Finite Element Method

Introduction, 1D heat conduction

Siemens Wind Power A/S
External lecturer, Ph.D. Jesper Winther Stærdahl

1
 Finite Element Method
Introduction, 1D heat conduction

Post doc. Johan Clausen

2
 Finite Element Method
Introduction, 1D heat conduction
Lecture plan Finite Element Method I
wind.civil.aau.dk/lecture/7sem_finite_element/Finite_element.htm

3
 Finite Element Method
Introduction, 1D heat conduction
Form and expectations
 To give the participants an understanding of the basic elements
of the finite element method as a tool for finding approximate
solutions of linear boundary value problems. These will be
exemplified with examples within stationary heat conduction.
After the course the participants should be able to set up their
own finite element solution of linear boundary value problems.
 We expect that you participate in the given lectures and
exercises and ask questions when something is unclear.
 Many of the exercises are programming problems, which
should be solved on a laptop. Preferably you should work in
groups of two. Each group should bring a laptop with MatLab
installed.
 We expect minor prior knowledge of scientific programming.
Basic MatLab programming will be repeated during the lectures
 You should bring pen and paper, and a copy of the lecture
notes.

4
 Finite Element Method
Introduction, 1D heat conduction

Choices we have made

 The lecture material is mainly slides. The referenced book is
additional information. The book is not a beginners book, but can be
used for additional information and also for advanced Finite Element
problems. We believe the basics can be understood from the slides
and attending the lectures.
 You will be programming into a rather extensive matlab toolbox
developed by the lecturers. I.e. the overview might be lost in the
beginning but you will have a working program, which also will be
used in later courses. You will program a simple Finite Element
program from nothing during the final lectures.
 We do everything in matlab. You will have your own program with
the source code and all the problems this gives. We believe that in
order to understand the Finite Element method you need to do some
programming instead of just using commercial solvers
 4 full lectures including exercise time and 1 self study, instead of 5
short lectures without exercises

5
 Finite Element Method
Introduction, 1D heat conduction

Literature

 Lectures will be given from the slides.

The theory and derivations are not
directly based on the chapters in the
book
 The book is chosen as a good overall
Finite Element book which can be used
also for more advanced problems
 During the lectures references will be
given to relevant chapters and pages
where the present subject is presented
with Cook in parenthesis. E.g. (Cook:
chapter 4). These pages are summarized
in the lecture plan.
 Relatively cheap!

6
 Finite Element Method
Introduction, 1D heat conduction

Beginners literature

 The formulations are based on this book,

however, it is expensive and does not
cover more advanced issues.
 What is covered in this book is essential
what we cover during the lectures and on
the slides.
 References to this book will be
referenced in parenthesis with OP. E.g.
(OP: chapter 4)

7
 Finite Element Method
Introduction, 1D heat conduction

What is the Finite Element Method (FEM)?

 A nummerical approach for solving partial differential equation,

boundary value problems
 Finite element method
 Finite difference method
 Finite volume method
 Boundary element method
 An approxmative solution.
 Simplification of geometry
 Mesh dependent
 Convergence issues
 Widely used within structural analysis
 deformation
 heat conduction

8
 Finite Element Method
Introduction, 1D heat conduction

Why use FEM?

 General approach for solving complex structures.

 Static undetermined systems causes no problems
 unsteady problems can be solved (not included in this course)
 Few basic building blocks (elements) can simulate complex
problems
 Very suiteable for computer solution.
 Fast solutions
 Many different load cases

9
 Finite Element Method
Introduction, 1D heat conduction
Basic steps of the finite-element method (FEM)
1. Establish strong formulation
 Partial differential equation
2. Establish weak formulation
 Multiply with arbitrary field and integrate over element
3. Discretize over space
 Mesh generation
4. Select shape and weight functions
 Galerkin method
5. Compute element stiffness matrix
 Local and global system
6. Assemble global system stiffness matrix
7. Apply nodal boundary conditions
 temperature/flux/forces/forced displacements
8. Solve global system of equations
 Solve for nodal values of the primary variables
(displacements/temperature)
9. Compute temperature/stresses/strains etc. within the element
 Using nodal values and shape functions

