stifness n forces

klocal1=(A*E)*[7/(3*L1) -8/(3*L1) 1/(3*L1);
-8/(3*L1) 16/(3*L1) -8/(3*L1);
1/(3*L1) -8/(3*L1) 7/(3*L1)];
klocal2=(A*E)*[7/(3*L2) -8/(3*L2) 1/(3*L2);

-8/(3*L2) 16/(3*L2) -8/(3*L2);
1/(3*L2) -8/(3*L2) 7/(3*L2)];
flocal1=[q*L1/6;2*q*L1/3;q*L1/6];
flocal2=[q*L2/6;2*q*L2/3;q*L2/6];
%Assemble
Kglobal=zeros(5,5);
KG=Kglobal;
Fglobal=zeros(5,1);
FG=Fglobal;
Fglobal(5,1) =Fglobal(5,1)+P; % Point Load
%Impose BC
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Fglobal(1,:)=[];

U=Kglobal\Fglobal

U1=[0;U(1,1);U(2,1)];

U2=[U(2,1);U(3,1);U(4,1)];
% Reaction
UU=[0;U(1,1);U(2,1);U(3,1);U(4,1)];
R=KG*UU-FG
%--------------------------------------------------------------------------
% Plot Result
%--------------------------------------------------------------------------
L=5;
Uexact=@(x) -q / E / A * x ^ 2 / 0.2e1 + (P + q * L) / E / A * x;
FEM3Nodex=[0;1.25;2.5;3.75;5];
FEM3Nodey=[0;U(1,1);U(2,1);U(3,1);U(4,1);];
plot(FEM3Nodex,FEM3Nodey,'-or')
hold on
fplot(Uexact,[0 L],'-.*b')
hold on
xlabel('x');
ylabel('u (x) ');
legend('FEM-Quadratic','Uexact')
LINEAR AND NONLINEAR FINITE ELEMENT ANALYSIS FOR SOLIDS AND
FLUIDS (unedited 1st draft)
2013
CHAPTER 3
GALERKIN FORMULATION: BEAM ELEMENT
3.1 INTRODUCTION
In the previous chapter, we have discussed the concept of Galerkin WRM hence FEM and
formulated the discretised equation for bar element. Herein, we are going to continue our
discussion by formulating the discretised equation for beam problem. We begin by deriving
the Euler-Bernoulli differential equation of beam which is given next.
3.1 ORDINARY DIFFERENTIAL EQUATION (ODE) OF EULER-BERNOULLI BEAM
An Euler-Bernoulli beam is a structural member that resists loads by bending and shearing.
The corresponding deformation would be rotation and translation. The differential equation
for such a beam can be derived for static loading by considering its differential element as
shown in Figure 3.1. Also shown is the typical arrangement of a beam structure.
As argued previously for bar element (Chapter 2), although we have moment force, M and
shear force, V at the left side of the differential element, due to the ‘disturbance’ along the
differential length, dx (due to external load, for example) the magnitude of these forces at
the right side of the differential element must change. However we don’t know the exact
change of these forces, else we would not have the problem in the first place would we? But
by assuming the change is continuous, we can say that the forces at the right-side of the
differential element can be represented by a Taylor series, as shown in Figure 3.1b).
By assuming higher order terms as insignificant and since this is a 1D problem (i.e. ( )
( )), a state as shown in Figure 3.1c) is considered. This is an important argument thus
must be grasped by the readers because as we are going to see later, the derivation of PDE
(or ODE) for other problems will be based on the same argument as well.
Having established the differential element and the corresponding forces acting on it, we
are in the position to derive the ODE for beam bar problem. However, it must be noted
again that the following differential equation does not consider axial deformation thus the
absent of axial forces. Beams allowing such forces are called beam-column which FEM
formulation is given in the next chapter. Also, present formulation assumes slope is equalled
to rotation. A more general formulation would be the Timoshenko’ beam theory as it allows
different values for the two entities.
2013
Figure 3.1: Beam structure and its differential element
Based on Figure 3.1c), the following equilibrium of forces can be employed:
∑ (3.1)
∑ (3.2)
∑ (3.3)
which yield
2013
( ) (3.4)
( ) (3.5)
( )
where
is the distributed transverse external loading
is the deflection of the beam
Eqns. (3.4) and (3.5) can be simplified into
(3.6)
(3.7)
( )
To arrive at Eqn. (3.7), higher order term has been assumed as insignificant thus
omitted. By differentiating Eqn. (3.7) once then inserting it into Eqn. (3.6) gives
(3.8)
Eqn. (3.8) is the differential equation of the beam in terms of moment force. But since our
FEM discussion focuses on displacement-based formulation, we need to express the ODE in
terms of displacement. This can be done by employing the constitutive equation which
relates the curvature of the beam with the moment force given as
(3.9)
where is the Young’s modulus of the material and is the second moment of area of the
beam’s cross-section. Together, they are known as the flexural stiffness of the beam, .
By differentiating Eqn. (3.9) twice and inserting into Eqn. (3.8), we obtain
(3.10)
Eqn. (3.10) is the ODE of the beam in terms of displacement variable as required. Like any
other differential equations, Eqn. (3.10) must be complemented by boundary condition
equations. There are two types of boundary conditions; force (natural) and displacement
(essential).
The force (natural) boundary conditions of the beam can be a single or a combination of the
following;
2013
(3.11)
( )|
(3.12)
( )|
(3.13)
( )|
(3.14)
( )|
where , are the specified end shear forces (edge transverse point loads) and ,
are the specified end moment forces. Note that the sign convention above is based on the
assumption that the boundary conditions are the reactions at support hence the opposite
directions to the internal forces.
On the other hand, the displacement (essential) boundary conditions of the beam can be a
single or a combination of the following;
(3.15)
( )|
(3.16)
( )|
( )| (3.17)
( )| (3.18)
where , are the specified end rotations and , are the specified end transverse
displacements.
Since the beam ODE is of 4th order respectively, four boundary conditions out of the six
given above (Eqns. (3.11) to (3.18)) must be known in prior so as to have a well-posed
problem.
Having established the beam ODE, we are now all set to discuss its discretization by FEM.
We begin with the derivation the shape functions and the beam’s degree of freedoms.
3.2 DOF, AND SHAPE FUNCTIONS, FOR BEAM ELEMENT
Figure 3.2 shows a beam element with two end nodes. At each node, two degree of
freedoms are introduced; a translational dof and a rotational dof.
2013
Figure 3.2: Degree of freedoms of beam element
We start the derivation of the shape function by assuming an interpolation function for the
transverse displacement, ( ) of the beam in the forms of polynomial. Thus;
( ) (3.19)
As frequently mentioned, it is preferable in finite element formulation, to deal with a special

set of functions, known as shape functions, and a special sets of coefficient, known as
degree of freedoms (dofs), . This means, Eqns. (3.19) must be equivalently expressed as:
( ) (3.20)
The above equations can be expressed compactly in component forms as:
( ) for (3.21)
Being referred as the degree of freedoms, , , and in Eqs. (3.20) (or in Eqs.
(3.21), are actually the nodal values of the transverse displacement, ( ) as graphically
described in Figure 3.2.
The conversion from Eqn. (3.19) into Eqn. (3.20), hence the derivation of the shape
functions, can be done as follows.
i) Evaluate Eq. (3.19) and its first derivative at the location of the nodes (i.e. at both
ends, ) and equate them to the dof to give:
( )|
( )
|
(3.22)
( )|
( )
|
2013
ii) Solving for the values of and then gives:
(3.23)
iii) Inserting Eqn. (3.23) into Eqn. (3.19) and by factorizing give:
( ) ( ) ( ) ( ) (3.24)
( )
By comparing Eqn. (3.24) with Eqn. (3.20), it can be concluded that:
( )
(3.25)
( )
( )
( )
The shape functions above are known as Hermite shape functions and they take the shapes
as shown in Figure 3.3. Take note of the shapes as they will be referred later in the
discussion of integration by parts and the resulting boundary terms.
2013
Figure 3.3: Hermite shape functions of beam element
3.3 DISCRETIZATION BY GALERKIN METHOD
Having established the shape functions and the degree of freedoms of the beam element,
we are all set to proceed with the discretization of the ODE.
By inserting Eqn. (3.25) into Eqn. (3.10) gives:
( )
(3.26)
Take note that, from now on, we are going to deal with the component forms as this would
eliminate the need to write (hence to read) long statement, thus Eqn. (3.26) can be given in
component forms as
( ) (3.27)
Since we have j number of unknowns (because we want to determine the values of the dofs,
) we thus need j number of simultaneous equation. In our present discussion, so we
require four simultaneous equations. These required equations can be obtained by
multiplying Eqn. (3.27) subsequently with a set of weight functions. As mentioned, since our
discussion focuses on Bubnov-Galerkin formulation, the weight functions are also the shape
functions, thus
( )
( )
2013
( ) (3.28)
( )
( )
( )
( )
( )
By observing Eqn. (3.28) it can be deduced that, in component forms, the required
simultaneous equations can be given as:
( )
( ) (3.29)
The continuous nature of Eqns. (3.29) (since it is still in ODE forms) can be converted into
the equivalent algebraic forms by integrating over the length of the beam. This action is
obtained by forcing the residual to zero. The algebraic forms of Eq. (3.29) are thus given as:
( )
∫ ( ) (3.30)
It can be seen that Eqn. (3.30) involves the integration of an inner product of functions, thus
integration by parts (IBP) can be employed. The reasons for employing the IBP, as
mentioned earlier in this chapter, are:
i. to reduce the continuity requirement since only second order derivatives will be dealt with
ii. to induce (make explicit) the natural boundary conditions as boundary terms resulting
from the IBP
Employing the first IBP to Eqn. (3.30) gives:
( )
∫ ( ) ∫ | (3.31)
In Eqn. (3.31) above, the dofs, is taken out from the integral sign because it represents
constants thus does not involve in both differentiation and integration. The last terms on
the right hand side of Eq. (3.31), are called boundary terms resulting from the first IBP. A
careful examination reveals that this boundary terms are also the force (natural) boundary
2013
conditions given by Eqns. (3.11) and (3.12), which specially related to end shear force.
Inserting these equations into Eqn. (3.31) modifies the latter into
∫ ( ) ∫ ( | | ) (3.32)
Next, we conduct IBP one more time on Eqn. (3.32). But why we want to do this? The
reasons are:
i. we want to optimize the continuity relaxation and since the order of derivatives between
the terms in the right hand side integral is still “unbalanced”(i.e. ), we have the
choice to balance them by another IBP
ii. Another IBP, as we are going to see, would induce another type of force (natural)
boundary conditions of the beam which are the end moments.
By conducting another IBP on Eqn. (3.32), we obtain
∫ ( ) (3.33)
( )
∫ ( | | ) |
The last terms on the right hand side of Eq. (3.33), are called boundary terms resulting from
the second IBP. A careful examination reveals that this boundary terms are also the force
(natural) boundary conditions given by Eqns. (3.13) and (3.14), which specially related to
end moment forces. Inserting these equations into Eqn. (3.33) modifies the latter into
∫ ( ) (3.34)
∫ ( | | )
( | | )
Eqn. (3.34) is the final weak statement of the beam problem but the last two terms on the
right hand side of the equation deserves further explanation. The prevalence of the
natural/force boundary conditions (whether they remain within the equation or not)
depends on the accompanying value of the shape functions when evaluated at the location
of the respective node i.e. | | | and | . As can be seen in Figure 3.3
the value of the shape functions and its derivatives at the node’s location can be given as:
2013
at
| | | |
| | | |
(3.35)
At
| | | |
| | | |
Based on the above characteristic of shape functions, we can observe what remains on the
right side of Eqn. (3.34) say for ,
∫ ( ) (3.36)
∫ ( | | )
( | | )
Employing Eqn. (3.35) to Eq. (3.36), we obtained
∫ ( ) ∫ ( ) ( ) (3.37)
Simplifying, Eqn. (3.37) becomes
∫ ( ) ∫ (3.38)
By observing the evolution of discussion for other values of , the general formulation in the
forms of Eqn. (3.38) can thus be given as
∫ ( ) ∫ (3.39)
where refers to boundary terms or nodal loads. Eqn. (3.39) completes the discretization
of the beam’s ODE by Galerkin Method.
2013
3.4 LOCAL LOAD VECTOR, { } AND STIFFNESS MATRIX, [ ]
Now, let’s discuss the other term on the right hand side of Eqn. (3.39), that is ∫ .
∫ (3.40)
load, that is acting on the domain; effects as perceived at the nodes. To better describe
such equivalence, let’s refer to Figure 3.4. As shown in the figure, the distributed loading,
( ) which is acting on the domain of the beam would deform the bar in the same manner
as if the beam is subjected to the four nodal loads which values are calculated by Eqn.
(3.40).
Figure 3.4: Equivalent nodal loads of beam element
Having defined Eqn. (3.40), Eqn. (3.39) can be rewritten as:
∫ ( ) (3.41)
Eqn. (3.41) has been derived for one particular bar element, thus it is called local
equilibrium equation. For a more compact statement, Eqn. (3.41) is usually written in
component forms as:
(3.42)
where is termed as the local stiffness matrix of the beam element. and , on the
other hand, are the local vector of dofs and the load vector of the beam element,
respectively. Thus, the stiffness matrix and the load vector of the beam element can be
given as:
2013
∫ ( ) (3.43)
[ ]{ } { } (3.44)
In an expanded matrix forms, the stiffness matrix of a beam element can be respectively
given as:
∫ ( )
(3.45)
[ ] [ ]
∫ ( )
[ ]
The stiffness matrix which is given by Eqn. (3.45) is still in integral forms. Generally, to carry
out the integration, numerical integration is used due to the complexity. However, since
Eqns. (3.45) are simple, direct integration is used herein. The main purpose is that, with
direct (analytical) integration, the evolution of the procedure can easier be followed since all
variables remained to be seen. For demonstration purposes, the integration of and
are shown below.
∫ ( )
∫ ( ) ( )
(3.46a)
∫ ( )
2013
∫ ( ) ( )
(3.46b)
The fully integrated stiffness matrix of the beam element is then given as;
(3.47)
[ ]
[ ]
∫
(3.48a)
∫
{ } { } { }
∫
∫
{ }
It is worth to note again that component forms and matrix forms will be used
interchangeably in our forthcoming discussion.
2013
For a uniform distributed load, the load vector can be given as
∫
(3.48a)
∫
{ } { } { }
∫
∫ { }
{ }
3.5 ASSEMBLY OF ELEMENTS: GLOBAL STIFFNESS MATRIX, [ ] AND LOAD VECTOR, { }
element to form the global stiffness matrix, [ ] and the global load vector, { }. Although in
practice, hundreds and thousands of elements might be assembled, herein, only the
assembly of two bar elements is demonstrated for easy tracing of the procedure. Figure 3.5
shows the two assembled beam elements together with the local and global numbering of
nodes and dofs.
Figure 3.5: Local and global numbering of beam element
2013
Based on the above and to distinguish different elements we introduce the following
superscript notation:
i) The local stiffness matrix of element e is denoted as or [ ] (3.49a)
ii) The local load vector of element e is denoted as or { } (3.49b)
iii) The vector of dofs of element e is denoted as or { } (3.49c)
thus be given as:
For element 1
[ ]{ } (3.50a)
[ ]{ } { }
For element 2
[ ]{ } (3.50b)
[ ]{ } { }
Having established the local matrices and their notation system, the assembly of the two
elements hence the global equilibrium equation is given as
(3.51)
[ ]{ }
{ }
[ ] { }
{ }
2013
3.6 IMPOSITION OF ESSENTIAL BOUNDARY CONDITIONS
Essential (or displacement) boundary conditions must be imposed before the equilibrium
equation of the beam can be solved. Basic concept of direct imposition of boundary
conditions has been described in detail in previous chapter.
Eqns. (3.15) to (3.18) give all the possible essential boundary conditions of a beam,
reproduced herein for ease of reading as:
(3.52a)
( )|
(3.52b)
( )|
( )| (3.52c)
( )| (3.52d)
Graphical representations of these boundary conditions are given in Figure 3.6.
Figure 3.6: Various beam’s essential boundary conditions
EXAMPLE 3.1
Let’s put what we have learned so far into practice. Figure 3.6 shows a cantilever beam
subjected to a distributed load of -2kN/m and an edge point load of -5kN.
Figure 3.6: Cantilever beam
2013
An analytical solution for the beam can be obtained by conducting integration directly on
the ODE given by Eqn. (3.10) and satisfying the all the relevant boundary conditions (both
force and displacement), which are;
i.
( )| ( )
ii.
( )|
iii.
( )|
iv. ( )| =0
The analytical or the closed form solution of the beam can be given as
( ) ( ) (E3.1)
Herein, we are going to solve the problem by the assembly of two beam elements of equal
length and compare the results against Eqn. (E3.1) to validate the FEM formulation. Due to
the symmetrically, element 1 and element 2 would have a similar local stiffness matrix thus,
(E3.2)
[ ]
and the load vectors are given as
(E3.3a)
{ }
{ } (E3.3b)
The assembled global stiffness matrix can be given as (based on Eqn. (3.51)):
2013
[ ]
(E3.4)
[ ]
and the assembled global load vector can be given as:
{ }
(E3.5)
{ } { }
To emphasize the global reactions forces, and are expressed as and ,

respectively. These are unknown because their corresponding dofs are the essential
boundary conditions which in turn, are known values. The whole equilibrium equation can
thus be given as
[ ]{ }
{ }
(E3.6)
By imposing the essential boundary conditions (i.e. ), Eqn. (E3.6) is reduced to
(E3.7)
[ ]{ } { }
2013
(E3.8a)
{ }
{ }
Eqn. (E3.8) is the assembled global solution. By inserting these values accordingly into Eqns.
(3.44) the local forces can then be determined. However, before let’s determine first the
global reaction forces, and which can be done by inserting Eqn. (E3.8a) into Eqn.
(E3.6) which yields
{ }
[ ]{ }
{ }
(E3.8b)
The global reaction forces can thus be given as
(E3.8c)
{ } { }
{ } { } { } (E3.9)
{ }
2013
{ } { } { } (E3.10)
{ }
Eqns. (E3.9) and (E3.10) are the elemental local solutions. Next, we determine the local
forces.
Element 1
By inserting Eqn. (E3.9) into Eq. (3.44), the local forces can be determined as follows.
{ } [ ]{ } { }
[ ]{ } { }
(E3.11)
{ } { }
{ }
Element 2
{ } [ ]{ } { }
[ ]{ } { }
(E3.12)
{ } { }
{ }
2013
Table 3.1: Values of beam global degree of freedoms, { },
Di D1 D2 D3 D4 D5 D6
Values 0 0 -0.0021 -0.0018 -0.0064 -0.0023
The validity of the formulation is assessed by comparing its result against the exact solution
as given by Eqn. (3.1) and whilst the reactions can as well be obtained analytically, herein
reactions obtained from commercial software are used for validation so as to have diversity
in data sources. The plot of FEM results against the exact solution (Eqn. (E3.1) is shown in
Figure 3.6 which numerical values at several locations along the beam are given in Table 3.2.
Table 3.3 gives the comparison of the reaction forces at support.
Figure 3.6: Plots of wFEM and wexact
Table 3.2: Comparison of values between wFEM and wexact
wFEM 0 - -
wexact 0 -
2013
Table 3.3: Comparison of values of reaction forces at fix support
wFEM 13 36
Software 13 36
(converged)
Based on the plot and the tables, it can be seen, the assembled two beam elements provide
quite a poor approximation except at the location of the nodes where the results agree with
the closed-form solution and the commercial software. A better hence a converged solution
is provided by the use of more elements. Figure 3.7, Table 3.4 and Table 3.5 show the
convergence if four elements are used.
Figure 3.7: Plots of w2-elements, w4-elements, and wexact
2013
Table 3.4: Comparison of values between w2-elements, w4-elements, and wexact
wexact 0 -0.0006 -0.0021 -0.0041 -0.0064

w2-elements 0 - -0.0021 - -0.0064
w4-elements 0 -0.0006 -0.0021 -0.0041 -0.0064
Table 3.3: Comparison of values of reaction forces at fix support
{ } { }
w2-elements
w4-elements
Software (converged)
The convergence of the results above is an immediate demonstration of the beneficial effect
of the use of higher order elements. This is an immediate demonstration to the statement
“FEM converges to the “accurate” solution with the increase in the order of polynomial of
the guessed functions (or degree of freedom and mesh density later on)” which was stated in
previous chapter.
3.7 SOURCE CODE
%Input
E=200e6; %kN/m^2
b=0.2; %m
h=0.2; %m
A=b*h; %m^2
I=(b*h^3)/12; %m^4
L1=2; %m
L2=2; %m
P=-5; %kN
q=-2; %kN/m
%Create Stifness
klocal1=[(12*E*I/L1^3) (6*E*I/L1^2) -(12*E*I/L1^3) (6*E*I/L1^2);
(6*E*I/L1^2) (4*E*I/L1) -(6*E*I/L1^2) (2*E*I/L1);
-(12*E*I/L1^3) -(6*E*I/L1^2) (12*E*I/L1^3) -(6*E*I/L1^2);
(6*E*I/L1^2) (2*E*I/L1) -(6*E*I/L1^2) (4*E*I/L1)];

(6*E*I/L2^2) (4*E*I/L2) -(6*E*I/L2^2) (2*E*I/L2);
2013
-(12*E*I/L2^3) -(6*E*I/L2^2) (12*E*I/L2^3) -(6*E*I/L2^2);

(6*E*I/L2^2) (2*E*I/L2) -(6*E*I/L2^2) (4*E*I/L2)];
%calculate equilvalent nodal forces

flocal1=[(q*L1/2);(q*L1^2/12);(q*L1/2);(-q*L1^2/12)];
flocal2=[(q*L2/2);(q*L2^2/12);(q*L2/2);(-q*L2^2/12)];
%Assemble
Kglobal=zeros(6,6);
KG=Kglobal;
Fglobal=zeros(6,1);
FG=Fglobal;
Fglobal(5,1) =Fglobal(5,1)+P;
%Impose BC
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Fglobal(2,:)=[];
Fglobal(1,:)=[];
%Solve
U=Kglobal\Fglobal;

u1=[0;0;U(1,1);U(2,1);];
f1=klocal1*u1-flocal2; %minus equilvalent nodal load

u2=[U(1,1);U(2,1);U(3,1);U(4,1)];
% Reaction
UU=[0;0;U(1,1);U(2,1);U(3,1);U(4,1)];
R=KG*UU-FG;
2013
CHAPTER 4
PLANE STRUCTURES: TRUSS AND FRAME
4.1 INTRODUCTION
In the previous chapters, we have discussed the derivation of FEM formulation for bar and
beam elements. However, these elements were arranged and assembled in a line. A more
general arrangement would require the elements to be arbitrarily oriented and assembled.
Assembly of such oriented elements would make up a truss system and a frame system,
respectively. As we are going to see, the orientation process requires the introduction of the
transformation matrix. The use of this matrix is to transform local entities into global
entities. Another point to emphasize is the introduction of two degree of freedoms and two
load components into the bar element’s global representation. Also, since the construction
of a frame would require the transfer of axial load/force, a beam element is supplied with
extra degree of freedoms in the axial direction. As can be seen, this will involve the
combination of previously derived bar and beam elements to form what is called beam-
column element.
4.1 TRUSS SYSTEM
A truss system is an assembled of inclined bar elements. Figure 4.1 shows a typical
arrangement of a plane truss system.
Figure 4.1: A typical truss system
2013
To allow for the inclination and the corresponding assembly, a bar element formulation
must be re-expressed in global manner. In a global frame, a bar element would have two
degree of freedoms per node.
An inclined bar is shown in Figure 4.2a together with the newly introduced elemental global
direction dofs and previously defined local dofs. Note that, to distinguish between local and
elemental global direction dofs, the former is primed.
Figure 4.2a: Inclined bar element’s degree of freedoms
For a linear bar (two-nodes bar), by considering the geometry of Figure 4.2 the relationship
between the local and the elemental global dofs can be given as,
(4.1 a)
(4.1 b)
In matrix forms, Eqns. (4.1) can be given as
(4.2)
{ } [ ]{ }
or
{ } [ ]{ } (4.3a)
where { } is the vector of local dofs, [ ] is from now on is called the bar transformation
matrix and { } is the elemental global dofs of the bar element.
To simplify the expression, [ ] is expressed as
2013
[ ] [ ] (4.3b)
So far, we have established the relationship between local dofs of bar and its elemental
global dofs. Now we are going to establish the relationship between local forces and the
elemental global forces. By referring to Figure 4.2b, the relationship can be as;
(4.4 a)
(4.4 b)
(4.4 c)
(4.4 d)
In matrix forms Eqns. (4.4) can be given as
(4.5)
{ } [ ]{ }
Figure 4.2b: Inclined bar element’s nodal forces
A comparison with Eqns. (4.2) and (4.3) would show that the 4 by 2 matrix on the right hand
side of Eqn. (4.5) is actually the transpose of the transformation matrix, [ ]. Realizing this
Eqn. (4.5) can be given as
{ } [ ] { } (4.6)
2013
where { } is the vector of elemental global forces and { } is the vector of the local forces of
the bar. Eqn. (4.6) is therefore the relationship between global and local forces of an
element.
Having established the relationship between elemental global and local entities for both
dofs and forces, we are now in the position to establish the elemental global equilibrium
equation for the inclined bar element from its local equation. First we write the local
equation of the bar (previously given as Eqn. (2.30) but with the prime superscript to
distinguish them from their corresponding elemental global entities):
[ ]{ } { } (4.7)
Note that, as given in Eqn. (2.29), { } { }. Inserting Eqn. (4.3) into Eqn. (4.7) would
then give
[ ][ ]{ } { } (4.8)
Now, we multiply the transpose of transformation matrix, [ ] to both sides of Eqn. (4.8) to
obtain
[ ] [ ][ ]{ } [ ] { } (4.9)
Observing Eqn. (4.9) we should notice that the right hand side terms is also the previously
given Eqn. (4.6). So, by inserting Eqn. (4.6) into (4.9), we obtain
[ ] [ ][ ]{ } { } (4.10)
Based on Eqn. (4.10) we can conclude that the elemental global stiffness matrix of the bar,
[ ]can be defined as
[ ] [ ] [ ][ ] (4.11 a)
which for a linear bar element, [ ]can explicitly be given as
(4.11 b)
[ ] [ ][ ]
Our discussion on the transformation of discretized equilibrium equation from local to

element global condition is completed by inserting Eqn. (4.11b) into Eqn. (4.10) to obtain
[ ]{ } { } (4.12)
Once the elemental global equilibrium equation is established, the assembly, the imposition
of essential boundary conditions and the solution process follow the same as outlined
previously. However once a solution is obtained, the corresponding local values becomes of
interest. The determination of the internal axial forces in the bar requires the solved dofs,
{ } to be inserted back into Eqn. (4.8) and the axial displacement, on the other hand can be
obtained by inserting { } into Eqn. (4.3a). However before such a procedure can be
executed, the truss elements must be assembled first.
2013
4.1.1 ASSEMBLY OF TRUSS SYSTEM
What we have derived so far are the elemental global entities. To solve the problem, these
entities must be assembled. The discussion on the assembly of a truss system can be best
done by considering a typical arrangement as shown in Figure 4.3.
Figure 4.3: Assembly of a typical truss system
Figure 4.3 shows both elemental global dofs { } for each element and the assembled global
{ } for the whole structure. The assembly of the elemental values into assembled matrix
and vectors can be done by correlating the elemental numbering subscript to the assembled
numbering. For example, by observing Figure 4.3, we can correlate that, say, of
elemental 2 to in the assembled matrix because it can be traced that elemental
subscript 3 corresponds to assembled subscript 5 (i.e. ) and elemental subscript 1
corresponds to assembled subscript 3 (i.e. ). Based on this, the following is the
correlation between elemental entities to assembled entities for typical truss system of
Figure 4.3.
For element 1
2013
Note that the correlation of element 1 seems to agree directly with the assembled
numbering because their numbering is accidental from the beginning. Such a direct
correlation will not exist for element 2 as we can see next.
For element 2
Having established the above correlation, we thus can build the assembled global stiffness
matrix, as
[ ] (4.13)
[ ]
Based on the same correlation argument, the assembled load vector, { } can thus be given
as
(4.14)
{ }
{ }
{ }
Finally, we complete our discussion on truss system formulation by writing down the
assembled global equilibrium equation as
2013
(4.15)
{ }
[ ] { }
or
[ ]{ } { } (4.16)
By imposing the essential boundary conditions, { } can then be solved.
Now, let’s nail our understanding with an example and the corresponding Matlab source
code.
EXAMPLE 4.1
In this example, we are required to determine the internal axial forces, reaction forces at
supports and displacements of a truss system as shown in Figure (E4.1)
Figure E4.1 Truss with single point load.
Based on the data given in the figure, the local stiffness matrix for each element,
determined based on Eqn. (2.33a), is given below. Note that, the external point load is
2013
inserted directly into the assembled global load vector once everything is assembled; an act
which omits the requirement for it to be considered at a local level.
Element 1
[ ] [ ] (E4.1)
Element 2
[ ] [ ] (E4.2)
Element 3
[ ] [ ] (E4.3)
Their corresponding element global entities can be determined using Eqns. (4.11) as follows.
Element 1
[ ] [ ] [ ][ ]
[ ][ ]
(E4.4)
[ ]
[ ]
[ ]
2013
Element 2
[ ]
[ ]
[ ]
[ ]
(E4.5)
Element 3
[ ] [ ]
[ ]
[ ]
(E4.6)
Having established the elemental global stiffness for each element, the assembled global
stiffness matrix, [ ] is thus given as (refer Eqn. (4.13)
[ ]
[ ]
(E4.7)
2013
The assembled global load vector is given as
{ }
(E4.8a)
{ } { }
The complete equilibrium equation can thus be given as
[ ]{ } { }
(E4.8b)
To note, , , and are the global reactions. These are unknown because their
corresponding dofs are the essential boundary conditions which are known values. By
imposing the essential boundary conditions ( ), the assembled global
equilibrium equation is reduced to
(E4.9)
[ ]{ } { }
assembled global dofs are thus obtained as:
(E4.10a)
{ }
{ }
Eqn. (E4.10) is the assembled global solution. By inserting these values accordingly into
Eqns. (4.3a) and (4.7) the local values can then be determined (i.e. axial displacements and
local forces). However, before that, let’s determine first the global reaction forces, , ,
and which can be done by inserting Eqn. (4.10) into Eqn. (4.8b) which yields
2013
{ }
[ ]{ }
{ }
The reaction forces at support can thus obtained as:
(E4.10b)
{ } { }
Now, let’s proceed with determination of the local values. The elemental global solution for
{ } { } { } (E4.11)
{ }
{ } { } { } (E4.12)
{ }
{ } { } { } (E4.13)
{ }
Eqns. (E4.11) to (E4.13) are the elemental global solutions. In practice, local values are of
interest. To obtain local axial displacements and local internal forces, Eqns. (4.3a) and (4.7)
and are respectively reemployed.
2013
Element 1
By inserting Eqn. (E.4.11) into Eqn. (4.3a), the local dofs (axial displacements) are
{ } [ ]{ }
{ } [ ]
{ }
[ ]{ }
{ } (E4.14)
Eqn. (4.14) give the local dofs of element 1. By inserting this equation into Eq. (4.7), the local
forces can be determined as follows.
{ } [ ]{ } { }
{ } [ ]{ } { }
(E4.15)
{ }
Note that in the above, { } is equalled to zero because no distributed load acts on the
element.
Element 2
By inserting Eqn. (E4.12) into Eqn. (4.3a), the local dofs (axial displacement) is determined as
follows.
{ } [ ]{ }
{ }
(E4.16)
2013
Eqn. (E4.16) gives the local dofs of element 2. By inserting this equation into Eqn. (4.7), the
local forces can be determined as follows.
{ } [ ]{ } { }
{ } [ ]{ } { }
(E4.17)
{ }
element.
Element 3
By inserting Eqn. (E4.13) into Eqn. (4.3a), the local dofs (axial displacement) is determined as
follows.
{ } [ ]{ }
{ }
(E4.18)
local force can be determined as follows.
{ } [ ]{ } { }
{ } [ ]{ } { }
(E4.19)
{ }
element.
The obtained results are validated herein against those obtained with commercial software.
Table E4.1 compares the results of local dofs whilst Table E4.2 and Table E4.3 compare the
results of the global reactions and internal forces, respectively.
2013
Table E4.1: Validation of local dofs
Node Present (m) Software (m)

1 0 0
0 0
2
3 0 0
Table E4.2: Validation of reaction forces at support
Global reactions Present (kN) Software (kN)
Table E4.3: Validation of truss internal forces
Element Nodal Present Software

forces Internal State Internal State
Force (compression/tension) Force (compression/tension)
1 8 compression 8 compression
2 12.806 tension 12.806 tension
3 10 compression 10 compression
Based on Tables E4.1, E4.2 and E4.3, it can be concluded that the formulation agrees well
with the results produced by the commercial software. However, it must be noted that the
tabulated values are those taken from nodal locations. To have better accuracy elsewhere,
normally denser mesh (more elements) is needed for convergence.
Figure E4.2 shows the deformation of the truss system, plotted based on the obtained
results (magnified for easy viewing).
2013
Figure 4.2: Plot of deformation of plane truss
2013
4.2 FRAME SYSTEM
In the same line of argument as in a truss system, a frame system is an assembled of

inclined beam-column elements as shown in Figure 4.4. As mentioned in the introduction of
this chapter, such an element is used because in addition to the bending and shearing, a
transfer of axial forces and displacement also occur in the structure. Therefore, before we
proceed, we must first establish the formulation of beam-column element.
Figure 4.4: A plane-frame system
4.2.1 BEAM-COLUMN ELEMENTS
Due to the need to cater for the axial forces and displacement, a plane beam-column
element can be formulated by combining bar element and Euler-Bernoulli beam element
together. Thus an arrangement of a beam-column element can be given as in Figure 4.5.
Figure 4.5: Beam-column as a combination of bar and beam elements
2013
Please note the numbering system set up for the beam-column element. This is for
convenience purposes which to be seen later. Based on such an arrangement, the local dof
vector of the beam is thus given as:
{ }
(4.17)
{ }
Note also that in Eqn. (4.17), the variables are primed since they represent local values.
Based on the arrangement of the dofs, the stiffness matrix and the load vector of the beam-
column are thus given as:
(4.18)
[ ]
[ ]
and the load vector is given as:
{ }
(4.19)
{ }
where and are the axial (or bar) related load values as given by Eqn. (2.34a), and
and bending as well shearing (Euler beam) related load values as given by Eqn.
(3.48). Note that { } { }.
As in bar, the assembly of the frame system requires the local entities i.e. [ ], { }, { } of
the beam-column element to be transformed into elemental global entities. Fortunately,
the general formulations previously derived for bar elements are still applied for beam-
column element thus the elemental global dofs, { } of beam-column is given as
{ } [ ]{ } (4.20)
the elemental global load vector { } as
{ } [ ] { } (4.21)
and the elemental global stiffness matrix [ ] as
2013
[ ] [ ] [ ][ ] (4.22)
All we need now is to establish the transformation matrix for the beam-column element.
This is done next.
4.2.2 BEAM-COLUMN TRANSFORMATION MATRIX, [ ]
The transformation matrix [ ] of the beam-column can be derived by considering an

inclined beam-column as shown in Figure 4.6. It can be noticed that such a discussion is just
an extension of previously given for bar.
Figure 4.6: Inclined beam-column element.
Based on Figure 4.6, the relationship between local dofs { } and its elemental global
counterpart { } can be given as
(4.22a)
(4.22b)
(4.22c)
(4.22d)
(4.22e)
(4.22f)
2013
Eqns. (4.22) can be arranged in matrix form as;
(4.23)
[ ]{ }
{ }
or
{ } [ ]{ } (4.24)
hence the attainment of the equation previously given as Eqn. (4.20). The attainment of
Eqn. (4.24) also yields the transformation matrix for beam-column element, that is;
(4.25)
[ ]
[ ]
Having established the transformation matrix, [ ] of the beam-column element, further
discussion on frame system analysis follows the same procedure previously outlined for
truss element.
EXAMPLE 4.2
In this example, a plane frame as shown in Figure E4.2 is going to be analysed. Both local
and global displacements (dofs) and nodal forces of the structure are going to be
determined.
Figure E4.2 Plane Frame with a single point load and uniform load.
2013
Based on the data given in, the local stiffness matrix and local load vector for each element,
determined based on Eqn. (4.18) and Eqn. (4.19) are as given below. Note that, the external
point load is inserted directly into the assembled global load vector once everything is
assembled; an act which omits the requirement for it to be considered at a local level.
Element 1
[ ]
[ ]
(E4.12)
and
(E4.13)
{ }
To note, and are the local reactions at support for element 1 and unknown
variables (to be solved) since the corresponding displacement (essential boundary
conditions) are known.
Element 2
[ ]
(E4.14)
[ ]
and
(E4.15)
{ }
2013
To note, and are the local reactions at support for element 2 and unknown
variables (to be solved) since the corresponding displacement (essential boundary
conditions) are known.
The corresponding elemental global stiffness matrix and elemental global load vector can be
determined using Eqn. (4.22) and Eqn. (4.21) given as follows.
Element 1
[ ] [ ] [ ][ ]
[ ][ ]
[ ]
[ ][ ]
[ ]
[ ]
(E4.16)
and
{ } [ ] { }
[ ] { }
{ } { } { }
(E4.17)
Take note that the unprimed i.e. and represent the elemental reaction forces. Due to
the transformation, it can be seen that, there are possibilities for what used to be in the x-
2013
direction to be the y-direction and vice versa and what to be in one direction to be in the
other direction.
Element 2
[ ] [ ] [ ][ ]
[ ][ ]
(E4.18)
[ ]
[ ]
and
{ } [ ] { }
(E4.19)
[ ] { }
{ }
Take note that the unprimed i.e. , and represent the element reaction forces which
description is similar to the one given for element 1.
2013
stiffness matrix, [ ] is thus given as
[ ]
[ ]
(E4.20)
{ }
(E4.21a)
{ }
The complete global equilibrium equation can be given as
[ ]{ }
{ }
(E4.21b)
To note, , , and are the global reactions. These are unknown because their
corresponding dofs are the essential boundary conditions which are known values. By
2013
imposing the essential boundary conditions ( ), the assembled

global equilibrium equation is reduced to
(E4.22)
[ ]{ } { }
assembled global dofs are thus obtained as:
(E4.23a)
{ }
{ }
Eqn. (E4.23) is the assembled global solution. By inserting these values accordingly into
Eqns. (4.24) and (4.7), the local values can then be determined (i.e. local displacements and
local forces). However, before let’s determine first the global reaction forces, , ,
, ,and which can be done by inserting Eqn. (E4.23a) into Eqn. (E4.21b) which yields
{ }
[ ]{ }
{ }
2013
The global reaction forces can thus be given as
(E4.23b)
{ } { }
Now, let’s proceed with determination of the local values. The elemental global solution for
(E4.24)
{ }
{ } { }
{ }
(E4.25)
{ }
{ } { }
{ }
Eqns. (E4.24) and (E4.25) are the elemental global solutions. In practice, local values are of
interest. To obtain local internal forces and local displacements, Eqns. (4.7) and (4.20) are
respectively reemployed.
Element 1
By inserting Eqn. (E.4.24) into Eqn. (4.20), the local dofs are determined as follows.
{ } [ ]{ }
[ ]
{ } { }
2013
[ ]{ }
{ } (E4.26)
Eqn. (E4.26) gives the local dofs of element 1. By inserting this equation into Eq. (4.7), the
local force (or reaction if at support) can be determined as follows.
{ } [ ]{ } { }
(E4.27)
[ ]{ } { }
{ }
element.
Element 2
By inserting Eqn. (E4.25) into Eqn. (4.20), the local dofs (axial displacement) is determined as
follows.
[ ]{ }
{ }
2013
{ }
(E4.28)
local force (or reactions if at support) can be determined as follows.
{ } [ ]{ } { }
[ ]{ }
(E4.29)
{ }
{ } { } { }
The obtained results are validated herein against those obtained from commercial software.
Table E4.4 compares the results of local dofs whilst Table E4.5 and Table E4.6 compare the
results of global reactions and internal forces, respectively.
2013
Table E4.4: Validation of local dofs
Node Present Software

