Finite Element Programming

with MATLAB

Chapter 12 includes a general introduction to MATLAB functions, selected topics in

linear algebra with MATLAB, and a collection of finite element programs for: trusses
(Chapter 2), general one-dimensional problems (Chapter 5), heat conduction in 2D
(Chapter 8) and elasticity in 2D (Chapter 9). This Chapter is published electronic format
only for several reasons:

1. the data structure of the finite element program will be periodically updated to
reflect emerging finite element technologies and MATLAB syntax changes;
2. to allow the course instructors to use their own MALAB or other finite element
codes.
3. to create a forum where students and instructors would exchange ideas and place
alternative finite element program data structures. The forum is hosted at
http://1coursefem.blogspot.com/

12.1 Using MATLAB for FEM1

12.1.1 The MATLAB Windows
Upon opening MATLAB you should see three windows: the workspace window, the
command window, and the command history window as shown in Figure 12.1. If you do
not see these three windows, or see more than three windows you can change the layout
by clicking on the following menu selections: View → desktop layout → default.

1
May not be covered in the class. Recommended as independent reading.
 Figure 12.1: Matlab Windows

12.1.2 The Command Window

If you click in the command window a cursor will appear for you to type and enter various
commands. The cursor is indicated by two greater than symbols (>>).

12.1.3 Entering Expressions

After clicking in the command window you can enter commands you wish MATLAB to execute.
Try entering the following: 8+4. You will see that MATLAB will then return: ans = 12.

12.1.4 Creating Variables

Just as commands are entered in MATLAB, variables are created as well. The general
format for entering variables is: variable = expression. For example, enter y = 1 in the
command window. MATLAB returns: y = 1. A variable y has been created and assigned
a value of 1. This variable can be used instead of the number 1 in future math operations.
For example: typing y*y at the command prompt returns: ans = 1. MATLAB is case
sensitive, so y=1, and Y=5 will create two separate variables.

2
12.1.5 Functions
MATLAB has many standard mathematical functions such as sine (sin(x)) and cosine
(cos(x)) etc. It also has software packages, called toolboxes, with specialized functions
for specific topics.

12.1.6 Getting Help and Finding Functions

The ability to find and implement MATLAB’s functions and tools is the most important
skill a beginner needs to develop. MATLAB contains many functions besides those
described below that may be useful.
There are two different ways obtain help:
• Click on the little question mark icon at the top of the screen. This will open up the
help window that has several tabs useful for finding information.

• Type “help” in the command line: MATLAB returns a list of topics for which it has
functions. At the bottom of the list it tells you how to get more information about a
topic. As an example, if you type “help sqrt” and MATLAB will return a list of
functions available for the square root.

12.1.7 Matrix Algebra with MATLAB

MATLAB is an interactive software system for numerical computations and graphics. As
the name suggests, MATLAB is especially designed for matrix computations. In addition,
it has a variety of graphical and visualization capabilities, and can be extended through
programs written in its own programming language. Here, we introduce only some basic
procedures so that you can perform essential matrix operations and basic programming
needed for understanding and development of the finite element program.

12.1.8 Definition of matrices

A matrix is an mxn array of numbers or variables arranged in m rows and n columns; such
a matrix is said to have dimension mxn as shown below

Bold letters will denote matrices or vectors. The elements of a matrix a are denoted by
, where i is the row number and j is the column number. Note that in both describing
the dimension of the matrix and in the subscripts identifying the row and column number,
the row number is always placed first.
An example of a 3x3 matrix is:

3
The above matrix a is is an example of a square matrix since the number of rows and
columns are equal.
The following commands show how to enter matrices in MATLAB (>> is the
MATLAB prompt; it may be different with different computers or different versions of
MATLAB.)

Notice that rows of a matrix are separated by semicolons, while the entries on a row are
separated by spaces (or commas). The order of matrix can be determined from

The transpose of any matrix is obtained by interchanging rows and columns. So for
example, the transpose of a is:

In MATLAB the transpose of a matrix is denoted by an apostrophe (‘).

If , the matrix is symmetric.

A matrix is called a column matrix or a vector if n=1, e.g.

