FIRST COURSE in FINITE ELEMENTS

Jacob Fish and Ted Belytschko

First Course in Finite Elements

Preface
This book is intended as a textbook for junior and senior level undergraduate students in science and engineering. The text material evolved from over 50 years of combined teaching experience of the authors in undergraduate finite element courses. The book deals with a formulation and application of the finite element method and it differs from the related finite element textbooks in the following three respects: 1. It is introductory and self-contained. Only a modest background covered by most engineering and science curricula in the first two years is required. The relevant topics in mathematics, such as matrix algebra, differential and integral equations, and in physics, such as conservation laws and constitutive equations, are reviewed as needed. reinforced. 2. It is generic. While most introductory finite element textbooks are application specific, such as being based on energy principles for linear elasticity, the finite element method in this book is formulated as a general purpose numerical procedure for solving partial differential equations. Consequently, students from various engineering and science disciplines will benefit equally from the exposition of the subject. 3. It is a hands-on experience. The book integrates finite element theory, finite element code development and application of commercial software packages. Finite element code development is introduced through MATLAB exercises and a MATLAB program, whereas ABAQUS is chosen for demonstrating the use of the finite element method in the commercial arena. The material in the book can be covered in a single semester. The course material is organized in three chronological units one month each: (1) the finite element formulation for one-dimensional problems, (2) the finite element formulation for scalar field problems in two dimensions, (3) finite element programming and application to scalar field problems; and finite element formulation for vector field problems in two dimensions and beams. The following course outline is recommended: Part 1: Finite element formulation for trusses and one-dimensional problems Chapter 1: Introduction Chapter 2: Direct Approach Chapter 3: Strong and weak forms for one-dimensional problems Chapter 4: Trial and weight functions for one-dimensional problems

Chapter 5: Finite element formulation for one-dimensional problems ******* Quiz 1********** Part 2: Finite element formulation for scalar field problems in two dimensions Chapter 6: Strong and weak forms for scalar field problems in two dimensions Chapter 7: Trial and Weight Functions in two dimensions Chapter 8: Finite element formulation for two-dimensional heat flow *******Quiz 2********** Part 3: Finite element program programming and application to scalar field problems Chapter 11: Finite element programming with MATLAB. *********Project assignment***** Chapter 12: Application of general purpose finite element software ABAQUS tutorial for linear heat conduction and elasticity problems Chapter 9: Finite element formulation for vector field problems in twodimensions Chapter 10: Finite element formulation for beams *******Project Due**********

Chapter

1
1.1 Background

Introduction

Many physical phenomena in engineering and science can be described in terms of partial differential equations. In general, solving these equations by classical analytical methods for arbitrary shapes is almost impossible. The finite element method is a numerical approach by which these partial differential equations can be solved approximately. From an engineering standpoint, the finite element method is a method for solving engineering problems such as stress analysis, heat transfer, fluid flow and electromagnetics by computer simulation.