1 0 0
0 0
0
2
0
3 0 0
0 0
0 0
* since and are in radian, the values are too small thus agree actually with null
values produced by the software.
Table E4.5: Validation of reaction forces at support
Reaction Present Software
Table E4.6: Validation of internal forces (only element 1 is compared)
Element Nodal Present Software

forces
Based on Tables E4.4, E4.5 and E4.6, it can be concluded that the formulation agrees well
with the results produced by the commercial software. However, it must be noted that the
tabulated values are those taken from nodal locations. To have better accuracy elsewhere,
normally denser mesh (more elements) is needed for convergence.
2013
Figure E4.3 shows the deformation of the frame system, plotted based on the obtained
results (magnified for easy viewing).
Figure 4.3: Plot of deformation of plane frame
4.3 SOURCE CODE

Example 4.1 Truss
%Input
E=200e6; %kN/m^2
A=0.04; %m^2
L1=4; %m
L2=6.403;
L3=5;
theta1=0; %rad
theta2=pi-atan(5/4);%rad
theta3=90*pi/180; %rad
P=-10; %kN
%Create Stifness
-(A*E)/L1 (A*E)/L1];
-(A*E)/L2 (A*E)/L2];
-(A*E)/L3 (A*E)/L3];
2013
%Transformation matrix
c=cos(theta1);
s=sin(theta1);
T1=[c s 0 0;
0 0 c s];
Kglobal1=T1'*klocal1*T1;
c=cos(theta2);
s=sin(theta2);
T2=[c s 0 0;
0 0 c s];
c=cos(theta3);
s=sin(theta3);
T3=[c s 0 0;
0 0 c s];
%Assemble
Kglobal=zeros(6,6);
Kglobal(1:4,1:4)=Kglobal(1:4,1:4)+Kglobal1;
Kglobal(1:2,1:2)=Kglobal(1:2,1:2)+Kglobal3(1:2,1:2);
Fglobal=[0;0;0;P;0;0];
KG=Kglobal;
%Impose BC
Kglobal(5,:)=[];
Kglobal(:,5)=[];
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Fglobal(5,:)=[];
Fglobal(2,:)=[];
Fglobal(1,:)=[];

U=Kglobal\Fglobal

U1=[0;0;U(1,1);U(2,1)];
f1=klocal1*T1*U1;

U2=[U(1,1);U(2,1);0;U(3,1)];
f2=klocal2*T2*U2;

U3=[0;0;0;U(3,1)];
f3=klocal3*T3*U3;
2013
% Reaction
UG=[0;0;U(1,1);U(2,1);0;U(3,1)];
R=KG*UG
Example 4.2 Plane Frame
%Input
E=200e6; %kN/m^2
b=0.2; %m
h=0.2; %m
A=b*h; %m^2
I=(b*h^3)/12; %m^4
L1=5;
L2=4;
theta1=90*pi/180; %rad
theta2=0; %rad
P=10; %kN
q=-5; %kNm
%Create Stifness
klocal1=[(A*E/L1) 0 0 -(A*E/L1) 0 0;
0 (12*E*I/L1^3) (6*E*I/L1^2) 0 -(12*E*I/L1^3) (6*E*I/L1^2);
0 (6*E*I/L1^2) (4*E*I/L1) 0 -(6*E*I/L1^2) (2*E*I/L1);
-(A*E/L1) 0 0 (A*E/L1) 0 0;
0 -(12*E*I/L1^3) -(6*E*I/L1^2) 0 (12*E*I/L1^3) -(6*E*I/L1^2);
0 (6*E*I/L1^2) (2*E*I/L1) 0 -(6*E*I/L1^2) (4*E*I/L1)];
klocal2=[(A*E/L2) 0 0 -(A*E/L2) 0 0;
0 (12*E*I/L2^3) (6*E*I/L2^2) 0 -(12*E*I/L2^3) (6*E*I/L2^2);
0 (6*E*I/L2^2) (4*E*I/L2) 0 -(6*E*I/L2^2) (2*E*I/L2);
-(A*E/L2) 0 0 (A*E/L2) 0 0;
0 -(12*E*I/L2^3) -(6*E*I/L2^2) 0 (12*E*I/L2^3) -(6*E*I/L2^2);
0 (6*E*I/L2^2) (2*E*I/L2) 0 -(6*E*I/L2^2) (4*E*I/L2)];
%calculate equilvalent nodal forces

flocal1=[0;0;0;0;0;0];
flocal2=[0;(q*L2/2);(q*L2^2/12);0;(q*L2/2);-(q*L2^2/12)];
%Transformation matrix
c=cos(theta1);
s=sin(theta1);
T1=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
Fglobal1=T1'*flocal1;
c=cos(theta2);
s=sin(theta2);
T2=[c s 0 0 0 0;
2013
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
Fglobal2=T2'*flocal2;
%Assemble
Kglobal=zeros(9,9);
KG=Kglobal;
Fglobal=zeros(9,1);
Fglobal(1:6,1)=Fglobal(1:6,1)+Fglobal1;
Fglobal(4:9,1)=Fglobal(4:9,1)+Fglobal2;
FG=Fglobal;
Fglobal(4,1) =Fglobal(4,1)+P;
%Impose BC
Kglobal(9,:)=[];
Kglobal(:,9)=[];
Kglobal(8,:)=[];
Kglobal(:,8)=[];
Kglobal(7,:)=[];
Kglobal(:,7)=[];
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Fglobal(9,:)=[];
Fglobal(8,:)=[];
Fglobal(7,:)=[];
Fglobal(2,:)=[];
Fglobal(1,:)=[];
%Solve
U=Kglobal\Fglobal

u1=T1*[0;0;U(1,1);U(2,1);U(3,1);U(4,1)];
f1=klocal1*u1;

u2=T2*[U(2,1);U(3,1);U(4,1);0;0;0];
% Reaction
UU=[0;0;U(1,1);U(2,1);U(3,1);U(4,1);0;0;0];
R=KG*UU-FG
2013
CHAPTER 5
EIGENPROBLEMS: FREE VIBRATION AND BUCKLING
5.1 INTRODUCTION
In previous chapters, we have been dealing with structures which are loaded in the direction
of the degree of freedoms; point loads (either equivalent or nodal loads) in the direction of
the translational dof or applied moment in the direction of the rotational dof. Herein, we
are going to discuss a quite different situation that is, the deformation which is
“ungoverned’ by loads that are acting on the structure.
5.1 “UNGOVERNED BY THE LOADS” AND EIGENPROBLEMS
The word “ungoverned”, however, requires further elaboration. The word refers to the
statement that the load vector, { } or { } is set to zero. Mathematically this means solving
only the homogenous part of the differential equation of the problem. The resulting values
would be some properties of the structure and their corresponding deformation.
For example, free vibration refers to the “vibration” of a structure which is described by the
natural frequencies and the corresponding mode shapes, without any explicit consideration
on the type of external loading. Only during the discussion of resonance, these frequencies
would then be compared with the incoming frequencies (external loads).
Same goes to the discussion of buckling of a structure. The load that would cause the
buckling will be specific critical values of compressive axial force termed as buckling loads.
As will be seen, since Euler-Bernoulli beam formulation able to capture such a phenomenon
and determine the buckling loads and their corresponding buckling modes without the need
to introduce axial dofs, it should be obvious that the bucking loads and their modes are not
governed by the external applied loads.
Since both values (natural frequencies, buckling loads) are not governed by the loads, they
must be some properties of the structure hence the name “eigen” which means “inherent”
or “characteristic” in German. Then, what governs their values? Something must affect their
values, must not they? As some properties, they are governed by other properties of the
structures; material and geometrical properties. In our discussion on bar, beam and their
inclined elements, these would be Young’s modulus, , cross-sectional area, , second
moment of area, and element's length, .
Also, since these values are “ungoverned’ by the loads, we will see that the FEM formulation
for both problems will involve with the discretization of their differential equations without
the forcing terms hence the setting up of load vector, { } or { } to zero.
2013
Accordingly, all physical problems that fall under the same argument are called
eigenproblems as their equilibrium equations can all be arranged into a standard
mathematical statement. If [ ] and [ ] are square matrices with known coefficients and is
an unknown constant, an eigenproblem is a problem that can be described by the following
typical mathematical statement;
[ ] [ ] { } (5.1)
where is termed as eigenvalue and { } is termed as eigenvector.
For free vibration analysis, matrix [ ] represents the stiffness matrix, [ ], matrix [ ]
represents the mass matrix, [ ], constant represents the square of natural frequencies
and vector { } represents the vector of mode shape dofs { ̂ }.
For buckling problem, matrix [ ] represents the stiffness matrix, [ ], matrix [ ] represents
the stress stiffness matrix [ ], constant represents the buckling load, and vector { }
represents the buckling modes. Table 5.1 summarizes these.
Table 5.1: Eigenproblem grouping
Free Vibration Buckling

[ ] [ ] [ ]
[ ] [ ] [ ]
{ } { ̂} { }
[ ] [ ] { } [ ] [ ] { } [ ] [ ] { }
Besides having a typical statement, eigenproblem also has a typical argument for the
solution that is;
“The solution of Eqn. (5.1) has two solutions, the trivial solution when { } =0 and the non-
trivial solution when the determinant of the coefficient matrices is zero i.e. [ ] [ ]
. Since a trivial solution would also mean that there will be no deformation, a non-trivial
solution is therefore the solution of interest.”
Since the non-trivial solution is of interest, the setting up of
|[ ] [ ]| (5.2)
will result in a characteristic polynomial. If the coefficient matrices are matrix, the
resulting characteristic polynomial will be in the forms;
(5.3)
where , , and are constants resulted from the process of setting up the
determinant to zero. For matrices, the resulting polynomial would be
2013
(5.4)
The four roots of Eqn. (5.3) and the roots of Eqn. (5.4) are thus the eigenvalues for the
problem. For free vibration, these would be the natural frequencies and for buckling
problem, these would be the buckling loads. Inserting one of the roots (or eigenvalues) back
into Eqn. (5.1) will allow the eigenvector for that particular eigenvalue to be solved hence
determined. For free vibration, the eigenvector is the mode shape associated with the
inserted natural frequency and for buckling problem, the eigenvector is the buckling mode.
However it must be noted that the whole process is lacking one equation. Say, coefficient
matrices of Eqn. (5.1) are matrix. This means vector { } would consist of four
unknowns i.e. , , and and with as another unknown. In this situation, we have
five unknowns in total with only four simultaneous equations. There are two ways to
proceed;
i) Employ continuity equation in providing the extra equation

ii) Solve { } by setting any of the dofs as unity
In the next section, we are going to discuss all these in more detail by focusing on the FEM
formulation of free vibration of bar and beam followed by a discussion on the buckling
analysis of beam. Formulation for plane frame is given last.
A.5.2 FREE VIBRATION OF BAR
Free vibration analysis of a bar concerns the determination of the value of the natural
frequencies of the bar and their corresponding mode shapes. Such information is important
especially in the consideration of resonance; a large deformation induced by an external
load (which can be small in magnitude if compared to the ‘strength’ of the bar) which
frequency is close to the natural frequency of the bar.
The discussion on free vibration of bar requires the derivation of the bar’s partial differential
equation subjected to time-dependent external loading. Whilst the homogenous part of the
PDE will be solved herein hence the term “free vibration”, the complete PDE will be solved
in the next chapter where we consider FE formulation for dynamic problem hence the form
“forced vibration” therein.
A.5.2.1 DERIVATION OF BAR’S PARTIAL DIFFERENTIAL EQUATION OF MOTION
The derivation of the bar’s PDE of motion is basically similar to the one outlined in Chapter
2, except that
i) The external loading is time-dependent

ii) Due to the time dependency of the load, it would set the beam in motion. Based on
D’Alembert principle, this motion is resisted by an intensity of inertial force of
2013
where represents the acceleration of the bar and is the mass per unit length of
the bar since is the density and is the cross-sectional of the bar respectively.
Having established the above, a typical bar configuration with time dependent loading is
given in Figure A5.1a) while Figure A5.1b) shows the corresponding differential element.
Figure A5.1: Bar differential element with inertial force
Based on Figure A5.1b), the equilibrium of forces in the axial direction is employed:
∑ (A5.1)
which yield
(A5.2)
where is the time-varying axially distributed external loading and is the axial
displacement.
Eqn. (A5.2) can be arranged into
(A5.3)
2013
From Hooks Law, we know that
(A5.4)
and since
(A5.5)
By differentiating Eqn. (A5.5) once and inserting into Eqn. (A5.3), we obtain
(A5.6)
Eqn. (A5.6) is the PDE of motion of the bar in terms of axial displacement as required. It
describes the behaviour of the bar when subjected to time-varying loading (or time-varying
force boundary conditions). The discussion on its solution will be given in the next chapter
where therein, we will discuss the topic of dynamic. However, herein, as mentioned, since
we are dealing with eigenproblem, we will discuss the solution of the homogenous part of
the equation thus;
(A5.7)
Since an eigenproblem does not involve “loading”, Eqn. (A5.7) does not require any
specification of the force (natural) boundary conditions. All it needs are the essential
boundary conditions which are given as
| (A5.8)
| (A5.9)
Having established the bar’s vibration PDE, we are now all set to discretize the equation by
FEM.
A5.2.2 DISCRETIZATION OF BAR’S EIGENPROBLEM BY GALERKIN METHOD
Like previous elements, the discretization process begins with the provision of the
interpolation function (for liner bar element)
(A5.10a)
or in component forms as
(A5.10b)
2013
where and are the shape functions and degree of freedoms as previously given in
Chapter 2 for linear bar element (Eqn. (2.13d)).
By inserting Eqn. (A5.10a) into Eqn. (A5.7) gives:
(A5.11a)
( ) ( ) (A5.11b)
Having established the PDE in terms of shape functions and dofs, we then multiply Eqn.
(A5.11b) with weight functions, consecutively and integrate the inner product so as to
obtain the discretised equation.
Thus
( ) ( ) (A5.12)
∫ ( )
Next, we conduct integration by parts (IBP) to the first term. It must be noted that such
integration is no longer conducted to induce naturally the force boundary conditions as this
is an eigenproblem but to optimize on the continuity relaxation only. Also, no IBP is
conducted to the second term because the term is not a spatial derivative but a time
derivative. In fact, the time derivative term requires further elaboration which will be given
next. But until then, let’s conduct IBP to Eqn. (A5.12) so as to obtain
( ) ( ) (A5.13)
∫ ∫
Whilst the first integral term is familiar to us from the discussion of bar in Chapter 2, the
second term, as mentioned, deserves further explanation. In this term, the shape functions,
is taken out from the derivative since it will not involve with the differentiation due to
the fact that the shape functions are not function of time, but function of only. But
something must be a function of time; else the second term would vanish. This leaves us
with nothing but the degree of freedoms, . So, must be a function of time, thus must
involve in the differentiation as shown in Eq. (A5.13). Now, by expressing the time second
derivative by a double dot, ̈ , Eq. (A5.13) can be given in matrix forms as
[ ]{ } [ ]{ ̈ } (A5.14)
where [ ] is the stiffness matrix of the bar (similar to Eqn. (2.31b)) and [ ] is termed as the
equivalent mass matrix of the beam. These matrices can be given as
(A5.15a)
[ ] ∫
2013
(A5.15b)
[ ] ∫
To note, [ ] given by Eqn. (A5.15b) is termed as equivalent mass matrix as it is derived by
the integration of shape functions in the same sense as the equivalent nodal loads hence
the term “equivalent”. An alternative is to simply lump the mass at the nodal location to
result in a lumped mass matrix. Due to its simplicity, lumped mass matrix is not discussed
herein.
For demonstration purpose the derivation of is shown below.
(A5.16)
∫
By inserting and of Eqn (2.13d), Eqn. (A5.16) will become
(A5.17)
∫ ( )( )
By conducting the integration to Eqn. (A5.17), is obtained as
(A5.18)
The complete integrated values of [ ] for a bar element can thus be given as
(A5.19)
[ ] [ ]
A5.2.3 EIGENPROBLEM STATEMENT FOR FREE VIBRATION OF BAR
Now, as mentioned, to derive the eigenproblem statement for the bar problem, we must
express { } as a function of time and we can do this by employing separation of variable
which gives
{ } { ̂} (A5.20)
Based on Eqn. (A5.20) we can say that in time dimension, our { } vary in a harmonic manner
with amplitudes { ̂} and frequencies, . Inserting Eqn. (A5.20) into Eqn. (A5.14) and by
conducting the differentiation in time twice would give
[ ]{ ̂} [ ]{ ̂} (A5.21)
By collecting and by expressing (the latter is to suit earlier definition), Eqn. (A5.21)
can be given as
[ ] [ ] { } (A5.22)
which is in a similar form as Eqn. (5.1) and as given in Table 5.1.
As mentioned, Eqn. (A5.22) has two solutions; trivial and non-trivial. The latter is of interest
to us and this is when the determinant of the multiplying matrix is set to zero;
2013
|[ ] [ ]|=0 (A5.23)
Once essential boundary conditions are imposed, Eqn. (A5.23) will yield a polynomial
characteristic equation in the forms of Eqn. (5.4) which roots would be the eigenvalues,
mathematically and square of the natural frequencies, physically. Consecutive insertion of
these roots back into Eqn. (A5.23) gives the corresponding eigenvector or mode shape; the
former is a mathematical terms whilst the latter is a physical terms.
However, with Matlab, this process can automatically be carried out using the built-in
function eig(). All we need to do is to establish stiffness matrix, [ ] and mass matrix, [ ],
assemble them and impose the essential boundary conditions. The rest of the process will
be taken care by the built-in function.
An assembled equation would be
|[ ] [ ]| (A5.24)
which the assembled nature is highlighted by the use of capital letters. The assembly
process is similar to the one outlined in Chapter 4.
5.2 FREE VIBRATION OF BEAM
Free vibration analysis of a beam concerns the determination of the value of the natural
frequencies of the beam and their corresponding mode shapes. Such information is
important especially in the consideration of resonance; a large deformation induced by an
external load (which can be small in magnitude if compared to the ‘strength’ of the beam)
which frequency is close to the natural frequency of the beam.
The discussion on free vibration of beam requires the derivation of the beam’s partial
differential equation subjected to time-dependent external loading. Whilst the homogenous
part of the PDE will be solved herein hence the term “free vibration”, the complete PDE will
be solved in the next chapter where we consider FE formulation for dynamic problem hence
the form “forced vibration” therein.
5.2.1 DERIVATION OF BEAM’S PARTIAL DIFFERENTIAL EQUATION OF MOTION
The derivation of the beam’s vibration partial differential equation (PDE) is basically similar
to the one outlined in Chapter 3, except that
iii) The external loading is time-dependent

iv) Due to the time dependency of the load, it would set the beam in motion. Based on
D’Alembert principle, this motion is resisted by an intensity of inertial force of
where represents the acceleration of the beam and is the mass per unit length
of the beam since is the density and is the cross-sectional of the beam respectively.
2013
Having established the above, a typical beam configuration with time dependent loading is
given in Figure 5.1a) while Figure 5.1b) shows the corresponding differential element.
Figure 5.3: Beam differential element with inertial force
Based on Figure 5.1b), the following equilibrium of forces are employed:
∑ (5.5)
∑ (5.6)
∑ (5.7)
which yield
(5.8)
( )
(5.9)
( )
where
is the time dependent distributed transverse external loading
is the deflection of the beam
2013
(5.10)
(5.11)
To arrive at Eqn. (5.11), higher order terms have been assumed as insignificant thus
omitted. By differentiating Eqn. (5.11) once and then inserting it into Eqn. (5.10) give
(5.12)
Eqn. (5.12) is the differential equation of the beam in terms of both moment force and
displacement. But since our FEM discussion focuses on displacement-based formulation, we
need to express the PDE completely in terms of displacement. This can be done by
employing the constitutive equation which relates the curvature of the beam with the
moment force given as
(5.13)
(5.14)
Eqn. (5.14) is the PDE of the beam in terms of displacement as required. It describes the
behaviour of the beam when subjected to time-varying loading (or time-varying force
boundary conditions). The discussion on its solution will be given in the next chapter where
therein, we will discuss the topic of dynamic. However herein, as mentioned, since we are
dealing with eigenvalue problems, we will discuss the solution of the homogenous part of
the equation only thus
(5.15)
Since eigenproblems do not involve with “loading”, Eqn. (5.15) does not require any
specification of the force (natural) boundary condition. All it needs are the essential
boundary conditions which can be a single or a combination of the following;
(5.16)
|
2013
(5.17)
|
| (5.18)
| (5.19)
where , are the specified end rotations and , are the specified end transverse
displacements. Having established the beam’s vibration PDE, we are now all set to discuss
its discretization by FEM.
5.2.2 DISCRETIZATION BEAM’S EIGENPROBLEM BY GALERKIN METHOD
interpolation function
(5.21a)
(5.21b)
Chapter 3 for Euler-Bernoulli beam.
By inserting Eqn. (5.21a) into Eqn. (5.15) gives:
(5.22a)
( ) ( ) (5.22b)
(5.22b) with weight functions, consecutively and integrate the inner product so as to obtain
the discretised equation.
Thus
( ) ( ) (5.23)
∫ ( )
Next, we conduct integration by parts (IBP) to the first term. It must be noted that such
integration is no longer conducted to induce naturally the force boundary conditions as this
is an eigenproblem but to optimize on the continuity relaxation only. Also, no IBP is
2013
derivative. In fact, the time derivative term requires further elaboration which will be given
next. But until then, let’s conduct IBP to Eqn. (5.23) so as to obtain
( ) ( ) (5.24)
∫ ∫
Whilst the first integral term is familiar to us from the discussion of beam bending in
Chapter 3, the second term, as mentioned, deserves further explanation. In this term, the
shape functions, is taken out from the derivative since it would not involve with the
differentiation due to the fact that the shape functions are not function of time, but
function of only. But something must be a function of time else the second term would
vanish. This leaves us with nothing but the degree of freedoms, . So, must be a function
of time, thus must involve in the differentiation as shown in Eq. (5.24). Now, by expressing
the time second derivative by a double dot, ̈ , Eq. (5.24) can be given in matrix forms as
[ ]{ } [ ]{ ̈ } (5.25)
where [ ] is the stiffness matrix of the beam (similar to Eqn. 3.47) and [ ] is termed as the
equivalent mass matrix of the beam. These matrices can be given as
(5.26a)
[ ] ∫
(5.26b)
[ ] ∫
To note, [ ] as given by Eqn. (5.26b) is termed as equivalent mass matrix as it is derived by
the integration of shape functions in the same sense as the equivalent nodal loads hence
the term “equivalent”. An alternative is to simply lump the mass at the nodal location to
result in a lumped mass matrix. Due to its simplicity, lumped mass matrix is not discussed
herein.
For demonstration purpose the derivation, of is shown below.
(5.26c)
∫
By inserting and of Eqn (3.25), Eqn. (5.26) will become
(5.26d)
∫ ( )( )
By conducting the integration to Eqn. (5.26d), is obtained as
(5.26e)
The complete integrated values of [ ] for a beam element can thus be given as
2013
(5.26f)
[ ] [ ]
5.2.3 EIGENPROBLEM STATEMENT FOR FREE VIBRATION OF BEAM
Now, as mentioned, to derive the eigenproblem statement for the problem, we must
express { } as a function of time and we can do this by employing separation of variable
which gives
{ } { ̂} (5.27)
Based on Eqn. (5.27) we can say that in time dimension, our { } vary in a harmonic manner
with amplitudes { ̂ } and frequencies, . Inserting Eqn. (5.27) into Eqn. (5.25) and by
conducting the differentiation in time twice would give
[ ]{ ̂ } [ ]{ ̂ } (5.28)
By collecting and by expressing (the latter is to suit earlier definition), Eqn. (5.28)
can be given as
[ ] [ ] { ̂} (5.29)
which is in a similar form as Eqn. (5.1) and as given in Table 5.1.
As mentioned, Eqn. (5.29) has two solutions; trivial and non-trivial. The latter is of interest
to us and this is when the determinant of the multiplying matrix is set to zero;
|[ ] [ ]|=0 (5.30)
Once essential boundary conditions are imposed, Eqn. (5.30) will yield a polynomial
characteristic equation in the forms of Eqn. (5.4) which roots would be the eigenvalues,
mathematically and square of the natural frequencies, physically. Consecutive insertion of
these roots back into Eqn. (5.29) gives the corresponding eigenvector or mode shape; the
former is a mathematical terms whilst the latter is a physical terms.
function eig(). All we need to do is to establish stiffness matrix, [ ] and mass matrix, [ ],
assemble them and impose the essential boundary conditions. The rest of the process will
be taken care by the built-in function.
An assembled equation would be
|[ ] [ ]| (5.30a)
which is highlighted by the use of capital letters. The assembly process is similar to the one
outlined in Chapter 4. We are going to employ the whole procedure hence the formulation
in the next example.
2013
EXAMPLE 5.1
Figure E5.1 shows a cantilever beam and its properties. Herein, the beam’s natural
frequencies and the corresponding mode shapes are of interest. The results are validated
against a closed form solution.
Figure E5.1: Cantilever beam
matrix and local mass matrix, thus,
(E5.1)
[ ]
and
(E5.2)
[ ] [ ] [ ]
The assembled global mass matrix can be given as:
[ ]
(E5.3)
[ ]
and
[ ] (E5.4)
[ ]
Using Eqn. (5.30a) the eigenproblem statement can be given as
2013
|
[ ] (E5.5)
|
=0
|
[ ]
|[ ]
[ ] |=0 (E5.6)
By solving Eqn. (E5.6) using Matlab command “ [ ] [ ] ”, (source code 5.1), the values
of are thus obtained as:
{ } (E5.7)
Since , thus the frequency, can be obtained by
√ { } (E5.8)
Table 5.2: Comparison of values between FEM and exact solution
(rad/sec) FEM Exact

2 Element 4 Element 8 Element
1st Mode 20.592 20.582 20.582 20.582
2nd Mode 130.077 129.133 128.993 128.956
The corresponding mode shapes are shown in Figure E5.1a.
2013
Figure E5.1a: Mode shapes of beam’s free vibration
5.3 BUCKLING OF BEAM
Beam buckling is a phenomenon of instability. The phenomenon is characterized by the out-

of-plane deformation; a deformation that is away from the plane in which the buckling load
is acting.
The analysis of buckling of structures, on the other hand, refers to the determination of
buckling (critical) load, and its corresponding mode shape, { }. If the compressive load
within a structure equals this buckling load, buckling will occur. In the context of ultimate
limit state design, occurrence of buckling is considered as premature failure as it prevents
the attainment of the ultimate resistances of the structure.
The determination of buckling load and its corresponding mode will be done in the same
manner as outlined previously for free vibration of beam as the analysis falls under the
same category of eigenproblem. By referring to Table 5.1, it can be seen that the process
requires the derivation of the stiffness matrix [ ] and stress stiffness matrix [ ] which can
be done by discretising the governing differential equation. Therefore, for better
understanding, it is best to start our discussion by deriving first the ODE of the problem; a
philosophy which has been and will always be upheld throughout this book.
Before we proceed with derivation of the ODE, it is worth to note that the stiffness matrix
[ ] will be the same as the one previously derived in Chapter 3 and Eqn. (5.26a).
5.3.1 DERIVATION OF BEAM’S DIFFERENTIAL EQUATION WITH CONSTANT AXIAL LOAD, P
The derivation of the ODE is basically similar to the one outlined in Chapter 3, except that
this time, we introduce a couple of constant axial loads, each at each side of the differential
element as can be seen in Figure 5.4b. Take note of the constant nature of the axial load,
evident by the fact that they are not affected by the disturbance within the differential
element (or by the varying boundary conditions). This is in contrast to the bar element
2013
formulation as described in Chapter 2 where the axial load on the right side of the element
is expanded by the Taylor expansion.
Another point to consider is the deformed shape of the differential element as shown in
Figure 5.4b which was absent in our previous discussions. This consideration is necessary to
allow for the moment contributed by the axial load. By considering the geometry, if the
total moment is taken about point and if the angle is assumed small, the axial load on the
right side of the differential element would have a lever arm of and would be
contributing a moment of .
Having established all the relevant differences, we are all set to derive the ODE for the
problem.
Figure 5.4: Beam differential element with constant axial load
Based on Figure 5.4b), the following equilibriums of forces are employed:
∑ (5.31)
∑ (5.32)
∑ (5.33)
which yield
(5.34)
(5.35)
2013
(5.36)
(5.37)
To arrive at Eqn. (5.37), higher order terms have been assumed as insignificant thus
omitted. By differentiating Eqn. (5.37) once then inserting it into Eqn. (5.36) gives
(5.38)
Eqn. (5.38) is the differential equation of the beam in terms of both moment force and
displacement. But since our FEM discussion focuses on displacement-based formulation, we
need to express the PDE completely in terms of displacement. This can be done by
employing the constitutive equation which relates the curvature of the beam with the
moment force given as
(5.39)
(5.40)
Eqn. (5.40) is the ODE for the bending problem of a beam with constant axial load and is yet
a buckling statement. To get the latter, under the argument of eigenproblem, we will
discuss the solution of the homogenous part of the equation only, thus
(5.41)
where now is treated as unknown and termed as the buckling load. Similar to free
vibration of beam, since eigenproblems do not involve with “loading”, Eqn. (5.41) does not
require any specification of the force (natural) boundary condition. All it needs are the
essential boundary conditions which similar to those described for free vibration of beam
(Eqns. (5.16) to (5.19)).
Having established the beam’s bucking ODE, we are now all set to discuss its discretization
by FEM.
2013
5.3.2 DISCRETIZATION OF BEAM’S BUCKLING PROBLEM BY GALERKIN METHOD
Like in free vibration of beam, the discretization process begins with the provision of the
interpolation function
(5.42a)
(5.42b)
(5.43a)
( ) ( ) (5.43b)
Having established the ODE in terms of shape functions and dofs, we then multiply Eqn.
Thus
( ) ( ) (5.44)
∫ ( )
Next, we conduct integration by parts (IBP) to both terms in Eqn. (5.44); twice to the first
and once to the second. As argued for free vibration, such integrations are no longer
conducted to induce natural (force) boundary conditions as this is an eigenproblem but to
optimize on the continuity relaxation only. By conducting the IBP gives
(5.45)
∫ ∫
As can be seen, while the first integral term is familiar to us from the discussion of beam
bending in Chapter 3 which is the stiffness matrix, the second integral term is herein termed
as stress stiffness matrix.
By collecting the dofs, , Eq. (5.45) can be given in matrix forms as
[ ] [ ] { } (5.46)
where [ ] is the stiffness matrix of the beam (similar to Eqn. 3.47) and [ ] is termed as the
stress stiffness matrix of the beam. These matrices can be given as
2013
(5.47a)
[ ] ∫
(5.47b)
[ ] ∫
For demonstration purpose the derivation, of is shown below.
By inserting and of Eqn (3.25), Eqn. (5.47b) will become
(5.48)
( ) ( )
∫ ∫
By conducting the the differentiation, Eq. (5.48) becomes
(5.49)
∫ ( )( )
Finally, by conducting the integration, is numerically obtained as
(5.50)
The complete integrated value of the stress stiffness matrix, [ ] for a beam element can
thus be given as
(5.51)
[ ] [ ]
5.3.3 EIGENPROBLEM STATEMENT FOR BUCKLING OF BEAM
There will be two solutions for Eqn. (5.46) either { } which is the trivial solution or the
following determinant is equalled to zero which is the non-trivial solution hence the solution
of interest.
|[ ] [ ]| (5.52a)
Eqn. (5.52) is the eigenproblem for the buckling problem of beam. Once essential boundary
conditions are imposed, Eqn. (5.52a) will yield a polynomial characteristic equation in the
forms of Eqn. (5.4) which roots would be the eigenvalues, mathematically and the buckling
load, physically. Consecutive insertion of these roots back into Eqn. (5.46) gives the
corresponding eigenvector or buckling mode { }; the former is a mathematical terms whilst
the latter is a physical terms.
function eig(). All we need to do is to establish stiffness matrix, [ ] and stress stiffness
2013
matrix, [ ], assemble them and impose the essential boundary conditions. The rest of the
process will be taken care by the built-function. An assembled equation would be
|[ ] [ ]| (5.52b)
outlined in Chapter 4. We are going to employ the whole procedure hence the FEM buckling
formulation in the next example.
EXAMPLE 5.2
Figure E5.2 shows a beam-column with pinned at both ends. The material properties and
the length of the element are similar with Example 5.1. The buckling loads and the buckling
modes are of interest. Results are validated against closed form solution.
Figure E5.2: Column with pinned at both ends.
Since the material and the length of element is similar, [ ] is obtain from Eqn. (E.5.3). All we
need to provide afresh is the global stress stiffness matrix. We begin by providing the local
stress stiffness matrix for each element. However like in Example 5.1, due to the
symmetrically, element 1 and element 2 would have a similar local stress stiffness matrix,
thus,
(E5.9)
[ ] [ ] [ ]
The assembled global stress stiffness matrix can be given as:
[ ] (E5.10)
[ ]
2013
Using Eqn. (5.52b) the eigenproblem statement can be given as
|
[ ] (E5.11)
|
=0
|
[ ]
|[ ]
[ ] |=0 (E5.12)
By solving Eqn. (E5.12) using Matlab command “ [ ][ ] ”, (source code 5.1), the
values of are thus obtained as:
{ } (E5.13)
Table 5.2: Comparison of values between FEM and exact solution
(N) FEM Exact