10
 Finite Element Method
Introduction, 1D heat conduction
MatLab FE-program
 main.m (main program runs until "return" plot functions located at the
bottom)
 Coordinates.m (defines node coordinates, done by user)
 Topology.m (defines the element topology, done by user)
 BoundaryConditions.m (defines the BC, done by user)
 calc_globdof.m (determines the global dof numbering, done by program)
 Assemblering.m (assemble the global stiffness matrix, done by program)
 solvequations.m (solve the system of equations for primary dof and forces,
done by program)
 Plot functions
 Visualize1Dheat.m (visualize the 1D heat problem with a temperature curve)
 Visualize2D (visualize all 2D problems including 2D heat conduction and 2D
structural elements)
 Visualize3Dbeam_struct.m (visualize the 3D beam problem with geometry,
deformation, beam normals, node and dof numbering)
 Visualize3Dbeam_struct_secForce.m (visualize the 3D beam problem with
geometry, deformation, beam normals, node numbering and section forces)

11
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Test that the program can run.

 Start main.m
 Run the plotting of 1D heat conduction
 Plotting functions are at the bottom of the program

 Output:
 u = [0 2]T
 K = [1 0 ; 0 1]

 Hints:
 Open the m-file in the editor F5 runs the program until it reaches a "return"-
command or a red circle next to the line numbers (debug mode)
 To debug use F11 to advance one line (will enter sub-functions). Use F10 to
advance one line without entering sub-functions
 F9 runs the marked section
 Enable cell-mode. Then ctrl-enter will run the present cell. Cells are divided
by %%

12
 Finite Element Method
Introduction, 1D heat conduction

Advanced plotting in MatLab using handles

 When a plot is generated in matlab corresponding handles are

created.
 a handle for the figure
 a handle for the axis
 a handle for each plot on the figure
 In this handle every information about the plot is defined
 Direct handles are created by putting a variable before the plot
>> hp = plot([0 1],[0 1])
 The information in the handle can be accessed by
>> get(hp)

 The information can be changed by

>> set(hp,'linewidth',3,'marker','o')
new value, is put between ' '
information is put between ' ' if it is none numerical
13
 Finite Element Method
Introduction, 1D heat conduction
 Handle for figures and axes can be accessed either by putting a
handle when generating them or by gcf (get-current-figure) or gca
(get-current-axis) as handles
>> hf = figure >> ha = axes
>> get(gcf) >> get(gca)
>> get(hf) >> get(ha)
 Changing values works as for the plot
>> set(hf,'color',[1 0 0])
>> set(ha,'fontsize',22)
 Position information, remember correct units
>> set(hf,'units','centimeters','postion',[5 5 10 10])
 position = [lower_left_corner_x lower_left_corner_y width height]
 Activating a figure for adding axes or axes for plotting
>> figure(hf) >> axes(hf)

14
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Make an advanced plot using handles

 figure 1 units in centimeters postion lower left corner (5,5) size (10,10)
 axis units in centimeters, position lower left corner (1,1) size (8,3)
 xcolor and ycolor is red, figure color is black [r,g,b]=[0,0,0]
 figure 2 units in centimeters postion lower left corner (15,5) size
(10,10)
 axis units in centimeters, position lower left corner (1,6) size (8,3)
 xcolor and ycolor is red, figure color is black [r,g,b]=[0,0,0]
 Hint
 you need to change units,
position and color of the figure-
handles
 you need to change the units,
position, xcolor and ycolor of the
axis-handle

15
 Finite Element Method
Introduction, 1D heat conduction

Make video using handles

 The coordinates for the part that moves are changed through the handle
(xdata,ydata) and a 'drawnow' command is made for the plot to be
updated
figure
hold on
x = 0:0.001:5;
y = sin(x);

hp1 = plot(x,y);
hp2 = plot(x(1),y(1),'ro','markersize',15);

for j=1:size(x,2)
set(hp2,'xdata',x(j),'ydata',y(j));
drawnow
end

16
 Finite Element Method
Introduction, 1D heat conduction
 Generating an avi-file (only every 100th frame is captured)
p=1;
for j= 1:100:size(x,2);
set(hp2,'xdata',x(j),'ydata',y(j));
drawnow; compression method: none,
Mov(p) = getframe(gcf); divx, xvid
p=p+1;
end
movie2avi(Mov,'test2.avi','Quality',100,'compression','none');

 Problem:
 Powerpoint cannot play the movie directly
 Solution:
 re-code using e.g. virtualdub
 www.virtualdub.org

17
 Finite Element Method
Introduction, 1D heat conduction

Plotting many lines in the same plot

 Every plot generates many informations, i.e. many plots become

heavy
 Often a FEM solution is plotted using many line pieces because we
don't know the geometry in advance. Hence, every element needs to
be plotted separately
 One line, one handle independent of the number of points
 By introducing NaN (Not a Number) in the coordinates the line will
be broken but still work as one line (i.e. one handle)