Figure 1.1: Geometry, loads and finite element meshes

Millions of engineers and scientists worldwide are using the Finite Element Method (FEM) to predict the behavior of structural, mechanical, thermal, electrical and chemical systems in both design and performance analysis. Its popularity can be gleaned by the fact that over $1 billion is spent annually in the United States on FEM software and computer time. More than 250 books and over 20,000 articles have been published on this subject. To explain the basic approach of the finite element method, consider a plate with a hole as shown in Figure 1.1. Suppose that our objective is to find the temperature distribution in the plate. In principle, it is possible to write a heat balance equation at each point in the plate. However, the solution of the partial differential equations that governs energy balance for a complicated geometry such as an engine block becomes almost impossible by classical methods like separation of variables. Numerical methods such as finite difference methods also are quite awkward for arbitrary shapes, software developers have not marketed finite difference programs that can deal with the arbitrary geometries that are commonplace in engineering. Similarly, in stress analysis problems, equilibrium is governed by complex partial differential equations that are very difficult to solve except for very simple geometries, like rectangles, and engineering problems seldom have such simple shapes. The basic idea behind the FE method is to divide the body or region into a finite number of finite elements, connected by nodes, and obtain an approximate solution in terms of the temperatures at these nodes. These elements may be one, two or threedimensional. A popular two-dimensional element is the triangular element. When a two dimensional region is divided into non-overlapping triangles, we can see that essentially any planar geometry can be easily represented by a union (in the language of mathematical sets) of these elements. Instead of determining the temperature at every point in the plate, let us consider determining the temperature at only a finite number of points, the vertices (or nodes) of the triangles. The finite element method provides a systematic methodology whereby the temperatures at the nodes (nodal temperatures) can be determined. Interpolation can then be used to obtain the temperature throughout the problem domain. For linear problems, we will need to solve a system of linear equations with the number of unknowns (nodal temperatures) equal to the number of nodes. To obtain a reasonably accurate solution, thousands of nodes are usually needed, so computers are essential in solving these equations. Generally, the accuracy of the solution improves as the number of elements (nodes) increases, but the computer time, and hence the cost, also increases. The computer program determines the temperature at each node and the heat flow through each element. The results are usually presented as computer visualizations, such as contour plots, although selected results are often printed. This information is then used in the engineering design process. Although we have just presented the concept in the context of heat transfer, the same basic concept is applicable in other areas. In stress analysis, the field variables are the displacements, in chemical systems the variables are material concentrations, and in electromagnetics, the potential field. This course is limited to linear finite element analysis. The preponderance of finite element analyses in engineering design is today still linear FEM. In heat conduction, linearity requires that the conductance be independent of temperature. In

stress analysis, linear FEM is only applicable if the material behavior is linear elastic and the displacements are small. Additional discussions of the assumptions underlying linear analysis are given later. The finite element method was developed in the 1950s in the aerospace industry. The major players were Boeing and Bell Aerospace (long vanished) in the U.S. and Rolls Royce in the U.K. M. J. Turner, R. W. Clough, H. C. Martin, and L. J. Topp published one of first papers that laid out the major ideas in 1956. It established the procedures of element matrix assembly and element formulations that you will learn in this course, but did not use the term finite elements. The second author in this paper, Ray Clough, was a professor at Berkeley, who was at Boeing for a summer job. Subsequently he wrote a paper that first used the name finite elements, and he is given much credit as one of the founders of the method. He only worked in finite elements for a few more years, and then turned to experimental methods, but his work instigated a tremendous effort at Berkeley, led by the younger professors, primarily E. Wilson and R.L. Taylor and graduate students such as T.J.R. Hughes, C. Felippa, and K.J. Bathe. These established Berkeley as the center of linear finite element research for many years. Their work coincided with the rapid growth of computer power, and the method quickly became widely used in the nuclear power, defense, automotive and aeronautics industries. Much of the academic community first viewed finite element methods very skeptically, and some journals refused to publish papers on FEM-the typical resistance of mankind (and particularly academic communities) to the new. Nevertheless, several capable researchers early recognized its potential, most notably O.C. Zienkiewicz (then at Northwestern) and R.H. Gallagher (at Cornell). O.C. Zienkiwiecz built a renowned group at Swansea in Wales that included B. Iron, R. Owen and many others who pioneered concepts like the isoparametric element and nonlinear analysis methds. Other important early contributors were J.H. Argyris and J.T. Oden. Subsequently mathematicians discovered a 1943 paper by R. Courant, in which he used triangular elements with variational principles to solve vibration problems. Consequently, many mathematicians have claimed that this was the original discovery of the method (though it is somewhat reminiscent of the claim that the Vikings discovered America instead of Columbus). It is interesting that for many years the finite element method lacked a theoretical basis, i.e. any logical underpinning that finite element solutions gave the right answer, and this original work of Courants was duplicated by the engineering community to provide a mathematical justification for the method. In the late 1960s, the field became of great interest to many mathematicians, who showed that for linear problems, such as we will deal with in this course, finite element solutions converge to the correct solution of the partial differential equation (provided that certain aspects of the problem are sufficiently smooth). In other words, it has been shown that as the number of elements is increased the solutions will improve and tend in the limit to the exact solution of the partial differential equations. The improvements in personal computers since the 1980s have had a major impact on the finite element applications, and hence engineering. Until about the early 1990s, personal computers were too feeble to handle the many equations needed for practical finite element analysis. However, around 1990, personal computers crossed that threshold, and PCs rapidly became capable of treating finite element analyses of linear problems required in engineering practice. The software that is most widely used by