2 Element 4 Element 8 Element
st
1 Mode
2nd Mode
3rd Mode
4th Mode
2013
Figure 5.6: The first five buckling modes of a column with pinned at both ends.
5.4 EIGENPROBLEM FOR PLANE FRAME (FREE VIBRATION)
Previous sub-sections detailed the eigenproblem formulation for free vibration of bar and
beam as well as buckling of beam. For a frame system, similar to the construction of the
stiffness matrix of beam-column element (as detailed in Subsection 4.2.1 and Eqn. (4.18),
the mass matrix of the beam-column can also be obtained by superposing the beam’s mass
matrix (Eqn. (5.26f)) and the bar’s mass matrix (A5.19). This can be given as (note that, the
matrix is primed so as to distinguish them from their elemental)
(A5.53)
[ ]
[ ]
2013
Having established the beam-column’s mass matrix as given by Eqn. (A5.53) whereas the
beam column stiffness matrix is already known (as given by Eqn. (4.18)), the discussion on
free vibration of a frame system would just be an extension of what have been discussed; all
we need to do is to insert Eqns. (4.18) and (A5.53) into Eqns. (5.30) and (5.52) which are
derived for local frame and transform them to their corresponding global equation. Similar
process as outlined in Chapter 4 applies where entities are multiplied accordingly by
transformation matrix [ ] of Eqn. (4.25). Once transformed, the elemental matrices are
assembled and after the imposition of the essential boundary conditions, the equation can
be solved by using Matlab command eig() to obtain the eigenvalues (i.e. natural
frequencies) and the eigenvectors (i.e. mode shapes) for the plane frame.
Accordingly the elemental global eigenproblem statement for free vibration of beam can be
given as;
|[ ] [ ]|=0 (5.53)
where
[ ] [ ] [ ][ ] (5.54a)
[ ] [ ] [ ][ ] (5.54b)
Note that, the primed matrices, [ ] and [ ] are actually the matrices given by Eqns.
(A5.53) and (4.18), respectively so as to distinguish them from their elemental global
counterparts. The global assembled eigenproblem statement for free vibration of the frame
can thus respectively given as
|[ ] [ ]|=0 (5.57)
outlined in Chapter 4. Note that, for buckling of frame, similar procedure applies. We are
going to employ the whole procedure in the next examples.
EXAMPLE 5.3
In this example, a plane frame as shown in Figure E5.3 is going to be analysed. The natural
frequencies and the corresponding mode shapes are of interest. Results are validated
against Cook et. al (2002).
2013
Figure E5.3 Plane Frame structure and its properties.
Based on the data given in, due to the symmetrically, element 1 and element 2 would have
a similar local mass matrix and local mass matrix, thus,
Element 1 and 2
[ ] [ ]
[ ]
(E5.14)
[ ] [ ]
[ ]
(E5.15)
Element 3 and element 4 also would have a similar local mass matrix and local mass matrix
due to the symmetrically, thus,
Element 3 and 4
[ ] [ ]
[ ]
(E5.16)
2013
[ ] [ ]
[ ]
(E5.17)
The corresponding elemental global stiffness matrix and global mass matrix can be
determined using Eqn. (5.54a) and Eqn. (5.54b) given as follows.
Element 1 and 2
[ ] [ ] [ ] [ ][ ]
[ ][ ]
[ ]
[ ]
(E5.18)
and
[ ] [ ] [ ] [ ][ ]
[ ][ ]
[ ] (E5.19)
[ ]
Element 3 and 4
[ ] [ ] [ ] [ ][ ]
2013
[ ][ ]
[ ]
[ ]
(E5.20)
and
[ ] [ ] [ ] [ ][ ]
[ ][ ]
[ ]
(E5.21)
[ ]
stiffness matrix[ ], and global mass matrix, [ ] is thus given as
2013
|
[ ]
|
[ ]
(E5.22)
2013
By imposing the essential boundary conditions ( ), the assembled global equilibrium equation is reduced to
|
[ ]
(E5.23)
|
[ ]
2013
By solving Eqn. (E5.23) using Matlab command “ [ ] [ ] ”, (source code 5.3), the
values of are thus obtained as:
(E5.24)
{ }
Since , thus the frequency, can be obtained by
(E5.25)
√ in rad/sec
{ }
Or
(E5.26)
in Hz
{ }
Table 5.2: Comparison of values between FEM and Cook et. al. (2002) (50 Elements)
1st Mode 2nd Mode 3rd Mode 4th Mode 5th Mode
FEM
Cook et. al
The corresponding mode shapes of the plane frame are shown in Figure E5.3a.
2013
Figure E5.3a: Mode shapes of plane frame (continues)
2013
Figure E5.3a: Mode shapes of plane frame (continued)
2013
SOURCE CODES
5.1 Free Vibration
%Input
E=200e6; %N/m^2
b=0.2; %m
h=0.2; %m
A=b*h; %m^2
I=(b*h^3)/12; %m^4
rho=76; %kg/m^3
L1=2; %m
L2=2; %m
% Create Stifness
(6*E*I/L1^2) (4*E*I/L1) -(6*E*I/L1^2) (2*E*I/L1);
-(12*E*I/L1^3) -(6*E*I/L1^2) (12*E*I/L1^3) -(6*E*I/L1^2);
(6*E*I/L1^2) (2*E*I/L1) -(6*E*I/L1^2) (4*E*I/L1)];

(6*E*I/L2^2) (4*E*I/L2) -(6*E*I/L2^2) (2*E*I/L2);
-(12*E*I/L2^3) -(6*E*I/L2^2) (12*E*I/L2^3) -(6*E*I/L2^2);
(6*E*I/L2^2) (2*E*I/L2) -(6*E*I/L2^2) (4*E*I/L2)];
% Calculate Consistent Mass Matrix

m1=rho*A*L1;
kmass1=(m1/420)*[156 22*L1 54 -13*L1;
22*L1 4*L1^2 13*L1 -3*L1^2;
54 13*L1 156 -22*L1;
-13*L1 -3*L1^2 -22*L1 4*L1^2];
m2=rho*A*L2;
kmass2=(m2/420)*[156 22*L2 54 -13*L2;
22*L2 4*L2^2 13*L2 -3*L2^2;
54 13*L2 156 -22*L2;
-13*L2 -3*L2^2 -22*L2 4*L2^2];
% Assemble
Kglobal=zeros(6,6);
KG=Kglobal;
KMASS=zeros(6,6);
KMASS(1:4,1:4)=KMASS(1:4,1:4)+kmass1;
KMASS(3:6,3:6)=KMASS(3:6,3:6)+kmass2;
%Impose BC
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
KMASS(2,:)=[];
2013
KMASS(:,2)=[];
KMASS(1,:)=[];
KMASS(:,1)=[];
% Solve Eigenvalue
[eigVe,eigVal]=eig(Kglobal,KMASS);
eigValue=diag(eigVal)
Freq = sqrt(eigValue) % circular freq (rad/sec)
fcycle = Freq/(2*pi); % cyclic freq (Hertz; cycles/sec)
5.2 Buckling
clc;clear all;
%Input
E=200e6; %N/m^2
b=0.2; %m
h=0.2; %m
A=b*h; %m^2
I=(b*h^3)/12; %m^4
L1=2; %m
L2=2; %m
P=1; %N
%Calculate Stifness
(6*E*I/L1^2) (4*E*I/L1) -(6*E*I/L1^2) (2*E*I/L1);
-(12*E*I/L1^3) -(6*E*I/L1^2) (12*E*I/L1^3) -(6*E*I/L1^2);
(6*E*I/L1^2) (2*E*I/L1) -(6*E*I/L1^2) (4*E*I/L1)];

(6*E*I/L2^2) (4*E*I/L2) -(6*E*I/L2^2) (2*E*I/L2);
-(12*E*I/L2^3) -(6*E*I/L2^2) (12*E*I/L2^3) -(6*E*I/L2^2);
(6*E*I/L2^2) (2*E*I/L2) -(6*E*I/L2^2) (4*E*I/L2)];
%Calculate Consistent Mass Matrix

kstress1=(P/(30*L1))*[36 3*L1 -36 3*L1;
3*L1 4*L1^2 -3*L1 -L1^2;
-36 -3*L1 36 -3*L1;
3*L1 -L1^2 -3*L1 4*L1^2];
kstress2=(P/(30*L2))*[36 3*L2 -36 3*L2;

3*L2 4*L2^2 -3*L2 -L2^2;
-36 -3*L2 36 -3*L2;
3*L2 -L2^2 -3*L2 4*L2^2];
%Assemble
Kglobal=zeros(6,6);
KSTRESS=zeros(6,6);
KSTRESS(1:4,1:4)=KSTRESS(1:4,1:4)+kstress1;
KSTRESS(3:6,3:6)=KSTRESS(3:6,3:6)+kstress2;
2013
%Impose BC
Kglobal(5,:)=[];
Kglobal(:,5)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
KSTRESS(5,:)=[];
KSTRESS(:,5)=[];
KSTRESS(1,:)=[];
KSTRESS(:,1)=[];
%Solve Eigenvalue
[eigVe,eigVa]=eig(Kglobal,-KSTRESS);
% or
Val=eig(Kglobal,KSTRESS);
Pcr=Val(1,1);
5.3 Frame vibration
clc;clear all; % Clear previous data
% Input
b = 0.1; % width (m)
h = 0.1; % height (m)
A = b*h; % Area (m^2)
I = b*h^3/12; % Second moment of area (m^4)
mass = 7860; % Density (kg/m^2)
E=200e9; % Young's Modulus (Pa)
L1=1.5; % Length of Element 1 (m)
theta1=90*pi/180; % Angle of transformation Element 1 (rad)

theta3=0; % Angle of transformation Element 3 (rad)
deltaT=0.00001; % Time Step (sec)

TotalTime=10000; % total time step
% Calculate Stifness
A1=A;
I1=I;
klocal1=[(A1*E/L1) 0 0 -(A1*E/L1) 0 0;
0 (12*E*I1/L1^3) (6*E*I1/L1^2) 0 -(12*E*I1/L1^3) (6*E*I1/L1^2);
0 (6*E*I1/L1^2) (4*E*I1/L1) 0 -(6*E*I1/L1^2) (2*E*I1/L1);
-(A1*E/L1) 0 0 (A1*E/L1) 0 0;
0 -(12*E*I1/L1^3) -(6*E*I1/L1^2) 0 (12*E*I1/L1^3) -(6*E*I1/L1^2);
0 (6*E*I1/L1^2) (2*E*I1/L1) 0 -(6*E*I1/L1^2) (4*E*I1/L1)];
A2=A;
I2=I;
klocal2=[(A2*E/L2) 0 0 -(A2*E/L2) 0 0;
0 (12*E*I2/L2^3) (6*E*I2/L2^2) 0 -(12*E*I2/L2^3) (6*E*I2/L2^2);
0 (6*E*I2/L2^2) (4*E*I2/L2) 0 -(6*E*I2/L2^2) (2*E*I2/L2);
2013
-(A2*E/L2) 0 0 (A2*E/L2) 0 0;
0 -(12*E*I2/L2^3) -(6*E*I2/L2^2) 0 (12*E*I2/L2^3) -(6*E*I2/L2^2);
0 (6*E*I2/L2^2) (2*E*I2/L2) 0 -(6*E*I2/L2^2) (4*E*I2/L2)];
A3=A;
I3=I;
klocal3=[(A3*E/L3) 0 0 -(A3*E/L3) 0 0;
0 (12*E*I3/L3^3) (6*E*I3/L3^2) 0 -(12*E*I3/L3^3) (6*E*I3/L3^2);
0 (6*E*I3/L3^2) (4*E*I3/L3) 0 -(6*E*I3/L3^2) (2*E*I3/L3);
-(A3*E/L3) 0 0 (A3*E/L3) 0 0;
0 -(12*E*I3/L3^3) -(6*E*I3/L3^2) 0 (12*E*I3/L3^3) -(6*E*I3/L3^2);
0 (6*E*I3/L3^2) (2*E*I3/L3) 0 -(6*E*I3/L3^2) (4*E*I3/L3)];
A4=A;
I4=I;
klocal4=[(A4*E/L4) 0 0 -(A4*E/L4) 0 0;
0 (12*E*I4/L4^3) (6*E*I4/L4^2) 0 -(12*E*I4/L4^3) (6*E*I4/L4^2);
0 (6*E*I4/L4^2) (4*E*I4/L4) 0 -(6*E*I4/L4^2) (2*E*I4/L4);
-(A4*E/L4) 0 0 (A4*E/L4) 0 0;
0 -(12*E*I4/L4^3) -(6*E*I4/L4^2) 0 (12*E*I4/L4^3) -(6*E*I4/L4^2);
0 (6*E*I4/L4^2) (2*E*I4/L4) 0 -(6*E*I4/L4^2) (4*E*I4/L4)];

mass1=mass;
cmass1=(mass1*A1*L1/420)*[140 0 0 70 0 0;
0 156 22*L1 0 54 -13*L1;
0 22*L1 4*L1^2 0 13*L1 -3*L1^2;
70 0 0 140 0 0;
0 54 13*L1 0 156 -22*L1;
0 -13*L1 -3*L1^2 0 -22*L1 4*L1^2];
mass2=mass;
cmass2=(mass2*A2*L2/420)*[140 0 0 70 0 0;
0 156 22*L2 0 54 -13*L2;
0 22*L2 4*L2^2 0 13*L2 -3*L2^2;
70 0 0 140 0 0;
0 54 13*L2 0 156 -22*L2;
0 -13*L2 -3*L2^2 0 -22*L2 4*L2^2];
mass3=mass;
cmass3=(mass3*A3*L3/420)*[140 0 0 70 0 0;
0 156 22*L3 0 54 -13*L3;
0 22*L3 4*L3^2 0 13*L3 -3*L3^2;
70 0 0 140 0 0;
0 54 13*L3 0 156 -22*L3;
0 -13*L3 -3*L3^2 0 -22*L3 4*L3^2];
mass4=mass;
cmass4=(mass4*A4*L4/420)*[140 0 0 70 0 0;
0 156 22*L4 0 54 -13*L4;
0 22*L4 4*L4^2 0 13*L4 -3*L4^2;
70 0 0 140 0 0;
0 54 13*L4 0 156 -22*L4;
0 -13*L4 -3*L4^2 0 -22*L4 4*L4^2];
% Transformation matrix
c=cos(theta1);
2013
s=sin(theta1);
T1=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
Mmass1=T1'*cmass1*T1;
c=cos(theta2);
s=sin(theta2);
T2=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
c=cos(theta3);
s=sin(theta3);
T3=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
c=cos(theta4);
s=sin(theta4);
T4=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
% Assemble global stiffness

Kglobal=zeros(15,15);
% Assemble global mass matrix

Mglobal=zeros(15,15);
Mglobal(1:6,1:6)=Mglobal(1:6,1:6)+Mmass1;
KGG=Kglobal;
MGG=Mglobal;
2013
%Impose BC
Kglobal(14,:)=[];
Kglobal(:,14)=[];
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
Mglobal(14,:)=[];
Mglobal(:,14)=[];
Mglobal(2,:)=[];
Mglobal(:,2)=[];
Mglobal(1,:)=[];
Mglobal(:,1)=[];
% Solve Eigenvalue
[eigVector,eigValue]=eig(Kglobal,Mglobal);
eigValue=diag(eigValue);
fcycle = Freq/(2*pi) % cyclic freq (Hertz; cycles/s
2013
CHAPTER 6
DYNAMICS: FORCED VIBRATION
6.1 INTRODUCTION
Dynamic analysis is a topic which main concern is to determine structure’s responses or

motions when subjected to time-varying loading (or/and time-varying boundary conditions).
It is an act of solving the equation of motion of a problem which discretised (matrix) forms is
typically given by Eq. (). Accordingly, in this chapter, we are going to focus our discussion on
the solution of forced vibration problems for bar and beam elements as well as frame
structures. Although we have derived the discretised equation of motion of beam element
in Chapter 5, for completeness, we are going to begin this chapter with the derivation of the
discretised equation for bar element, starting from its PDE derivation.
6.2 BAR’S PDE OF MOTION
Derivation of the bar’s PDE of motion has been given in Section A5.2.1. We are not going to
repeat the procedure but to give directly the PDE as (previously given by Eqn. Eqn. (A5.6)
(6.6)
( )
However, although we have discretised this PDE in Section A5.2.2, but for completeness, we
are going to detail the discretization all over again, this time with the existence of the
forcing terms i.e. ( ). As a reminder, in previous discretization, this forcing term was
omitted because we were considering the homogenous solution only (eigenproblem).
Before we proceed, below are the natural and essential boundary conditions of the problem
(previously given by Eqns. (2.9a) to (2.9d)). However, since this is a dynamic analysis, there
is possibility for the boundary conditions to vary with time, thus
( ) (6.7a)
| ( )
( )
| ( )
(6.7b)
© AIRIL YASREEN YASSIN, PhD, ERWAN HAFIZI KASIMAN, MEng

AND MOHD AL-AKBAR NOOR, MEng Page 140
2013
| ( ) (6.8c)
| ( ) (6.9d)
Note that the sign convention for the natural (force) boundary conditions above is based on
the assumption that the boundary conditions are the reactions at support hence the
opposite directions to the internal forces. Since Eqn. (6.6) is a 2nd order ODE in , two
boundary conditions out of the four given above must be known in prior so as to have a
well-posed problem. Also, since Eqn. (6.6) consists of second order time derivatives, two
initial conditions must be known. We are going to detail on this later during the discussion
of time integration by finite difference method.
6.2.1 DISCRETIZATION OF BAR’S PDE OF MOTION BY GALERKIN METHOD
interpolation function (for liner bar element)
( ) (6.10a)
( ) (6.10b)
Chapter 2 for linear bar element (Eqn. (2.13d)).
( ) ( ) (6.11a)
( )
or component forms as
( ) ( ) (6.11b)
( )
Thus
(6.12)
( ) ( )
∫ ( ( ))

2013
Next, we conduct integration by parts (IBP) to the first term. It must be noted that, no IBP is
derivative. By conducting the IBP, we obtain
( ) ( ) (6.13)
∫ ∫ ∫ ( ) ( )
where ( )is the boundary terms (conditions) which can be either Eqn. (6.7a) or (6.7b) as
explained in detailed in Section 2.2.5 except that now they can vary with time, ( ). While
the first integral term of Eqn. (6.13) is familiar to us from the discussion of bar element
(static) in Chapter 2, the second term, as mentioned, deserves further explanation. In this
term, the shape functions, is taken out from the derivative since it would not involve with
the differentiation because shape functions are not function of time, but function of
only. But something must be a function of time else the second term would vanish. This
leaves us with nothing but the degree of freedoms, . So, must be a function of time
thus must involve in the differentiation as shown in Eq. (6.13). Now, by expressing the time
second derivative by a double dot, ̈ , Eq. (6.13) can be given in matrix forms as
[ ]{ } [ ]{ ̈ } { ( )} { ( )} { ( )} (6.14)
where [ ] is the stiffness matrix of the are and [ ] is termed as the equivalent mass matrix
of the ar. These matrices are already given by Eqns. (2.31b) and (A5.19), respectively.
Discussion on time-integration in solving Eqns. (6.14) will be discussed later. Next, we are
going to discuss the forced vibration analysis of beam.
6.3 BEAM’S PDE OF MOTION
Derivation of the beam’s PDE of motion has been given in Section 5.2.1. We are not going to
repeat the procedure but to give directly the PDE as (previously given as Eqn. (5.14));
(6.20)
( )
However, although we have discretised this PDE in Section 5.2.2, but for completeness, we
are going to detail the discretization all over again, this time with the existence of the
forcing terms i.e. ( ). As a reminder, in previous discretization, this forcing term was
omitted because we were considering the homogenous solution only (Eigenproblem).
Before we proceed, below are the natural and essential boundary conditions of the
problem. There are similar to the ones in Eqns. (3.11) to (3.18) but since this is a dynamic
analysis, there is possibility for the boundary conditions to vary with time.
The force (natural) boundary conditions of the beam as function of time can be given as
(6.21)
( )| ( )

2013
(6.22)
( )| ( )
(6.23)
( )| ( )
(6.24)
( )| ( )
where ( ), ( ) are the specified end shear forces (edge transverse point loads) and
( ), ( ) are the specified end moment forces. Note that the sign convention above is
based on the assumption that the boundary conditions are the reactions at support hence
the opposite directions to the internal forces.
On the other hand, the displacement (essential) boundary conditions of the beam can be
given as function of time as;
(6.25)
( )| ( )
(6.26)
( )| ( )
( )| ( ) (6.27)
( )| ( ) (6.28)
where ( ), ( ) are the specified end rotations and ( ), ( ) are the specified end
transverse displacements. Also, since Eqn. (6.20) consists of second order time derivatives,
two initial conditions must be known. We are going to detail on this later during the
discussion on time integration by finite difference method.
6.3.1 DISCRETIZATION OF BEAM’S PDE OF MOTION BY GALERKIN METHOD
Like previous elements, the discretization process begins with the provision of the beam’s
interpolation function which can be given as;
( ) (6.29)
( ) (6.30)
Chapter 3 for Euler-Bernoulli beam.
2013
By inserting Eqn. (6.29) into Eqn. (6.20) gives:
( ) ( ) (6.31)
or component forms as
( ) ( ) (6.32)
( )
(6.32) with weight functions, consecutively and integrate the inner product so as to obtain the
discretised equation.
Thus
(6.33)
( ) ( )
∫ ( ( ))
Next, we conduct integration by parts (IBP) to the first term. No IBP is conducted to the
second term because the term is not a spatial derivative but a time derivative. In fact, the
time derivative term requires further elaboration which will be given next. But until then,
let’s conduct IBP to Eqn. (5.23) so as to obtain
( ) ( ) (6.34)
∫ ∫ ∫ ( ) ( )
where ( )is the boundary terms (conditions) which can be any two of the four of Eqns.
(6.21) to (6.28) as explained in detailed in Section 3.3 except that now they can vary with
time, ( ). While the first integral term is familiar to us from the discussion of beam bending
in Chapter 3, the second term, as mentioned, deserves further explanation. In this term, the
shape functions, is taken out from the derivative since it would not involve with the
differentiation due to the fact that the shape functions are not function of time but function
of only. But something must be a function of time else the second term would vanish. This
leaves us with nothing but the degree of freedoms, . So, must be a function of time,
thus must involve in the differetiation as shown in Eqn. (6.34). Now, by expressing the time
second derivative by a double dot, ̈ , Eq. (6.34) can be given in matrix forms as
[ ]{ } [ ]{ ̈ } { ( )} { ( )} { ( )} (6.35)
where [ ] is the stiffness matrix of the beam and [ ] is termed as the equivalent mass
matrix of the beam. These matrices are already given by Eqns. (3.47) and (5.26f),
respectively. As can be noticed, Eqn. (6.35) is exactly in similar forms as Eqn. (6.14) derived
for bar problem. This highlights that the same time-integration algorithm can be used for
both problems, in fact for all problems, provided that they are arranged in similar forms.
Discussion on time-integration in solving Eqns. (6.14) and (6.35) will be discussed next. But,
before that, we will discuss first the analysis of forced vibration for truss and frame system.

2013
6.4 FORCED VIBRATION OF PLANE FRAME (AND TRUSS)
The discussion on forced vibration of a frame system would just be an extension of what
have been discussed; all we need to do is to insert Eqns. (4.18) and (A5.53) into Eqns. (6.35)
which are derived for local frame and transform them to their corresponding global
equation. Similar process as outlined in Chapter 4 applies where entities are multiplied
accordingly by transformation matrix [ ] of Eqn. (4.25). Once transformed, the elemental
matrices are assembled. Accordingly the elemental global discretized equation of forced
vibration of plane frame can be given as;
[ ]{ } [ ]{ ̈ } { ( )} (6.36)
where
[ ] [ ] [ ][ ] (6.37)
[ ] [ ] [ ][ ] (6.38)
{ ( )} [ ] { ( )} (6.39)
Note that, the primed matrices, [ ], [ ] and { ( )} are actually the matrices and vector
given by Eqns. (A5.53), (4.18) and (6.35), respectively. These are primed so as to distinguish
them from their elemental global counterparts. As a reminder, detailed explanations on
transformation are given in Sections 4.1 and 4.2.1.
Having established the elemental global equation, the assembled global discretized
equation of forced vibration of plane frame can thus be given as;
[ ]{ ̈ } [ ]{ } { ( )} (6.40)
which their assembled states are highlighted by the use of capital letters. The assembly
process is similar to the one outlined in Chapter 4.
By observing Eqn. (6.40), we should realize by now that the equation poses a new problem
which we have never encountered before that is, now we must deal with time second
derivative terms symbolized by the double dots i.e. { ̈ }. To solve this kind of problem, we
need time integration. This is discussed next.
6.5 TIME INTEGRATION AND THE SOLUTION OF DISCRETIZED EQUATION OF MOTION
Eqn. (6.40) involves the second derivative of the dof’s in time, expressed by { ̈ }. As a
derivative, it can be discretized, which the common practice would be, by finite difference
method. Basically there are three ways to do this;
i) Forward difference
ii) Backward difference
iii) Central difference

2013
Despite the three available techniques, herein, we will concern only with central difference
technique. Readers who are interested to know more about the other techniques can refer
to books on finite difference method and finite element method.
A central difference time discretization of the derivative terms; { ̈ } can be given as
(6.41)
{ ̈} { }
where refers to the time at which the problem is solved, is time of the previous
solution and is the solution just before . refers to time step taken.
Inserting Eqn. (6.41) into Eqn.(6.40) gives
(6.42)
[ ]{ } [ ]{ } { }
Eqn. (6.42) is the fully discretized equation of motion for the problem. As can be seen,
spatial dimension terms are discretized by FEM whilst time derivative term is discretized by
finite difference method or FDM.
It must be noted that, the vector of dof’s associates with the stiffness matrix (i.e. [ ]{ } )
has been chosen as known values; solved values of previous time, (i.e. { } ). Such a
selection is actually a matter of choice. As will be seen, this will ease the solution process
where the problem can be solved in an explicit manner (forward marching). However this
comes at a cost. Due to such a selection, despite the ease it provides, the process is not
always stable. Therefore, the process requires that the time step, to be smaller than the
critical time step, , else the analysis will “blow-up”. The critical time step can be
calculated as
(6.43)
where is the highest natural frequency of the system, obtained from eigenproblem
analysis discussed in the previous chapter.
By rearranging Eqn. (6.42) we can obtain;
(6.44)
[ ]{ } { } [ ]{ } [ ]{ } [ ]{ }
It is a common practice to simplify Eq. (6.44) further as
[ ]{ } { } [ ]{ } [ ]{ } [ ]{ } (6.45)
where and .

2013
Having established Eqn. (6.45), we are now in the position to solve the problem. The straight
forward nature of Eqn. (6.45) should be obvious to us because as we can see, all the terms
on the right hand side of the equation are known values, obtained from previous analysis
(previous time step). Having everything known on the right hand side the equation, { }
now can be solved by inversing the matrix [ ] (or by employing Gauss Elimination or any
other solvers). This is best demonstrated by examples. However, before we do that, we
must discuss the initial conditions for the problem; a compulsory discussion and provision
for all dynamic (time-dependent) problems.
6.5.1 INITIAL CONDITIONS
Examining Eqn. (6.45), we can see that, to kick start the procedure, we must provide two
known values of { } and { } . To provide these values and since we have second order
( )
derivatives in time (i.e. ), two initial conditions must be known which can either be,
displacement, { } , velocity { ̇ } or the acceleration { ̈ } . Knowing any two of these is
sufficient because the third one can be obtained consequently. As an example, for a system
at rest, { } and { ̇ } can be taken as zero (i.e. { } { ̇} . With these conditions,
{ ̈ } can be determined by solving Eqn. (6.40) as below
[ ]{ ̈ } [ ]{ } { } (6.46)
or
[ ]{ ̈ } { } (6.47)
thus
{ ̈} [ ] { } (6.48)
Based on these initial conditions, values of { } and { } at the first time step (values to
kick start the recursion), for the case of a system at rest prior to the imposition of loading,
can be given as
{ } { } (6.49)
and { } can be obtained from the following relationship
{ } { } { ̇} { ̈} (6.50)
For a system at rest, Eqn. (6.50) means
{ } { ̈} (6.51)
Having established the values of { } and { } for the first time step (to kick start the
recursion), we are now all set for some examples.

2013
EXAMPLE 6.2
Let’s solve the plane frame problem of EXAMPLE 5.3. This time, the frame is subjected to a
time-varying point load as shown in Figure E6.3
Figure E6.2 Plane Frame structure and its properties.

2013
Figure E6.3: Time-dependent loading
[ ] and [ ] are obtained from Eqn. (4.18) and (A5.53), respectively which transformation
and assembly have been discussed in EXAMPLE 5.3. All we need to do now is to provide the
global displacement and load vector. Since the external point load is inserted directly into
the assembled global load vector, the global load vector, { } can be given as
(E6.1)
{ }
{ }

2013
The complete global equilibrium can be given as
[ ]{ }
̈
̈
(E6.2)
̈
̈
̈
̈
̈
̈
̈
̈
̈
̈
̈ { }
[ ] ̈
{ ̈ }
By imposing the essential boundary conditions ( ), the assembled global equilibrium equation is reduced to

2013
[ ]{ }
̈
̈ (E6.3)
̈
̈
̈
̈
̈
̈
̈
̈
{ }
[ ] ̈
{ ̈ }

2013
for :
I. With , assume { } , thus { ̈ } can be solved using Matlab command “\”
[ ]{ ̈ } { } (E6.4)
{ }
[ ]
̈
̈
̈
̈
̈
̈
̈
̈
̈
̈
{ }
[ ] ̈
{ ̈ }

2013
̈
̈
̈
̈
̈
̈ (E6.5)
̈
̈
̈
̈
̈ { }
{ ̈ }
II. solve { }
(E6.6)
{ } { } { ̈}
{ } { }
{ } { }
III. substitute { } ,and { } into Eqn. (6.44) or (6.45)
(E6.7)
[ ]{ } { } ([ ] [ ]) { } [ ]{ }

2013
Let , , and as
[ ]
( )
[ ]
[ ]
(E6.8)

2013
[ ] [ ]
[ ]
[ ]
[ ]
(E.6.9)

2013
[ ]
( )
[ ]
[ ]
(E6.10)

ANALYSIS FOR SOLIDS AND FLUIDS (unedited 1st
draft)
IV. solve { } using Matlab command “\”, thus
{ } { } { } { } (E6.11)
{ } (E6.12)
{ }
for :
by referring Eqn. (6.11), subs { } with { } and { } with { } , yield the following form
{ } { } { } { } (E6.13)
solve for { }
{ } (E6.14)
{ }
for :
{ } { } { } { } (E6.15)
solve for { }
draft)
{ } (E6.16)
{ }
for :
{ } { } { } { } (E6.17)
solve for { }
{ } (E6.18)
{ }
Figure E6.4 shows the vibration (rotation) of the frame taken 1 m from the roller support.
The results can be validated by comparing Figure E6.4 with Figure 11.18-2(d), page 441 of
Cook et.al (2002).
draft)
Figure E6.4: vibration (rotation) of the frame taken 1 m from the roller support.
draft)
SOURCE CODE
Example 6.2
clc;clear all; % Clear previous data
% Input
b = 0.1; % width (m)
h = 0.1; % height (m)
A = b*h; % Area (m^2)
I = b*h^3/12; % Second moment of area (m^4)
mass = 7860; % Density (kg/m^2)
E=200e9; % Young's Modulus (Pa)

deltaT=0.00001; % Time Step (sec)

TotalTime=10000; % total time step
% Calculate Stifness
A1=A;
I1=I;
klocal1=[(A1*E/L1) 0 0 -(A1*E/L1) 0 0;
0 (12*E*I1/L1^3) (6*E*I1/L1^2) 0 -(12*E*I1/L1^3) (6*E*I1/L1^2);
0 (6*E*I1/L1^2) (4*E*I1/L1) 0 -(6*E*I1/L1^2) (2*E*I1/L1);
-(A1*E/L1) 0 0 (A1*E/L1) 0 0;
0 -(12*E*I1/L1^3) -(6*E*I1/L1^2) 0 (12*E*I1/L1^3) -(6*E*I1/L1^2);
0 (6*E*I1/L1^2) (2*E*I1/L1) 0 -(6*E*I1/L1^2) (4*E*I1/L1)];
A2=A;
I2=I;
klocal2=[(A2*E/L2) 0 0 -(A2*E/L2) 0 0;
0 (12*E*I2/L2^3) (6*E*I2/L2^2) 0 -(12*E*I2/L2^3) (6*E*I2/L2^2);
0 (6*E*I2/L2^2) (4*E*I2/L2) 0 -(6*E*I2/L2^2) (2*E*I2/L2);
-(A2*E/L2) 0 0 (A2*E/L2) 0 0;
0 -(12*E*I2/L2^3) -(6*E*I2/L2^2) 0 (12*E*I2/L2^3) -(6*E*I2/L2^2);
0 (6*E*I2/L2^2) (2*E*I2/L2) 0 -(6*E*I2/L2^2) (4*E*I2/L2)];
A3=A;
I3=I;
klocal3=[(A3*E/L3) 0 0 -(A3*E/L3) 0 0;
0 (12*E*I3/L3^3) (6*E*I3/L3^2) 0 -(12*E*I3/L3^3) (6*E*I3/L3^2);
0 (6*E*I3/L3^2) (4*E*I3/L3) 0 -(6*E*I3/L3^2) (2*E*I3/L3);
-(A3*E/L3) 0 0 (A3*E/L3) 0 0;
0 -(12*E*I3/L3^3) -(6*E*I3/L3^2) 0 (12*E*I3/L3^3) -(6*E*I3/L3^2);
0 (6*E*I3/L3^2) (2*E*I3/L3) 0 -(6*E*I3/L3^2) (4*E*I3/L3)];
A4=A;
draft)
I4=I;
klocal4=[(A4*E/L4) 0 0 -(A4*E/L4) 0 0;
0 (12*E*I4/L4^3) (6*E*I4/L4^2) 0 -(12*E*I4/L4^3) (6*E*I4/L4^2);
0 (6*E*I4/L4^2) (4*E*I4/L4) 0 -(6*E*I4/L4^2) (2*E*I4/L4);
-(A4*E/L4) 0 0 (A4*E/L4) 0 0;
0 -(12*E*I4/L4^3) -(6*E*I4/L4^2) 0 (12*E*I4/L4^3) -(6*E*I4/L4^2);
0 (6*E*I4/L4^2) (2*E*I4/L4) 0 -(6*E*I4/L4^2) (4*E*I4/L4)];

mass1=mass;
cmass1=(mass1*A1*L1/420)*[140 0 0 70 0 0;
0 156 22*L1 0 54 -13*L1;
0 22*L1 4*L1^2 0 13*L1 -3*L1^2;
70 0 0 140 0 0;
0 54 13*L1 0 156 -22*L1;
0 -13*L1 -3*L1^2 0 -22*L1 4*L1^2];
mass2=mass;
cmass2=(mass2*A2*L2/420)*[140 0 0 70 0 0;
0 156 22*L2 0 54 -13*L2;
0 22*L2 4*L2^2 0 13*L2 -3*L2^2;
70 0 0 140 0 0;
0 54 13*L2 0 156 -22*L2;
0 -13*L2 -3*L2^2 0 -22*L2 4*L2^2];
mass3=mass;
cmass3=(mass3*A3*L3/420)*[140 0 0 70 0 0;
0 156 22*L3 0 54 -13*L3;
0 22*L3 4*L3^2 0 13*L3 -3*L3^2;
70 0 0 140 0 0;
0 54 13*L3 0 156 -22*L3;
0 -13*L3 -3*L3^2 0 -22*L3 4*L3^2];
mass4=mass;
cmass4=(mass4*A4*L4/420)*[140 0 0 70 0 0;
0 156 22*L4 0 54 -13*L4;
0 22*L4 4*L4^2 0 13*L4 -3*L4^2;
70 0 0 140 0 0;
0 54 13*L4 0 156 -22*L4;
0 -13*L4 -3*L4^2 0 -22*L4 4*L4^2];
% Transformation matrix
c=cos(theta1);
s=sin(theta1);
T1=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
c=cos(theta2);
draft)
s=sin(theta2);
T2=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
c=cos(theta3);
s=sin(theta3);
T3=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
c=cos(theta4);
s=sin(theta4);
T4=[c s 0 0 0 0;
-s c 0 0 0 0;
0 0 1 0 0 0;
0 0 0 c s 0;
0 0 0 -s c 0;
0 0 0 0 0 1;];
% Assemble global stiffness

Kglobal=zeros(15,15);
% Assemble global mass matrix

Mglobal=zeros(15,15);
KGG=Kglobal;
MGG=Mglobal;
%Impose BC
Kglobal(14,:)=[];
Kglobal(:,14)=[];
Kglobal(2,:)=[];
Kglobal(:,2)=[];
Kglobal(1,:)=[];
Kglobal(:,1)=[];
draft)
Mglobal(14,:)=[];
Mglobal(:,14)=[];
Mglobal(2,:)=[];
Mglobal(:,2)=[];
Mglobal(1,:)=[];
Mglobal(:,1)=[];
% Solve Eigenvalue
[eigVector,eigValue]=eig(Kglobal,Mglobal);
eigValue=diag(eigValue);
fcycle = Freq/(2*pi) % cyclic freq (Hertz; cycles/sec)
%Check deltaTmax
deltaTmax=2/max(Freq);
if deltaT <=deltaTmax
display ('deltaT smaller than critical time: Pass!')
else
display ('deltaT bigger than critical time: Fail!')
end
% solve static problem for UU at t;

time=0;
P=100000;
Fglobal=zeros(12,1);
Fglobal(11,1)=P;
Acc=Mglobal\Fglobal;
% Calculate Equation Constant

a0=1/(deltaT^2);
a1=1/(2*deltaT);
a2=2*a0;
a3=1/a2;
MLHS=(a0*Mglobal);
Ut=zeros(12,1);
Utminus1=Ut+a3*(Acc);
% Saving Data
DataTime(1) =0;
DataForces(1)=P;
Result(:,1) =Ut;
% Loop over total time

for i=1:TotalTime
time=(i-1)*deltaT;
if time <0.01
P=100000;
draft)
else
P=0;
end
Fglobal(11,1)=P;
RHS=Fglobal-((Kglobal-a2*Mglobal)*Ut)-(a0*Mglobal)*Utminus1;
Utplus1=MLHS\RHS;
Utminus1=Ut;
Ut=Utplus1;
% Save Result
DataTime(i) =time;
DataForces(i) =P;
Result(:,i) =Utplus1;
end
%--------------------------------------------------------------------------
% Plot Tip Displacement vs Time
%--------------------------------------------------------------------------
% Result from Example

figure(1)
XXX1=DataTime(:);
YYY1=Result(10,:);
plot(XXX1,YYY1,'-r')
title('Y-displacement with time at dof1')
xlabel('Time (sec)'); % Title of x-axis
ylabel('Y-displacement (in)'); % Title of y-axis
LINEAR AND NONLINEAR FINITE ELEMENT FOR 2013
SOLIDS AND FLUIDS
SCALAR ELEMENT
HEAT TRANSFER
7. INTRODUCTION
This chapter marks the beginning of the discussion on continuum elements hence
continuum mechanics. Whilst previous chapters mainly discussed line elements, herein we
are going to discuss the first type of continuum element that is scalar element.
Scalar element is an element which node has only one degree of freedom (dof). It is
employed in various physical problems such as heat transfer, potential flow, groundwater
flow, electrostatics and magnetostatics. Despite its various applications, herein, for
demonstration purposes on the use of scalar element and its corresponding FEM
formulation, steady heat transfer problem is considered.
7.2 DERIVATION OF STEADY HEAT TRANSFER PARTIAL DIFFERENTIAL EQUATION
Heat transfer analysis concerns with the determination of temperature distribution within a
body due to internal heat generation and temperature differences (as well external flux,
convection and radiation) at the boundaries.
Since it is customary in this book to start a discussion from first principle, we begin our
discussion by deriving the partial differential equation (PDE) of the problem. We limit our
derivation to two-dimensional only.
The PDE can be derived by employing the conservation of energy principle to the differential
element as shown in Figure 7.1.
As shown in the figure, the dependent variables to be considered are the heat flux in x-
direction, and in y-direction, . Similar to previous chapters, the variables are expanded
by Taylor series at all sides (surfaces).
The conservation of energy principle requires that the time rate of change of internal energy
inside the differential element must be equalled to the net heat flowing into the differential
element due to conduction plus the heat generated inside the differential element.
The first can be given as;
The time rate of change of internal energy= (7.1)
where is the density, is the specific heat and is temperature.
© AIRIL, AL-AKHBAR, ERWAN AND ZHAFRI Page 165

SOLIDS AND FLUIDS
Figure 7.1: Heat flow differential element
Next, the determination of the net heat flow can be done by considering first the net heat
flow in each direction.
In x-direction, by balancing terms on the right and left sides of the differential element, we
obtain;
(7.2)
( )
By similar argument, the net heat flow in y-direction is given as;
(7.3)
( )
Having established the net heat flow in both directions, the net heat flowing into the
differential element due to conduction through its surface can thus be given as
Net heat flow= (7.4)
Finally, if is the rate of heat generation per unit volume, for the differential element, the
heat generated inside the element can be given as
Generated heat= (7.5)

SOLIDS AND FLUIDS
The principle of energy conservation requires Eqn. (7.1) to be balanced Eqns. (7.4) and (7.5)
thus;
(7.6)
which, by cancelling terms gives
(7.7)
Eqn. (7.7) describes the problem in terms of heat fluxes ( , ) as the dependent variables
as well as the temperature, . To completely describe the problem in terms of temperature
alone, we employ Fourier’s Law of heat conduction where
(7.8)
and
(7.9)
where and are thermal conductivity in x-direction and in y-direction, respectively. By

differentiating Eqns. (7.8) and (7.9) once and inserting into Eqn. (7.7), we obtain;
(7.10)
( ) ( )
For time independent problem, hence steady heat transfer, the time derivative terms on the
left hand side ofEqn. (7.10) can be omitted thus reducing the equation to
(7.11)
( ) ( )
Equation (7.11) is the PDE of heat transfer for solid. For fluid problems, the time derivative
terms must remain and if Eulerian formulation is adopted, it must be replaced by material
derivative (i.e. { }{ } where { }is the velocity vector and { } { }) which
discussion will be given later during the discussion of fluid dynamic. Also, for heat transfer in
fluid, internal energy due to fluid viscosity must be included. For simplicity and due to its
introductory nature, all these are not considered herein; only the steady heat transfer
problem for solid as given by Eqn. (7.11) is considered instead. Readers who are interested
to know more should consult comprehensive text on heat transfer and FEM elsewhere.