In MATLAB, single subscript matrices are considered row matrices, or row vectors.
Therefore, a column vector in MATLAB is defined by

4
Note the transpose that is used to define b as a column matrix. The components of the
vector b are . The transpose of b is a row vector

or in MATLAB

A matrix is called a diagonal matrix if only the diagonal components are nonzero,
i.e., . For example, the matrix below is a diagonal matrix:

A diagonal matrix in MATLAB is constructed by first defining a row vector b = [1 5 6],

and then placing this row vector on the diagonal

A diagonal matrix where all diagonal components are equal to one is called an identity or
unit matrix and is denoted by I. For example, identity matrix is given by

5
The MATLAB expression for an order n unit matrix is

Thus, the MATLAB expression gives the above matrix.

A matrix in which all components are zero is called a zero matrix and is denoted by 0. In
MATLAB, B = zeros (m, n) creates matrix B of zeros. A random matrix can
be created by rand (m,n).

In finite element method, matrices are often sparse, i.e., they contain many zeros.
MATLAB has the ability to store and manipulate sparse matrices, which greatly increases
its usefulness for realistic problems. The command sparse (m, n) stores an zero
matrix in a sparse format, in which only the nonzero entries and their locations are sorted.
The nonzero entries can then be entered one-by-one or in a loop.

Notice that the display in any MATLAB statement can be suppressed by ending the line
with a semicolon.

The inverse of a square matrix is defined by

if the matrix is not singular. The MATLAB expression for the inverse is . Linear
algebraic equations can also be solved by using backslash operator as shown in Section
1.3.10, which avoids computations of the inverse and is therefore faster.
The matrix is nonsingular if its determinant, denoted by , is not equal to
zero. A determinant of a 2x2 matrix is defined by

6
The MATLAB expression for the determinant is

For example,

12.1.9 Operation with matrices

Addition and Subtraction

An example of matrix addition in MATLAB is given below:

7
Multiplication

1. Multiplication of a matrix by a scalar

2. Scalar product of two column vectors

In MATLAB the scalar product as defined above is given by either or .

The length of a vector a is denoted by |a| and is given by

The length of a vector is also called its norm.

3. Product of two matrices

The product of two matrices a and b is defined as

Alternatively we can write the above as

8
 Note the the i,j entry of c is the scalar product of row i of a and column j of b.

The product of two matrices a and b c is defined only if the number of columns in
a equals the number of rows in a. In other words, if a is an matrix, then
b must be an matrix, where k is arbitrary. The product c will then have
the same number of rows as a and the same number of columns as b, i.e. it will be
an matrix.

An important fact to remember is that matrix multiplication is not commutative,

i.e. except in unusual circumstances.

The MATLAB expression for matrix multiplication is

Consider the same matrices a and c as before. An example of matrix

multiplication with MATLAB is:

4. Other matrix operations

a) Transpose of product:
b) Product with identity matrix:
c) Product with zero matrix:

12.1.10 Solution of system of linear equations

Consider the following system of n equations with n unknowns, ,

9
We can rewrite this system of equations in matrix notation as follows:

where

The symbolic solution of the above system of equation can be found by multiplying both
sides with inverse of K, which yields

MATLAB expression for solving the system of equations is

or

An example of solution of system of equations with MATLAB is given below:

10
As mentioned before, the backslash provides a faster way to solve equations and should
always be used for large systems. The reason for this is that the backslash uses
elimination to solve with one right hand side, whereas determining the inverse of an nxn
matrix involves solving the system with n right hand sides. Therefore, the backslash
should always be used for solving large system of equations.

12.1.11 Strings in MATLAB

MATLAB variables can also be defined as string variables. A string character is a text
surrounded by single quotes. For example:

It is also possible to create a list of strings by creating a matrix in which each row is a
separate string. As with all standard matrices, the rows must be of the same length. Thus:

11
Strings are used for defining file names, plot titles, and data formats. Special built-in
string manipulation functions are available in MATLAB that allow you to work with
strings. In the MATALB codes provided in the book we make use of strings to compare
functions. For example the function strcmpi compares two strings

A true statment results in 1 and a false statement in 0. To get a list of all the built-in
MATLAB functions type

Another function used in the codes is fprintf. This function allows the user to print to the
screen (or to a file) strings and numeric information in a tabulated fasion. For example

The first argument to the function tells MATLAB to print the message to the screen. The
second argument is a string, where %d defines a decimal character with the value of 10
and the \n defines a new line. To get a complete description type

12.1.11 Programming with MATLAB

MATLAB is very convenient for writing simple finite element programs. It provides the
standard constructs, such as loops and conditionals; these constructs can be used
interactively to reduce the tedium of repetitive tasks, or collected in programs stored in
''m-files'' (nothing more than a text file with extension ``.m'').