engineers in stress analysis, such as NASTRAN, ANSYS, ABAQUS, and LSDYNA, were adapted to PCs. Today a large part of finite element analysis in industry is done by PCs, identical to yours.

1.2 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 ideas so that you can perform essential matrix operations and basic programming needed for understanding and development of the finite element program.

1.2.1 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

a11 a12 a21 a22 a = a m 1

a1n amn

Bold letters will denote matrices or vectors. The elements of a matrix a are denoted by aij , where i is the row number and j is the column number. Note that in both describing the dimension of the matrix and in column number, the row is always placed first. An example of a 3x3 matrix is: 1 2 a = 4 5 7 8 the subscripts identifying the row and 3 6 0

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.)

>> a = [1 2 3; 4 5 6; 7 8 0] a= 1 4 7 2 5 8 3 6 0

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 a can be determined from

size(a )
The transpose of any matrix is obtained by interchanging rows and columns. So for example, the transpose of a is: 1 4 7 T a = 2 5 8 3 6 0 In MATLAB the transpose of a matrix is denoted by an apostrophe (). Square matrices may be symmetric or diagonal. If aT = a we say a is symmetric. A matrix is called a column matrix or a vector if n=1, e.g. b 1 b = b2 b3 A column vector in MATLAB is defined as
>> b = [1 2 3]' b= 1 2 3

The components of the vector b are b1, b2 , b3 . The transpose of b is given by a row vector

bT = b1 b2 b3

or in MATLAB

>> b = [1 2 3] b= 1 2 3

A matrix is called a diagonal matrix if only the diagonal components are nonzero, i.e., aij = 0, i j . For example, the matrix below is a diagonal matrix:
1 0 0 a = 0 5 0 0 0 6

Diagonal matrix in MATLAB is constructed by first defining a row vector b = [1 5 6], then >> b = [1 5 6];
>> a = diag (b) a= 1 0 0 0 5 0 0 0 6

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, 2 2 identity matrix is given by

1 0 I = 0 1
The MATLAB expression for an order n unit matrix is

eye(n )
Thus, the MATLAB expression I = eye(2) gives the matrix above. 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 mn matrix B of zeros. A random mn matrix can be created by rand (m,n).

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

>> a = sparse (3,2) a= All zero sparse: 3-by-2 >> a(1,2)=1; >> a(3,1)=4; >> a(3,2)=-1; >> a a= (3,1) (1,2) (3,2) 4 1 -1

Notice that the display can be suppressed by ending the line with a semicolon. The inverse of a square matrix is defined such that

a 1a = aa 1 = I
provided that matrix a is not singular. The MATLAB expression is inv(a ) . The inverse can also be obtained using backslash operator as shown in Section 1.2.3. Matrix a is nonsingular if the determinant, denoted by det (a ) , is not equal to zero. A determinant of order 2 is defined as

a11 a12 a = a 22 a22

det (a ) = a11a22 a12a21

The MATLAB expression for determinant is

det(a )
For example,

10

>> a = [1 3; 4 2]; >> det (a) ans = -10

1.2.2 Operation with matrices

1.2.2.1 Addition and Subtraction

a11 b11 a12 b12 a b a22 b22 21 21 c=ab= am 1 bm 1

In example of matrix addition in MATLAB is given below:
>> a = [1 2 3;4 5 6;7 8 9]; >> a = [1 1 1;2 2 2;3 3 3]; >> c = [1 2;3 4;5 6]; >> a+b ans = 2 6 10 >> a+c ??? Error using ==> + 3 7 11 4 8 12

a1n b1n amn bmn

Matrix dimensions must agree

1.2.2.2 Multiplication 1. Scalar-matrix (vector) multiplication

a11 a12 a21 a22 ca = c a m 1

a1n ca11 ca12 ca21 ca22 = amn cam 1

11

ca1n camn

2. Scalar product

b1 b n T a a a 2 = a b= 1 2 aibi n i =1 bn In MATLAB the scalar product as defined above is either a * b or dot(a, b) .