SOLIDS AND FLUIDS
It is convenient to express Eqn. (7.11) in matrix forms especially in the view of FEM
formulation. This can be achieved as follows.
(7.12)
{ }[ ]{ }
or
{ } [ ]{ } (7.13)
where
(7.14)
[ ] [ ]
Such representation is convenient because as we are going to see later in the derivations of
plane stress and fluid elements, some of their terms will resemble Eqn. (7.13).
7.2.1 Boundary Condition Equations
Like other PDEs, the heat transfer PDE as given by Eqn. (7.11) (or 7.13) must be
supplemented by boundary condition equations so as to have a well-posed problem. There
are two types of boundary conditions; Newmann (natural) and Dirichlet (essential) boundary
conditions.
Neumann (natural) boundary conditions
For general heat transfer problem, Newmann (natural) boundary conditions can be either
(or combination of) specified flux, convection or/and radiation. However, for simplicity, only
specified flux is considered herein.
i. in x-direction
(7.15)
( | )
ii. in y-direction
(7.16)
( | )
where, and are the specified (known) flux whilst , and are the components of
the unit normal vector of the surface under consideration. Symbol | means evaluated at
x-coordinate of the boundary. Same goes to | .

SOLIDS AND FLUIDS
It worth to note that, for rectangular domain hence rectangular elements (meshes),
simplification to the above equations can be obtained as for this case, in x-direction,
and in y-direction, .
It can be seen later that the secondary terms produced by the integration by parts
conducted on Eqn. (7.11) (or 7.13) will be in similar forms as in Eqns. (7.15) and (7.16). Due
to this, boundary equations above are also known as natural boundary conditions.
Dirichlet (essential) boundary conditions
Dirichlet boundary condition requires that the temperature at the boundary where natural
conditions are unknown (to be solved) must be known and specified. Thus;
| ̅ (7.17)
| ̅ (7.18)
7.3 FEM FORMULATION FOR HEAT TRANSFER
Having established the PDE of the problem, we are all set to discretize the equations so as to
obtain the FEM algebraic formulation. To do this, we must first derive the relevant shape
functions.
7.3.1 Degree of Freedoms and Shape Functions for Scalar Element
Herein, we are going to derive shape function for scalar element (heat transfer) having 4
nodes and 8 nodes.
4-NODES ELEMENT
Figure 7.2 shows a rectangular element with four nodes;
Figure 7.2: 4-nodes scalar element

SOLIDS AND FLUIDS
For this element, we take a polynomial interpolation function as
(7.19)
Next we evaluate Eqn. (7.19) at the location of the nodes and equate the evaluated values
to the corresponding degree of freedom, , as follows.
| ( ) ( ) ( )( ) ̂
| ( ) ( ) ( )( ) ̂
(7.20)
| ( ) ( ) ( )( ) ̂
| ( ) ( ) ( )( ) ̂
By solving the simultaneous equations of Eqns. (7.20), the polynomial coefficients can be
obtained as
̂
̂ ̂
( ) (7.21)
̂ ̂
( )
̂ ̂ ̂ ̂
By inserting Eqn. (7.21) into (7.19) and by collecting, we obtain;
( )̂ ( )̂ ( )̂ ( )̂ (7.22)
Based on Eqn. (7.22), the shape functions can thus be given as
(7.23)

SOLIDS AND FLUIDS
Therefore, Eqn. (7.22) can also be given as
̂ ̂ ̂ ̂ (7.24)
8-NODES ELEMENT
Figure 7.3 shows a rectangular element with 8 nodes.
Figure 7.3: 8-nodes scalar elements
For this element, we take the polynomial interpolation function as
(7.24)
where the monomials are taken from the following Pascal triangle
𝑥 𝑦
𝑥 𝑥𝑦 𝑦
𝑥 𝑦
𝑥 𝑦 𝑥𝑦
Figure 7.4: Pascal triangle of cubic terms
Since Eqn. (7.18) is relatively of high order, it would be efficient to discuss its shape function
derivation from a more general approach which applies to all other cases. To do this, it is
convenient to deal with matrix and vector representations. Thus, in vector forms, eqn.
(7.24) can be given as

SOLIDS AND FLUIDS
{ } (7.25)
{ }
or
{ }{ } (7.26)
Next we evaluate Eqn. (7.26) at the location of the nodes and equate the evaluated values
to the corresponding dof, . This can be represented in matrix forms as
[ |]{ } { ̂} (7.27)
where [ |] is a 8 X 8 matrix which contains the evaluated values of the monomials and { }
is the vector of dof given as
{ ̂} {̂ ̂ ̂ ̂ ̂ ̂ ̂ ̂ } (7.28)
Solving Eqn. (7.27) for { } gives
{ } [ |] { ̂ } (7.29)
Inserting Eqn. (7.29) into Eqn. (7.26) gives
{ }[ |] { ̂ } (7.30)
Based on Eqn. (7.30), the shape functions can be given in vector forms as
{ } { }[ |] (7.31)
By conducting the multiplication in Eqn. (7.31), individual shape function can thus be given
as
(7.32)
( ) ( )
(7.33)
( ) ( )
(7.32)
( )

SOLIDS AND FLUIDS
(7.35)
( )
(7.36)
( ) ( )
(7.37)
( )
(7.38)
(7.39)
( )
Therefore, the temperature distribution, can also be given as

̂ ̂ ̂ ̂ ̂ ̂ ̂ ̂ (7.40)
Referring to Eqns. (7.24) and (7.40), it can be observed that, for both elements, a general
representation can be given in vector forms as;
{ }{ ̂ } (7.41)
where for 4-nodes element, { } { } and { ̂ } {̂ ̂ ̂ ̂ } whilst

for 8-nodes element, { } { } and { ̂ } { ̂ ̂ }.
7.3.2 Discretization by Galerkin WRM
By having established the shape functions, we are all set to discretize the PDE into algebraic
equations. By inserting Eqns. (7.41) into Eqn. (7.13) we obtain
{ } [ ]{ } { }{ ̂ } (7.42)
By observing Figure 7.2 and Eqn. (7.42), we can identify that, for a 4-nodes element, we
have 4 unknown dofs, i.e. { ̂ ̂ ̂ ̂ } with only 1 equation. To obtain sufficient
number of equations, we will multiply Eqn. (7.42) with the vector of shape function { }
then integrate to produce four independent equations as required. Similar argument applies
to 8-nodes element, resulting in eight independent equations as required.
Multiplying Eqn. (7.42) with the vector of shape functions { } and integrating give
∫ ∫ { } ({ } [ ]{ } { }{ ̂ } ) (7.43)

SOLIDS AND FLUIDS
7.3.3 Integration by Parts (IBP)
As mentioned in Chapter 2, the intentions of IBP are:
i) To relax the continuity requirement

ii) To induce explicit the natural boundary conditions
Therefore, by conducting IBP to Eqn. (7.43), we obtain
(7.44)
∫ ∫ { } { } [ ]{ } { }{ ̂ } ∫ ∫{ } ∫[ ] { }
Rearranging gives
(7.45)
∫ ∫ { } { } [ ]{ } { }{ ̂ } ∫ ∫{ } ∫[ ] { }
Now, let’s introduce the following matrix and vectors.
7.3.4 Matrix, Vectors and Discretized Equation of Heat Transfer
Since the conduction terms are similar in forms to those to be derived for plane stress
problem; let’s call the corresponding matrix as conductance matrix [ ] where
[ ] ∫ ∫ { } { } [ ]{ } { } (7.46)
Next, let’s introduce a vector for the heat generation and call it as equivalent load vector,
{ } where
(7.47)
{ } ∫ ∫{ }
Finally, let’s introduce a vector for the boundary terms induced by the IBP and call it as
natural boundary condition vector, { } where
(7.48)
{ } ∫[ ] { }

SOLIDS AND FLUIDS
It can be shown that, for rectangular domain hence rectangular element (mesh), Eqn. (7.48)
is given by Eqn. (7.15) and (7.16), respectively depending on the direction and location of
evaluation.
Having established these, the discretized equation of heat transfer can thus be given as
[ ]{ ̂ } { } { } { } (7.49)
It must be noted that, since IBP is conducted, the multiplication in Eqn. (7.46) between
terms inside the integral must be carried out from left to right. To demonstrate this and to
provide example on how to build the above matrix, the determination of for 4-nodes
scalar element is shown below. Although in practice, numerical integration is employed
which detailed procedure is given in later chapter, herein analytical integration is employed
so as to allow for easy tracing of the procedure and to allow for its immediate application.
Such analytical integrations are possible herein because we limit our discussion to
rectangular domain hence rectangular elements only.
Step-by-step procedure for the determination of
Step 1
(7.50)
{ }{ }
Step 2
(7.51)
{ }[ ]
Step 3
(7.52)
{ }
{ }
Step 4
(7.53)
{ }{ }
Step 5 (analytical integration)
Now by inserting and as given by Eqn. (7.23) into (7.53), is given as
∫∫( ) (7.54)

SOLIDS AND FLUIDS
For a rectangular element with a width of and a height of , as shown in Figure 7.2,
can be re-given as
∫∫( ) (7.55)
The complete integrated values of [ ] for 4-nodes element can be given as
( ) ( ) ( ) ( ) (7.56)
( ) ( ) ( ) ( )
[ ]
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ]
Repeating the procedure above (Step 1 to Step 5), the complete integrated values of [ ] for
8-nodes element can be given as
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
(7.57)
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( )
[ ]

SOLIDS AND FLUIDS
EXAMPLE 7.1
Herein, we are going to demonstrate the FEM formulation for heat transfer equations by
solving the problem as shown in Figure E7.1. For a step-by-step demonstration (hand-
calculation), only four assembled 4-nodes elements and two 8-nodes elements are used, so
as to allow for easy tracing of the procedure. However, it must be noted that such usage is
for demonstration purpose only as the result can be poor. After the step-by-step calculation,
the procedure is run using a sufficient number of elements (denser mesh). The results are
then validated against results obtained from commercial software for proof of correctness.
Figure E7.1: Properties and boundary conditions for fin heat transfer problem.
a) 4-nodes element
Figure E7.2 Arrangement of element and global dofs numbering for 4-nodes element
Due to the symmetrically, each element would have similar conductance matrix and internal
heat generation forces. Based on Eqn. 7.56, the conductance matrix is given as

SOLIDS AND FLUIDS
[ ] [ ] [ ] [ ] [ ] (E7.1)
and the internal heat generation are given as
( )
( )
(E7.2)
{ } { } { } { } { }
( )
{( )}
[ ]{ } { }
(E7.3)
By imposing the essential boundary conditions ( ), the
assembled global equilibrium equation is reduced to
[ ]{ } { } (E7.4)
By solving Eqn. (E7.4) using Matlab command “\”,the values of the temperature, are thus
obtained as:
{ } { } (E7.5)
Table E7.1: Validation of global temperature for 4-nodes element
4-nodes ele. (2x2) 0.0021 0.0024 0.0016 0.0018

4-nodes ele. (8x8) 0.0019 0.0023 0.0015 0.0017
Software 0.0019 0.0023 0.0015 0.0017

SOLIDS AND FLUIDS
b) 8-nodes element
Figure E7.3 Arrangement of element and global dofs numbering for 8-nodes element
Due to the symmetrically, each element would have similar conductance matrix and internal
heat generation forces. Based on Eqn. 7.57, the conductance matrix is given as
[ ] [ ]
[ ]
(E7.6)

SOLIDS AND FLUIDS
and the internal heat generation are given as
( )
( )
( )
( )
{ } { } (E7.7)
( )
( ) { }
( )
{ ( ) }
Since the size of an assembled stiffness would be very large when 8-nodes elements are
used, herein, reduced size of complete global equilibrium equation with boundary
conditions have been imposed, are presented ( ).
[ ]{ } { }
(E7.8)
By solving Eqn. (E7.8) using Matlab command “\”,the values of the temperature, are thus
obtained as:
(E7.9)
{ } { }
Table E7.2: Validation of global temperature for 8-nodes element
( ) ( ) ( ) ( )
8-nodes (2x1) 0.0019 0.0023 0.0015 0.0017
Software 0.0019 0.0023 0.0015 0.0017
Figures E7.4 and E7.5 show the contour of the temperature distribution obtained from 4-
nodes element (64 elements) and from commercial software, respectively. The former is

SOLIDS AND FLUIDS
potted in Matlab environment. Despite the coarseness of the former, close agreement is still
evident between the two contours.
Figure E7.4 Temperature contour from 4-nodes element (64 elements)
Figure E7.5 Temperature contour from commercial software

SOLIDS AND FLUIDS
SOURCE CODE (4-NODES ELEMENT)
clc;clear all;
%--------------------------------------------------------------------------
% Input
%--------------------------------------------------------------------------
kx1=400; % Btu/(hr-ft-F)
ky1=400; % Btu/(hr-ft-F)
Qa=50;
h1=0; % Btu/(hr-ft-F)
t1=0.1; % Thickness of element (in)
dx=8; % No of Element in x-dir

dy=8; % No of Element in y-dir
Lx=0.4; % Width of element in x-dir

Ly=0.2; % Height of element in y-dir
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
deltax=Lx/dx; % Width of single element
deltay=Ly/dy; % Height of single element
% Connectivity matrix
dofNo=[];
NodeNo=[];
index=0;
for i=1:dy
for j=1:dx
index=index+1;
val1=(dx+1)*(i-1)+j;
val2=(dx+1)*(i)+j;
Nd1=val1;
Nd2=val1+1;
Nd3=val2+1;
Nd4=val2;
NodeNo(index,:)=[Nd1 Nd2 Nd3 Nd4];
dofNo(index,:)=NodeNo(index,:);
end
end
TTDOF=max(max(dofNo)); % Total DOF
%--------------------------------------------------------------------------
% Stiffness from Maple
%--------------------------------------------------------------------------
syms a b t kx ky h
k1= [t * kx / a * b / 0.3e1 + t * ky * a / b / 0.3e1 + 0.2e1 / 0.9e1 * h *

a * b t * ky * a / b / 0.6e1 - t * kx / a * b / 0.3e1 + h * a * b / 0.9e1 -
t * kx / a * b / 0.6e1 - t * ky * a / b / 0.6e1 + h * a * b / 0.18e2 t * kx
/ a * b / 0.6e1 - t * ky * a / b / 0.3e1 + h * a * b / 0.9e1;

SOLIDS AND FLUIDS
t * ky * a / b / 0.6e1 - t * kx / a * b / 0.3e1 + h * a * b / 0.9e1 t *
kx / a * b / 0.3e1 + t * ky * a / b / 0.3e1 + 0.2e1 / 0.9e1 * h * a * b t *
kx / a * b / 0.6e1 - t * ky * a / b / 0.3e1 + h * a * b / 0.9e1 -t * kx / a
* b / 0.6e1 - t * ky * a / b / 0.6e1 + h * a * b / 0.18e2;
-t * kx / a * b / 0.6e1 - t * ky * a / b / 0.6e1 + h * a * b / 0.18e2 t
* kx / a * b / 0.6e1 - t * ky * a / b / 0.3e1 + h * a * b / 0.9e1 t * kx /
a * b / 0.3e1 + t * ky * a / b / 0.3e1 + 0.2e1 / 0.9e1 * h * a * b t * ky *
a / b / 0.6e1 - t * kx / a * b / 0.3e1 + h * a * b / 0.9e1;
t * kx / a * b / 0.6e1 - t * ky * a / b / 0.3e1 + h * a * b / 0.9e1 -t
* kx / a * b / 0.6e1 - t * ky * a / b / 0.6e1 + h * a * b / 0.18e2 t * ky *
a / b / 0.6e1 - t * kx / a * b / 0.3e1 + h * a * b / 0.9e1 t * kx / a * b /
0.3e1 + t * ky * a / b / 0.3e1 + 0.2e1 / 0.9e1 * h * a * b;];
klocal=subs(k1,{a,b,t,kx,ky,h},{deltax,deltay,t1,kx1,ky1,h1})
%--------------------------------------------------------------------------
% Assemble stifness matrix
%--------------------------------------------------------------------------
kglobal=zeros(TTDOF,TTDOF);
flocal=zeros(TTDOF,1);
for i=1:size(dofNo,1)
for j=1:4
dof1=dofNo(i,j);
for k=1:4
dof2=dofNo(i,k);
kglobal(dof1,dof2)=kglobal(dof1,dof2)+klocal(j,k);
end
end
end
%--------------------------------------------------------------------------
% Forces
%--------------------------------------------------------------------------
a=deltax;
b=deltay;
for i=1:size(NodeNo,1)
f1=[(Qa*a*b*t1)/4;(Qa*a*b*t1)/4;(Qa*a*b*t1)/4;(Qa*a*b*t1)/4];
for j=1:4
dof1=dofNo(i,j);
flocal(dof1,1)=flocal(dof1,1)+f1(j,1);
end
end
%--------------------------------------------------------------------------
% Impose BC
%--------------------------------------------------------------------------
% dofe=[9;8;7;4;1]; % 2x2 element
% dofe=[25;24;23;22;21;16;11;6;1]; % 4x4 element
dofe=[81;80;79;78;77;76;75;74;73;64;55;46;37;28;19;10;1];% 8x8 element
for i=1:size(dofe,1)
val1=dofe(i,1);
kglobal(val1,:)=[];
kglobal(:,val1)=[];

SOLIDS AND FLUIDS
flocal(val1,:)=[];
end
%--------------------------------------------------------------------------
% Solve d=k\f
%--------------------------------------------------------------------------
T=kglobal\flocal
%--------------------------------------------------------------------------
% plotting result (deflection) % 8x8 element
%--------------------------------------------------------------------------
Tori=zeros(TTDOF,1);
Tori(2:9,1)=T(1:8,1);
Tori(11:18,1)=T(9:16,1);
Tori(20:27,1)=T(17:24,1);
Tori(29:36,1)=T(25:32,1);
Tori(38:45,1)=T(33:40,1);
Tori(47:54,1)=T(41:48,1);
Tori(56:63,1)=T(49:56,1);
Tori(65:72,1)=T(57:64,1);
Zx=[];
index=0;
for i=1:dy+1
for j=1:dx+1
index=index+1;
x_val=Tori(index,1);
Zx(i,j)=x_val;
end
end
[X,Y] = meshgrid(0:deltax:Lx, 0:deltay:Ly);
figure (1)
surf(X,Y,Zx)
title('Temperature (K)')
colorbar
SOURCE CODE (8-NODES ELEMENT)
clc;clear all;
%--------------------------------------------------------------------------
% Input ref pg 83 schaum's
%--------------------------------------------------------------------------
kx1=400; % Btu/(hr-ft-F)
ky1=400; % Btu/(hr-ft-F)
Qa=50;
h1=0; % Btu/(hr-ft-F)
t1=0.1; % Thickness of element (in)

SOLIDS AND FLUIDS
Lx=0.4; % Width of element in x-dir

Ly=0.2; % Height of element in y-dir
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
dofNo=[];
NodeNo=[];
index=0;
for i=1:dy
for j=1:dx
index=index+1;
val1=2*(j-1)+1;
val2=(i-1)*((3*dx)+2);
val3=i*((3*dx)+2);
val4=j-dx;
Nd1=val1+val2;
Nd4=val1+val3;
Nd6=val3+val4;
Nd5=Nd1+1;
Nd2=Nd1+2;
Nd3=Nd4+2;
Nd7=Nd4+1;
Nd8=Nd6-1;
NodeNo(index,:)=[Nd1 Nd2 Nd3 Nd4 Nd5 Nd6 Nd7 Nd8];
dof1=2*(Nd1-1)+1; dof9=2*(Nd5-1)+1;
dof2=2*(Nd1-1)+2; dof10=2*(Nd5-1)+2;
dof3=2*(Nd2-1)+1; dof11=2*(Nd6-1)+1;
dof4=2*(Nd2-1)+2; dof12=2*(Nd6-1)+2;
dof5=2*(Nd3-1)+1; dof13=2*(Nd7-1)+1;
dof6=2*(Nd3-1)+2; dof14=2*(Nd7-1)+2;
dof7=2*(Nd4-1)+1; dof15=2*(Nd8-1)+1;
dof8=2*(Nd4-1)+2; dof16=2*(Nd8-1)+2;
dofNo(index,:)=NodeNo(index,:);
end
end
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
syms a b t kx ky h

SOLIDS AND FLUIDS
klocal_raw= [0.26e2 / 0.45e2 * t * kx / a * b + 0.26e2 / 0.45e2 * t * ky *
a / b 0.14e2 / 0.45e2 * t * kx / a * b + 0.17e2 / 0.90e2 * t * ky * a / b
0.23e2 / 0.90e2 * t * kx / a * b + 0.23e2 / 0.90e2 * t * ky * a / b 0.17e2
/ 0.90e2 * t * kx / a * b + 0.14e2 / 0.45e2 * t * ky * a / b -0.8e1 / 0.9e1
* t * kx / a * b + t * ky * a / b / 0.15e2 -t * kx / a * b / 0.15e2 - 0.4e1
/ 0.9e1 * t * ky * a / b -0.4e1 / 0.9e1 * t * kx / a * b - t * ky * a / b /
0.15e2 t * kx / a * b / 0.15e2 - 0.8e1 / 0.9e1 * t * ky * a / b; 0.14e2 /
0.45e2 * t * kx / a * b + 0.17e2 / 0.90e2 * t * ky * a / b 0.26e2 / 0.45e2
* t * kx / a * b + 0.26e2 / 0.45e2 * t * ky * a / b 0.17e2 / 0.90e2 * t *
kx / a * b + 0.14e2 / 0.45e2 * t * ky * a / b 0.23e2 / 0.90e2 * t * kx / a
* b + 0.23e2 / 0.90e2 * t * ky * a / b -0.8e1 / 0.9e1 * t * kx / a * b + t
* ky * a / b / 0.15e2 t * kx / a * b / 0.15e2 - 0.8e1 / 0.9e1 * t * ky * a
/ b -0.4e1 / 0.9e1 * t * kx / a * b - t * ky * a / b / 0.15e2 -t * kx / a *
b / 0.15e2 - 0.4e1 / 0.9e1 * t * ky * a / b; 0.23e2 / 0.90e2 * t * kx / a *
b + 0.23e2 / 0.90e2 * t * ky * a / b 0.17e2 / 0.90e2 * t * kx / a * b +
0.14e2 / 0.45e2 * t * ky * a / b 0.26e2 / 0.45e2 * t * kx / a * b + 0.26e2
/ 0.45e2 * t * ky * a / b 0.14e2 / 0.45e2 * t * kx / a * b + 0.17e2 /
0.90e2 * t * ky * a / b -0.4e1 / 0.9e1 * t * kx / a * b - t * ky * a / b /
0.15e2 t * kx / a * b / 0.15e2 - 0.8e1 / 0.9e1 * t * ky * a / b -0.8e1 /
0.9e1 * t * kx / a * b + t * ky * a / b / 0.15e2 -t * kx / a * b / 0.15e2 -
0.4e1 / 0.9e1 * t * ky * a / b; 0.17e2 / 0.90e2 * t * kx / a * b + 0.14e2 /
0.45e2 * t * ky * a / b 0.23e2 / 0.90e2 * t * kx / a * b + 0.23e2 / 0.90e2
* t * ky * a / b 0.14e2 / 0.45e2 * t * kx / a * b + 0.17e2 / 0.90e2 * t *
ky * a / b 0.26e2 / 0.45e2 * t * kx / a * b + 0.26e2 / 0.45e2 * t * ky * a
/ b -0.4e1 / 0.9e1 * t * kx / a * b - t * ky * a / b / 0.15e2 -t * kx / a *
b / 0.15e2 - 0.4e1 / 0.9e1 * t * ky * a / b -0.8e1 / 0.9e1 * t * kx / a * b
+ t * ky * a / b / 0.15e2 t * kx / a * b / 0.15e2 - 0.8e1 / 0.9e1 * t * ky
* a / b; -0.8e1 / 0.9e1 * t * kx / a * b + t * ky * a / b / 0.15e2 -0.8e1 /
0.9e1 * t * kx / a * b + t * ky * a / b / 0.15e2 -0.4e1 / 0.9e1 * t * kx /
a * b - t * ky * a / b / 0.15e2 -0.4e1 / 0.9e1 * t * kx / a * b - t * ky *
a / b / 0.15e2 0.16e2 / 0.9e1 * t * kx / a * b + 0.8e1 / 0.15e2 * t * ky *
a / b 0 0.8e1 / 0.9e1 * t * kx / a * b - 0.8e1 / 0.15e2 * t * ky * a / b 0;
-t * kx / a * b / 0.15e2 - 0.4e1 / 0.9e1 * t * ky * a / b t * kx / a * b /
0.15e2 - 0.8e1 / 0.9e1 * t * ky * a / b t * kx / a * b / 0.15e2 - 0.8e1 /
0.9e1 * t * ky * a / b -t * kx / a * b / 0.15e2 - 0.4e1 / 0.9e1 * t * ky *
a / b 0 0.8e1 / 0.15e2 * t * kx / a * b + 0.16e2 / 0.9e1 * t * ky * a / b 0
-0.8e1 / 0.15e2 * t * kx / a * b + 0.8e1 / 0.9e1 * t * ky * a / b; -0.4e1 /
0.9e1 * t * kx / a * b - t * ky * a / b / 0.15e2 -0.4e1 / 0.9e1 * t * kx /
a * b - t * ky * a / b / 0.15e2 -0.8e1 / 0.9e1 * t * kx / a * b + t * ky *
a / b / 0.15e2 -0.8e1 / 0.9e1 * t * kx / a * b + t * ky * a / b / 0.15e2
0.8e1 / 0.9e1 * t * kx / a * b - 0.8e1 / 0.15e2 * t * ky * a / b 0 0.16e2 /
0.9e1 * t * kx / a * b + 0.8e1 / 0.15e2 * t * ky * a / b 0; t * kx / a * b
/ 0.15e2 - 0.8e1 / 0.9e1 * t * ky * a / b -t * kx / a * b / 0.15e2 - 0.4e1
/ 0.9e1 * t * ky * a / b -t * kx / a * b / 0.15e2 - 0.4e1 / 0.9e1 * t * ky
* a / b t * kx / a * b / 0.15e2 - 0.8e1 / 0.9e1 * t * ky * a / b 0 -0.8e1 /
0.15e2 * t * kx / a * b + 0.8e1 / 0.9e1 * t * ky * a / b 0 0.8e1 / 0.15e2 *
t * kx / a * b + 0.16e2 / 0.9e1 * t * ky * a / b;];
klocal1=subs(klocal_raw,{a,b,t,kx,ky,h},{deltax,deltay,t1,kx1,ky1,h1})
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------

SOLIDS AND FLUIDS
for j=1:8
for k=1:8
dof1=dofNo(i,j);
dof2=dofNo(i,k);
kglobal(dof1,dof2)=kglobal(dof1,dof2)+klocal1(j,k);
end
end
end
%--------------------------------------------------------------------------
% UDL
%--------------------------------------------------------------------------
a=deltax;
b=deltay;
t=t1;
Q=Qa;
for i=1:size(NodeNo,1)
f1= [-Q * t * a * b / 0.12e2; -Q * t * a * b / 0.12e2; -Q * t * a * b /
0.12e2; -Q * t * a * b / 0.12e2; Q * t * a * b / 0.3e1; Q * t * a * b /
0.3e1; Q * t * a * b / 0.3e1; Q * t * a * b / 0.3e1;];
for j=1:8
dof1=dofNo(i,j);
flocal(dof1,1)=flocal(dof1,1)+f1(j,1);
end
end
%--------------------------------------------------------------------------
% Impose BC
%--------------------------------------------------------------------------
dofe=[13;12;11;10;9;6;1];% 1 element
val=dofe(i,1);
kglobal(val,:)=[];
kglobal(:,val)=[];
flocal(val,:)=[];
end
%--------------------------------------------------------------------------
% Solve d=k\f
%--------------------------------------------------------------------------
T=kglobal\flocal

SOLIDS AND FLUIDS
PLANE ELASTICITY
PLANE STRESS
8. INTRODUCTION
Plane stress element is used to model thin body or structure that is subjected to in plane
loading (or boundary stresses) thus correspondingly stressed in the plane. In practical, it is
used to model shear and load bearing walls, steel web and many more.
The study of plane stress FEM formulation is important, not only for its direct application to
physical problems but also to further the study on FEM itself. For example, in the discussion
of shell elements, the formulation can be viewed as combining the plane stress element and
the plate element together. Also, in fluid dynamics, with some modifications, the plane
stress element can be used to model fluid flow.
Herein, partial differential equation (PDE) of plane stress elasticity will be first derived
followed by the FEM formulation.
8.1 DERIVATION OF PLANE STRESS PARTIAL DIFFERENTIAL EQUATION
The PDE can be derived by employing the conservation of linear momentum principle to the
differential element shown in Figure 8.1.
Figure 8.1: Plane stress differential element
As shown in the figure, the dependent variables to be considered are the normal stresses
(i.e. , ) and the shear stresses (i.e. , ). and , are the known body forces in
x-direction and y-direction, respectively. Similar to previous chapters, the dependent
variables are expanded by Taylor series at all sides (surfaces).

SOLIDS AND FLUIDS
The conservation of linear momentum principle requires that, for a static condition, the
total forces to be zero; thus
∑ (8.1)
∑ (8.2)
In x-direction, by taking equilibrium of forces, Eqn. (8.1) gives
(8.3)
∑ (( ) ) (( ) )
By expanding and cancelling, the following is obtained;
(8.4)
In y-direction, by taking equilibrium of forces, Eqn. (8.2) gives
(8.5)
∑ (( ) ) (( ) )
By expanding and cancelling, the following is obtained;
(8.6)
Eqns. (8.4) and (8.6) are the PDEs for the plane stress problem, described in terms of
stresses as the dependent variables. In matrix forms, the equations can be given as
{ } (8.7)
{ }
Based on conservation of angular momentum, it can be shown that . Thus, Eqn.