12
12.1.11.1 Conditional and Loops
MATLAB has a standard if-elseif-else conditional.
The general form An example
>> t = 0.76;
if expression1
>> if t > 0.75
statements1
s = 0;
elseif expression2
elseif t < 0.25
statements2
s = 1;
…
else
…
s = 1-2*(t-0.25);
…
end
else
>> s
statements
s=
end
0

MATLAB provides two types of loops, a for-loop (comparable to a Fortran do-loop or a

C for-loop) and a while-loop. A for-loop repeats the statements in the loop as the loop
index takes on the values in a given row vector; the while-loop repeats as long as the
given expression is true (nonzero):

The general form Examples

>> for i=1:1:3
disp(i^2)
for index = start:increment:end
end
statements
1
end
4
9
>> x=1;
>> while 1+x > 1
while expression x = x/2;
statements end
end >> x
x=
1.1102e-16

12.1.11.2 Functions
Functions allow the user to create new MATLAB commands. A function is defined in an
m-file that begins with a line of the following form:

function [output1,output2,...] = cmd_name(input1,input2,...)

The rest of the m-file consists of ordinary MATLAB commands computing the values of
the outputs and performing other desired actions. Below is a simple example of a
function that computes the quadratic function . The following
commands should be stored in the file fcn.m (the name of the function within MATLAB
is the name of the m-file, without the extension)

13
12.1.12 Basic graphics
MATLAB is an excellent tool for visualizing and plotting results. To plot a graph the user
specifies the x coordinate vector and y coordinate vector using the following syntax

The above will generate

Figure 12.2 Typical outpout of plot(x,y) function

Various line types, plot symbols and colors may be obtained with plot(x,y,s) where s is a
character string consisting of elements from any combination of the following 3 columns:

b blue . point - solid

g green o circle : dotted
r red x x-mark -. dashdot
c cyan + plus -- dashed
m magenta * star (none) no line
y yellow s square
k black d diamond

14
To add a title, x and y labels, or a grid, the user should use the following MATLAB
functions. Note that the arguments to the functions are strings

In the MATLAB Finite Element code provided in the book, we also use two specialized
plots. The first plot is the patch function. This function is used to visualize 2D polygons
with colors. The colors are interpolated from nodes of the polygon to create a colored
surface. The following example generates a filled square. The colors along the x axis are
the same while the colors along the y axis are interpolated between the values [0,1].

Figure 12.3 Typical outpout of patch(x,y,c) function

We will use the patch function to visualize temperatures, stresses and other variables
obtained at the finite element solutions. Another specialized plot function is the quiver.
This function is used to visualize gradients of functions as an arrow plot. The following

15
example demonstrates the use of quiver function for plotting the gradients to the function
y=x2

Figure 12.4 Typical outpout of quiver(x,y,cx,cy) function

The hold on command is used to hold the current plot and all axis properties so that
subsequent graphing commands will executed on the existing graph.
Using the text function, the user can add to a plot a text message. For example

The first and second arguments define the position of the text on the plot, while the
string gives the text.

12.1.13 Remarks
a) In practice the number of equations n can be very large. PCs can today solve
thousands of equations in a matter of minutes if they are sparse (as they are in
FEM analysis-you will learn about this later) but sometimes millions of
equations are needed, as for an aircraft carrier or a full model of an aircraft;
parallel computers are then needed.
b) Efficient solution techniques that take advantage of the sparsity and other
advantageous properties of FEM equations are essential for treating even

16
 moderately large systems. The issue of how to efficiently solve large systems
will not be considered in this course.
c) In this course, we will see that
 The matrix corresponding to the system of equations arising from
FEM (denoted as K) is non-singular (often called regular), i.e.,
exists if the correct boundary conditions are prescribed and the
elements are properly formulated. Furthermore, for good models it is
usually well-conditioned, which means it is not very sensitive to
roundoff errors.
 K is symmetric, i.e. .
 K is positive definite, i.e., (meaning for any value of x)
Alternatively, K is said to be positive definite if all the eigenvalues are
strictly positive. The eigenvalue problem consists of finding nonzero
eigenvectors and the corresponding eigenvalues satisfying

The MATLAB expression for the eigenvalues problem is:

12.2 Finite element programming with MATLAB for trusses

In Chapter 2 the basic structure of the finite element method for truss structures
has been illustrated. In this section we present a simple finite element program using
MATLAB programming language. Since MATLAB manipulates matrices and vectors
with relative ease the reader can focus on fundamentals ideas rather than on algorithmic
details.
The code is written to very closely follow the formulation given in this chapter.
In order to better understand how the program works Figure 2.8 and Example Problem
2.2 in Chapter 2 have been included as examples solved by the program. Going through
the code along with this guide and the example problems is an effective method to
comprehend the program.