18
 Finite Element Method
Introduction, 1D heat conduction
 Try the following
figure
hold on
hp1=plot([0 0],[0 1])
hp2=plot([1 1],[0 1])
hp3=plot([-0.2 1.4],[0.2 0.4])

figure
hold on
hp4=plot([0 0 1 1 -0.2 1.4],[0 1 0 1 0.2 0.4])

figure
hold on
hp4=plot([0 0 NaN 1 1 NaN -0.2 1.4],[0 1 NaN 0 1 NaN 0.2 0.4])
 Try to mark the lines on the three figures
 figure 1: The lines can be marked separately, i.e. changed separately
 figure 2: One line is drawn and can only be changed as one, but this is
not what we want. We don't want the connecting lines
 figure 3: The lines are drawn separately but are marked as one, i.e. one
handle
19
 Finite Element Method
Introduction, 1D heat conduction

Call a function given in a string variable (used in Assembling.m)

 Create 2 files cube.m and square.m
 cube.m
function out=cube(x)
out = x^3;
 square.m
function out=square(x)
out = x^2;

 Now try
FuncName = ‘cube’
Func = str2func(FuncName);
Output = Func(5)
 Change the first line and run the second and third again
FuncName = ‘square’

 We can call many different functions by changing a string variable

20
 Finite Element Method
Introduction, 1D heat conduction
Ways to store data in matlab
 Matrices: use square brackets [] index in parenthesis af matrix name
e.g. A(row column) columns are separated by “ “ rows by “new line”
or “;”
 Advantages: when the same type and number of data should be stored,
can count over index. Should ALWAYS be initialised (A = zeros(2,2);)
A = [3 4 ; 5 6]
A(2,1) = ?
 Cells: use curly brackets {}
 Advantages: when the same type of data are stored but the number of
data varies. Can count over index in both curly and squared brackets
B = [3 4 5; 5 6 7]
C{1} = A; C{2} = B;
C{2}(2,1) = ?
 Structures:use “.” in name
 Advantages: When data size and type varies. You can call the data by
name instead of just by index
S.A = A; S.B = B; S.C = C; S.info = ‘test’
S.info(3) = ?

21
 Finite Element Method
Introduction, 1D heat conduction
Example problem 1D stationary heat conduction (Cook:
p21-22), (OP chapter 9)

 Constant area, A, thermal conductivity, k, and heat supply Q

 Boundary a: constant temperature T
 Boundary b: constant flux q

22
 Finite Element Method
Introduction, 1D heat conduction
Step 1: Establish strong formulation for 1D heat conduction
(OP: Chapter 4, p48-52)

23
 Finite Element Method
Introduction, 1D heat conduction
 Balance or conservation equation
 Heat inflow equal heat outflow
dH
H + Qdx = H + dH ) =Q
dx
 By definition

H = Aq
 Material property or constitutive relation

 one-dimensional heat equation

24
 Finite Element Method
Introduction, 1D heat conduction
 Second order differential equation needs two boundary conditions

 Possible boundary conditions: temperature or temperature gradient

(flux)

 This is the strong formulation for stationary 1D heat conduction

 Constant A, k ,Q with T(a)=Ta , q(b)=qb , analytical solution can be
found (Differential equations and nummerical methods 5th semester,
Zill&Cullen)

25
 Finite Element Method
Introduction, 1D heat conduction
Step 2: Establish weak formulation
(OP: chapter 4, p56-62)

 Strong form

 Multiply with an arbitrary function v(x) (weight function) and integrate

over the pertinent region

26
 Finite Element Method
Introduction, 1D heat conduction
 Use integration by parts of the first term to obtain the same
derivative of the weight function and primary variable T

 Weak formulation of 1D stationary heat conduction

boundary conditions
distributed load
 v is not completly arbitrary since the above derivation should hold (i.e. v
should be ones differentiable and defined in the region of integration)

27
 Finite Element Method
Introduction, 1D heat conduction

Why are we interested in the weak form?