(8.7) can further be expressed as
(8.8)
{ } { }
{ } [ ]

SOLIDS AND FLUIDS
or
[ ]{ } { } (8.9)
where [ ] is termed from now on as differential operator matrix.
Eqn. (8.9) (or (8.4) and (8.6)) describes the problem in terms of stresses as the dependent
variables. Since we are focusing on the displacement-based formulation, it would be
necessary to express the equation in terms of displacement variables. This can be done by
considering the following constitutive relationship.
(8.10)
( )
( )
which can be arranged in matrix forms as
{ } { } [ ( )] { }
(8.11)
or
{ } [ ]{ } (8.12)
where is the Young’s (elasticity) modulus and is the Poisson’s ratio of the material.
and are the axial strain in x and y directions, respectively whilst is the shear strain.
The strains can, on the other hand, be expressed in terms of displacements by employing
the following strain-displacement relationship;
(8.13)

SOLIDS AND FLUIDS
(8.14)
{ } { } { }
[ ]
Observe that the matrix on the right hand side of the equation above is also the transpose
of the differential operator matrix, [ ] previously defined in Eqn. (8.8). Based on this Eqn.
(8.14) can be given as;
{ } [ ] { } (8.15)
where { } is the vector which contains both displacement components, and . By

inserting Eqn. (8.15) into Eqn. (8.12), the stresses can thus be given in terms of
displacements as
{ } [ ][ ] { } (8.16)
Finally, by inserting Eqn. (8.16) into Eqn. (8.9) the PDEs can be given in terms of
displacement variables as
[ ][ ][ ] { } { } (8.17)
Like other PDEs, the plane stress PDE given by Eqn. (8.17) (or (8.4) and (8.6)) must be
supplemented by boundary condition equations so as to have a well-posed problem. There
are two types of boundary conditions; Newmann (natural) and Dirichlet (essential) boundary
conditions.
i. in x-direction
(8.18)
ii. in y-direction
(8.19)
where, , and are the components of the unit normal vector of the surface under
consideration. It worth to note that, for rectangular domain hence rectangular elements

SOLIDS AND FLUIDS
(meshes), major simplification to the above equations can be obtained because for this
case, , in x-direction and , in y-direction, thus;
(8.20)
(8.21)
It can be seen later that the secondary terms produced by the integration by parts on the
weighted PDEs will be in similar forms to (8.20) and (8.21). Due to this, equations above are
also known as natural boundary conditions.
Dirichlet boundary condition requires that the displacement components at the boundary
where natural conditions are unknown (to be solved) must be known thus specified.
̅ (8.22)
̅ (8.23)
8.2 FEM FORMULATION FOR PLANE STRESS
Having established the PDE of the problem, we are all set to discretize the equations so as to
obtain the FEM algebraic formulation.
8.2.1 Degree of Freedoms and Shape Functions for Plane Stress Element
In a plane stress element, two translational degree of freedoms exist at each node;
represents the nodal displacement in x-direction whilst represents the nodal
displacement in y-direction. Their arrangements in an element are shown in Figure 8.2 for 4-
nodes element and in Figure 8.3 for 8-nodes element.
Figure 8.2: 4-nodes plane stress element

SOLIDS AND FLUIDS
Figure 8.3: 8-nodes plane stress element
For 4-nodes element, and are interpolated by the shape functions in such a way that
(8.24)
(8.25)
whilst for 8-nodes element, they can be given as
(8.26)
(8.27)
For 4-nodes element, the shape functions, are similar to those given by Eqn. (7.23) whilst
for 8-nodes element, they are similar to those given by Eqns. (7.32) to (7.39). Observing
Eqns. (8.24) to (8.27), it can be noticed that they can be given in vector forms as follows;
For 4-nodes element;
{ } [ ] (8.28)
{ }

SOLIDS AND FLUIDS
and for 8-nodes element;
{ } [ ] (8.29)
{ }
Observing Eqns. (8.28) and (8.29), it can be noticed that, for both elements, a general
representation can be provided which is
{ } [ ]{ ̂} (8.30)
8.2.2 Discretization by Galerkin WRM
By having established the shape functions, we are all set to discretize the PDE into algebraic
equations. By inserting Eqns. (8.30) into Eqn. (8.17) we obtain
[ ] [ ][ ] [ ]{ ̂} { } (8.31)
By observing Figure 8.2 and Eqns. (8.24) and (8.25), we can identify that, for a 4-nodes
element, we have 8 unknown dofs, i.e. { ̂ ̂ ̂ ̂ } with only 1 equation, whilst
for 8-nodes element we have 16 unknown dofs i.e. { ̂ ̂ ̂ ̂ }. To obtain
sufficient number of equations, we will multiply Eqn. (8.31) with the matrix of shape
function [ ] then integrate to produce eight independent equations as required for 4-nodes
element. Similar argument applies to 8-nodes element, resulting in sixteen independent
equations as required.
Multiplying Eqn. (8.31) with the matrix of shape functions [ ] and integrating give
∫ ∫ [ ] ([ ] [ ][ ] [ ]{ ̂} { }) (8.32)
8.2.3 Integration By Parts (IBP)


SOLIDS AND FLUIDS
Therefore, by conducting IBP to Eqn. (8.32), we obtain
∫ ∫ [ ] [ ] [ ][ ] [ ]{ ̂} ∫ ∫[ ] { } ∫[ ] { } (8.33)
Rearranging gives
∫ ∫ [ ] [ ] [ ][ ] [ ]{ ̂} ∫ ∫[ ] { } ∫[ ] { } (8.34)
Now, let’s introduce the following matrix and vectors.
8.4.4 Matrix, Vectors and Discretized Equation of Heat Transfer
The left hand side terms of Eqn. (8.34) can be given in matrix forms. Let’s call the
corresponding matrix as stiffness matrix [ ] where
[ ] ∫ ∫ [ ] [ ] [ ][ ] [ ] (8.35)
Next, let’s introduce a vector for the body force, { } where
(8.36)
{ } ∫ ∫[ ] { }
Finally, let’s introduce a vector for the boundary terms induced by the IBP and call it as
natural boundary condition vector, { } where
(8.37)
{ } ∫[ ] { }
It can be shown that, for rectangular domain hence rectangular element (mesh), Eqn. (8.37)
is given by Eqn. (8.20) and (8.21), respectively depending on the direction and location of
evaluation.
Having established these, the discretized equation of heat transfer can thus be given as
[ ]{ ̂} { } { } { } (8.36)

SOLIDS AND FLUIDS
It must be noted that, since IBP is conducted, the multiplication in Eqn. (8.35) between
terms inside the integral must be carried out from left to right. To demonstrate this and to
provide example on how to build the above matrix, the determination of for 4-nodes
element is shown below. Although in practice, numerical integration is employed which
detailed procedure is given in later chapter, herein analytical integration is employed so as
to allow for easy tracing of the procedure and to allow for its immediate application. Such
analytical integrations are possible herein because we limit our discussion to rectangular
domain hence rectangular elements only.
Step-by-step procedure for the determination of
Step 1
( )
[ ] [ ]
( )
[ ]
(8.37)
Step 2
( ) ( )
[ ] [ ]
[ ]
(8.38)
Step 3
( )
∫∫[ ] [ ] (8.39)
( ) ( )
The complete integrated values of [ ] for 4-nodes element can be given as
(8.40)
[ ]
[ ]

SOLIDS AND FLUIDS

SOLIDS AND FLUIDS

SOLIDS AND FLUIDS
Due to the very large size of the matrix, the integrated values of [ ] are not given herein but
directly provided in the source code instead.
EXAMPLE 8.1
Herein, we are going to demonstrate the FEM formulation for plane stress by solving the
cantilever beam problem as shown in Figure E8.1. For a step-by-step demonstration (hand-
calculation), only four assembled 4-nodes elements and two 8-nodes elements are used, so
as to allow for easy tracing of the procedure. However, it must be noted that such usage is
for demonstration purpose only as the result can be poor. After the step-by-step calculation,
the procedure is run using a sufficient number of elements (denser mesh). The results are
then validated against results obtained from commercial software for proof of correctness.
Figure E8.1 Plane stress and its properties.

SOLIDS AND FLUIDS
a) 4-nodes Element
Figure E8.2 The arrangement of element and global dofs numbering for 4-nodes element
Due to the symmetrically, each element would have similar local stiffness matrix. Based on
Eqn. 8.??, the local stiffness matrix is given as
[ ]
[ ]
(E8.1)
and for the load vector for element 2 and element 4 given as follow
{ } { }
(E8.2)
{ }
{ }
where and . The assembled global stiffness matrix, [ ] is thus given as

SOLIDS AND FLUIDS
[ ]
[ ]
(E8.3)
(E8.4)
{ }
{ }

SOLIDS AND FLUIDS
{ }
[ ]{ }
(E8.5)
By imposing the essential boundary conditions ( ), the assembled global equilibrium equation is reduced
to
(E8.6)
{ }
[ ]{ }

SOLIDS AND FLUIDS
By solving Eqn. (E8.6) using Matlab command “\”, the values of the assembled global dofs
are thus obtained as:
{ } (E8.7)
{ }
Table E8.1: Validation of global displacement
Q4 (2x2) -0.042 -0.131 0.042 -0.131

Q4 (16x16) -0.061 -0.189 0.061 -0.189
Q4 (80x80) -0.062 -0.192 0.062 -0.192
Software -0.062 -0.192 0.062 -0.192
b) 8-nodes Element
Figure E8.3 The arrangement of element and global dofs numbering for 8-nodes element

SOLIDS AND FLUIDS
Due to the symmetrically, each element would have similar local stiffness matrix. Based on Eqn. 8.??, the local stiffness matrix is given as
[ ] [ ]
[ ]
(E8.8)
and for the load vector for element 2 is given as follow
{ }
(E8.9)
{ }
{ }

SOLIDS AND FLUIDS
where and . Since the size of an assembled stiffness matrix would be

very large when 8-nodes elements are used, herein, the reduced size of assembled global
equilibrium equation with boundary conditions have been imposed, are presented (i.e.
)
[ ]{ } { }
(E8.10)
By solving Eqn. (E7.10) using Matlab command “\”, the values of the assembled global dof
are thus obtained as:
(E8.11)
{ }
{ }
Q8 (2x1) -0.058 -0.178 0.058 -0.178

Q8 (4x4) -0.061 -0.190 0.061 -0.190
Q8 (30x30) -0.062 -0.192 0.062 -0.192
Software -0.062 -0.192 0.062 -0.192

SOLIDS AND FLUIDS
Figures E8.4 and E8.5 show the contour of the displacement distribution obtained from 4-
nodes element potted in Matlab environment whilst Figures E8.6 and E8.7 show the contour
obtained from commercial software. As can be seen, close agreement are evident between
the two sources.
Figure E8.4 Displacement contour in x-direction of 4-nodes element.

SOLIDS AND FLUIDS
Figure E8.5 Displacement contour in y-direction of 4-nodes element.
Figure E8.6 Displacement contour in x-direction from commercial software

SOLIDS AND FLUIDS
Figure E8.7 Displacement contour in y-direction from commercial software

SOLIDS AND FLUIDS
SOURCE CODE
SOURCE CODE (4-NODES ELEMENT) – SIMPLE
clc;clear all;
%--------------------------------------------------------------------------
% Input
%--------------------------------------------------------------------------
E1=200e6;% Pa (N/m)
v1=0.33; % Poison ratio
t1=0.01; % Thickness of element (m)

Lx=4; % Width of element in x-dir

Ly=2; % Height of element in y-dir
qy=-5000; % N
% Node to impose BC (for cantilever problem only)

BCNode=[];
index=0;
for i=1:dy+1
val=(dx+1)*(i-1)+1;
index=index+1;
BCNode(index,1)=val;
end
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
dofNo=[];
NodeNo=[];
index=0;
for i=1:dy
for j=1:dx
index=index+1;
val1=(dx+1)*(i-1)+j;
val2=(dx+1)*(i)+j;
Nd1=val1;
Nd2=val1+1;
Nd3=val2+1;
Nd4=val2;
dof1=2*(Nd1-1)+1;
dof2=2*(Nd1-1)+2;
dof3=2*(Nd2-1)+1;

SOLIDS AND FLUIDS
dof4=2*(Nd2-1)+2;
dof5=2*(Nd3-1)+1;
dof6=2*(Nd3-1)+2;
dof7=2*(Nd4-1)+1;
dof8=2*(Nd4-1)+2;
dofNo(index,:)=[dof1 dof2 dof3 dof4 dof5 dof6 dof7 dof8];
end
end
% Coordinate of nodes
index=0;
NodeCoor=[];
for i=1:dy+1
for j=1:dx+1
index=index+1;
xcoor=deltax*(j-1);
ycoor=deltay*(i-1);
NodeCoor(index,:)=[xcoor ycoor];
end
end
% dof to be eliminate
dofe1=[];
index=0;
for i=1:size(BCNode,1)
NdBC=BCNode(i);
ddofeli1=2*(NdBC-1)+1;
ddofeli2=2*NdBC;
index=index+1;
dofe1(index,1)=ddofeli1;
index=index+1;
end
dofe=sort(dofe1,'descend');
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
syms a b t E v
klocal_raw= [-0.2e1 / 0.3e1 * t / a * b * E / (1 - v ^ 2) - 0.2e1 / 0.3e1 *

t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) + t * ((1 / a ^ 2
* E / (1 - v ^ 2)) + 0.1e1 / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * a * b (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 /
(a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^
2) / 0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^
2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1

SOLIDS AND FLUIDS
+ t * (-(1 / a ^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E /
(1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 + t * ((1
/ a / b * E / (1 - v ^ 2) * v) + 0.1e1 / a / b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * a * b t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -
v / 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^
2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * ((1 / a ^ 2
/ b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1) -(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 - (t * E / (1 - v ^ 2)
* v) / 0.2e1 (t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v
^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E /
(1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a) * (b ^ 2) / 0.2e1 - (t * E / (1 - v ^ 2) * v) / 0.2e1;
(t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
* (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^ 2) * E / (1 -
v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1 + t * (-(1 / a
^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 + t * ((1 / a / b * E /
(1 - v ^ 2) * v) + 0.1e1 / a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a * b -0.2e1 / 0.3e1 * t / a * b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) - 0.2e1 / 0.3e1 * t * a / b * E / (1 - v ^ 2) + t * ((1
/ b ^ 2 * E / (1 - v ^ 2)) + 0.1e1 / (a ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * a * b (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) *
v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * (a ^ 2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) /
0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - (t *
E / (1 - v ^ 2) * v) (t * a / b * E / (1 - v ^ 2)) / 0.6e1 - t / a * b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^
2) * v)) * (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 -t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1 (t * (-(1 / a ^ 2 / b ^ 2 * E
/ (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a) * (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1;
t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t
/ a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^
2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * (a ^ 2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2)
/ 0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b

SOLIDS AND FLUIDS
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - (t

* E / (1 - v ^ 2) * v) (t / a * b * E / (1 - v ^ 2)) / 0.3e1 + t * a / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 + t * (-(1 / a ^ 2
/ b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 (t / a * b * E / (1 - v ^ 2)) /
0.6e1 - t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t
* (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1
+ (t * E / (1 - v ^ 2) * v) / 0.2e1 -(t / a * b * E / (1 - v ^ 2)) / 0.6e1
- t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t *
((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2
* E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1 + (t * E / (1 - v ^ 2) * v) /
0.2e1;
(t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) /
0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - t *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) (t * a / b * E / (1 - v ^ 2))
/ 0.6e1 - t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1
t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 + t * (-(1 / a ^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 t / a
* b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 + (t * a / b *
E / (1 - v ^ 2)) / 0.3e1 t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) -
0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1))
* (a ^ 2) * (b ^ 2) / 0.4e1 + t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E
/ (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.2e1 -t / a * b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1;
-(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2)
/ 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.2e1 (t / a
* b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.3e1 t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v)
- 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 + t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1) / 0.2e1 (t / a * b * E / (1 - v ^ 2)) / 0.3e1 + t * a / b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t * ((1 / a ^ 2 / b ^ 2
* E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 t * a / b * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^
2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^
2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) /

SOLIDS AND FLUIDS
0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1;

(t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) /
0.2e1 - (t * E / (1 - v ^ 2) * v) / 0.2e1 -t / a * b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1
t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 + (t * E / (1 - v ^ 2) * v) / 0.2e1 t / a * b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1
t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 +
(t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 -
v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 (t * a / b * E / (1 - v ^ 2)) / 0.6e1
- t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1;
(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v
^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -
v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^ 2) * v)
+ 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a)
* (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.2e1 -(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2)
* (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v
^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -
v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.2e1 t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2
/ b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 (t / a * b * E /
(1 - v ^ 2)) / 0.3e1 + t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a
^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) /
0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^ 2) *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1;
(t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
* ((1 / a / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v
^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1 - (t * E / (1 -
v ^ 2) * v) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * ((1 / a ^ 2 / b
^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^
2) * v)) * (b ^ 2) / 0.2e1 + (t * E / (1 - v ^ 2) * v) / 0.2e1 -t / a * b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1
- v ^ 2)) / 0.6e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1
/ (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^
2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1 (t * a /
b * E / (1 - v ^ 2)) / 0.6e1 - t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) +

SOLIDS AND FLUIDS
0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1))

* (a ^ 2) / 0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) /
0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 +
(t * a / b * E / (1 - v ^ 2)) / 0.3e1;];
klocal1=subs(klocal_raw,{a,b,E,t,v},{deltax,deltay,E1,t1,v1});
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
for j=1:8
for k=1:8
dof1=dofNo(i,j);
dof2=dofNo(i,k);
end
end
end
%--------------------------------------------------------------------------
% UDL
%--------------------------------------------------------------------------
eleNo=size(dofNo,1);
for i=1:dy
ele=dx*i;
dof_v2=dofNo(ele,4);
flocal(dof_v2,1)=flocal(dof_v2,1)+qy*deltay/2;
end
%--------------------------------------------------------------------------
% Impose BC
%--------------------------------------------------------------------------
KGG=kglobal;
val=dofe(i,1);
kglobal(val,:)=0;
kglobal(:,val)=0;
kglobal(val,val)=1;
flocal(val,:)=0;
KGG(val,:)=[];

SOLIDS AND FLUIDS
KGG(:,val)=[];
end
%--------------------------------------------------------------------------
% Solve d=k\f
%--------------------------------------------------------------------------
d=kglobal\flocal;
%--------------------------------------------------------------------------
% Show result for verification
%--------------------------------------------------------------------------
val=size(d,1);
u2=d(2*dx+1)
v2=d(2*dx+2)
u3=d(val-1)
v3=d(val)
%--------------------------------------------------------------------------
% plotting result (deflection)
%--------------------------------------------------------------------------
Zx=[];
Zy=[];
index=0;
for i=1:dy+1
for j=1:dx+1
index=index+1;
x_val=d(2*(index-1)+1,1);
y_val=d(2*index,1);
Zx(i,j)=x_val;
Zy(i,j)=y_val;
end
end
figure (1)
surf(X,Y,Zx)
title('x-displacement(m)')
colorbar
figure (2)
surf(X,Y,Zy)
colorbar
title('y-displacement(m)')
SOURCE CODE (4-NODES ELEMENT) – AUTOMATED

SOLIDS AND FLUIDS
clc;clear all;
%--------------------------------------------------------------------------
% Input
%--------------------------------------------------------------------------
E1=200e6;% Pa (N/m)


qy=-5000; % N

BCNode=[];
index=0;
for i=1:dy+1
val=(dx+1)*(i-1)+1;
index=index+1;
BCNode(index,1)=val;
end
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
dofNo=[];
NodeNo=[];
index=0;
for i=1:dy
for j=1:dx
index=index+1;
val1=(dx+1)*(i-1)+j;
val2=(dx+1)*(i)+j;
Nd1=val1;
Nd2=val1+1;
Nd3=val2+1;
Nd4=val2;
dof1=2*(Nd1-1)+1;
dof2=2*(Nd1-1)+2;
dof3=2*(Nd2-1)+1;
dof4=2*(Nd2-1)+2;
dof5=2*(Nd3-1)+1;
dof6=2*(Nd3-1)+2;
dof7=2*(Nd4-1)+1;

SOLIDS AND FLUIDS
dof8=2*(Nd4-1)+2;
dofNo(index,:)=[dof1 dof2 dof3 dof4 dof5 dof6 dof7 dof8];
end
end
% Coordinate of nodes
index=0;
NodeCoor=[];
for i=1:dy+1
for j=1:dx+1
index=index+1;
xcoor=deltax*(j-1);
ycoor=deltay*(i-1);
NodeCoor(index,:)=[xcoor ycoor];
end
end
dofe1=[];
index=0;
NdBC=BCNode(i);
ddofeli2=2*NdBC;
index=index+1;
index=index+1;
end
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
syms a b t E v
klocal_raw= [-0.2e1 / 0.3e1 * t / a * b * E / (1 - v ^ 2) - 0.2e1 / 0.3e1 *

t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) + t * ((1 / a ^ 2
* E / (1 - v ^ 2)) + 0.1e1 / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * a * b (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 /
(a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^
2) / 0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^
2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1
+ t * (-(1 / a ^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E /
(1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 + t * ((1
/ a / b * E / (1 - v ^ 2) * v) + 0.1e1 / a / b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * a * b t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -

SOLIDS AND FLUIDS
v / 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^

2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * ((1 / a ^ 2
/ b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1) -(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 - (t * E / (1 - v ^ 2)
* v) / 0.2e1 (t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v
^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E /
(1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a) * (b ^ 2) / 0.2e1 - (t * E / (1 - v ^ 2) * v) / 0.2e1;
(t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
* (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^ 2) * E / (1 -
v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1 + t * (-(1 / a
^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 + t * ((1 / a / b * E /
(1 - v ^ 2) * v) + 0.1e1 / a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a * b -0.2e1 / 0.3e1 * t / a * b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) - 0.2e1 / 0.3e1 * t * a / b * E / (1 - v ^ 2) + t * ((1
/ b ^ 2 * E / (1 - v ^ 2)) + 0.1e1 / (a ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * a * b (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) *
v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * (a ^ 2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) /
0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - (t *
E / (1 - v ^ 2) * v) (t * a / b * E / (1 - v ^ 2)) / 0.6e1 - t / a * b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^
2) * v)) * (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 -t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1 (t * (-(1 / a ^ 2 / b ^ 2 * E
/ (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * a) * (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1;
t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t
/ a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^
2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * (a ^ 2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2)
/ 0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - (t
* E / (1 - v ^ 2) * v) (t / a * b * E / (1 - v ^ 2)) / 0.3e1 + t * a / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t * ((1 / a ^ 2 / b ^
2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *

SOLIDS AND FLUIDS
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 + t * (-(1 / a ^ 2

/ b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b * E / (1 - v ^ 2) * (0.1e1
/ 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 (t / a * b * E / (1 - v ^ 2)) /
0.6e1 - t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t
* (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1
+ (t * E / (1 - v ^ 2) * v) / 0.2e1 -(t / a * b * E / (1 - v ^ 2)) / 0.6e1
- t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t *
((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2
* E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1 + (t * E / (1 - v ^ 2) * v) /
0.2e1;
(t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) /
0.2e1 + t * ((1 / a ^ 2 / b * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 - t *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) (t * a / b * E / (1 - v ^ 2))
/ 0.6e1 - t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1
t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 + t * (-(1 / a ^ 2 / b * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * b / 0.2e1 t / a
* b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 + (t * a / b *
E / (1 - v ^ 2)) / 0.3e1 t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) -
0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1))
* (a ^ 2) * (b ^ 2) / 0.4e1 + t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E
/ (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.2e1 -t / a * b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1;
-(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^
2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2)
/ 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.2e1 (t / a
* b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.3e1 t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v)
- 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 + t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1) / 0.2e1 (t / a * b * E / (1 - v ^ 2)) / 0.3e1 + t * a / b * E
/ (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 t * ((1 / a ^ 2 / b ^ 2
* E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) / 0.4e1 t * a / b * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^
2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^
2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) /
0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1;
(t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - t
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) /

SOLIDS AND FLUIDS
0.2e1 - (t * E / (1 - v ^ 2) * v) / 0.2e1 -t / a * b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.6e1
t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 + (t * E / (1 - v ^ 2) * v) / 0.2e1 t / a * b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1
t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2)
* E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) * (b ^ 2) /
0.4e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 +
(t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 -
v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 (t * a / b * E / (1 - v ^ 2)) / 0.6e1
- t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1;
(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v
^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -
v / 0.2e1)) * (a ^ 2) / 0.2e1 + t * ((1 / a / b ^ 2 * E / (1 - v ^ 2) * v)
+ 0.1e1 / a / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a)
* (b ^ 2) / 0.2e1 - t * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) /
0.2e1 -(t / a * b * E / (1 - v ^ 2)) / 0.6e1 - t * a / b * E / (1 - v ^ 2)
* (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v
^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 -
v / 0.2e1)) * (a ^ 2) / 0.2e1 - t / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 + t * E / (1 - v ^ 2) * (0.1e1 /
0.2e1 - v / 0.2e1) / 0.2e1 t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v
/ 0.2e1) / 0.6e1 - (t / a * b * E / (1 - v ^ 2)) / 0.3e1 (t * (-(1 / a ^ 2
/ b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^
2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t / (b ^ 2) * E / (1
- v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (b ^ 2) / 0.2e1 (t / a * b * E /
(1 - v ^ 2)) / 0.3e1 + t * a / b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a
^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) /
0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a / (b ^ 2) *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1;
(t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / (a ^ 2) / (b
^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 + t
* ((1 / a / b ^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / a / (b ^ 2) * E / (1 - v
^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) / 0.2e1 - (t * E / (1 -
v ^ 2) * v) / 0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v /
0.2e1) / 0.6e1 - (t * a / b * E / (1 - v ^ 2)) / 0.3e1 (t * ((1 / a ^ 2 / b
^ 2 * E / (1 - v ^ 2) * v) + 0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) *
(0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^ 2) / 0.2e1 - (t / b ^ 2 * E / (1 - v ^
2) * v)) * (b ^ 2) / 0.2e1 + (t * E / (1 - v ^ 2) * v) / 0.2e1 -t / a * b *
E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.6e1 - (t * a / b * E / (1
- v ^ 2)) / 0.6e1 (t * (-(1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1
/ (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * (a ^
2) / 0.2e1 + (t / b ^ 2 * E / (1 - v ^ 2) * v)) * (b ^ 2) / 0.2e1 (t * a /
b * E / (1 - v ^ 2)) / 0.6e1 - t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1
- v / 0.2e1) / 0.3e1 (t * ((1 / a ^ 2 / b ^ 2 * E / (1 - v ^ 2) * v) +
0.1e1 / (a ^ 2) / (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1))
* (a ^ 2) / 0.2e1 + t * (-(1 / a / b ^ 2 * E / (1 - v ^ 2) * v) - 0.1e1 / a
/ (b ^ 2) * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1)) * a) * (b ^ 2) /
0.2e1 t / a * b * E / (1 - v ^ 2) * (0.1e1 / 0.2e1 - v / 0.2e1) / 0.3e1 +

SOLIDS AND FLUIDS
(t * a / b * E / (1 - v ^ 2)) / 0.3e1;];
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
for j=1:8
for k=1:8
dof1=dofNo(i,j);
dof2=dofNo(i,k);
end
end
end
%--------------------------------------------------------------------------
% UDL
%--------------------------------------------------------------------------
for i=1:dy
ele=dx*i;
end
%--------------------------------------------------------------------------
% Impose BC
%--------------------------------------------------------------------------
KGG=kglobal;
val=dofe(i,1);
kglobal(val,:)=0;
kglobal(:,val)=0;
kglobal(val,val)=1;
flocal(val,:)=0;
KGG(val,:)=[];
KGG(:,val)=[];
end
%--------------------------------------------------------------------------

SOLIDS AND FLUIDS
% Solve d=k\f
%--------------------------------------------------------------------------
d=kglobal\flocal;
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
val=size(d,1);
u2=d(2*dx+1)
v2=d(2*dx+2)
u3=d(val-1)
v3=d(val)
%--------------------------------------------------------------------------
% plotting result (deflection)
%--------------------------------------------------------------------------
Zx=[];
Zy=[];
index=0;
for i=1:dy+1
for j=1:dx+1
index=index+1;
x_val=d(2*(index-1)+1,1);
y_val=d(2*index,1);
Zx(i,j)=x_val;
Zy(i,j)=y_val;
end
end
figure (1)
surf(X,Y,Zx)
title('x-displacement(m)')
colorbar
figure (2)
surf(X,Y,Zy)
colorbar
title('y-displacement(m)')
SOURCE CODE (8-NODES ELEMENT) – AUTOMATED
clc;clear all;
%--------------------------------------------------------------------------
% Input ref pg 83 schaum's
%--------------------------------------------------------------------------

SOLIDS AND FLUIDS
E1=200e6;% Pa (N/m)


qy=-5000; % N

BCNode=[];
index=0;
for i=1:dy
val2=(i-1)*((3*dx)+2);
val3=i*((3*dx)+2);
val4=1-dx;
Nd1=1+val2;
Nd5=1+val3;
Nd4=val3+val4;
Nd8=Nd4-1;
index=index+1;
BCNode(index,1)=Nd1;
index=index+1;
index=index+1;
end
BCNode=unique(BCNode);
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
dofNo=[];
NodeNo=[];
index=0;
for i=1:dy
for j=1:dx
index=index+1;
val1=2*(j-1)+1;
val2=(i-1)*((3*dx)+2);
val3=i*((3*dx)+2);
val4=j-dx;

SOLIDS AND FLUIDS
Nd1=val1+val2;
Nd4=val1+val3;
Nd6=val3+val4;
Nd5=Nd1+1;
Nd2=Nd1+2;
Nd3=Nd4+2;
Nd7=Nd4+1;
Nd8=Nd6-1;
dof1=2*(Nd1-1)+1; dof9=2*(Nd5-1)+1;
dof2=2*(Nd1-1)+2; dof10=2*(Nd5-1)+2;
dof3=2*(Nd2-1)+1; dof11=2*(Nd6-1)+1;
dof4=2*(Nd2-1)+2; dof12=2*(Nd6-1)+2;
dof5=2*(Nd3-1)+1; dof13=2*(Nd7-1)+1;
dof6=2*(Nd3-1)+2; dof14=2*(Nd7-1)+2;
dof7=2*(Nd4-1)+1; dof15=2*(Nd8-1)+1;
dof8=2*(Nd4-1)+2; dof16=2*(Nd8-1)+2;
dofNo(index,:)=[dof1 dof2 dof3 dof4 dof5 dof6 dof7 dof8...

dof9 dof10 dof11 dof12 dof13 dof14 dof15 dof16];
end
end
dofe1=[];
index=0;
NdBC=BCNode(i);
ddofeli2=2*NdBC;
index=index+1;
index=index+1;
end
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
syms a b t E v
klocal_raw= “NOT SHOWN BECAUSE TOO BIG”

SOLIDS AND FLUIDS
%--------------------------------------------------------------------------
%--------------------------------------------------------------------------
for j=1:16
for k=1:16
dof1=dofNo(i,j);
dof2=dofNo(i,k);
end
end
end
%--------------------------------------------------------------------------
% UDL
%--------------------------------------------------------------------------
for i=1:dy
ele=i*dx;
flocal(dof_v6,1)=flocal(dof_v6,1)+2*qy*deltay/3;
end
%--------------------------------------------------------------------------
% Impose BC
%--------------------------------------------------------------------------
val=dofe(i,1);
kglobal(val,:)=[];
kglobal(:,val)=[];
flocal(val,:)=[];
end
%--------------------------------------------------------------------------
% Solve d=k\f
%--------------------------------------------------------------------------
d=kglobal\flocal;
%--------------------------------------------------------------------------

SOLIDS AND FLUIDS

%--------------------------------------------------------------------------
val=size(d,1);
u3=d(2*(2*dx)+1-2)
v3=d(2*(2*dx)+2-2)
u5=d(val-1)
v5=d(val)

NONLINEAR FE FORMULATIONS (1-D) 2013
NONLINEAR BAR ELEMENT
10. INTRODUCTION
This chapter describes the material nonlinearity finite element formulation of bar elements.
It begins with the discretization of the bar nonlinear ordinary differential equation (ODE) by
the Galerkin Method which is then followed by the discussion on Picard’s and Newton-
Raphson’s iterative schemes. Numerical examples are given in the forms of two assembled
bar elements for easy tracing of the procedures. Matlab source-codes are also given at the
end of the chapter.
10. 1 ORDINARY DIFFERENTIAL EQUATION (ODE) OF BAR ELEMENT
Bar element represents one dimensional structure which deforms only in the axial direction.
A graphical description of a bar is given in Fig. 10.1 below.
Fig. 10.1: Bar Element
Fig. 10.1 shows that, when subjected to a distributed axial load, q(u,x), the bar will deforms
axially in the forms of axial displacement, u(x). The distribution of u(x) throughout the
domain (length, L) is typically depicted by the graph u(x) versus distance, x as also shown in
the figure. Such behaviour is mathematically described by the solution of the ODE which is
given as
( ) (10.1)
( ) ( ) ( )
where ( ) and ( ) are the material property and the cross-sectional area of the bar
respectively. Both variables can vary throughout the length of the bar and can also depend
© AIRIL YASREEN MOHD YASSIN, PhD & ERWAN HAFIZI KASIMAN, MEng Page 228
on the axial displacement, ( ). Such a variation can also be applied to the axial distributed
loading, ( ).
For every ODE (or PDE for that matter), the domain equation/s must be accompanied by
boundary equations and for this particular case, the equations are given as below:
( ) (10.2a)
( ) ( ) |
( )
( ) ( ) | (10.2b)
( )| (10.2c)
( )| (10.2d)
Since Eq. (10.1) is a 2nd order ODE, two boundary conditions out of the four given above
must be known in prior so as to have a well-posed problem.
Eq. (10.1) is general in the sense it contains both material and geometrical non-linearities. In
reducing the problem to material nonlinear only, the following is assumed:
i) The axial distributed load and the cross sectional area of the bar do not depend
on the axial displacement thus, q(x) and A(x), respectively
To further reduce the problem, let’s assume that:
ii) The cross-sectional area of the bar does not change throughout the length thus,
A is a constant in the axial direction, x
Based on the above assumptions, Eq. (10.1) can be reduced to
( ) (10.3)
( ) ( )
and by omitting the writing of the independent variable, x for ease of reading gives
(10.4)
( )
10.2 SHAPE FUNCTIONS AND DEGREE OF FREEDOMS
The finite element formulation of the bar can be obtained through various ways i.e. Direct
Method, Weighted Residual Method, Principle of Virtual Works, Minimum Principle of Total
Potential Energy etc. Herein, however, the formulation is obtained by employing Galerkin’s
Weighted Residual Method or simply stated as Galerkin Method.
The method starts by assuming an interpolation function, u(x) in the forms of polynomial.
( ) (10.5a)
( ) (10.5b)
Despite the familiar use of polynomial functions, it is preferable however, in finite element
formulation, to deal with a special set of functions, known as shape functions, and a
special sets of coefficient, known as degree of freedoms (dof), . This means, Eqs. (10.5)
must be equivalently expressed as:
( ) (10.6a)
( ) (10.6b)
The above equations can be expressed compactly in components forms as follows:
( ) for (10.7a)
( ) for ,3 (10.7b)
Being referred to as the degree of freedoms, , , and in Eqs. (10.5) (or in Eqs.
(10.6), are actually the nodal values of the axial displacement, u(x). These are graphically
described in the following figure.
Fig. 10.2: Degree of freedoms of bar elements
The conversion from Eqs. (10.5) to Eqs. (10.6), hence the derivation of the shape functions,
can be done as follows.
For linear element
i) Evaluate Eq. (10.5a) at the location of the nodes (i.e. at both ends, )
and equate them according to the dof give:
( )| ( )
(10.8a)
( )| ( )
ii) Solving for the values of then gives:
(10.8b)
iii) Inserting Eq. (10.8b) into Eq. (10.5a) and by factorizing give:
(10.8c)
( )
By comparing Eq. (10.8c) and Eq. (10.6a), it can be concluded that:
(10.8d)
Both and take the shapes as shown in Fig. 10.3. Take note of the shapes as they will
Fig. 10.3: Linear shape functions
For quadratic element
i) Evaluate Eq. (10.5b) at the location of the nodes (i.e. at both ends,
) and equate them according to the dof give:
( )| ( ) ( )
( )| ( ) ( ) (10.9a)
( )| ( ) ( )
ii) Solving for the values of then gives:
(10.9b)
iii) Inserting Eq. (10.9b) into Eq. (10.5b) and by factorizing give:
(10.9c)
( ) ( )( ) ( ) ( )
By comparing Eq. (10.9c) and Eq. (10.6b), it can be concluded that:
( )( )
( )
(10.9d)
( )
, and take the shapes as shown in Fig. 10.4. Take note of the shapes as they will be
referred later in the discussion of integration by parts and the resulting boundary terms.
Fig. 10.4: Quadratic shape functions
10.3 DISCRETIZATION BY GALERKIN METHOD
Having established the shape functions and the degree of freedoms of the bar element, we
are all set to proceed with the discretization of the ODE.
For linear element, inserting Eq. (10.8c) into Eq. (10.4) gives:
( )
( ) (10.10a)
whilst for quadratic element, inserting Eq. (10.9c) into Eq. (10.4) gives:
( ) (10.10b)
( )
Now, for the purpose of making easy the tracing of the procedure, let’s assume that
( ) takes the following linear constitutive relationship:
( ) (10.11)
where is taken as a constant. By inserting Eq. (10.8c) and (10.9c) subsequently, Eq. (10.11)
is then modified to become:
( ) ( ) (10.12a)
( ) ( ) (10.12b)
By inserting Eqs. (10.12) into Eqs. (10.10), the following ODEs are obtained:
( ) (10.13a)
( )
( ) (10.13b)
( )
Observing Eqs. (10.13), we can see that both can be compactly expressed in the forms of
components as:
( ) (10.14)
Take note that, from now on, we are going to deal frequently with the component forms
like Eq. (10.14), as this would eliminate the necessity to distinguish between linear and
quadratic elements. With this, it should also demonstrate the advantage of such notation
system.
Since we have i number of unknowns (because we want to determine the values of the dof,
) we thus need i number of simultaneous equation. These required equations can be
obtained by multiplying Eq (10.14) subsequently with a set of weighting functions. For
Bubnov-Galerkin formulation, the weighting functions are also the shape functions. For
example, for a linear bar element, the two simultaneous equations can be obtained as
follows:
( )
( )
(10.15a)
( )
( )
So, in component forms, the required i simultaneous equations can be given as:
( )
( ) (10.15b)
The continuous nature of Eqs. (10.15) (since it is still in ODE forms) can be converted into
the equivalent algebraic forms by integrating over the length of the bar. This action is
obtained by forcing the residual to zero. The algebraic forms of Eq. (10.15b) are thus given
as:
( )
∫ ( ) (10.16)
It can be seen that Eq. (10.16) involves the integration over a product of functions, thus
integration by parts (IBP) can be employed. The reasons for employing the IBP are:
i) To reduce the continuity requirement since only first and lower derivatives will
be dealt with
ii) To induce (make explicit) the natural boundary conditions as boundary terms
resulting from the IBP
Employing IBP to Eq. (10.16) gives:
( )
∫ ( ) ∫ | (10.17)
In Eq. (10.17) above, the dof, is taken out from the integral sign because it represents
constants thus does not involve in both differentiation and integration. The last terms on
the right hand side of Eq. (10.17), are called boundary terms resulting from the IBP.
By inserting Eqs. (10.7) and Eq. (10.11), the boundary terms can be given as:
( ) ( )
| | (10.18)
And further comparison with Eqs. (10.2a) and (10.2b), one can identify that Eq. (10.18) is
actually the natural boundary conditions of the bar; hence it can be given that:
( ) ( )
| | | | (10.19)
Eq. (10.19) deserves further explanation. The prevalence of the natural/force boundary
conditions, i.e. and within the equation depends on the accompanying value of the
shape functions, being evaluated at the location of the respective node i.e. | and
| . As can be seen, say from Fig. 10.3 (linear element), the value of the shape functions
at the location can be given as:
at | |
(10.20)
at | |
As a result, for linear bar element, the boundary terms of Eq. (10.19) can be specifically
given as:
| | ( ) ( )
(10.21)
| | ( ) ( )
But, for quadratic bar element, there is one particular feature worth of attention, that is, the
evaluated value of always null at the location of the edge nodes (refer Fig. 10.4). By
taking this into consideration, the boundary terms of Eq. (10.19) for a quadratic bar element
are thus given as:
| | ( ) ( )
(10.22)
| | ( ) ( )
| | ( ) ( )
Due to these various conditions and combination, it is common in the discussion of finite
element formulation to write Eq. (10.17) simply as:
∫ ( ) ∫ (10.23)
where refers to boundary terms or nodal loads. Eq. (10.23) completes the discretization
of the bar’s ODE by Galerkin Method.
10.4 LOCAL LOAD VECTOR, { } AND STIFFNESS MATRIX, [ ]
Now, let’s discuss the other term on the right hand side of Eq. (10.23), that is ∫ .
∫ (10.24)
load, q that is acting on the domain; effects as perceived at the nodes. To better describe
such equivalence, let’s refer to Fig. 10.5. As shown in the figure, the distributed loading, q(x)
which is acting on the domain of the bar would deform the bar in the same manner as if the
bar is subjected to the two nodal loads (or three for quadratic bar) which values are
calculated by Eq. (10.24).
Fig. 10.5: Equivalent nodal loads
Having defined Eq. (10.24), Eq. (10.23) can be rewritten as:
∫ ( ) (10.25)
Eq. (10.25) has been derived for one particular bar element, thus it is called local equilibrium
equation. For a more compact statement, Eq. (10.25) is usually written in component forms
as:
(10.26)
where is termed as the local stiffness matrix of the bar element. and , on the other
hand, are the local vector of dof and the load vector of the bar element, respectively. Thus,
the stiffness matrix and the load vector of the bar element can be given as:
∫ ( ) (10.27)
[ ]{ } { } (10.28)
In an expanded matrix forms, the stiffness matrix of linear and quadratic elements can be
respectively given as:
(10.29a)
[ ] [ ]
∫ ( )( ) ∫ ( )( )
∫ ( )( ) ∫ ( )( )
[ ]
[ ] [ ]
∫ ( )( )
(10.29b)
∫ ( )( )
[ ]