The main routines in the finite element code are:

17
 1. Preprocessing including input data and assembling the proper arrays, vectors,
and matrices.
2. Calculation of element stiffness matrices and force vectors
3. Direct assembly of matrices and vectors
4. Partition and solution
5. Postprocessing for secondary variables

Explanation for various MATLAB routines (stored in *.m files) are described as
comments within each subroutine.

12.2.1 Notations and definitions

12.2.1.1 User provided

nsd: number of space dimension (1 for 1D problems)
ndof: number of degrees-of-freedom per node
nnp: number of nodal points
nel: number of elements
nen: number of element nodes (2 in this case)
nd: number of prescribed (known) displacements
CArea: cross-sectional area
Area = CArea(element number)
E: Young’s Modulus
Young = E(element number)
leng: element length
Length = leng(element number)
phi: angle from axis to x axis for each element specified in degrees. Remember,
is
always from local node 1 to 2
phi = phi(element number)
IEN: connectivity information matrix
global node number = IEN (local node number, element number)
d_bar: prescribed displacement vector - in Eq. .
f_hat: given force vector - in Eq. .

plot_truss: string for output control: [‘yes’] to plot truss elements

plot_nod: string for output control: [‘yes’] to plot truss global node numbers
plot_stress: string for output control: [‘yes’] to plot stresses

12.1.1.2 Calculated or derived by program

neq: total number of equations
K: global stiffness matrix

18
d: global displacement vector is stored as:

for 1-D problems for 2-D problems

f: global force vector (excluding the reactions) is stored as:

for 1-D problems for 2-D problems

e: element number
ke: element stiffness matrix
de: element nodal displacement vector:

for 1-D problems for 2-D problems

=de = de

LM: gather matrix

The gather matrix is used to extract the element and local degrees-of-freedom. It
has the following structure:

global degree-of-freedom=LM (local degree-of-freedom, element number)

When ndof = 1 (see example in Figure 2.8) IEN and LM are defined as follows:

= IEN = LM

When ndof = 2 (example Problem 2.2), IEN and LM are defined as:

19
 = IEN = LM

In both examples, columns indicate the elements and rows indicate global degrees-of-
freedom.

K_E: partition of the global stiffness matrix K based on Eq.

K_EF: partition of the global stiffness matrix K based on Eq.
K_F: partition of the global stiffness matrix K based on Eq.
d_F: unknown (free) part of the global displacement vector d based on Eq.
d_E: prescribed (essential) part of the global displacement vector d based on
Eq.
f_E: reaction force (unknown) vector based on Eq.
stress: stress for each element

Remark: In this chapter nodes where the displacements are prescribed have to be
numbered first.

12.21.2 MATLAB Finite element code for trusses

truss.m
%%%%%%%%%%%%%%%%%%%%%%
% 2D Truss (Chapter 2) %
% Haim Waisman, Rensselaer %
%%%%%%%%%%%%%%%%%%%%%%
clear all;
close all;

% include global variables

include_flags;

% Preprocessor Phase
[K,f,d] = preprocessor;

% Calculation and assembly of element matrices

for e = 1:nel
ke = trusselem(e);
K = assembly(K,e,ke);
end

% Solution Phase
[d,f_E] = solvedr(K,f,d);

% Postprocessor Phase
postprocessor(d)

20
include_flags.m
% file to include global variables
global nsd ndof nnp nel nen neq nd
global CArea E leng phi
global plot_truss plot_nod plot_stress
global LM IEN x y stress

preprocessor.m
% preprocessing– read input data and set up mesh information
function [K,f,d] = preprocessor;
include_flags;

% input file to include all variables

input_file_example2_2;
%input_file_example2_8;

% generate LM array
for e = 1:nel
for j = 1:nen
for m = 1:ndof
ind = (j-1)*ndof + m;
LM(ind,e) = ndof*IEN(j,e) - ndof + m;
end
end
end

input_file_example2_2.m
% Input Data for Example 2.2
nsd = 2; % Number of space dimensions
ndof = 2; % Number of degrees-of-freedom per node
nnp = 3; % Number of nodal points
nel = 2; % Number of elements
nen = 2; % Number of element nodes

neq = ndof*nnp; % Number of equations

f = zeros(neq,1); % Initialize force vector

d = zeros(neq,1); % Initialize displacement matrix
K = zeros(neq); % Initialize stiffness matrix