 The FE-method is based on the weak form

 The FE-method is an approximate method, i.e. in the strong form we
need an approximation to the second derivative of the temperature,
where only the first direvative is needed in the weak form at the cost
of introducing an arbitrary field.
 The weak form holds when discontinuities occur, the strong form
requires a modification (see OP: p60-62)

28
 Finite Element Method
Introduction, 1D heat conduction
Step 3: Discretize over space
 Original problem

 Discretized problem. Define: nodes, unknown (degree-of-freedom dof)

numbering, element numbering, element topology (which nodes define
the element)
dof

Nodes

Node number
coordinate Elements

29
 Finite Element Method
Introduction, 1D heat conduction
Step 4: Select shape and weight functions
(OP: chapter 7, p90-94, 98-106)

 The finite element is an approximative method

 The discrete values Ti of the unknowns are evaluated at the nodes
 The continuous field T(x) is interpolated using shape functions.

30
 Finite Element Method
Introduction, 1D heat conduction

Approximation requirements

 Convergence criteria
 In the limit when the elements are infinitely small, the approximation
should be infinitely close to the exact solution
 Completeness requirements
 The approximation must be able to represent an arbitrary constant
temperature gradient
 The approximation must be able to represent an arbitrary constant
temperature
 i.e. the approximation of the temperature should include terms
corresponding to a linear polynomial
 Compatibility or conforming requirement
 The approximation of the temperature over element boundaries must be
continues
 convergence = completeness + compatibility

31
 Finite Element Method
Introduction, 1D heat conduction
 Assuming nodal values to be known
 Linear variation of temperature allows a
constant temperature gradient
 Simplest one-dimensional element (OP:
p98-99)

 Test

 Matrix notation
nodal values (dof)

shape functions
32
 Finite Element Method
Introduction, 1D heat conduction
 Weight functions (see OP: chapter 8,p142-156)
arbitrary field

weight functions arbitrary unknown

 FE-method uses the Galerkin method (OP: p152-153)
 weight functions = shape functions

 Transpose of a scalar field equals the field

33
 Finite Element Method
Introduction, 1D heat conduction

Step 5: Compute element stiffness matrix

 If the weak formulation holds for the entire field, it also holds for part
of the field, i.e. integration is done over one element

 Insert the temperature field and arbitrary field into the weak
formulation
 a and c are constants, i.e. they can be taken outside the integrals

 cT cancels from the equation!!!!

34
 Finite Element Method
Introduction, 1D heat conduction
 In compact form

35
 Finite Element Method
Introduction, 1D heat conduction
Exercise: Compute by hand the element stiffness matrix
and force vector

 Assume constant A, k, Q
 q(xi) = qi, q(xj) = qj
 Need the derivative of the shape functions

36
 Finite Element Method
Introduction, 1D heat conduction

Local and global coordinates (OP: p191-193)

 Integration and shape functions are defined in a local coordinate

system, i.e. along the beam axis
 Often we wish to normalize the integration. In this case it is not
nessesary but for the case of argument and introduction to 2D and
numerical integration.

Global coordinate system

Local coordinate system

37
 Finite Element Method
Introduction, 1D heat conduction
 Transformation between coordinate systems

 Test

38
 Finite Element Method
Introduction, 1D heat conduction
 Integration in local coordinate system

 constant A and k

39
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Program the shape functions

 make a function shape_1d_heat.m

function [N,B] = shape_1d_heat(x,eCoord)

% Shape functions for 1D heat element

% INPUT
% x = coordinate where N and B is evaluated [1x1]
% eCoord = [x1 y1 z1 ; x2 y2 z2] Nodal coordinates
%
% OUTPUT
% N = shape functions evaluated at coordinate x [1x2]
% B = derivative of shape functions evaluated at coordinate x [1x2]

40
 Finite Element Method
Introduction, 1D heat conduction
step 6: Assemble global system stiffness matrix
(OP: p184-191)

 Global system of equations

 The global system stiffness matrix [ndof x ndof] is assembled from

all the element stiffness matrices [2x2] according to the global
numbering of the degrees-of-freedom.

41
 Finite Element Method
Introduction, 1D heat conduction
 Example: 4 dofs, 3 elements

From node number the dof-number is found

From node coordinates the length is defined

 Need information (input) about
 node coordinates
 element definition or topology (which nodes are used, section (A) and
material (k))

42
 Finite Element Method
Introduction, 1D heat conduction
 Local dof numbering vs. global dof numbering

Global numbering

Local numbering

 Local system of equations, see exercise slide 36

43
 Finite Element Method
Introduction, 1D heat conduction
 Global system of equations
 In a linear system we can add stiffnesses and loads (in this case
heat supply)

 Element 1: Global numbering

44
 Finite Element Method
Introduction, 1D heat conduction
 Element 2: Global numbering

 Element 3: Global numbering

45
 Finite Element Method
Introduction, 1D heat conduction
 The assembling procedure is as follows

1. Determine the local stiffness matrix

2. determine the global number of dof corresponding to the local dof for
the element
3. add the components of the local stiffness matrix to the rows and
columns of the global stiffness matrix corresponding to the global
dof numbers
4. repeat 1-3 until all contributions from all elements have been added.