Stiffness matrices which are given in Eqs (10.29) are still in integral forms. Generally, to carry
out the integration, numerical integration is used due to the complexity. However, since
those in Eqs (10.29) are simple, direct integration is used herein. The main purpose is that,
with direct (analytical) integration, the evolution of the procedure can easier be followed
since all variables remained to be seen. For demonstration purposes, the integration of
are shown below for both linear and quadratic elements. The fully integrated stiffness
matrix of both elements is then given.
(10.30a)
∫ ( )( )
∫ (( ) ( ) ) (( ) ( ))
( )
(10.30b)
∫ ( )( )
∫ (( )( ) ( )
( ) ) (( )( ))
( )
The fully (or analytically) integrated stiffness matrix for both elements can be given as:
( ) ( ) (10.31a)
[ ] [ ]
( ) ( )
[ ]
[ ]
(10.31b)
(10.32a)
∫
{ } { }
∫
{ }
(10.32b)
∫
{ } { } { } ∫
∫
{ }
Component forms and matrix forms will be used interchangeably in our forthcoming
discussion.
10.4 NUMERICAL INTEGRATION
So far, the determination of the stiffness matrix terms, as have been given by Eqs. (10.31a)
and (10.31b), were carried out by analytical integration. For more general elements, such a
determination is carried out by numerical integration, for reasons as follows:
i) To cater for irregular domain (i.e. isoparametric, subparametric,

superparamteric)
ii) Suits computer programming; it is preferred to ‘evaluate’ than ‘integrate’
iii) To employ reduced integration in reducing the effect of shear locking
These points, however, will be detailed in upcoming chapters. In fact, a numerical

integration is not even needed for the problem at hand as exact integration of the stiffness
matrix terms (and tangent stiffness matrix, as will be shown later) have been obtained
already (as given by Eqs. (10.31a) and (10.31b)). But, since numerical integration would be
the common approach to determine the stiffness matrix values (thus a very essential
feature of finite element), it is therefore wise to introduce the concept while it is still in a
simple context.
The basic concept of numerical integration is to map the actual domain (or physical domain)
which is usually irregular (but not in the present case) into a regular domain known as
natural domain. If Gauss-Legendre quadrature is employed, this natural domain usually in
the particular forms of rectangular with a side length of 2 units for two-dimensional
problem and a line of 2 units length for one-dimensional problem. This mapping is shown in
Figure 10.5a.
Figure 10.5a: Mapping of domain
As can be seen, the mapping process requires the coordinates of the physical domain to be
expressed as an interpolation functions in terms of shape functions and nodal coordinates,
which, for one-dimensional problem, can be given as:
( ) (10.33)
where, ( ) are the shape functions derived for the natural domain. The procedure to
derive these shape functions is similar to the procedure outlined previously (Eqs. (10.8) and
(10.9) except that the nodal values are evaluated at for linear bar element and
for quadratic elements. Accordingly, ( ) can be given as follows (for
quadratic element only to ease the discussion):
(10.34)
( )
(10.35)
(10.36)
( )
Now for demonstration purposes, let’s get the value of one of the stiffness matrix terms, say
(which was previously given by Eq. (10.30b)):
∫ ( )( ) (10.37)
It must be noted that in , all terms including the shape functions, are expressed in
terms of the physical coordinates, . To express everything in terms of the natural
coordinates, we must first obtain the equivalent differential. Expanding by Taylor
expansion:
(10.38)
where is the Jocabian. From Eqs. (10.33) to (10.36), the Jacobian, can be determined as
follows:
( ( ) ( ) ( ) ) (10.39)
( )
For the case where node 2 is at mid-point of the physical domain (actual element) i.e.
, the Jacobian, can be shown as simply:
(10.40)
Inserting Eq. (10.40) into Eq. (10.38), the differential can be given as:
(10.41)
Now, since we have established the differential terms, let’s determine the equivalent
derivatives terms. Based on chain-rule, the differential operator, can be expressed as:
(10.42)
Based on Eq. (10.39) and Eq. (10.40), Eq. (10.41) can be given as:
(10.43)
Havinq established the differential and the differential operator in the natural domain, the
equivalent can be given as:
∫ ( )( )( ) (10.44)
∫ ( )( )( )
As can be seen, once converted into the natural domain, the integration limits will then
range from -1 to 1. For this limits, Gauss-Legendre quadrature provides the numerical
integration as:
∑ ( )| | |
(10.45)
where and are the number and the location of Gauss points, respectively and is the
Gauss weight. Their values are given in Table 10.1 below.
Table 10.1
Gauss Point, Location of Gauss Point, Gauss Weight,
2 ( )( ) ( )( )
3 ( )( )( ) ( )( )( )
4 ( )( ) ( )( )
Now, to have a better understanding towards the whole procedure, let’s demonstrate the numerical integration of for . Inserting
the shape functions of Eqs. (10.34), (10.35) and (10.36) into Eq. (10.45) yields:
( ( ))
| ( )
∑ (( ( )) ( ) ( ( )) )| | |
(10.46)
Based on the values from Table 1 for the case , Eq. (10.46) can be expanded as:
( ) (( ( )) ( ) ( ( )) )| ( )| ( )|
( ) (( ( )) ( ) ( ( )) )| ( )| ( )|
( ) (( ( )) ( ) ( ( )) )| ( )| ( )|
(10.47)
or
( ) (( ( ( ) ( ) )) ( ( ) ) ( (( ) ( ) )) )
( ( )) ( ( ))
( ) (( ( ( ) ( ) )) ( ( ) ) ( (( ) ( ) )) )
( ( )) ( ( ))
( ) (( ( ( ) ( ) )) ( ( ) ) ( (( ) ( ) )) ) ( ( )) ( ( ))
(10.48)
thus;
( ) (10.49)
Eq. (10.49) is similar to the exact integration as given by Eq. (10.30b). This is due to the fact that was chosen as the number of
quadrature. A full integration will be obtained if the function to be integrated is of degree and the number of quadrature chosen is at
least, .
If quadrature higher than is chosen, it will not increase the accuracy of the integration
whilst quadrature lesser than will result in the so called reduced integration. In the
present case, due to the nonlinearity, the degree is 4 (i.e. ) thus must be at least 3 if
exact integration is to be obtained.
It must be noted however that since for this problem analytical integration is available for
the matrices, the accompanying MATLAB computer programs use these analytically
integrated values (Eqs. (10.31)) for the source coding. Numerical integration will be
employed for higher dimensions (i.e. two dimensional problems) in forthcoming chapters.
10.5 ASSEMBLY OF ELEMENTS: GLOBAL STIFFNESS MATRIX, [ ] AND LOAD VECTOR, { }
element to form the global stiffness matrix, [ ] and load vector, { }. Although in practices,
hundreds and thousands of elements might be assembled, herein, only the assembly of two
bar elements is demonstrated for easy tracing of the procedure. Fig. 10.6 shows the two
assembled bar elements together with the local and global numbering of nodes and dof.
Fig. 10.6a: Local and global numbering of linear bar element
Fig. 10.6b: Local and global numbering of quadratic bar element
Based on the above, it is obvious that we need to have extra notation to distinguish
different elements. In doing this, we introduce the use of superscript as follows:
i) The local stiffness matrix of element e is denoted as or [ ] (10.50a)
ii) The local load vector of element e is denoted as or { } (10.50b)
iii) The vector of dof of element e is denoted as or { } (10.50c)
thus be given as:
[ ]{ } { } [ ]{ } { }
(10.51a)
Element 1 Element 2
[ ]{ } { } [ ]{ } { } (10.51b)
Element 1 Element 2
Now the assembly of the elements can be done as follows.
[ ]{ } { }

(10.52a)
[ ]{ } { }
[ ]{ } { }
[ ]{ } { }
(10.52b)
[ ]{ } { }
[ ]{ } { }
Based on Eqs. (10.52), it can be deduced that[ ], { } and { } are the global stiffness matrix,
global vector of dof and global load vector, respectively.
10.6 CONNECTIVITY MATRIX
The assembly of local matrices into global matrix (as given by Eqs. (10.52a) and (10.52b)),
was carried out rather manually and intuitively. This approach however, would not be an
efficient way when it comes to computer programming. For a more general and systematic
approach, connectivity matrix is employed. For the assembly of two linear elements and by
referring to Figure (10.6a), the connectivity matrix can be given as:
1 2
1 2 1
(10.53a)
2 3 2
and for the assembly of two quadratic elements and by referring to Figure (10.6b), the
connectivity matrix can be given as:
1 2 3
1 2 3 1
(10.53b)
3 4 5 2
As can be seen from Eqs. (10.53a) and (10.53b), connectivity matrix is bracketed by a
horizontal array of dof numbering and a vertical array of element numbering. The matrix
basically maps the local dof numbering (i.e horizontal array) to the global numbering (i.e.
elements within the matrix) according to the elements (i.e. vertical array).
For example, based on the Eq. (10.53b) and for quadratic element, of element 1 would
‘overlap’ with of element 2. Also, of element 2 is mapped to globally and so on. A
careful examination would reveal that all these relationships agree with Figure (10.6b) and if
utilized, will result in an assembled global matrix similar to Eq. (10.52b).
The automatic generation of the connectivity matrix and its utilization for the assembly of
the global matrix will be demonstrated later during the discussion of the accompanying
MATLAB source codes.
10.7 IMPOSITION OF ESSENTIAL BOUNDARY CONDITIONS
As mentioned earlier, to have a well-posed problem, essential (or displacement) boundary

conditions must be imposed prior to the solution of the equilibrium equation. There are
various ways to impose boundary conditions such as row-column condensation, penalty
method, Lagrange multiplier method etc. Herein, row-column condensation is employed.
The basic concept of the row-column condensation can be explained as follows. Assume
that we have a set of three simultaneous equations given as:
(10.54a)
Now, let’s assume that is known (the boundary conditions) i.e. ̅ . By inserting this
known value into Eq. (10.54a) gives:
̅ (10.54b)
Since ̅ is known, terms associated with it can be moved to the right hand side of Eq.
(10.54b), thus:
̅ (10.54c)
By examining Eq. (10.54c), one can realize that we have two unknowns with three
equations. Since we need only two equations for the solution, we can just use say, the first
and the third equations. (We can also decide to use the second equation together with
either first or third equation), thus
̅ (10.54d)
Now, the condensation can be made more obvious if we discuss the above in matrix forms.
Eq. (10.54a) can be given in matrix forms as:
[ ]{ } { } (10.54f)
If ̅ is known, Eq. (10.54f) can be condensed (thus the employment of the boundary
conditions) as follows:
̅
[ ]{ } { } (10.54g)
̅
The above explains the basic concept of row-column condensation where it can be
concluded that, an imposition of number of boundary conditions to a stiffness matrix of
size and to a load vector of , would condense the former to a size of
( ) ( ) and the latter to as size of ( ) . Once condensed, the solution
of the matrix system can proceed.
Now, let’s employ this method directly on our global equilibrium equation as given by Eqs.
(10.52). For the purpose of discussion, let’s assume that the right end of the bar is having a
specified displacement. Such a condition leads to the following essential boundary
conditions:
For linear element: ̅
For quadratic element: ̅
The imposition of the above boundary conditions to the global equilibrium equation can be
done as follows.
[ ]{ } { }
̅
̅
[ ]{ } { ̅}
(10.55a)
(10.55b)
[ ]{ ̅ } { }
̅
̅
̅
̅
[ ]{ } { } (10.55c)
Eqs. (10.55) are now ready to be solved. There are various ways to solve a matrix equation
such as direct inversion, Gauss Elimination, Cholesky Decomposition, etc. In the
accompanying MATLAB source codes, Gauss Elimination is adopted. Symbolically, the
solution of the matrix equation is represented as
{ } [ ] { } (10.55d)
10.8 NONLINEARITY AND THE NEED FOR ITERATIVE SCHEMES
As mentioned, herein, we concern with the material nonlinear finite element formulation of
bar element. The source of this nonlinearity is the dependence of the value of the material
property, ( ) on the solution (or displacement), ( ) itself. Such dependency hence the
nonlinearity can be made obvious if we compare the ODEs and the equation of equilibriums
between linear and nonlinear bar problems. This is done as follows.
For a linear problem, the ODE of a bar is given as:
(10.56)
( )
whilst for nonlinear bar, the ODE is as previously given by Eq. (10.4), which is rewritten
herein as:
(10.57)
( )
As can be seen, the difference between Eq. (10.56) and Eq. (10.57) is on the constitutive
relationship of the material property, E where it is a constant in the former and a function of
( ) in the latter.
Discretizing both equations by the Galerkin method, say for linear bar element, gives:
a) For linear problem
(10.58a)
[ ]{ } { }
b) For nonlinear problem (as given by Eq. (10.31a))
( ) ( )
(10.58b)
[ ]{ } { }
( ) ( )
By examining Eqs. (10.58), it can be seen that, whilst the stiffness matrix for the linear
problem consists of constant coefficients, the stiffness matrix for the nonlinear problem
contains the unknown coefficients (or the dof), i.e. , . The nonlinearity can further be
made obvious if we expand Eqs. (10.58)) as follows.
a) For linear problem
(10.59a)
b) For nonlinear problem (as given by Eq. (10.31a))
( ) ( )
(10.59b)
( ) ( )
Simultaneous equation as given by Eq. (10.59a) is linear thus it can be solved explicitly for
and . In contrast, Eq. (10.59b) is quadratic in and thus cannot be solved directly.
Based on this, it can be said that, solving a nonlinear problem is also an act of solving for the
roots of a set of polynomial equation, thus is in need of iterative schemes.
Finally, due to the difference between linear and nonlinear problems, there is a need to
express the dependency on ( ) in the expression of the stiffness matrix of the latter, thus
the equilibrium equation of a nonlinear problem is written from now on as:
a) For local equilibrium

(10.60a)
[ ( )]{ } { }
b) For global equilibrium

(10.60b)
[ ( )]{ } { }
In the next discussion, we are going to consider two iterative schemes; Piccard’s and
Newton-Raphson’s.
10.9 BASIC CONCEPT OF ITERATIVE SCHEME
Consider the nonlinear equilibrium equation as given in Eq. (10.60b); rewritten herein as:
[ ( )]{ } { } (10.61)
Basically, we are in a dilemma as the vector { }that we want to determine (i.e. by inversing
the stiffness matrix, [ ( )]) in turn, is what is needed for the determination of the matrix in
the first place. So, the best option for us to proceed, (in fact the only option that we have),
is to use a known (or assumed) set of { } in calculating [ ( )] and once calculated, the
latter is inversed to solve for{ }. One can expect that this is going to be a sequential
process; a process which sequence can be mathematically expressed with the assistance of
a new superscript system, detailed as:
[ ( ) ]{ } { } (10.62)
where refers to the present stage or present iteration. This then makes ( ) to
represent the immediate previous stage or iteration. Since this is a material nonlinear
formulation, no superscript needs to be associated with the load vector as the forces acting
on the system is independent of { } hence the iteration. And, since in general { } and
{ } are not similar (except when the solution has converged), the left hand side terms of
Eq. (10.62) cannot be equalled to the load vector, { } hence the inequality sign. In fact,
the difference between these terms would yield a residual, thus:
{ } [ ( ) ]{ } { } (10.63)
The aim of the iteration process is to reduce the residual, { } to a point of convergence,
that is { } is small enough for us to accept that { } and { } can be considered as
‘equal’ thus { } can be considered as the solution of the equation of equilibrium. The
measure of the residual can be given in two types of norm, firstly by:
(10.64)
√∑
where is a specified negligible value termed convergence criterion. Alternative, Euclidean

norm is used, given as:
∑ | |
√ (10.65)
∑ | |
At the first iteration, i.e. (thus, ), { } is termed as initial condition and it

should not be mistaken as boundary conditions although the latter are also known in priori.
The differences between the initial and the boundary conditions are as follows:
i) Whilst boundary conditions are known and applied throughout the process, the
known set { } is only employed once at the beginning so as to kick start the
iteration
ii) Whilst boundary conditions is the known conditions at the boundary nodes, the
known set { } is the value at all nodes; both in the domain and at the
boundaries
Despite some of the variations, all iterative schemes basically share the same basic concept
as detailed above. To be specific, most of the available schemes would have the following
features:
i) The use of ‘initial conditions’ to kick start the iteration

ii) Looping of ‘solve and insert back’ the value of { } from and into the equilibrium
equation
iii) The calculation of the residual, { } and it’s assessment against the convergence
criterion
10.10 PICCARD SCHEME
Picard scheme is also known as the direct iteration method. It is called direct because the
scheme is not far from the basic concept as previously outlined. But, its simplicity is not
without a price. Piccard scheme usually works for mild nonlinear problems but diverges for
severe nonlinearity.
The step by step procedure of the scheme is as follows.
STEP 1: take or assume a known value of { }
STEP 2: use the known { } to calculate [ ( ) ] using Eqs. (10.31)
STEP 3: solve { }
STEP 4: calculate the residual, { } using Eq. (10.63) and check for convergence using
either Eq. (10.64) or Eq. (10.65).
STEP 5: if converge, stop iteration else solve { } using Eq. (10.62), else
STEP 6: repeat STEP 2 by using the just solved { } as { }
The flowchart of the above procedure is shown in Fig. 10.7.
INPUT
For
r=1:m
YES
r=1 {𝑈}𝑖𝑛𝑖𝑡𝑖𝑎𝑙
NO
{𝑈}𝑟
[𝐾]𝑟
Impose essential
boundary conditions
{𝑈}𝑟 [𝐾]𝑟 {𝐹}
{𝑅}𝑟
YES
𝑒𝑟𝑟𝑜𝑟 < PRINT STOP
NO
YES NO WRITE A
r<1 STOP
MESSAGE
Fig. 10.7: Piccard flowchart
10.11 NEWTON-RAPHSON SCHEME
An alternative iterative scheme which performs better when the nonlinearity is more severe
is the Newton-Raphson scheme. The basic concept of the scheme is based on the Taylor’s
series expansion of the equilibrium equation. The residual of the equilibrium equation can
be given as:
{ ( )} [ ( )]{ } { } (10.66)
Expanding Eq. (10.66) by Taylor’s series about the known ( ) solution gives:
{ ( ) } (10.67)
{ ( )} { ( ) } { }
{ }
where the series has been truncated up to linear terms only. Rearranging gives:
{ ( ) } (10.68)
{ } { ( ) }
{ }
Let’s define a new matrix and call it as tangent stiffness matrix, [ ( ) ], thus;
{ ( ) } (10.69)
[ ( ) ]
{ }
By inserting Eq. (10.69) into Eq. (68), the following is obtained.
[ ( ) ]{ } { ( ) } (10.70)
Examining Eq. (10.70), one can observed that { } is now solvable since the other terms i.e.
[ ( ) ] and { ( ) } are known because these terms use the results from the previous
iteration i.e. ( ) solution. By inversing [ ( ) ], { } is thus can be obtained as:
{ } [ ( ) ] ( { ( ) }) (10.71)
Once { } is solved, the solution at the iteration is thus:
{ } { } { } (10.72)
In the next iteration, { } of Eq. (10.72) will be inserted back into Eq. (10.70) and take the
role as { } in the determination of [ ( ) ] and { ( ) }. This process is repeated
until the solution converges.
10.12 TANGENT STIFNESS MATRIX, [ ( ) ]
In the previous discussion, definition of tangent stiffness has been given. Herein, the
detailed formulation of the matrix is detailed out. For this purpose, Eq (10.69) is rewritten as
below.
{ ( ) } (10.73)
[ ( ) ]
{ }
The nominator on the right hand side terms of Eq. (10.73) is actually the residual of the
equilibrium equation at the ( ) iteration which can be given as:
{ ( ) } [ ( ) ]{ } { } (10.74)
By inserting Eq. (10.74) into Eq. (10.73) gives:
{[ ( ) ]{ } { }} (10.75)
[ ( ) ]
{ }
By employing chain-rule and since { } does not depend on { } (we are not considering
geometric nonlinearity), Eq. (10.75) can be expanded to:
[ ( ) ] { }
[ ( ) ] { } [ ( ) ]
{ } { }
[ ( ) ]
{ } [ ( ) ] (10.76)
{ }
[ ̂( ) ] [ ( ) ]
Eq. (10.76) is derived at global level. It is more convenient to derive the tangent stiffness
matrix at local. Therefore, for local and in terms of components, the tangent stiffness matrix
of an element can be given as:
[ ( ) ]
[ ( ) ] { } [ ( ) ]
{ }
(10.77)
[̂ ( ) ] [ ( ) ]
whilst ( ) has been derived as the stiffness matrix of the element as given by Eqs.
(10.31), ̂ ( ) deserves further and more detailed formulation. This is given next.
Based on Eq. (10.77), it can be given that:
[ ( ) ]
̂ ( ) { } (10.78)
{ }
By inserting Eq. (10.27) into Eq. (10.78) gives:
[∫ ( ) ]
̂ ( ) { } (10.79)
{ }
but it can be shown that:
( )
(10.80)
{ }
For example, consider linear bar element and take , thus
( )
(10.81)
{ }
A more mathematically elegant way is by the use of Kronecker delta, . It can be shown
that
{ }
{ (10.82)
{ }
By inserting Eq. (10.82) into Eq, (10.80) and since Kronecker delta, acts as substitution
operator, it can thus be obtained that:
( )
(10.83)
{ }
By inserting either Eq. (10.80) or Eq. (10.83) into Eq. (10.79), the following is obtained:
̂ ( ) ∫ ( ) { } (10.84)
Since we are freely can place { } either inside or the outside of the integral (since it is a
vector of constant), it is preferable to place it inside the integral; a preference which is to be
obvious later. Thus;
( )
̂ ( ) ∫ ( ) (10.85)
The association of with in the Eq. (10.85) above is appropriate because together,
they would represent the interpolation function, ( ) . Having established all
these, we are now in the position to formulate the fully integrated tangent stiffness matrix,
( ) for both linear and quadratic bar element. Using Eq. (10.85) and the appropriate
shape functions as given by Eq. (10.8d) and Eq. (10.9d), it can be obtained that:
̂ ̂
[ ] [ ] [ ]
̂ ̂
( ) ( )
∫ ( ) ∫ ( )
( ) ( )
∫ ( ) ∫ ( )
[ ]
[ ]
(10.86a)
̂ ̂ ̂
[ ] [̂ ̂ ̂ ] [ ]
̂ ̂ ̂
( )
∫ ( ) ̂ ̂
̂ ̂ ̂
( )
∫ ( ) ̂ ̂
[ ]
[ ]
(10.86b)
Stiffness matrices which are given in Eqs. (10.86) are still in integral forms. For
demonstration purposes, the integration of ̂ are shown below for both linear and
quadratic elements. The fully integrated stiffness matrix of both elements is then given.
( ) (10.87a)
̂ ∫ ( )
∫ ( ( )( ) (( ) ( ) ))
( )
( ) (10.87b)
̂ ∫ ( )
∫ ( ( )( ) (( ) ( )
( ) ))
( )
By repeating the process for other elements of the matrix, the fully integrated stiffness
matrix for both elements can be given as:

̂ ̂
[ ] [ ] [ ]
̂ ̂
( ) ( )
( ) ( )
[ ]
( ) ( )
( ) ( )
[ ]
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ]
(10.88a)
a) For quadratic bar element
̂ ̂ ̂
[ ] [̂ ̂ ̂ ] [ ]
̂ ̂ ̂
[ ]
[ ]
and by inserting Eq. (10.31b), the tangent stiffness, [ ] is thus:
[ ]
[ ]
(10.88b)
The step by step procedure of the Newton Raphson scheme is as follows.
STEP 1: take or assume a known value of { }
STEP 2: use the known { } to calculate the global tangent stiffness matrix,
[ ] using Eqs. (10.88)
STEP 3: calculate the residual, { } using Eq. (10.66)
STEP 4: solve the vector of incremental dof, { } using Eq. (10.71)
STEP 5: calculate the present iteration of solve { } using Eq. (10.72)
STEP 6: check for convergence using either Eq. (10.64) or Eq. (10.65).
STEP 7: if converge, stop iteration else repeat STEP 2 by using the just solved { } as
{ }
The flowchart of the above procedure is shown in Fig. 10.8.
INPUT
For r=1:m
YES
r=1 {𝑈}𝑖𝑛𝑖𝑡𝑖𝑎𝑙
NO
{𝑈}𝑟
[𝐾]𝑟
[𝑇]𝑟
{𝑅}𝑟
Impose essential boundary

conditions
{ 𝑈} [𝑇]𝑟 \ {𝑅}𝑟
{𝑈}𝑟 {𝑈}𝑟 { 𝑈}
YES PRINT
𝑒𝑟𝑟𝑜𝑟 < STOP
NO
YES r<m NO WRITE A STOP

MESSAGE
Fig. 10.8: Newton-Raphson flowchart
10.12 NUMERICAL EXAMPLES
Herein, we are going to provide the numerical examples to demonstrate the use of the
procedures which we have discussed previously. Three examples will be given as follows:
i. Newton-Raphson scheme: linear bar

ii. Piccard scheme: quadratic bar
iii. Newton-Raphson scheme: quadratic bar
The examples solve the problem of a bar structure as shown in Fig. 10.9 by discretizing it (or
meshing) it with two assembled bar elements.
Fig. 10.9: Bar problem example
The material and the geometrical data for the analysis are given as follows:
Cross-sectional area, = 1 length2
Material coefficient, =1
Total length of structure (bar) = 1 length
Length of a bar element, = 0.5 length
Boundary conditions:
Natural (or force) : a distributed loading with an intensity of -1 is acting

throughout the length of structures the structures and no
nodal loads are considered, thus
Essential (or displacement) : a specified displacement of 1.4142 is applied at the far-right

end/edge of the structure, thus
(for linear element)
(for quadratic elements)
‘Initial’ conditions:
Linear element : whilst
Quadratic element : whilst
Convergence criterion:
Exact solutions (or closed –form solution) of the ODE (for comparison purpose):
at
at
I. Newton-Raphson scheme: linear bar
a) First iteration, ( )
By using previously given data and Eq (88a) the tangent stiffness matrices at the first
iteration (i.e ( )), [ ] and [ ] for element 1 and 2, respectively are thus given
as:
̂ ̂
[ ] [ ] [ ]
̂ ̂
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ( ) ( ) ] [ ( ) ( ) ]
) ( ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ( ) ( ) ] [ ( ) ( ) ]
[ ]
(10.89a)
̂ ̂
[ ] [ ] [ ]
̂ ̂
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ( ) ( ) ] [ ( ) ( ) ]
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ( ) ( ) ] [ ( ) ( ) ]
[ ] [ ]
[ ]
(10.89b)
And the load vector as well as the vector of dof can be given as:
Element 1
∫
{ } { } { } { } { } { } { }
∫
{ } (10.90a)
{ }
{ } { }
Element 2
∫
{ } { } { } { } { } { } { }
∫
{ }
{ }
{ } { } { }
(10.90b)
By assembling the two elements, we thus obtain the following. This is done by inserting the
values derived in Eqs. (10.89) and (10.90) into Eq. (10.70).
(10.91)
[ ] { }
[[ ] { }
{ } ]
Expanding the right hand side of Eq. (10.91) gives:
(10.92)
[ ] { } { }
Take note that is a specified boundary condition thus does not change. This unchanged
value means there is no variation in thus is zero throughout the analysis, as shown
in Eqs. (10.91) and (10.92) above. A condensed system will then give us:
(10.93)
[ ] { } { }
Solving Eq. (10.93) we thus obtain the solution as:
(10.94)
{ } { }
And the complete solution as given by Eq. (10.72) can be obtained as:
(10.95)
{ } { } { } { } { } { }
b) Second iteration, ( )
In this second iteration, values solved in the first iteration (10.95) are used for the
determination of the relevant matrices thus the tangent stiffness matrices for element 1
and element 2 are given as follows.
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
[ ]
( ) ( ) ( ) ( )
[ ( ) ( ) ] [ ( ) ( ) ]
[ ]
(10.96a)
( ) ( )
( ) ( )
[ ]
( ) ( )
[ ( ) ( ) ]
( ) ( )
( ) ( )
( ) ( )
[ ( ) ( ) ]
[ ]
(10.96b)
By inserting the values derived in Eqs. (10.90) and (10.95) into Eq. (10.70).
(10.97)
[ ] { }
[[ ] { } { } ]
{ }
A condensed system will then give us:
(10.98)
[ ] { } { }
(10.99)
{ } { }
(10.100)
{ } { } { } { }
Next, we checked for convergence and this is done by calculating the root-mean square
error taken about the solution obtained from the previous iteration (1 st iteration) using Eq.
(10.64), thus:
(10.101)
√∑
Based on the above, it can be seen that the solutions still have not converged so the
iteration must be continued. Solution as given in Eq. (10.100) is then used in the next
iteration( ). If we repeat the whole procedure hence proceed with the iteration, the
following results can be obtained:
For the 3rd iteration: (10.102)