% Element properties
CArea = [1 1 ]; % Elements area
leng = [1 sqrt(2)]; % Elements length
phi = [90 45 ]; % Angle
E = [1 1 ]; % Young’s Modulus

% prescribed displacements
% displacement d1x d1y d2x d2y
d = [0 0 0 0]';
nd = 4; % Number of prescribed displacement degrees-of-freedom

21
% prescribed forces
f(5) = 10; % Force at node 3 in the x-direction
f(6) = 0; % Force at node 3 in the y-direction

% output plots
plot_truss = 'yes';
plot_nod= 'yes';

% mesh Generation
truss_mesh_2_2;

truss_mesh_2_2.m
% geometry and connectivity for example 2.2
function truss_mesh_2_2
include_flags;

% Nodal coordinates (origin placed at node 2)

x = [1.0 0.0 1.0 ]; % x coordinate
y = [0.0 0.0 1.0 ]; % y coordinate

% connectivity array
IEN = [1 2
3 3];

% plot truss
plottruss;

input_file_example2_8.m
% Input Data from Chapter 2 Figure 2.8
nsd = 1; % Number of spatial dimensions
ndof = 1; % Number of degrees-of-freedom per node
nnp = 3; % Total number of global nodes
nel = 2; % Total number of elements
nen = 2; % Number of nodes in each element

neq = ndof*nnp; % Number of equations

f = zeros(neq,1); % Initialize force vector

d = zeros(neq,1); % Initialize displacement vector
K = zeros(neq); % Initialize stiffness matrix

% Element properties
CArea = [.5 1]; % Elements cross-sectional area
leng = [2 2]; % Elements length
E = [1 1]; % Young’s Modulus

% prescribed displacements
d(1) = 0;
nd = 1; % Number of prescribed displacement degrees of freedom

% prescribed forces

22
f(3) = 10; % force at node 3 in the x-direction

% output controls
plot_truss = 'yes';
plot_nod = 'yes';

% mesh generation
truss_mesh_2_8;

truss_mesh_2_8.m
% geometry and connectivity for example problem in Figure 2.8
function truss_mesh_2_8;
include_flags;

% Node coordinates (origin placed at node 1)

x = [0.0 1.0 2.0 ]; % x coordinate
y = [0.0 0.0 0.0 ]; % y coordinate

% connectivity array
IEN = [1 2
2 3];

% plot truss
plottruss;

Plottruss.m
% function to plot the elements, global node numbers and print mesh parameters
function plottruss;
include_flags;

% check if truss plot is requested

if strcmpi(plot_truss,'yes')==1;
for i = 1:nel
XX = [x(IEN(1,i)) x(IEN(2,i)) x(IEN(1,i)) ];
YY = [y(IEN(1,i)) y(IEN(2,i)) y(IEN(1,i)) ];
line(XX,YY);hold on;

% check if node numbering is requested

if strcmpi(plot_nod,'yes')==1;
text(XX(1),YY(1),sprintf('%0.5g',IEN(1,i)));
text(XX(2),YY(2),sprintf('%0.5g',IEN(2,i)));
end
end
title('Truss Plot');
end

% print mesh parameters

fprintf(1,'\tTruss Params \n');
fprintf(1,'No. of Elements %d \n',nel);
fprintf(1,'No. of Nodes %d \n',nnp);

23
fprintf(1,'No. of Equations %d \n\n',neq);

trusselem.m
% generate the element stiffness matrix for each element
function ke = trusselem(e)
include_flags;

const = CArea(e)*E(e)/leng(e); % constant coefficient within the truss element

if ndof == 1
ke = const * [1 -1 ; % 1-D stiffness
-1 1];
elseif ndof == 2
p = phi(e)*pi/180; % Converts degrees to radians

s = sin(p); c = cos(p);
s2 = s^2; c2 = c^2;

ke = const*[c2 c*s -c2 -c*s; % 2-D stiffness

c*s s2 -c*s -s2;
-c2 -c*s c2 c*s;
-c*s -s2 c*s s2];

end

assembly.m
% assemble element stiffness matrix
function K = assembly(K,e,ke)
include_flags;

for loop1 = 1:nen*ndof

i = LM(loop1,e);
for loop2 = 1:nen*ndof
j = LM(loop2,e);
K(i,j) = K(i,j) + ke(loop1,loop2);
end
end

solvedr.m
% partition and solve the system of equations
function [d,f_E] = solvedr(K,f,d)
include_flags;