The length of a vector a is denoted by |a| and is given by
2 2 a = a1 + a 2 + 2 + an

3. Product of two matrices The product of two matrices a (m k ) and b (k n ) is defined as

k a1 jbj 1 j =1 k a2 jbj 1 c = ab = j =1 k amjbj 1 j =1

a1jbj 2
j =1 k

a
j =1

2j j2

b j =1 k amjbjn j =1

1 j jn

The components of matrix c are given as

cij = aikbkj
k =1 n

The product of two matrices a and b can be taken only if the number of columns in a equals the number of rows in a. In other words, if a is an (m k ) matrix, then b must be an (k n ) 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 m n matrix. An important fact to remember is that matrix multiplication is not commutative, i.e. in general ab ba . The MATLAB expression for matrix multiplication is

12

c = a *b
Consider the same matrices a and c as before. An example of matrix multiplication with MATLAB is:
>> a*c ans = 22 49 >> c*c ??? Error using ==> * Inner matrix dimensions must agree. 28 64

76 100

4. Other matrix operations a) Transpose of product: (ab) = bT aT b) Product with identity matrix: aI = a c) Product with zero matrix: a0 = 0
T

1.2.3 Solution of system of linear equations

Consider the following system of n equations with n unknowns, dk , k = 1,2,

,n :

K 11d1 + K 2d2 + + K 1ndn = f1 K d + K d + + K d = f 21 1 22 2 2n n 2 K d + K d + + K d = f n1 1 n2 2 nn n n

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

Kd = f
where

13

 K 11 K 12 K 21 K 22 K= K n 1

K 1n K nn

f1 f 2 f= fn

d1 d 2 d= dn

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

d = K1f
MATLAB expression for solving the system of equations is

d =K \f
or

d = inv (K ) * f
An example of solution of system of equations with MATLAB is given below:
>> A = rand (3,3) A= 0.2190 0.0470 0.6789 b= 0.0535 0.5297 0.6711 >> x = A\b x= -159.3380 314.8625 -344.5078 0.6793 0.9347 0.3835 0.5194 0.8310 0.0346

>> b = rand (3,1)

14

Remarks
a) In practical FEM analysis, 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 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 system matrix for a finite element model K is non-singular (often called regular), i.e., K1 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. KT = K . K is positive definite, i.e., xT Kx > 0 x (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 y and the corresponding eigenvalues satisfying

Ky = y
The MATLAB expression for the eigenvalues problem is:

>> K=[2 -2;-2 4]; >> [y, lamda]=eig(K) y= 0.8507 -0.5257 -0.5257 lamda = 0.7639 0 0 5.2361 0.8507

15

1.3 Mathematical Preliminaries

1.3.1 Scalar Peoduct Theorem
Consider scalar product of two vetors w and p equal to zero for any vector w denoted as w wT p = 0 w or w1p1 + w2 p2 + + wn pn = 0 w . Since the vector component wi are arbitrary it is convenient to make the following choices:
w1 = 1 w2 = w 3 = = wn = 0 p1 = 0 w 2 = 1 w1 = w 3 = = w n = 0 p2 = 0 wn = 1 w1 = w 2 = = wn 1 = 0 pn = 0

which yields p = 0 .