{ } { }
The iteration process is stopped at the 3rd iteration because it can be shown that the
convergence criterion has been met. In fact, these values coincide perfectly with the exact
solutions (i.e. those obtained from the closed-form solution of the ODE).
II. Picard scheme: quadratic bar
By using previously given data and Eq (10.31b) the local stiffness matrices at the first iteration (i.e ( )), [ ] for element 1 and 2,
respectively are thus given as:
[ ] [ ]
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ] (10.103a)
[ ] [ ]
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ]
(10.103b)
The load vector and the vector of degree of freedoms can be given as:
Element 1
{ } { } { } { } { } ∫ { } { }
∫ { }
{ }
{ } { } (10.104a)
Element 2
{ } { } { } { } { } ∫ { } { }
∫ { }
{ }
{ } { } { } (10.104b)
values derived in Eqs. (10.103) and (10.104) into Eq. (10.52b).
[ ] { } { }
(10.105)
( ) (10.106)
( )
[ ] { } { } { } { }
(10.107)
{ } { }
In this second iteration, values solved in the first iteration (10.107) are used for the determination of the relevant matrices thus the local
stiffness matrices for element 1 and element 2 are given as follows.
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ] (10.108a)
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ] (10.108b)
By inserting the values derived in Eqs. (10.108) into Eq. (10.52b) gives:
(10.109)
[ ] { } { }
( ) (10.110a)
( )
[ ] { } { } { } { }
Solving Eq. (110) we thus obtain the solution as:
(10.110b)
{ } { }
Next, we checked for convergence and this is done by calculating the root-mean square
error taken about the solution obtained from the previous iteration (1 st iteration) using Eq.
(10.64), thus:
√∑
iteration must be continued. Solution as given in Eq. (10.110b) is then used in the next
For the 3rd iteration:
{ } { }
For the 4th iteration:
{ } { }
The iteration process is stopped at the 4th iteration because it can be shown that the
III. Newton-Raphson scheme: quadratic bar
By using previously given data and Eq. (10.88b) the tangent stiffness matrices at the first iteration (i.e ( )), [ ] and [ ] for element
1 and 2, respectively are thus given as:
̂ ̂ ̂
[ ] [̂ ̂ ̂ ] [ ]
̂ ̂ ̂
[ ]
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ]
[
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ] [ ] [ ]
(10.111a)
̂ ̂ ̂
[ ] [̂ ̂ ̂ ] [ ]
̂ ̂ ̂
[ ]
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ]
[
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ] [ ]
[ ]
(10.111b)
And the load vector as well as the vector of degree of freedoms can be given as:
Element 1
{ } { } { } { } { } ∫ { }
(10.122a)
∫ { }
{ }
{ }
{ } { }
Element 2
{ } { } { } { } { } ∫ { } (10.122b)
∫ { }
{ }
{ }
{ } { } { }
values derived in Eqs. (10.111) and (10.112) into Eq. (10.70).
[ ] { }
(10.113)
[[ ] { } { } ]
Expanding the right hand side of Eq. (10.113) gives:
(10.114)
[ ] { } { }
Take note that is a specified boundary condition thus does not change. This unchanged value means there is no variation in thus is
zero throughout the analysis, as shown in Eqs. (10.113) and (10.114) above. A condensed system will then give us:
(10.115)
[ ] { } { }
(10.116)
{ } { }
(10.117)
{ } { } { } { } { } { }
In this second iteration, values solved in the first iteration (10.117) are used for the determination of the relevant matrices thus the tangent
stiffness matrices for element 1 and element 2 are given as follows.
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ]
(10.118a)
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ]
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
[ ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )]
[ ]
(10.118b)
By inserting the values derived in Eqs. (10.112) and (10.117) into Eq. (10.70).
[ ] { }
(10.119)
[ ] { }
{ } { }
[ ]
(10.120)
[ ] { } { }
(10.121)
{ } { }
(10.122)
{ } { } { } { }
Next, we check for convergence and this is done by calculating the root-mean square error
taken about the solution obtained from the previous iteration (1st iteration) using Eq.
(10.64), thus:
(10.123)
√∑
iteration must be continued. Solution as given in Eq. (10.122) is then used in the next
For the 3rd iteration: (10.124)
{ } { }
The iteration process is stopped at the 3rd iteration because it can be shown that the
10.12 THINGS TO KNOW ABOUT MATLAB AND THE ASSEMBLY OF TWO ELEMENTS
Herein, features of MATLAB employed in the accompanying source codes are briefly
reviewed.
1. Vector and Its Operation
i) The statement for a row vector, v is "v=[a,b,c]" or "v=[a b c]" whilst the
statement of for a column vector is "v=[a;b;c]". Take note of the use of comma (and
spacing) in the specification of the row vector and the use of semi colon for the column
vector. The above statements would return, respectively
v =
a b c
v =
ii) If a vector, say "v=[a,b,c]" has been specified, the statement, "v(2)" would return
ans =
iii) Usually a zero vector with a size of (1xn) is generated in prior to the future operation
on the vector. This is done by the statement, "v=zeros(1,n)". For example, statement
"v=zeros(1,3)" would return
v =
0 0 0
iii. A statement such as "w=subs (v,{a,b,c}, {4,7,3})" would return

w =
4 7 3
2. Matrix and Its Operations
i) The statement of a matrix, m is "m=[a,b,c;d,e,f]" which would return
m =
[ a, b, c]
[ d, e, f]
ii) If a matrix, say "m=[a,b,c;d,e,f]" has been specified, the statement, "m(2,3)"
would return
ans =
whilst a statement "m(2,:)" would return
ans =
[ d, e, f]
iii). If a matrix, say "m=[a,b,c;d,e,f]" has been specified, the statement, "m(2,:)=[]"
would return
ans =
[ a, b, c]
As can be seen, the above argument eliminates row 2 from the matrix. This argument
will be used in the imposition of essential boundary conditions.
iv). A statement such as "q=subs (m,{a,b,c,d,e,f}, {3,6,3,7,7,8})" would return
m =
3 6 3
7 7 8
The command "subs ()" will be used in the computer programming to replace the
symbolic degree of freedom (i.e. , and ) in the stiffness matrices with its
corresponding numerical values.
v) Usually a zero matrix with a size of (nxp) is generated in prior to the future operation
on the matrix. This is done by the statement, "m=zeros(n,p)". For example, statement
"m=zeros(2,3)" would return
m =
0 0 0
0 0 0
To create a square zeros matrix of (nxn), only a single argument is needed i.e.
"m=zeros(n)".
vi) Say two vectors have been specified; "v1=[2,3]" and "v2=[1,4]" and a matrix has
also been specified, say
m =
17 24 1 8 15
23 5 7 14 16
4 6 13 20 22
10 12 19 21 3
11 18 25 2 9
The statement "m(v1,v2)" and "m([2,3],[1,4])", both would return the same
answer, that is
ans =
23 14
4 20
Based on the above, it can be observed that, such arguments refer to the components
of matrix which are , , and , hence the answer. Observe this
relationship carefully because this important feature of MATLAB is frequently employed
during the assembly of the global matrices when utilizing connectivity matrix. In this
context, say there is another matrix been defined, "p=zeros(5)" which would return
p =
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
Note that, "p" is defined so as to have the same size as matrix "m". A statement
"p(v1,v2)=m(v1,v2)" would then return
p =
0 0 0 0 0
23 0 0 14 0
4 0 0 20 0
0 0 0 0 0
0 0 0 0 0
The above operation must be discussed in detailed. What happen is, the statement
actually maps each components of "m(v1,v2)" which has been shown as having the
values of
23 14
4 20
to become the components , , and of the matrix . And the following

recursive operation
p(v1,v2)=m(v1,v2)
p(v1,v2)=m(v1,v2)+ p(v1,v2)
would double up the values to

p =
0 0 0 0 0
46 0 0 28 0
8 0 0 40 0
0 0 0 0 0
0 0 0 0 0
This recursion and the resulting overlapping process will be the basic methodology for
the assembly of the global matrices. However, whilst herein the recursion has been
implemented manually, in the actual source coding this is done by using the loop "for-
end". This is discussed next.
3. Loop "for-end" and Recursion
In this nonlinear programming of steady bar problem, there are two loops involved;
loop for the assembly of the global matrices and iteration for the convergence of the
solution of the equilibrium equation. The former is referred as the internal loop whilst
the latter is referred as the external loop. We will encounter these loops in the
accompanying source codes thus discuss about it later. Herein, an introductory example
of a looping process is shown.
In MATLAB, iteration or looping is carry out by the function "for-end". Here is an

example. The previous recursive statement
p(v1,v2)=m(v1,v2)
p(v1,v2)=m(v1,v2)+ p(v1,v2)
can be made automated by the following statement
for i=1:2
p(v1,v2)=m(v1,v2)+ p(v1,v2)
end
To make obvious, the following three recursive actions
p(v1,v2)=m(v1,v2)
p(v1,v2)=m(v1,v2)+ p(v1,v2)
p(v1,v2)=m(v1,v2)+ p(v1,v2)
which would triple up the answer as follows
p =
0 0 0 0 0
69 0 0 42 0
12 0 0 60 0
0 0 0 0 0
0 0 0 0 0
can actually be made automated simply by increasing the running of i to 3, as follows
for i=1:3
p(v1,v2)=m(v1,v2)+ p(v1,v2)
end
4. Assembly of Global Matrix for Two Elements
i) Having established the knowledge on MATLAB vector's and matrix's commands as well
as the looping "for-end", we are ready to put them into practice in the assembly of two
quadratic bar elements. The local stiffness matrix is as shown in Eq. (103a) and Eq.
(103b) for Element 1 and Element 2, respectively. Herein, we are going to assemble the
global stiffness matrix of Eq. (105) alternatively using the connectivity matrix as
previously given by Eq. (53b) given again herein as
1 2 3
1 2 3 1
3 4 5 2
As mentioned, the connectivity matrix is bracketed by the horizontal array of local

degree of freedom and the vertical array of numbering of elements. These arrays are
specified as vectors in the source code as "LocalDOFNumbering" and
"ElementNumbering", respectively.
Specifically for this example, these vectors are defined as
"LocalDOFNumbering=[1,2,3]" and "ElementNumbering=[1;2]" which both would
return
LocalDOFNumbering =
1 2 3
ElementNumbering =
1
2
Now, we are in the position to define the connectivity matrix. First we specify a zero
connectivity matrix as "ConnectivityMatrix
=zeros(length(ElementNumbering),length(LocalDOFNumbering))" which returns
ConnectivityMatrix =
0 0 0
0 0 0
To note, in the statement above the command "length()" would produce the total
number of components that exists in a vector thus statement
"length(LocalDOFNumbering)" for example, would return the value of 3.
Employing the loop "for-end", the connectivity matrix can now be generated by the
following loop
for i=1:length(ElementNumbering)
ConnectivityMatrix(i,1:length(LocalDOFNumbering)) = ...
2*(i-1)+LocalDOFNumbering;
end
for "i=1", the above would return
1 2 3
0 0 0
and for "i=2", the above would return
1 2 3
3 4 5
With this, we have successfully generated automatically the connectivity matrix. Now
we can proceed with the assembly of the global stiffness matrix which previously given
by Eq. (105). To do this, we must define first, the local stiffness matrix of Element 1 and
Element 2.
The local stiffness matrix for Element 1 and Element 2 (with initial condition been
imposed) as given by Eq. (103a) and Eq. (103b) is defined herein as "k1=[4.6667,-
5.3333,0.6667;-5.3333,10.6667,-5.3333;0.6667,-5.3333,4.6667]" and
"k2=[4.5838,-5.4438,0.8600;-5.4438,11.9921,-6.5483;0.8600,-
6.5483,5.6884]" which both would return
k1 =
4.6667 -5.3333 0.6667

-5.3333 10.6667 -5.3333
0.6667 -5.3333 4.6667
k2 =
4.5838 -5.4438 0.8600

-5.4438 11.9921 -6.5483
0.8600 -6.5483 5.6884
Before we proceed with the automated assembly of the global stiffness matrix with the
use of loop "for-end", let's do it manually. A statement
K1Global =zeros(DofNumber)
K1Global(ConnectivityMatrix (1,:), ConnectivityMatrix (1,:))=k1
would return
K1Global =
4.6667 -5.3333 0.6667 0 0

-5.3333 10.6667 -5.3333 0 0
0.6667 -5.3333 4.6667 0 0
0 0 0 0 0
0 0 0 0 0
and a statement
would return
K2Global =
0 0 0 0 0
0 0 0 0 0
0 0 4.5838 -5.4438 0.8600
0 0 -5.4438 11.9921 -6.5483
0 0 0.8600 -6.5483 5.6884
Finally, the global stiffness matrix can be obtained through summation as follows
KGlobal=K1Global+K2Global
which would return
KGlobal =
4.6667 -5.3333 0.6667 0 0

-5.3333 10.6667 -5.3333 0 0
0.6667 -5.3333 9.2505 -5.4438 0.8600
0 0 -5.4438 11.9921 -6.5483
0 0 0.8600 -6.5483 5.6884
Thus the attainment of the global stiffness matrix at the left-hand side of Eq. (105).
Now, let's construct the global stiffness matrix automatically by employing the loop
"for-end". What basically we want to do is to make automatic recursion of the previous
statement, given again herein as
KGlobal =zeros(DofNumber)
KGlobal=K1Global+K2Global
Tracing the running of the arguments (i.e. (1,:), (2,:)) and the notation (i.e.
K1Global, K1Global ), the recursion of the above statement is obviously feasible.
However, to make everything automatic, this also means that generation of the local
stiffness matrix must be automatic as well. As given by Eq (31b), the general equation
for the local stiffness matrix can be given as
k = E*A*[((37*u1)/30 +(6*u2)/5 - u3/10)/L, ...

-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L;
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((8*u1)/5 +(32*u2)/15 + (8*u3)/5)/L, ...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L;
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L,...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L,...
(-u1/10 +(6*u2)/5 + (37*u3)/30)/L]
which would return
k =
[ ((37*u1)/30 + (6*u2)/5 - u3/10)/L, -((22*u1)/15 +

(16*u2)/15 + (2*u3)/15)/L, ((7*u1)/30 - (2*u2)/15 +
(7*u3)/30)/L]
[ -((22*u1)/15 + (16*u2)/15 + (2*u3)/15)/L, ((8*u1)/5 +
(32*u2)/15 + (8*u3)/5)/L, -((2*u1)/15 + (16*u2)/15 +
(22*u3)/15)/L]
[ ((7*u1)/30 - (2*u2)/15 + (7*u3)/30)/L, -((2*u1)/15 +
(16*u2)/15 + (22*u3)/15)/L, ((6*u2)/5 - u1/10 +
(37*u3)/30)/L]
As can be seen, the general statement of the local stiffness matrix, “ k” is expressed
symbolically in terms of the degree of freedoms. The insertion of the appropriate values
of the degree of freedoms is what differentiates between elements hence the
materialization of k1 and k2, for instance. To insert these value, the aforementioned
command "subs ()" is used and a vector which contains known degree of freedom
must be created. Also, the length of the element, “L” must also be specified.
For the first iteration, this vector contains the initial conditions. Since herein we are
going to demonstrate the automatic generation of the previously given global stiffness
matrix in Eq. (105); the ones obtained with the initial condition being imposed, what we
are going to do first is to create an initial condition vector and equate this to the vector
of global degree of freedom. The initial conditions vector is created by the following
statement
Uinitial = zeros(1,ElementNumber*2+1);
Uinitial(1:end-1) = 1;
Uinitial(end) = sqrt(2)
Argument “(end)” yields the last element in a vector and since the essential boundary
condition in the Example is specified at the far-right end of the bar with a value of √ ,
statement “Uinitial(end)= sqrt(2)” would prescribe such a boundary conditions.
The above statement would return the following
Uinitial =
1.0000 1.0000 1.0000 1.0000 1.4142

Now, equate this to the known (or solved) global degree of freedom vector by
UGlobal = Uinitial
which would return
UGlobal =
1.0000 1.0000 1.0000 1.0000 1.4142
Before we proceed, length of the element, “L” must also be assigned and this can be
done by stating that
TotalLength = 1;
ElementNumber = 2;
ElementLength = TotalLength/ElementNumber;
data which as given in the Example. The statement returns
TotalLength =
ElementNumber =
ElementLength =
0.5000
Having established “UGlobal” and “ElementLength”, the next step is to loop the
assembly of the global stiffness matrix. This can be done by the following loop
DofNumber = length(ElementNumbering)*2 + 1;
KGlobal =zeros(DofNumber);
for e = 1:length(ElementNumbering)
Klocal_Global = zeros(DofNumber);
uLocal = UGlobal(ConnectivityMatrix(e,:));
kLocal = subs(k,{u1,u2},{uLocal(1:2)});
kLocal = subs(kLocal,{u3,L},{uLocal(end), ElementLength});
Klocal_Global(ConnectivityMatrix(e,:),ConnectivityMatrix(e,:)) ...
= kLocal;
KGlobal= Klocal_Global + KGlobal
end
for "i=1", the above statement would return
kLocal =
4.6667 -5.3333 0.6667

-5.3333 10.6667 -5.3333
0.6667 -5.3333 4.6667
Klocal_Global =
4.6667 -5.3333 0.6667 0 0

-5.3333 10.6667 -5.3333 0 0
0.6667 -5.3333 4.6667 0 0
0 0 0 0 0
0 0 0 0 0
KGlobal =
4.6667 -5.3333 0.6667 0 0

-5.3333 10.6667 -5.3333 0 0
0.6667 -5.3333 4.6667 0 0
0 0 0 0 0
0 0 0 0 0
and for "i=2", the above would return
uLocal =
1.0000 1.0000 1.4142
kLocal =
4.5838 -5.4438 0.8600

-5.4438 11.9922 -6.5484
0.8600 -6.5484 5.6884
Klocal_Global =
0 0 0 0 0
0 0 0 0 0
0 0 4.5838 -5.4438 0.8600
0 0 -5.4438 11.9922 -6.5484
0 0 0.8600 -6.5484 5.6884
KGlobal =
4.6667 -5.3333 0.6667 0 0

-5.3333 10.6667 -5.3333 0 0
0.6667 -5.3333 9.2505 -5.4438 0.8600
0 0 -5.4438 11.9922 -6.5484
0 0 0.8600 -6.5484 5.6884
Again, the attainment of the stiffness matrix in Eq (105) but this time rather
automatically. To put everything into perspective, below is the source code of what
have been discussed above; provided in such a way that a user can simply cut and paste
them onto MATLAB and run the program straight away to obtain the final global
stiffness matrix above. Try them, just cut and paste.
5. MATLAB Source Code for Global Stiffness Matrix Assembly
syms u1 u2 u3 L
A=1
E=1
TotalLength = 1
ElementNumber = 2
ElementLength = TotalLength/ElementNumber
Uinitial(end) = sqrt(2)
k = E*A*[((37*u1)/30 +(6*u2)/5 - u3/10)/L, ...

-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L;
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((8*u1)/5 +(32*u2)/15 + (8*u3)/5)/L, ...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L;
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L,...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L,...
(-u1/10 +(6*u2)/5 + (37*u3)/30)/L];
ElementNumbering = zeros(ElementNumber,1);
for i=1:ElementNumber
ElementNumbering(i,1) = i;
end
for i=1:3
LocalDOFNumbering(1,i) = i;
end
ConnectivityMatrix = ...
zeros(length(ElementNumbering),length(LocalDOFNumbering));
2*(i-1)+LocalDOFNumbering
end
DofNumber = 2*length(ElementNumbering)+1;
UGlobal = Uinitial;
for e = 1:length(ElementNumbering)
uLocal = UGlobal(ConnectivityMatrix(e,:))
kLocal = subs(kLocal,{u3,L},{uLocal(end), ElementLength})
Klocal_Global(ConnectivityMatrix(e,:),ConnectivityMatrix(e,:))...
= kLocal;
KGlobal= Klocal_Global + KGlobal

end
It must be noted that the above given source code is general enough in terms of the
number of element. We can create the assembly of this initial stiffness matrix (initial
since the matrix has been generated by initial conditions only) for any desired number
of element just by setting the statement “ElementNumber = ()” to the desired
number.
Play with this as this can stress out the advantage of automated algorithm as compared
to manual process. For example, by setting “ElementNumber = 3” which means the
assembly of three elements, the following global stiffness matrix is obtained
KGlobal =
7.0000 -8.0000 1.0000 0 0 0 0

-8.0000 16.0000 -8.0000 0 0 0 0
1.0000 -8.0000 14.0000 -8.0000 1.0000 0 0
0 0 -8.0000 16.0000 -8.0000 0 0
0 0 1.0000 -8.0000 13.8757 -8.1657 1.2899
0 0 0 0 -8.1657 17.9882 -9.8225
0 0 0 0 1.2899 -9.8225 8.5326
And by setting “ElementNumber = 4” which means the assembly of four elements,

the following global stiffness matrix is obtained
KGlobal =
9.3333 -10.6667 1.3333 0 0 0 0 0 0

-10.6667 21.3333 -10.6667 0 0 0 0 0 0
1.3333 -10.6667 18.6667 -10.6667 1.3333 0 0 0 0
0 0 -10.6667 21.3333 -10.6667 0 0 0 0
0 0 1.3333 -10.6667 18.6667 -10.6667 1.3333 0 0
0 0 0 0 -10.6667 21.3333 -10.6667 0 0
0 0 0 0 1.3333 -10.6667 18.5010 -10.8876 1.7199
0 0 0 0 0 0 -10.8876 23.9843 -13.0967
0 0 0 0 0 0 1.7199 -13.0967 11.3768
Can you see the advantage now? It is just cool, isn’t? Imagine to assemble four elements
manually, it is definitely going to be a heck of a job.
So far, what we have discussed in this section is the algorithm to make automatic the
assembly of the stiffness matrix but we still have not discussed the full solution of the
nonlinear discretized equilibrium equation. We will do this in the next sections where
we will discuss the full source codes for the problem, involving the discussion on the
iterative schemes, imposition of essential boundary conditions, solution of the
equilibrium equation by Gauss Elimination through the use of command “\”,
convergence check and many more. But having this section as a ‘soft-landing’ would
definitely be of great help.
PICCARD SOURCE CODE (to be explained during course)
% -------------------------------------------------------------------------
% CLEAR DATA
% -------------------------------------------------------------------------
clc; clear all;
syms u1 u2 u3 L
% -------------------------------------------------------------------------
% 1. INPUT
% -------------------------------------------------------------------------
% 1.1 Physical Input
q = -1;
E = 1;
A = 1;
TotalLength = 1;
% 1.2 Discretization Input

ElementNumber = 3;
% 1.3 Initial Condition

Uinitial(end) = sqrt(2);
% -------------------------------------------------------------------------
% 2.LOCAL STIFFNESS AND TANGENT STIFFNESS MATRICES (Quadratic Element)
% AND LOCAL LOAD VECTORS
% -------------------------------------------------------------------------
% 2.1 Local Stiffness Matrix
k = E*A*[((37*u1)/30 +(6*u2)/5 - u3/10)/L, ...
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L;
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((8*u1)/5 +(32*u2)/15 + (8*u3)/5)/L, ...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L;
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L,...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L,...
(-u1/10 +(6*u2)/5 + (37*u3)/30)/L];
% 2.2 Local Load Vectors

f = q*[1/6*L;2/3*L;1/6*L];
% -------------------------------------------------------------------------
% 3. CONNECTIVITY MATRIX
% -------------------------------------------------------------------------
% 3.1 Element Numbering
end
% 3.2 Local Degree of Freedom Numbering

for i=1:3
end
% 3.3 Connectivity Matrix

end
% -------------------------------------------------------------------------
% 4. ASSEMBLY OF STIFFNESS MATRIX, IMPOSITION OF
% BOUNDARY CONDITION AND PICCARD SOLUTION
% -------------------------------------------------------------------------
UGlobal = Uinitial';
for r = 1:4
RGlobal =zeros(DofNumber,1);
FGlobal =zeros(DofNumber,1);
for e = 1:length(ElementNumbering);
Flocal_Global =zeros(DofNumber,1);
= kLocal;
KGlobal= Klocal_Global + KGlobal;
fLocal = subs(f,{L},{ElementLength});
Flocal_Global(ConnectivityMatrix(e,:),1) = fLocal;
FGlobal= Flocal_Global + FGlobal;
end
% 4.1 Imposition of Boundary Condition
BoundaryCondition = Uinitial(end);
RGlobal = FGlobal-KGlobal(:,end)*BoundaryCondition;
RGlobal(end) = [];
KGlobal(end,:) = [];
KGlobal(:,end) = [];
UGlobalPrevious = UGlobal;
% 4.2 Solution of Equilibrium Equation
UGlobal = KGlobal\RGlobal;
% 4.3 Returning KGlobal, FGlobal and UGlobal into its Original Size
KGlobal(:,end+1)=0;
KGlobal(end+1,:)=0;
FGlobal(end+1,1) = 0;
UGlobal(end+1,1) = Uinitial(end);
% 4.4 Convergence check

residual = norm(UGlobal-UGlobalPrevious)
if residual <1e-3
break
end
% 4.5 Data Storage for plotting purposes

disp(['r = ' num2str(r)])
UGlobalAll(r,:) = UGlobal';
end
UGlobal
% -------------------------------------------------------------------------
% 6. PLOT OF SOLUTIONS (ANALYTICAL VS FINITE ELEMENT CONVERGENCE)
% -------------------------------------------------------------------------
% 6.1 Analytical Result
figure(1);set(gcf,'color','w')
box on
xInterval = 0:ElementLength/2:TotalLength;
for i=1:length(xInterval)
x = xInterval(i);
U(i) = sqrt(1 + x^2);
end
line(xInterval,U,'linestyle','-.','color','r','marker','*')
% 6.2 Finite Element Results
for i=1:size(UGlobalAll,1)
if i==1
marker = 's';
elseif i==2
marker = 'o';
elseif i==3
marker = '+';
elseif i==4
marker = '^';
end
line(xInterval,UGlobalAll(i,:),'color','k','marker',marker)
hold on
end
legend('Analytical','r = 1','r = 2','r = 3','r = 4')
1.5
Analytical
r=1
1.4 r=2
r=3
r=4
1.3
1.2
1.1
0.9
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
NEWTON-RAPHSON SOURCE CODE (to be explained during course)
% -------------------------------------------------------------------------
% CLEAR DATA
% -------------------------------------------------------------------------
clc; clear all;
syms u1 u2 u3 L
% -------------------------------------------------------------------------
% 1. INPUT
% -------------------------------------------------------------------------
% 1.1 Physical Input
q = -1;
E = 1;
A = 1;
TotalLength = 1;
% 1.2 Discretization Input

ElementNumber = 3;
% 1.3 Initial Condition

Uinitial(end) = sqrt(2);
% -------------------------------------------------------------------------
% 2.LOCAL STIFFNESS AND TANGENT STIFFNESS MATRICES (Quadratic Element)
% AND LOCAL LOAD VECTORS
% -------------------------------------------------------------------------
% 2.1 Local Stiffness Matrix
k = E*A*[((37*u1)/30 +(6*u2)/5 - u3/10)/L, ...
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L;
-((22*u1)/15 +(16*u2)/15 + (2*u3)/15)/L,...
((8*u1)/5 +(32*u2)/15 + (8*u3)/5)/L, ...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L;
((7*u1)/30 -(2*u2)/15 + (7*u3)/30)/L,...
-((2*u1)/15 +(16*u2)/15 + (22*u3)/15)/L,...
(-u1/10 +(6*u2)/5 + (37*u3)/30)/L];
% 2.2 Local Tangent Stiffness Matrix

t = E*A*[ ((37*u1)/30 - (22*u2)/15 + (7*u3)/30)/L,...
-((16*u2)/15 - (6*u1)/5 + (2*u3)/15)/L,...
-(u1/10 + (2*u2)/15 - (7*u3)/30)/L;
-((22*u1)/15 - (8*u2)/5 + (2*u3)/15)/L,...
-((16*u1)/15 - (32*u2)/15 + (16*u3)/15)/L,...
-((2*u1)/15 - (8*u2)/5 + (22*u3)/15)/L;
-((2*u2)/15 - (7*u1)/30 + u3/10)/L,...
-((2*u1)/15 + (16*u2)/15 - (6*u3)/5)/L,...
((7*u1)/30 - (22*u2)/15 + (37*u3)/30)/L] + k;
% 2.3 Local Load Vectors

f = q*[1/6*L;2/3*L;1/6*L];
% -------------------------------------------------------------------------
% 3. CONNECTIVITY MATRIX
% -------------------------------------------------------------------------
% 3.1 Element Numbering
end
% 3.2 Local Degree of Freedom Numbering

for i=1:3
end
% 3.3 Connectivity Matrix

end
% -------------------------------------------------------------------------
% 4. ASSEMBLY OF STIFFNESS MATRIX, TANGENT STIFFNESS MATRIX, IMPOSITION OF
% BOUNDARY CONDITION AND NEWTON-RAPHSON SOLUTION
% -------------------------------------------------------------------------
UGlobal = Uinitial';
for r = 1:4
TGlobal =zeros(DofNumber);
FGlobal =zeros(DofNumber,1);
for e = 1:length(ElementNumbering);
Tlocal_Global = zeros(DofNumber);
Flocal_Global =zeros(DofNumber,1);
= kLocal;
KGlobal= Klocal_Global + KGlobal;
tLocal = subs(t,{u1,u2},{uLocal(1:2)});
tLocal = subs(tLocal,{u3,L},{uLocal(end), ElementLength});
Tlocal_Global(ConnectivityMatrix(e,:),ConnectivityMatrix(e,:)) ...
= tLocal;
TGlobal= Tlocal_Global + TGlobal;
fLocal = subs(f,{L},{ElementLength});
Flocal_Global(ConnectivityMatrix(e,:),1) = fLocal;
FGlobal= Flocal_Global + FGlobal;
end
% 4.1 Imposition of Boundary Condition (direct elimination can be
% implemented because Delta_U always zero
R = (KGlobal)*UGlobal - FGlobal;
R(end) = [];
TGlobal(end,:) = [];
TGlobal(:,end) = [];
% 4.2 Solution of Equilibrium Equation

Delta_U = TGlobal\-(R);
% 4.3 Returning KGlobal, TGlobal and Delta_U into its Original Size
TGlobal(:,end+1)=0;
TGlobal(end+1,:)=0;
Delta_U(end+1,1)= 0;
% 4.4 Convergence check

UGlobalPrevious = UGlobal;
UGlobal = UGlobal + Delta_U;
residual = norm(UGlobal-UGlobalPrevious)
if residual <1e-3
break
end
% 4.5 Data Storage for plotting purposes

disp(['r = ' num2str(r)])
UGlobalAll(r,:) = UGlobal';
end
UGlobal
% -------------------------------------------------------------------------
% 6. PLOT OF SOLUTIONS (ANALYTICAL VS FINITE ELEMENT CONVERGENCE)
% -------------------------------------------------------------------------
% 6.1 Analytical Result
figure(1);set(gcf,'color','w')
box on
xInterval = 0:ElementLength/2:TotalLength;
for i=1:length(xInterval)
x = xInterval(i);
U(i) = sqrt(1 + x^2);
end
line(xInterval,U,'linestyle','-.','color','r','marker','*')
% 6.2 Finite Element Results
for i=1:size(UGlobalAll,1)
if i==1
marker = 's';
elseif i==2
marker = 'o';
elseif i==3
marker = '+';
elseif i==4
marker = '^';
end
line(xInterval,UGlobalAll(i,:),'color','k','marker',marker)
hold on
end
legend('Analytical','r = 1','r = 2','r = 3','r = 4')
1.5
Analytical
r=1
1.4 r=2
1.3
1.2
1.1
0.9
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
SOLIDS AND FLUIDS
FLUID DYNAMICS
NAVIER-STOKES EQUATION
12. INTRODUCTION
The behaviour of flow of fluids and gases obey the three conservation laws; mass,
momentum and energy. The complete modelling of these laws hence the actual flow
behaviour is extremely difficult. It is a practice, therefore, to reduce the complexity of the
problem by introducing appropriate assumptions. Herein, it is assumed that;
i. flow is incompressible
ii. flow is isothermal
iii. flow behaviour in width direction is constant
iv. fluid is materially linear (Newtonian) and homogenous isotropic.
These assumptions allow us to deal with a simpler set of equations but sufficient enough for
its expansion to more general cases becomes obvious.
NOTE: Section 12.1 is provided for completeness. Readers which main interest is on the FEM
formulation for the problem can skip these subchapters and go straight to Section 12.2.
12.1 DERIVATION OF NAVIER-STOKES PARTIAL DIFFERENTIAL EQUATIONS
By employing assumption ii) which means that there is no variation in temperature in the
flow, a flow problem can be completely described by the mass (continuity) and momentum
equations alone as the energy equation is now uncoupled from the latter. The derivation of
mass (continuity) equation is given next, followed by the derivation of the momentum
equation. To note, from now on, mass equation will be referred to continuity equation.
12.1.1 Continuity Equation
The continuity equation can be derived by considering a differential element of the flow as
shown in Figure 12.1. As shown in the figure, the variables to be considered are the density,
, velocities in x-direction, and y-direction, as well in z-direction, Similar to previous
chapters, the variables are expand by Taylor series at all sides (surfaces).

SOLIDS AND FLUIDS
Figure 12.1: Flow mass differential element
The conservation of mass principle requires that the time rate of decrease of mass inside
the differential element must be equalled to the net mass flow out of the differential
element through its surface. The former can be given as;
The time rate of decrease of mass= ( ) (12.1)
The determination of the net mass flow can be done by considering first the net outflow of
the flow in each direction.
In x-direction, by balancing terms on the right and left sides of the differential element, we
obtain;
( ) ( ) (12.2)
( ) ( )

SOLIDS AND FLUIDS
By similar argument, the net outflow in y-direction is given as;
( ) ( ) (12.3)
( ) ( )
and in z-direction as;
( ) ( ) (12.4)
( ) ( )
Having established the new flow in all direction, the net mass flow of the fluid out of the
differential element through its surface can thus be given as
( ) ( ) ( ) (12.5)
Net mass flow= ( )
The principle of mass conservation requires Eqn. (12.5) to be equalled to (12.1) thus;
( ) ( ) ( ) (12.6)
=( )
or
( ) ( ) ( ) (12.7)
or in vector forms as;
(12.8)
{ }{ }
where
(12.9)
{ } { }
{ } { } (12.10)
Eqns. (12.7) and (12.8) are the continuity equation for the flow given in the form termed as
conservation forms. In this form, the product of , and are themselves treated as
the dependent variables.
However, sometime, it would be more convenient to deal directly with the velocities
components, , and w, termed as primitive variables as this would be more familiar from
FEM point of view, at least from what we have encountered so far.

SOLIDS AND FLUIDS
To express Eqn. (12.7) (or (12.8)) explicitly in terms of the primitive variables, we employ the
following derivative called material derivative. We omit the theoretical derivation and
discussion of this derivative but suffice to say, it is a natural product of Eulerian formulation;
a formulation usually employed in fluid mechanics whilst its counterpart, the Lagrangian
being usually employed in solid mechanics. The material derivative is given as;
(12.11)
{ }{ }
Now, since Eqn. (12.7) or involve the derivation of the product of two functions (more
obvious by looking as Eqn. (12.8), we can expand this derivative by chain-rule as;
(12.12)
{ }{ } { }{ }
If we observe carefully Eqn. (12.12), we will notice that the differential operators which
operate on the density, , can be replaced by the material derivative. By inserting Eqn.
(12.11) into (Eqn. (12.12) we obtain;
(12.13)
{ }{ }
Having established Eqn. (12.13), we thus obtained the continuity equation of the flow in
terms of the primitive variables. In its expanded form, the equation can be given as;
(12.14)
( )
Now, we employ assumption i) that is, the flow is incompressible. Incompressible fluid is a
fluid having constant density. Accordingly, incompressible flow refers to the state where the
rate of change of density is assumed zero. It is worth to note that, while incompressible flow
does not really refer to incompressible fluid as the latter is an ideal state of fluid, the terms
are usually interchangeable. Due to assumption i), Eqn. (12.14) is reduced to;
(12.15)
since the assumption would mean .
We have completed our discussion on the derivation of continuity equation (mass

conservation equation) for the time being. Next, we are going to discuss the derivation of
the momentum equation.

SOLIDS AND FLUIDS
12.1.2 Momentum Equation
The momentum equation is the principle based on Newton’s 2 nd Law which basic
form is
(12.16)
where is the force acting on a particle whilst and are the mass and the acceleration of
the particle, respectively. For a continuum of higher dimension, Eqn. (12.16) is expanded
into
∑ (12.17a)
∑ (12.17b)
∑ (12.17c)
Eqns. (12.17) represent the equilibrium of forces in all direction. Since we base our
discussion on Eulerian formulation, the accelerative terms on the right and side of the
equations should be expressed in terms of the material derivatives, thus;
(12.18a)
∑
(12.18b)
∑
(12.18c)
∑
since . Terms on the left hand side of Eqns. (2.18) can be determined by
considering the differential element of the flow as shown in Figure 12.2.

SOLIDS AND FLUIDS
Figure 12.2: Flow forces in differential element
As shown in the figure, the variables to be considered are pressure, and viscous stress;
normal and shear stresses in x-direction, y-direction and in z-direction Similar to previous
chapters, the variables are expand by Taylor series at all sides (surfaces).
By considering the equilibrium of forces in x-direction, we obtain;
(12.19)
∑ ( ( )) (( ) )
(( ) )
(( ) )
where is the body force in the x-direction. By inserting Eqn. (12.19) into Eqn.
(12.18a), expanding and cancelling, the following is obtained;
(12.20)
Eqn. (12.20) is the momentum equation for the flow in the x-direction. Applying the same
procedure to the other directions, momentum equations in y-direction and z-direction can
be given, respectively as
(12.21)

SOLIDS AND FLUIDS
(12.22)
Eqns. (12.20) to (12.22) are the complete momentum equations for 3 dimensional flow. By
inserting the material derivative as given by Eqn. (12.11) into the equations, we can
expressed the equations in terms of the primitive variables as
(12.23)
( )
(12.24)
( )
(12.25)
( )
Eqns. (12.15) and (12.23) to (12.25) are the 3 dimensional continuity and the momentum
equations of the flow which are famously known as Navier-Stokes equations. These
equations are supplemented by boundary condition equations which discussion is going to
be given next. Also, since the equations can be time-dependent or unsteady (if time
derivative terms are not omitted), the solution of the PDE would require initial conditions to
be prescribed.
Like other PDEs, the continuity and the momentum equations must be supplemented by
boundary condition equations so as to have a well-posed problem. There are two types of
boundary conditions; Newmann (natural) and Dirichlet (essential) boundary conditions.
i. in x-direction
(12.26)
ii. in y-direction
(12.27)
iii. in z-direction
(12.28)
where, , and are the components of the unit normal vector of the surface under
consideration. It worth to note that, for rectangular domain hence rectangular elements

SOLIDS AND FLUIDS
(meshes), major simplification to the above equations can be obtained as for this case, in x-
direction, , and , in y-direction, , and , and in z-
direction, , and , thus reducing the equations to
(12.29)
(12.30)
(12.31)
It can be seen later that the secondary terms produced by the integration by parts
conducted on the viscous and the pressure terms of the momentum equations will be in
similar forms as in Eqns. (12.26) to (12.28) and (12.29) to (12.31); the latter are for
rectangular domain. Due to this, equations above are also known as natural boundary
conditions.
Dirichlet boundary condition requires that the velocity components at the boundary where
natural conditions are unknown (to be solved) must be known thus specified.
̅( ) (12.32)
̅( ) (12.33)
̅( ) (12.34)
For no slip condition, normally the specified value of the velocities can either be zero or
equalled to the velocity of the boundary.
There are other possible boundary conditions for a flow such as slip condition and free
surface condition. However, these are not covered herein.
12.2 TWO-DIMENSIONAL NAVIER-STOKES EQUATIONS
Eqns. (12.15), (12.23) to (12.25) are the three-dimensional Navier-Stokes equations together
with their boundary conditions equations given by Eqns. (12.26) to (12.28) and (12.32) to
(12.34). If we employ assumption iii) that the flow behaviour is assumed constant in the
width direction, the equations can be reduced to two-dimensional which can be given as
follows.
Continuity equation
(12.35)

SOLIDS AND FLUIDS
Momentum equation
(12.36)
( )
(12.37)
( )
(12.38)
(12.39)
̅( ) (12.40)
̅( ) (12.41)
It is convenient to express the continuity equation (Eqn. (12.35) and the momentum
equations (Eqns. (12.36) and (12.37)) in vector and matrix forms, as we did for heat and
plane stress elements so as to allow for easy tracing of the extension of the procedure. This
away, the evolution of the procedure can easier be followed.
In matrix and vector forms, the continuity equation (Eqn. (12.35) can be given as;
(12.42)
{ }{ }
or
{ }{ } (12.43)
where { } { }. When expressed in vector forms, is taken as representing both

velocities, and . Transpose of the vector is employed because from now on, all vectors
and matrices are treated as if they are originally arranged in a row-wise manner; a normal
practice in computer programming works.
To express the momentum equations in matrix and vector forms, it would be an advantage
if we highlight the similarity of the viscous terms to previously discussed plane stress
formulation as this, not only allow for easy tracing of the evolution of the procedure,
emphasizes the applicability of the algorithm developed for the former to be directly used in
the latter, maybe with some slight modification.

SOLIDS AND FLUIDS
To highlight the similarity, the viscous terms are boxed in the following equation.
(12.44)
( )
(12.45)
( )
BOX A
By observing BOX A, we can notice how similar the terms inside the box to the ones we
encountered in the plane stress problem. Due to this, similar matrix and vector forms can
be employed, thus;
(12.46)
{ }
{ } [ ]
or
(12.47)
[ ]{ }
{ }
since it can be shown that . [ ] is termed as differential operator matrix.
Now, by employing assumption iv) that is, the fluid is assumed as materially linear
(Newtonian) and isotropic, as well as assumption i) that is, the flow is incompressible, the
terms ( ), which initially exists in the normal stress relationship where is the
second viscosity coefficient, is excluded.
The constitutive equations of the viscous stresses can thus be given as
(12.48)
( )

SOLIDS AND FLUIDS
(12.49)
{ } [ ]
{ }
or
{ } [ ] { } (12.50)
[ ]
Observe that the second matrix on the right hand side of the equation above is also the
transpose of the differential operator matrix, [ ] previously defined in Eqns. (12.46) and
(12.47). Based on this and by introducing the constitutive matrix, [ ], Eqn. (12.50) can be
given as;
{ } [ ][ ] { } (12.51)
By inserting Eqn. (12.51) into Eqn. (12.47), the viscous terms can be completely expressed in
terms of the primitive variables, and as
[ ][ ][ ] { } (12.52)
{ }
Observing Eqn. (12.52), we can notice that the equation is exactly similar to the ones
derived for the plane stress problem.