% partition the matrix K, vectors f and d

K_E = K(1:nd,1:nd); % Extract K_E matrix
K_F = K(nd+1:neq,nd+1:neq); % Extract K_E matrix
K_EF = K(1:nd,nd+1:neq); % Extract K_EF matrix
f_F = f(nd+1:neq); % Extract f_F vector
d_E = d(1:nd); % Extract d_E vector

24
% solve for d_F
d_F =K_F\( f_F - K_EF'* d_E);

% reconstruct the global displacement d

d = [d_E
d_F];

% compute the reaction r

f_E = K_E*d_E+K_EF*d_F;

% write to the workspace

solution_vector_d =d
reactions_vector = f_E

postprocessor.m
% postprocessing function
function postprocesser(d)
include_flags;

% prints the element numbers and corresponding stresses

fprintf(1,'element\t\t\tstress\n');
% compute stress vector
for e=1:nel
de = d(LM(:,e)); % displacement at the current element
const = E(e)/leng(e); % constant parameter within the element

if ndof == 1 % For 1-D truss element

stress(e) = const*([-1 1]*de);
end
if ndof == 2 % For 2-D truss element
p = phi(e)*pi/180; % Converts degrees to radians
c = cos(p); s = sin(p);
stress(e) = const*[-c -s c s]*de; % compute stresses
end

fprintf(1,'%d\t\t\t%f\n',e,stress(e));
end

12.3 Shape functions and Gauss quadrature with MATLAB

In Chapter 2 the basic finite element programming structure was introduced for
one- and two-dimensional analysis of truss structures. In this section we give the
functions for the construction of element shape functions in one-dimension and their
derivatives. The shape functions are defined in the physical coordinate system.

12.3.1 Notations and definitions

xe: element nodal x-coordinates
xt: x coordinate at which the functions are evaluated

25
N: array of shape functions
B: array of derivatives of the shape functions
gp: array of position of Gauss points in the parent element domain -

W: array of weights -

12.3.2 MATLAB code for shape functions and derivatives

Nmatrix1D.m
% shape functions computed in the physical coordinate - xt
function N = Nmatrix1D(xt,xe)
include_flags;

if nen == 2 % linear shape functions

N(1) = (xt-xe(2))/(xe(1)-xe(2));
N(2) = (xt-xe(1))/(xe(2)-xe(1));
elseif nen == 3 % quadratic shape functions
N(1)=(xt-xe(2))*(xt-xe(3))/((xe(1)-xe(2))*(xe(1)-xe(3)));
N(2)=(xt-xe(1))*(xt-xe(3))/((xe(2)-xe(1))*(xe(2)-xe(3)));
N(3)=(xt-xe(1))*(xt-xe(2))/((xe(3)-xe(1))*(xe(3)-xe(2)));
end

Bmatrix1D.m
% derivative of the shape functions computed in the physical coordinate - xt
function B = Bmatrix1D(xt,xe)
include_flags;

if nen == 2 % derivative of linear shape functions (constant)

B = 1/(xe(1)-xe(2))*[-1 1];
elseif nen == 3 % derivative of quadratic shape functions
B(1)=(2*xt-xe(2)-xe(3))/((xe(1)-xe(2))*(xe(1)-xe(3)));
B(2)=(2*xt-xe(1)-xe(3))/((xe(2)-xe(1))*(xe(2)-xe(3)));
B(3)=(2*xt-xe(1)-xe(2))/((xe(3)-xe(1))*(xe(3)-xe(2)));
end

12.3.3 MATLAB code for Gauss quadrature

gauss.m
% get gauss points in the parent element domain [-1, 1] and the corresponding weights
function [w,gp] = gauss(ngp)

if ngp == 1

26
 gp = 0;
w = 2;
elseif ngp == 2
gp = [-0.57735027, 0.57735027];
w = [1, 1];
elseif ngp == 3
gp = [-0.7745966692, 0.7745966692, 0.0];
w = [0.5555555556, 0.5555555556, 0.8888888889];
end