1.3.2 Taylors formula with remainder and the mean value theorem
Taylors formula with remainder plays a key role in understanding the behavior of the finite element method discussed in Chapter 5.7. A function f (x ) defined on interval 0 x l can be expanded around a point 0 x 0 l as follows: 1 2 f (x ) = f (x 0 ) + (x x 0 ) f,x (x 0 ) + (x x 0 ) f,xx (x 0 ) 2 (1.1) 1 + (x x 0 )k f,x ...x (c) k! k times where c is some point in the interval 0 c x 0 . The mean value theorem is obtained as a special case for k=1
f (x ) = f (x 0 ) + (x x 0 ) f,x (c)

(1.2)

Consider a special case of x 0 = 0 and x = l then

f,x (c) =

f (l ) f (0) l

(1.3)

16

which means that for any differentiable function f (x ) there is a point in the interval 0 c l which has a slope defined by Eq. (1.3). See problem 1.3 for a specific choice of function f (x ) .

1.4 Programming with MATLAB

MATLAB is very convenient to write simple finite element programs written in its own programming language. 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'').

1.4.1 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 for index = start:increment:end statements end Examples >> for i=1:1:3 disp(i^2) end 1 4 9 >> x=1; >> while 1+x > 1 x = x/2; end

while expression statements end

17

>> x x= 1.1102e-16

1.4.2 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. Here is a simple example of a function; it computes the quadratic function f (x ) = x 2 3x 1 . 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):

function y = fcn( x ) y=x^2-3*x-1; Then type command: >> fcn(0.1) ans = -1.2900

Problems Chapter 1
Problem 1.1: Write a MATLAB program to generate a set of linear algebraic equations Ax=b where A is an n n matrix given by

-1 if i = j 1 or i = j + 1 2 if i = j Aij = 0 otherwise 1 1 and compute A . Then check how closely A A corresponds to I . Do this for n = 5,10,1000 and a larger value you choose. The accuracy of the computed product
can be compared to the correct results by computing a norm of the error given by

err =

1 n2

( B
n n i =1 j =1

ij

I ij )

Repeat the above with the matrix A defined by

Aij =

1 i+ j

18

Repeat the above with n = 3,4,5,6,7... . Stop when the error is greater than one, since the solution is then meaningless. The top equation is of the form we will see in finite element equations. They can be accurately solved even for very large system for they are not susceptible to roundoff error. They are known as well-conditioned. The second matrix is called a Hilbert matrix. It is an example of an estremely ill-conditioned matrix. Problem 1.2: Consider a system of linear equations

a) Write the system of equations in matrix notation Ax = b and solve for the unknown x using MATLAB b) Suppose we impose and additional constraint on the solution g (x) = x 1 + x 2 + x 1 = 0 . Using MATLAB find a new vector xnew so that it will satisfy exactly the constraint equation g (x
new

8x + x + 6x = 7.5 1 2 3 3x 1 + 5x 2 + 7x 3 = 4 4x 1 + 9x 2 + 2x 3 = 12

and will minimize the error

err (x

new

) = (Ax

new

b) (Ax

new

b)

Problem 1.3: Consider a cubic function f (x ) =

x 2 (3l 2x ) defined on the interval l3 0 x l and its linaer approximation f (x ) defined such that f (0) = f (0) and f (l ) = f (l ) . Verify Eq. (1.3) and find a point c, where f,x (c) = f,x

f(x)
1

f (x )

x l
Figure 1.2: Cubic function f(x) and its linear approximation g(x) Problem 1.4: Consider the following symmetrix matrices

k1 + k2 k2 ; k k2 2

k 0 0 k ;

k 0 1 k1 k1 + k2 k2 ; 0 k2 k2

K: k1

where k, k1, k2 are positive constants. a) Check if the above three matrices are positive definite. Recall that if for any vector

x 0 we have xT Kx > 0 then matrix K is called Symmetric Positive Definite

19

(SPD). If, on the other hand, x Kx 0 for any vector x 0 then the matrix K is symmetric semi-positive definite. Choose one of the semi-positive definite matrices shown above and show that for any right hand side vector, f , the system of equations Kd = f has no unique solution. b) Verify your results by computing the eigenvalues for the above three matrices

20