SOLIDS AND FLUIDS
Now let’s deal with the rest of Eqns. (12.44) and (12.45). To maintain the same line of
arguments, boxes are also provided for other terms as shown below.
(12.53)
( )
(12.54)
( )
B C D E
Terms in BOX B can be in vector forms given as;
(12.55)
{ } { }
{ }
Terms in BOX C can be given in vector forms as
(12.56)
{ } { } { }
Terms in BOX D can be given in vector forms as
(12.57)
̇
{ } { } { } { ̇}
̇
Finally, terms in BOX E can be given in matrix forms as
( )
{ }
( )
{ } [ ]
(12.58)
{ }{ }{ }
{ } { }{ }

SOLIDS AND FLUIDS
Now by using Eqns. (12.52),(12.55) – (12.58), the momentum equations of Eqns. (12.36) and
(12.37) can be given in matrix and vector forms as
{ } [ ][ ][ ] { } { } { ̇} { } { }{ } (12.59)
Reminder: please note the difference between { } and [ ].
Eqns. (12.43) and (12.59) are the continuity and momentum equation for 2 dimensional
viscous incompressible fluid problems.
12.3 DESCRITIZATION BY GALERKIN WRM
Having established the PDE of the problem, we are all set to descritize the equations so as to
obtain the FEM algebraic formulation. There are various ways to this; two famous
formulations are mixed and penalty formulations. However, herein we will concern only
with mixed formulation.
The term ‘mixed’ refers to the fact that we have variables from different families; velocities
and pressure. Whilst the former is displacement related quantities, the latter is force-
related quantity. Previous studies have shown that, for the formulation to work well, the
order of interpolation of velocities must be higher than the pressure. Due to this, to
interpolate the velocities, quadratic shape functions are employed while for pressure,
bilinear shape functions are used. For rectangular elements these mean
(12.60)
{ } { } [ ]{ ̂}
{ }
{ }{ } { }{ ̂ }
(12.61)
[ ] and { } are similar to those previous derived for scalar (heat transfer) and plane stress
elements. Due to this the arrangement of nodes for a rectangular element is shown in
Figure 12.3.

SOLIDS AND FLUIDS
Figure 12.3: Rectangular element for fluid
For a rectangular element, it is still convenient to provide the shape functions in the physical
coordinates ( ) and conduct analytical integration. For demonstration purposes and to
ease the tracing of the formulation, herein these are pursued. But it must be noted that, in
practical, the shape function functions are expressed in natural coordinates, ( ) and
numerical integration is employed as these would allow for easier treatment of elements
with irregular shape. Discussion on these will be given later.
Therefore, the shape functions associated velocities, are
(A12.61a)
( ) ( )
(A12.61b)
( ) ( )
(A12.61c)
( )
(A12.61d)
( )
(A12.61e)
( ) ( )
(A12.61f)
( )
(A12.61g)

SOLIDS AND FLUIDS
(A12.61h)
( )
whilst, the shape functions associated with the pressure, are
(B12.61a)
(B12.61b)
(B12.61c)
(B12.61d)
Inserting Eqns. (12.60) and (12.61) into (12.59) and (12.43),we obtain
Continuity
{ }[ ]{ ̂} (12.62)
Momentum
{ }{ }{ ̂ } [ ][ ][ ] [ ]{ ̂} { } ̇
[ ]{ ̂} [ ]{ ̂} { }[ ]{ ̂} (12.63)
Note that, in the time derivative terms, since the shape function [ ] are not functions of
̇ .
time, the dofs are thus differentiated by time, i.e { ̂}
By observing Figure 12.3 and Eqn. (12.60) as well as Eqn. (12.61), we can identify that, for an
element, we have 20 unknown dofs, i.e. { } with
only 3 equations. To obtain sufficient number of equations, we will multiply then integrate
mass equation (Eqn. (12.62)) with pressure shape function { } to produce 4 independent
equations and the momentum equations (Eqn. (12.63)) to produce 16 independent
equations. Together, they will make up 20 simultaneous algebraic equations as required.
Multiplying Eqn. (12.62) with pressure shape functions { } and integrating give
∫ ∫ { } { }[ ]{ ̂} (12.64)
By multiplying the velocity shape functions [ ] to the momentum equations (Eqn. (12.63))
and integrating give

SOLIDS AND FLUIDS
∫ ∫ [ ] { }{ }{ ̂ } ∫ ∫ [ ] [ ][ ][ ] [ ]{ ̂}
∫ ∫[ ] { } (12.65)
̇
∫ ∫ [ ] [ ]{ ̂}
∫ ∫ [ ] [ ]{ } { }[ ]{ ̂}
12.3.1 INTEGRATION BY PARTS (IBP)

Such intentions will not be fulfilled if IBP is conducted to the continuity equation. In the
context of point ii) even if we conduct IBP to the continuity equation, the resulting boundary
terms, would not have physical meaning.
Also, based on point i) and ii), IBP should be conducted onto pressure and viscous terms
only as these will induce the natural boundary conditions as previously given by Eqns.
(12.38) and (12.39). Therefore, by conducting IBP to Eq (12.65) but to the pressure and
viscous terms only, results in
∫ ∫ [ ] { }{ }{ ̂ } ∫ ∫ [ ] [ ][ ][ ] [ ]{ ̂}
∫ ∫[ ] { }
(12.66)
̇
∫ ∫ [ ] [ ]{ ̂}
∫ ∫ [ ] [ ]{ ̂} { }[ ]{ ̂} ∫[ ] { }
Now, let’s introduce the following matrices.

SOLIDS AND FLUIDS
12.3.2 MATRICES AND THE DISCRETIZED EQUATION OF MOTION OF FLUIDS
Since the viscous terms are similar in forms to those derived for plane stress problem; let’s
call the corresponding matrix as stiffness matrix [ ] where
[ ] ∫ ∫ [ ] [ ]{ }[ ] [ ] (12.67)
Due to the similarity, stiffness matrix algorithms previously established for plane stress
problem can thus be applied. Next, by observing terms associated with time derivative of
dofs, one can notice that they are in similar forms to the mass matrix discussed in Chapter 6.
So let’s call the corresponding matrix as mass matrix [ ] where
(12.68)
[ ] ∫ ∫[ ] [ ]
Next, by considering the pressure terms, let’s introduce a coefficient matrix, [ ] where
(12.69)
[ ] ∫ ∫ [ ] { }{ }
The second last terms on the right hand side of Eqn. (12.66) exhibits nonlinearity.
Accordingly, we introduce a matrix termed as nonlinear matrix [ ] where
(12.70)
[ ] ∫ ∫ [ ] [ ]{ } { }[ ]
Finally, we introduce a body force vector { }; and natural force vector { } as
(12.71)
{ } ∫ ∫[ ] { }
(12.72)
{ } ∫[ ] { }
Having established these corresponding matrices and vectors, the continuity and the
momentum equations can be compactly expressed. For the continuity equation, it can be
shown that;

SOLIDS AND FLUIDS
(12.73)
[∫ ∫ [ ] { }{ } ] ∫ ∫ { } { }[ ]
By inserting Eqns. (12.67) to (12.73) into Eqns. (12.64) and (12.66) and by extracting the dofs
from the integral sign, we obtain the following
Continuity
[ ] { ̂} (12.74)
Momentum
[ ]{ } [ ]{ ̂} { } ̇
[ ]{ ̂} [ ]{ ̂} { } (12.75)
Since IBP is conducted only on some terms whilst there are others which are not, it is
important to note that, those terms which are subjected to IBP, the multiplication between
terms inside the integral must be carried out from left to right while for those which are not,
the multiplication must be carried out from right to left. To demonstrate this and to provide
example on how to build the above matrices, the determination of element (3,4) for each
matrix and element (3,1) for each vector above are shown below. Complete formula for
every element in every matrix and vector are given in the Appendix. Although in practice,
numerical integration is employed which detailed procedure is given in later chapter, herein
analytical integration is employed so as to allow for easy tracing of the procedure and to
allow for its immediate application. Such analytical integrations are possible herein because
we limit our discussion to rectangular domain thus rectangular elements only.
i.
We begin our demonstration by showing the derivation for element of the mass matrix,
[ ]. Since this term has no derivative, no dilemma on whether to conduct IBP or not. Thus,
∫∫[ ][ ] ∫∫ ( ) ( )( ) (12.76)

SOLIDS AND FLUIDS
ii.
Stiffness matrix, [ ] involves IBP thus the matrix multiplication must be carried out from left
to right. We will do this in step-by-step manner.
Step 1
(12.77)
[ ] [ ]
[ ]
Step 2
[ ][ ] [ ] (12.78)
Step 3
(12.79)
[ ] [ ( ) ( ) ]
[ ]
Step 4
[ ( ) ( ) ][ ]
(12.80)
( ) ( ) ( ) ( )
Now by employing Eqn. (12.67), is given as
∫∫ ( ) (12.81)
For a rectangular element with a width of and a height of , as shown in Figure 12.3,
can be re-given as

SOLIDS AND FLUIDS
∫ ∫ (( )( ( ) )) (12.82)
iii.
Coefficient matrix, [ ] involves IBP thus the matrix multiplication must be carried out from
left to right.
Step 1
(12.83)
[ ][ ] [ ]
Step 2
[ ][ ] [ ] (12.84)
∫∫( )
∫∫( ) (12.85)
iv.
Nonlinear matrix, [ ] does not involves IBP thus the matrix multiplication must be carried
out from right to left.
Step 1
[ ][ ] [ ]
(12.86)

SOLIDS AND FLUIDS
Step 2
(12.87)
[ ]
{ } [ { } ]
Step 3
[ ]
[ { } ]
{ } { } { } { } { } { } { } { }
{ } { } { } { } { } { } { } { }
[ ]
(12.88)

SOLIDS AND FLUIDS
Step 4
{ } { } { } { } { } { } { } { }
[ ]
{ } { } { } { } { } { } { } { }
[ ]
[ { } { } { } { } { } { }
{ } { }]
(12.89)
Step 5(analytical integration)
∫∫ { } { } { } { }
{ } { } { } { }
{ } { } { } { } { } { }
(12.90)
{ } { }
12.3.3 ALL THE MATRICES ARE COMBINED TOGETHER
The continuity equation and the momentum equations, as given by Eqns. (12.74) and
(12.75), respectively can be combined together in the following way. If the dof are grouped
together in such way that
{ ̂} (12.91)
{ } { }
{ ̂}
then eqns. (12.74) and (12.75) can be arranged as
[ ] ̇
{ ̂} [ ] [ ] [ ] { ̂} { } { } (12.92)
[ ]{ } [ ]{ } { } { }
{ ̇̂} [ ] { ̂}
Finally Eqn. (12.92) can be given as
[ ]{ ̇ } [ ( )]{ } { } (12.93)
where the use of capital letters are not only to highlight the representation of the various
sub-matrices by a combined matrix where

SOLIDS AND FLUIDS
[ ] [ ] [ ] (12.94)
[ ( )] [ ]
[ ]
and
[ ] (12.95)
[ ]{ ̇ } [ ]
but also to highlight that, our discussion from hereafter assumes that the elements have
been assembled.
To note, [ ( )] is expressed in such way so as to remind that the matrix contains nonlinear
matrix [ ] which in turn contains the dofs, { } hence the nonlinear nature.
Next, we are going to discuss about time integration and nonlinear scheme required in the
solution of Eqn. (12.93).
12.4 TIME INTEGRATION
Eqn. (12.93) involves the first order derivative of the dof’s in time, expressed by { ̇ }. As a
derivative, it can be discretized, which the common practice would be, by finite difference
method. Basically there are two ways to do this;
i) Forward Euler scheme

ii) Backward Euler scheme
Despite the two available techniques, herein, we will concern only with Backward Euler
scheme. Readers who are interested to know about the other technique can refer to books
on finite difference method and finite element method.
Based on Backward Euler, Eqn. (12.78) can be discretized in time as
(12.96)
[ [ ] [ ( )]] { } [ ]{ } { }
Eqn. (12.96) is the discretized equation of Navier-Stokes in time by Backward Euler (finite
difference). However, the equation is yet to be solved as it is nonlinear. We must resort to a
nonlinear solution scheme which is to be discussed next.

SOLIDS AND FLUIDS
12.4 PICCARD NONLINEAR SCHEME (DIRECT SUBSTITUTION)
Eqn. (12.96) can be given conveniently for nonlinear discussion as
[ ̅( )]{ } {̅ ( )} (12.97)
where
(12.98)
[ ̅( )] [ [ ] [ ( )]]
(12.99)
{̅ ( )} [ ]{ } { }
There are many nonlinear schemes which are available; the most basic ones are Piccard and
Newton-Raphson schemes. Herein, we will only discuss Piccard scheme. It is the most
straightforward scheme and also known as direct substitution method. To solve Eqn. (12.97)
by Piccard scheme, we use previous iteration values of i.e. { } where represent
last iteration, in the construction of the matrix, [ ̅ ( )]. With this, Eqn. (12.97) can be
modified as
[ ̅( )]{ } {̅ } (12.100)
where represent current stage of iteration. Having established Eqn. (12.85), we are
now all set to solve the problem. We need to iterate until the solution { } is converged.
Once converged, we then go to the next time step. A step-by-step procedure is given as
follows.
STEP 1: take or assume known values of { }
STEP 2: use these known values of { } to calculate [ ̅ ( )]
STEP 3: solve { }
STEP 4: calculate residual and check for convergence
STEP 5: if converged, stop iteration and go the next time step, else repeat STEP 1 by
using the just solved { } as { }
We are going to demonstrate the above procedure with an example.

SOLIDS AND FLUIDS
EXAMPLE 12.1
Herein, we are going to demonstrate the FEM formulation of Navier-Stokes equations by

solving a benchmark problem known as lid-driven cavity. Herein, for step-by-step
demonstration (hand-calculation), only four assembled elements are used, so as to allow for
easy tracing of the procedure. However, it must be noted that such usage is for
demonstration purpose only as the result will be very poor. After the step-by-step
calculation, the procedure is run using a sufficient number of elements (denser mesh). The
results are then validated against an established data for proof of correctness.
Figure E12.1 Lid-driven cavity problem
The parameters used in this problem are:
 Density,
 Viscosity,
 Lid velocity,
 Time step,

SOLIDS AND FLUIDS
Figure E12.2 The arrangement of node and element numbering
Figure E12.3 The arrangement of global dofs

SOLIDS AND FLUIDS
Time step 1, Iteration 1:
Due to the symmetrically, each element would have similar local stiffness matrix [ ], mass matrix [ ], coefficient matrix, [ ], and nonlinear
matrix [ ]. Based on Eqn. 12.67, local stiffness matrix [ ] is given as
[ ]
[ ]
(E12.1)

SOLIDS AND FLUIDS
Based on Eqn. 12.68, local mass matrix [ ] is given as
[ ]
[ ]
(E12.2)
Based on Eqn. 12.69, coefficient matrix [ ] is given as
[ ] [ ]
(E12.3)

SOLIDS AND FLUIDS
By assume { } , based on Eqn. 12.70, nonlinear matrix [ ] . Since the size of an assembled stiffness matrix would be very large,
herein, reduced size of assembled stiffness matrix with boundary conditions have been imposed, are presented. (i.e.
[ ] { }
{ }
(E12.4)

SOLIDS AND FLUIDS
By solving Eqn. (E12.4) using Matlab command “\”,{ }are thus obtained as:
(E.12.5)
{ }
{ }
{ }
The maximum residual between current and previous result are 2.5431.

SOLIDS AND FLUIDS
Using the result from previous iteration, the nonlinear matrix [ ] is given as
[ ]
[ ]
(E12.6)

SOLIDS AND FLUIDS
Thus, the assembled stiffness matrix can be given as
[ ] { }
{ }
(E12.7)

SOLIDS AND FLUIDS
(E.12.8)
{ }
{ }
{ }
The maximum residual between current and previous result are 0.0034. For the purpose of
demonstration, let assume that the result is convergence. We can now break the nonlinear
iteration loop and calculate the value at the second time step using the current result as
initial value.

SOLIDS AND FLUIDS
[ ]
[ ]
(E12.9)

SOLIDS AND FLUIDS
[ ] { }
{ }
(E12.10)

SOLIDS AND FLUIDS
(E.12.11)
{ }
{ }
{ }
The maximum residual between current and previous result are 0.1938.

SOLIDS AND FLUIDS
[ ]
[ ]
(E12.12)

SOLIDS AND FLUIDS
[ ] { }
{ }
(E12.13)

SOLIDS AND FLUIDS
(E.12.14)
{ }
{ }
{ }
The maximum residual between current and previous result are 0.0012. The obtained
results after total time are validated herein for various mesh number against those
obtained from commercial software.
Mesh (4x4)
Mesh (20x20)
Mesh (40x40)
Software

SOLIDS AND FLUIDS
Figure E12.4 Velocity contour of 40 x 40 elements (a) and commercial software (b).
Figure E12.5 Pressure contour of 40 x 40 elements (a) and commercial software (b).

SOLIDS AND FLUIDS
Figure E12.6 Velocity profile at x=0.5
Figure E12.7 Pressure profile at y=0.5

SOLIDS AND FLUIDS
Main.m (Main file)
clc;clear all;
%--------------------------------------------------------------------------
% Input
%--------------------------------------------------------------------------
Nx=2; % No of Element in x-dir

Ny=2; % No of Element in y-dir

re = 1;
rho = 1; % Fluid density [kg / m^3]
miu = 1.1; % Fluid viscosity [kg / (m.s)]
T = 10; % Total time [s]
dt = 0.1; % Time increment [s]
maxIter = 100; % Maximum nonlinear iteration
maxTol = 1e-4; % Maximum nonlinear error tolerance
Vel = 2; % velocity [m/s]
%--------------------------------------------------------------------------
% Preparing data
%--------------------------------------------------------------------------
[deltax,deltay,NodeNo,dofNo_UV,dofNo_P,numVariables,nodeXY,...
numNode,numElement,nodeXY_U,nodeXY_V,nodeXY_P]=PrepareData...
(Lx,Ly,Nx,Ny);
% Initial condition (assume zero field)

uvp0 = zeros(numVariables, 1); % Dof at previous time step
uvp = zeros(numVariables, 1); % Dof at current time step
t = 0; % Start time
Data_uvp=[];
index_save=0;
%-------------------------------------------------------------------------
% Euler Backward Scheme
%-------------------------------------------------------------------------
for iTime = 1:round(T/dt)
% Time increment
t = t + dt;
%---------------------------------------------------------------------
% Picard Scheme
%---------------------------------------------------------------------
for iter = 1:2%maxIter
% Preallocate
K = zeros(numVariables);
M = zeros(numVariables);
R = zeros(numVariables,1);
%-----------------------------------------------------------------
% Loop for every element
%-----------------------------------------------------------------

SOLIDS AND FLUIDS
for iele = 1:numElement
NonLinear_paper =NonLinearMatrix(rho,deltax,deltay,uvp);
Mass_paper =MassMatrix(rho,deltax,deltay);
Stiffness_paper =StiffnessMatrix(miu,deltax,deltay);
Coefficient_paper =CoefficientMatrix(deltax,deltay);
%-------------------------------------------------------------
% Assemble Matrix
%-------------------------------------------------------------
for assem_i=1:16
dofNo_UV_i=dofNo_UV(iele,assem_i);
for assem_j=1:16
dofNo_UV_j=dofNo_UV(iele,assem_j);
K(dofNo_UV_i,dofNo_UV_j)=K(dofNo_UV_i,dofNo_UV_j)+...
Stiffness_paper(assem_i,assem_j)+...
NonLinear_paper(assem_i,assem_j);
M(dofNo_UV_i,dofNo_UV_j)=M(dofNo_UV_i,dofNo_UV_j)+...
Mass_paper(assem_i,assem_j);
end
end
Coeff1=-Coefficient_paper;
Coeff2=Coeff1';
for assem_i=1:16
dofNo_UV_i=dofNo_UV(iele,assem_i);
for assem_j=1:4
dofNo_P_j=dofNo_P(iele,assem_j);
K(dofNo_UV_i,dofNo_P_j)=K(dofNo_UV_i,dofNo_P_j)+...
Coeff1(assem_i,assem_j);
K(dofNo_P_j,dofNo_UV_i)=K(dofNo_P_j,dofNo_UV_i)+...
Coeff2(assem_j,assem_i);
end
end
%-------------------------------------------------------------
end
%-----------------------------------------------------------------
% Effective stiffness & force vector (equation 12.83 & 12.84)

Keff = 1/dt*M + K;
Reff = 1/dt*(M*uvp0) + R;
Keffeli=Keff;
Reffeli=Reff;
%-----------------------------------------------------------------
% Impose BC
%-----------------------------------------------------------------
dofloc_eliuv=[];
dofloc_Velo=[];

SOLIDS AND FLUIDS
indx_e=0;
indx_velo=0;
for i = 1: numNode
% Node coordinate
x = nodeXY(i, 1);
y = nodeXY(i, 2);
% Global dof numbering for each variables

iu = nodeXY_U(i,3);
iv = nodeXY_V(i,3);
ip = nodeXY_P(i,3);
% Top lid
if y == 1
% Horizontal velocity bc (Top lid moves with velocity u0)

Keff(iu, :) = 0;
Keff(iu, iu) = 1;
Reff(iu, :) = Vel;
indx_velo=indx_velo+1;
dofloc_Velo(indx_velo,1)=iu;
% Vertical velocity bc
Keff(iv, :) = 0;
Keff(iv, iv) = 1;
Reff(iv, :) = 0;
end
% Bottom wall | Left wall | Right wall

if y == 0 || x == 0 || x == 1
% Horizontal velocity bc
Keff(iu, :) = 0;
Keff(iu, iu) = 1;
Reff(iu, :) = 0;
indx_e=indx_e+1;
dofloc_eliuv(indx_e,1)=iu;
% Vertical velocity bc
Keff(iv, :) = 0;
Keff(iv, iv) = 1;
Reff(iv, :) = 0;
indx_e=indx_e+1;
dofloc_eliuv(indx_e,1)=iv;
indx_e=indx_e+1;
dofloc_eliuv(indx_e,1)=ip;
end
% Pressure bc (bottom right node)

if (x == 1 && y==0)

SOLIDS AND FLUIDS
% Pressure bc
if ip>0
Keff(ip, :) = 0;
Keff(ip, ip) = 1;
Reff(ip, :) = 0;
end
end
end
%-----------------------------------------------------------------
for ivel=1:size(dofloc_Velo,1)
Velloc=dofloc_Velo(ivel,1);
Reffeli(Velloc,1)=Vel;
end
dofloc_eliuv=sort(dofloc_eliuv,'descend');
for ieli=1:size(dofloc_eliuv,1)
elival=dofloc_eliuv(ieli,1);
if elival >0
Keffeli(elival,:)=[];
Keffeli(:,elival)=[];
Reffeli(elival,:)=[];
end
end
% Solving system of simulateneous equation
unote=Keffeli\Reffeli;
unoteR=RepairAns(unote,dofloc_eliuv,numVariables);
% Calculate residual between current & previous iteration

err = norm(unoteR - uvp);
% Assign variables for next iteration

uvp = unoteR;
% Break loop if residual less than tolerance

if err<maxTol
break
end
end
%---------------------------------------------------------------------
% Assign variables for next time iteration
uvp0 = uvp;
%---------------------------------------------------------------------
% Save result
%---------------------------------------------------------------------
index_save=index_save+1;
Data_uvp(:,index_save)=uvp;
end
%-------------------------------------------------------------------------

SOLIDS AND FLUIDS
% Plot Velocity
%-------------------------------------------------------------------------
figure(1)
FU = TriScatteredInterp(nodeXY(:,1), nodeXY(:,2), uvp(nodeXY_U(:,3)));
FV = TriScatteredInterp(nodeXY(:,1), nodeXY(:,2), uvp(nodeXY_V(:,3)) );
% Meshgrid
[X, Y] = meshgrid(0:deltax:Lx, 0:deltay:Ly);
% Evaluate velocity at meshgrid points

U = FU(X,Y);
V = FV(X,Y);
% Plot velocity contour

set(gcf, 'color', 'w')
contourf(X, Y, sqrt( U.^2 + V.^2 ), 'LineColor', 'none')
axis equal tight
title ('Velocity contour')
colorbar
drawnow
PrepareData.m (subroutine files)
function [deltax,deltay,NodeNo,dofNo_UV,dofNo_P,TDOF,nodeXY,numNode,...
numElement,nodeXY_U,nodeXY_V,nodeXY_P]=...
PrepareData(Lx,Ly,Nx,Ny)
deltax=Lx/Nx; % Width of single element

deltay=Ly/Ny; % Height of single element
dofNo_UV=[];
dofNo_P=[];
NodeNo=[];
index=0;
for i=1:Ny
for j=1:Nx
index=index+1;
val1=2*(j-1)+1;
val2=(i-1)*((3*Nx)+2);
val3=i*((3*Nx)+2);
val4=j-Nx;
Nd1=val1+val2;
Nd7=val1+val3;
Nd4=val3+val4;
Nd2=Nd1+1;
Nd3=Nd1+2;
Nd5=Nd7+2;
Nd6=Nd7+1;
Nd8=Nd4-1;

SOLIDS AND FLUIDS
dof1=2*(Nd1-1)+1; dof9=2*(Nd2-1)+1;
dof2=2*(Nd1-1)+2; dof10=2*(Nd2-1)+2;
dof3=2*(Nd3-1)+1; dof11=2*(Nd4-1)+1;
dof4=2*(Nd3-1)+2; dof12=2*(Nd4-1)+2;
dof5=2*(Nd5-1)+1; dof13=2*(Nd6-1)+1;
dof6=2*(Nd5-1)+2; dof14=2*(Nd6-1)+2;
dof7=2*(Nd7-1)+1; dof15=2*(Nd8-1)+1;
dof8=2*(Nd7-1)+2; dof16=2*(Nd8-1)+2;
dofNo_UV(index,:)=[dof1 dof2 dof3 dof4 dof5 dof6 dof7 dof8...

dof9 dof10 dof11 dof12 dof13 dof14 dof15 dof16];
end
end
TTDOF_UV=max(max(dofNo_UV));
% Correction on dofNo_P
index=0;
for i=1:Ny
for j=1:Nx
index=index+1;
val1=(Nx+1)*(i-1)+j;
val2=(Nx+1)*(i)+j;
Nd1=val1+TTDOF_UV;
Nd2=val1+1+TTDOF_UV;
Nd3=val2+1+TTDOF_UV;
Nd4=val2+TTDOF_UV;
dofNo_P(index,:)=[Nd1 Nd2 Nd3 Nd4];
end
end
TDOF=max(max(dofNo_P)); % Total DOF
% Preliminary nodes coordinates

x = linspace(0, Lx, 2*Nx+1);
y = linspace(0, Ly, 2*Ny+1);
[X, Y] = ndgrid(x,y);
% Serendepity element (no center node)

n = 0;
nodes = zeros(2*Nx+1, 2*Ny+1);
nodes(2:2:end, 2:2:end) = NaN;
for j = 1: 2*Ny+1
for i = 1: 2*Nx+1
% Skip center element
if isnan( nodes(i,j) )
else
n = n+1;
nodes(i,j) = n;
end
end
end
% Elements number

SOLIDS AND FLUIDS
numElement = Nx * Ny;
n = 0;
elementNode = zeros(numElement, 8);
elementCenter = zeros(numElement, 2);
for j = 2:2:length(y)
for i = 2:2:length(x)
n = n+1;
elementNode(n,:) = [ nodes(i-1, j-1), ...
nodes(i+1, j-1), ...
nodes(i+1, j+1), ...
nodes(i-1, j+1), ...
nodes(i, j-1), ...
nodes(i+1, j), ...
nodes(i, j+1), ...
nodes(i-1, j), ...
];
% Element center coordinate
elementCenter(n,:) = [X(i,j), Y(i,j)];
end
end
% Nodes coordinate
nodeXY = [ X(:) Y(:) ];
nodeXY( isnan( nodes(:) ), : ) = [];
% Node number
numNode = length(nodeXY);
nodeXY_U=zeros(numNode,3);
nodeXY_V=zeros(numNode,3);
nodeXY_P=zeros(numNode,3);
nodeXY_U(1:numNode,1:2)=nodeXY;
nodeXY_V(1:numNode,1:2)=nodeXY;
nodeXY_P(1:numNode,1:2)=nodeXY;
% Connectivity
for i=1:numElement
Nd1=NodeNo(i,1);
Nd2=NodeNo(i,2);
Nd3=NodeNo(i,3);
Nd4=NodeNo(i,4);
Nd5=NodeNo(i,5);
Nd6=NodeNo(i,6);
Nd7=NodeNo(i,7);
Nd8=NodeNo(i,8);
u1=dofNo_UV(i,1); v1=dofNo_UV(i,2);

SOLIDS AND FLUIDS
p1=dofNo_P(i,1);
p2=dofNo_P(i,2);
p3=dofNo_P(i,3);
p4=dofNo_P(i,4);
nodeXY_U(Nd1,3)=u1;
nodeXY_V(Nd1,3)=v1;
nodeXY_P(Nd1,3)=p1;
nodeXY_U(Nd2,3)=u2;
nodeXY_V(Nd2,3)=v2;
nodeXY_P(Nd2,3)=p2;
nodeXY_U(Nd3,3)=u3;
nodeXY_V(Nd3,3)=v3;
nodeXY_P(Nd3,3)=p3;
nodeXY_U(Nd4,3)=u4;
nodeXY_V(Nd4,3)=v4;
nodeXY_P(Nd4,3)=p4;
nodeXY_U(Nd5,3)=u5;
nodeXY_V(Nd5,3)=v5;
nodeXY_U(Nd6,3)=u6;
nodeXY_V(Nd6,3)=v6;
nodeXY_U(Nd7,3)=u7;
nodeXY_V(Nd7,3)=v7;
nodeXY_U(Nd8,3)=u8;
nodeXY_V(Nd8,3)=v8;
end
NonLinearMatrix.m (subroutine files)

function NonLinear_Eqn=NonLinearMatrix(rho,a,b,DataU)
u1=DataU(1,1);
v1=DataU(2,1);
u2=DataU(3,1);
v2=DataU(4,1);
u3=DataU(5,1);
v3=DataU(6,1);
u4=DataU(7,1);
v4=DataU(8,1);
u5=DataU(9,1);
v5=DataU(10,1);
u6=DataU(11,1);
v6=DataU(12,1);
u7=DataU(13,1);
v7=DataU(14,1);
u8=DataU(15,1);
v8=DataU(16,1);
NonLinear_Eqn= not shown because too big
MassMatrix.m (subroutine files)

function Mass_Eqn=MassMatrix(rho,a,b)

SOLIDS AND FLUIDS
Mass_Eqn= not shown because too big
StiffnessMatrix.m (subroutine files)

function Stiffness_Eqn=StiffnessMatrix(miu,a,b)
Stiffness_Eqn= not shown because too big
CoefficientMatrix.m (subroutine files)

function Coefficient_Eqn=CoefficientMatrix(a,b)
Coefficient_Eqn= [-0.7e1 / 0.36e2 * b b / 0.36e2 b / 0.18e2 -b / 0.18e2;

-0.7e1 / 0.36e2 * a -a / 0.18e2 a / 0.18e2 a / 0.36e2;
-b / 0.36e2 0.7e1 / 0.36e2 * b b / 0.18e2 -b / 0.18e2;
-a / 0.18e2 -0.7e1 / 0.36e2 * a a / 0.36e2 a / 0.18e2;
-b / 0.18e2 b / 0.18e2 0.7e1 / 0.36e2 * b -b / 0.36e2;
-a / 0.18e2 -a / 0.36e2 0.7e1 / 0.36e2 * a a / 0.18e2;
-b / 0.18e2 b / 0.18e2 b / 0.36e2 -0.7e1 / 0.36e2 * b;
-a / 0.36e2 -a / 0.18e2 a / 0.18e2 0.7e1 / 0.36e2 * a;
0.2e1 / 0.9e1 * b -0.2e1 / 0.9e1 * b -b / 0.9e1 b / 0.9e1;
-a / 0.6e1 -a / 0.6e1 -a / 0.6e1 -a / 0.6e1;
b / 0.6e1 b / 0.6e1 b / 0.6e1 b / 0.6e1;
a / 0.9e1 0.2e1 / 0.9e1 * a -0.2e1 / 0.9e1 * a -a / 0.9e1;
b / 0.9e1 -b / 0.9e1 -0.2e1 / 0.9e1 * b 0.2e1 / 0.9e1 * b;
a / 0.6e1 a / 0.6e1 a / 0.6e1 a / 0.6e1;
-b / 0.6e1 -b / 0.6e1 -b / 0.6e1 -b / 0.6e1;
0.2e1 / 0.9e1 * a a / 0.9e1 -a / 0.9e1 -0.2e1 / 0.9e1 * a;];
RepairAns.m (subroutine files)

function unoteR=RepairAns(unote,dofloc_eliuv,numVariables)
dof_rem=[];
count=0;
for i_rem=1:numVariables
iexis=0;
for ieli=1:size(dofloc_eliuv,1)
elival=dofloc_eliuv(ieli,1);
if i_rem==elival
iexis=1;
end
end
if iexis==0
count=count+1;
dof_rem(count,1)=i_rem;
end
end
unoteR=zeros(numVariables,1);
for i_unoteR=1:numVariables
for j_unoteR=1:size(dof_rem,1)
loc_R=dof_rem(j_unoteR,1);
if i_unoteR==loc_R

SOLIDS AND FLUIDS
unoteR(loc_R,1)=unote(j_unoteR,1);
end
end
end

Linear and Nonlinear FEM Airil Et Al

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear and Nonlinear FEM Airil Et Al

Uploaded by

Copyright:

Available Formats

LINEAR AND

PART 1-LINEAR FORMULATION (A: 1D)

CHAPTER 1: BASIC CONCEPT OF NUMERICAL TECHNIQUES 2-16

CHAPTER 2: GALERKIN FORMULATION: BAR ELEMENT 17-45

CHAPTER 3: GALERKIN FORMULATION: BEAM ELEMENT 46-69

CHAPTER 4: PLANE STRUCTURES: TRUSS AND FRAME 70-101

CHAPTER 5: EIGENPROBLEMS: FREE VIBRATION AND BUCKLING 102-139

CHAPTER 6: DYNAMICS: FORCED VIBRATION 140-164

PART 1-LINEAR FORMULATION (B: 1D)

CHAPTER 7: SCALAR ELEMENT (HEAT TRANSFER) 165-188

CHAPTER 8: PLANE ELASTICITY (PLANE ELASTICITY) 189-227

PART 2-NONLINEAR FORMULATION (A: 1D)

CHAPTER 10: NONLINEAR BAR ELEMENT 228-312

PART 2-NONLINEAR FORMULATION (B: 2D)

CHAPTER 12: FLUID DYNAMICS (NAVIER-STOKE EQUATIONS) 313-363

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

BASIC CONCEPT OF NUMERICAL TECHNIQUES

1.1 INTRODUCTION: WHAT IS FINITE ELEMENT METHOD (FEM)?

1.2 BASIC CONCEPT OF NUMERICAL TECHNIQUES

i. Finite Element Method (FEM)

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Figure 1.1 Bar element / structure

Boundary conditions (equations)

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

However, observing Eqn. (1.9) we can notice that:

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Note: , , and are the alternative notations known as indicial/tensorial notations. We

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Table 1.1: Values of guessed solution’s constant

Figure 1.3 Comparison between unumerical and uexact

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Table 1.2: Comparison of values between unumerical and uexact

x=1m x=2m x=3m x=4m x=5m

1.3 BASIC CONCEPT OF FEM: WEIGHTED RESIDUAL METHOD

1.3.1 BASIC CONCEPT OF WRM

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

In WRM, Eqn. (1.14 a) is written as

Figure 1.4: Typical plot of residual error function, R

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Herein, to provide the sufficient number of algebraic equations, what we do is to firstly

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

By conducting the integration we obtain:

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

In matrix forms, Eqns. (1.20) can be given as

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Table 1.3: Values of guessed solution’s constant

Figure 1.7 Comparison between uWRM , unumerical and uexact

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

Table 1.4: Comparison of values between uWRM , unumerical and uexact

x=1m x=2m x=3m x=4m x=5m

1.5 SOURCE CODE

0 0 2*E*A 6*E*A*(L/2) 12*E*A*(L/2)^2;

fz=[ 0; P ;-q;-q;-q]; % Ritz

© AIRIL YASREEN MOHD YASSIN, PhD, MOHD AL AKHBAR NOOR, MEng,

GALERKIN FORMULATION: BAR ELEMENT

2.2 ORDINARY DIFFERENTIAL EQUATION (ODE) OF BAR PROBLEM

Figure 2.1: Bar structure and its differential element

From Hooks Law, we know that

By inserting Eqns. (2.4) and (2.5) into Eqn. (2.3) gives

a) Natural/force boundary conditions

b) Essential/displacement boundary conditions

The Galerkin WRM differs in the following aspects:

0 0 2EA 6EA(L/2) 12EA(L/2)^2;