12.4 Finite element programming in 1D with MATLAB

In Section 12.2 the basic finite element programming structure was introduced
for one- and two- dimensional analysis of truss structures. In 12.3, the program functions
for the calculation of the element shape functions, their derivatives and Gauss quadrature
in one-dimension were introduced. In this section we introduce a more general finite
element program structure for one-dimensional problems that in principle is similar to
that in multidimensions to be developed in Sections 12.5 and 12.6 for heat conduction
and elasticity problems, respectively.
In Chapter 2 we discussed various methodologies for imposing boundary
conditions. In the partition-based approach, the so-called E-nodes (where displacements
are prescribed) are numbered first. In general, however, node and element numberings
are initially defined by mesh generators and subsequently renumbered to maximize
efficiency of solving a system of linear equations. In our implementation we tag nodes
located on the natural boundary or essential boundary. Nodes on a natural boundary are
assigned flag=1, while nodes on an essential boundary are tagged as flag=2.
Subsequently, nodes are renumbered by the program so that E-nodes are numbered first.
This is accomplished by constructing the ID and LM arrays in the function
setup_ID_LM. With some minor modifications the program for the one-dimensional
elasticity problems can be modified to analyze heat conduction problems.

Explanation for various MATLAB routines is given as comments within each function.

Only the nomenclature and definitions which have been modified from the previous
chapters are included below. Much of the code is either identical or very similar to the
code developed in Section 12.2. An input file for the Example 5.2 in Chapter 5 modeled
with two quadratic elements is given below. Additional input files for one quadratic
element mesh and four quadratic elements mesh are provided in the disk.

12.4.1 Notations and definitions

User provided
nd: number of nodes on the essential boundary (E-nodes)
ngp: number of Gauss points
body: vector of values of body forces – defined at the nodes and then interpolated using
shape functions

27
E: vector of nodal values of Young’s modulus
CArea: vector of nodal values of cross-sectional area
flags: Flag array denoting essential and natural boundary conditions

flags(Initial global node number) = flag value

Flag values are: 1 – natural boundary; 2 – essential boundary

x: vector of nodal x-coordinates

y: vector of nodal y-coordinates (used for the plots only)
e_bc: vector of essential boundary conditions (displacements or temperatures)
n_bc: vector of natural boundary conditions (tractions or boundary fluxes)
P: vector of point forces (point sources in heat conduction)
xp: vector of the x-coordinates where the point forces are applied
np: number of point forces (point sources in heat conduction)
nplot: number of points used to plot displacements and stresses (temperatures and fluxes
in heat conduction)
IEN: location matrix
The location matrix relates initial global node number and element local node
numbers. Subsequently nodes are renumbered (see setup_ID_LM.m) so that
E-nodes are numbered first. IEN matrix has the following structure:

Calculated by FE program:
ID: Destination array

LM: Location matrix

Note that LM matrix is related to IEN matrix by

12.4.2 MATLAB Finite element code for one-dimensional problems

bar1D.m
%%%%%%%%%%%%%%%%%%
% 1D FEM Program (Chapter 5) %
% Haim Waisman, Rensselaer %
%%%%%%%%%%%%%%%%%%
clear all;
close all;

28
% include global variables
include_flags;

% Preprocessing
[K,f,d] = preprocessor;

% Element matrix computations and assembly

for e = 1:nel
[ke,fe] = barelem(e);
[K, f] = assembly(K,f,e,ke,fe);
end

% Add nodal boundary force vector

f = NaturalBC(f);

% Partition and solution

[d,f_E] = solvedr(K,f,d);

% Postprocessing
postprocessor(d);

% plot the exact solution

ExactSolution;

include_flags.m
% Include global variables
global nsd ndof nnp nel nen neq nd CArea E
global flags ID IEN LM body x y
global xp P ngp xplot n_bc e_bc np
global plot_bar plot_nod nplot

preprocessor.m
% preprocessing– reads input data and sets up mesh information
function [K,f,d] = preprocessor;
include_flags;

% input file to include all variables

input_file5_2_2ele;
%input_file5_2_1ele;
%input_file5_2_4ele;

% generate LM and ID arrays

d = setup_ID_LM(d);

input_file5_2_2ele.m
% Input Data for Example 5.2 (2 elements)

nsd = 1; % number of space dimensions

ndof = 1; % number of degrees-of-freedom per node

29
nnp = 5; % number of nodal points
nel = 2; % number of elements
nen = 3; % number of element nodes

neq = ndof*nnp; % number of equations

f = zeros(neq,1); % initialize nodal force vector

d = zeros(neq,1); % initialize nodal displacement vector
K = zeros(neq); % initialize stiffness matrix

flags = zeros(neq,1); % initialize flag vector

e_bc = zeros(neq,1); % initialize vector of essential boundary condition
n_bc = zeros(neq,1); % initialize vector of natural boundary condition

% element and material data (given at the element nodes)

E = 8*ones(nnp,1); % nodal values Young's modulus
body = 8*ones(nnp,1); % nodal values body forces
CArea = [4 7 10 11 12]'; % nodal values of cross-sectional area

% gauss integration
ngp = 2; % number of gauss points

% essential boundary conditions

flags(1) = 2; % flags to mark nodes located on the essential boundary
e_bc(1) = 0; % value of essential B.C
nd = 1; % number of nodes on the essential boundary

% natural boundary conditions

flags(5) = 1; % flags to mark nodes located on the natural boundary
n_bc(5) = 0; % value of natural B.C

% point forces
P = 24; % array of point forces
xp = 5; % array of coordinates where point forces are applied
np = 1; % number of point forces

% output plots
plot_bar = 'yes';
plot_nod = 'yes';
nplot = nnp*10; % number of points in the element to plot displacements and stresses

% mesh generation
bar_mesh5_2_2ele;

bar_mesh5_2_2ele.m
function bar_mesh5_2_2ele
include_flags;

% Node: 1 2 3 4 5
x = [2.0 3.5 5.0 5.5 6.0 ]; % x coordinate
y = 2*x; % y is used only for the bar plot

% connectivity array

30
IEN = [1 3
2 4
3 5];
plotbar;

setup_ID_LM.m
% setup ID and LM arrays
function d = setup_ID_LM(d);
include_flags;

count = 0; count1 = 0;
for i = 1:neq
if flags(i) == 2 % check if essential boundary
count = count + 1;
ID(i) = count; % number first the nodes on essential boundary
d(count)= e_bc(i); % store the reordered values of essential B.C
else
count1 = count1 + 1;
ID(i) = nd + count1;
end
end

for i = 1:nel
for j = 1:nen
LM(j,i)=ID(IEN(j,i)); % create the LM matrix
end
end

barelem.m
% generate element stiffness matrix and element nodal body force vector
function [ke, fe] = barelem(e);
include_flags;

IENe = IEN(:,e); % extract local connectivity information

xe = x(IENe); % extract element x coordinates
J = (xe(nen) - xe(1))/2; % compute Jacobian
[w , gp] = gauss(ngp); % extract Gauss points and weights

ke = zeros(nen,nen); % initialize element stiffness matrix

fe = zeros(nen,1); % initialize element nodal force vector

for i = 1:ngp
xt = 0.5*(xe(1)+xe(nen))+J*gp(i); % Compute Gauss points in physical coordinates

N = Nmatrix1D(xt,xe); % shape functions matrix

B = Bmatrix1D(xt,xe); % derivative of shape functions matrix

Ae = N*CArea(IENe); % cross-sectional area at element gauss points

Ee = N*E(IENe); % Young's modulus at element gauss points
be = N*body(IENe); % body forces at element gauss points
ke = ke + w(i)*(B'*Ae*Ee*B); % compute element stiffness matrix

31
 fe = fe + w(i)*N'*be; % compute element nodal body force vector
end
ke = J*ke;
fe = J*fe;

% check for point forces in this element

for i=1:np % loop over all point forces
Pi = P(i); % extract point force
xpi = xp(i); % extract the location of point force within an element
if xe(1)<=xpi & xpi<xe(nen)
fe = fe + Pi*[Nmatrix1D(xpi,xe)]'; % add to the nodal force vector
end
end

assembly.m
% assemble element stiffness matrix and nodal force vector
function [K,f] = assembly(K,f,e,ke,fe)
include_flags;

for loop1 = 1:nen

i = LM(loop1,e);
f(i) = f(i) + fe(loop1); % assemble nodal force vector
for loop2 = 1:nen
j = LM(loop2,e);
K(i,j) = K(i,j) + ke(loop1,loop2); % assemble stiffness matrix
end
end

naturalBC.m
% compute and assemble nodal boundary force vector
function f = naturalBC(f);
include_flags;

for i = 1:nnp
if flags(i) == 1
node = ID(i);
f(node) = f(node) + CArea(node)*n_bc(node);
end
end

postprocessor.m
% postprocessing
function postprocessor(d)
include_flags;

fprintf(1,'\n Print stresses at the Gauss points \n')

fprintf(1,'Element\t\t x(gauss1) \t\t x(gauss2) \t\t stress(gauss1) \t\t stress(gauss2)\n')
fprintf(1,'--------------------------------------------------------------------------------- \n')

32
% loop over elements to compute the stresses
for e = 1:nel
% compute stresses and displacements for the current element
disp_and_stress(e,d);
end

33