 In MatLab this is done in assemblering.m

K(gDof,gDof) = K(gDof,gDof) + Ke;

46
 Finite Element Method
Introduction, 1D heat conduction
 In order to be able to create a global system matrix we need to give
information about
 Material for each element
 Section dimensions for each element
 Coordinates for each node (and a numbering)
 Topology for each element (which nodes are in the element)
 dof numbering is given from the node numbering and number of dofs per node
(one in this case)
 See calc_globdof.m for numbering of global dof
 GlobDof = [nDof1 GDof1 GDof2 ...]
 dof numbering of each node
 nDof = total number of Dof

47
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Define the following problem in the program

 Discretize into 3 elements

 change coordinates.m, topology.m, material and section definitions
 Explain how the structure of section and material is defined
 plot the solution

48
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Enter the correct stiffness matrix into the program

 Look through assemblering.m and identify the steps in slide 46

 Used functions:
 elemInfo: provides element type,section number,Name of stiffness function
(see slide 20), coordinates for the element, global Dof numbering
 secType: provides the name of the section function giving the constitutive
relation
 Modify K_1D_heat.m
 Ke = K_1D_heat(eCoord,ConstRel)
 Input eCoord, ConstRel
 Output Ke [2x2]

 Output K [4x4]

49
 Finite Element Method
Introduction, 1D heat conduction
Step 7: Apply nodal boundary conditions
temperature/flux

 Why do we need boundary conditions?

 K is known from material and geometric input [ndof x ndof]

 a contains the unknown temperatures [ndof x 1]
 f contains the unknown loads (heat supply) [ndof x 1]
 i.e. 2 x ndof unknowns but only ndof equations
 We need to define ndof of the unknowns,
 We define either temperature or load in each node and solve for the
remaining unknowns, i.e. for the temperature where the load is
defined and for the load where the temperature is defined.

50
 Finite Element Method
Introduction, 1D heat conduction
 The force vector, see slide 45

 Q is the heat supply assumed constant, qi are the boundary nodal

fluxes

51
 Finite Element Method
Introduction, 1D heat conduction
Exercise: Determine the boundary conditions and enter
them into the program

 modify BoundaryCondition.m
 Type 1 is a temperature, type 2 is a heat
supply
 Calculate the load vector by hand and
enter nodal values

52
 Finite Element Method
Introduction, 1D heat conduction

Step 8: Solve global system of equations

 Global system with boundary conditions entered

 Left hand unknowns are called primary unknowns (temperature)

 First we identify the equations with primary unknowns (in this case
row 2-4)
 The defined temperatures are multiplied with their respective
columns of K and transferred to the right-hand side

53
 Finite Element Method
Introduction, 1D heat conduction
 The reduced system can be solved

 Finally, the unknown heat supplies can be determined from the

remaining equations
 The reduced solutions should enter the full solution, i.e. the
boundary conditions and solved temperature should be combined in
one vector for plotting.

 This is done in MatLab in solvequations.m

ured = Kred\Fred

54
 Finite Element Method
Introduction, 1D heat conduction

Step 9: Compute temperature/fluxes within elements

 Nodal values of dofs (temperature) are given from the solution of the
global system
 Internal values are determined using shape functions and nodal
values corresponding to the relevant element. See slide 32

 Fluxes are determined from the derivative of the temperature. See

slide 24.

 I.e. internal values are determined from nodal values multiplied with
shape functions evaluated at the respective coordinate

55
 Finite Element Method
Introduction, 1D heat conduction

Overall steps in MatLab FE-program

 Input
 section, material, node coordinates, topology, boundary conditions
 Generate global dof numbering
 Assemble global stiffness matrix
 loop over elements
 element section and material gives the constitutive relation
 Calculate element stiffness matrix from coordinates and constitutive relation
 global numbering of element dofs
 add element stiffness matrix to the system stiffness matrix according to the
global dof numbering
 Setup the global system of equations by introducing boundary
conditions and solve for dofs and forces
 Post processing
 determine internal fields (temperature/flux) from nodal values and shape
functions
 visualize results

56
 Finite Element Method
Introduction, 1D heat conduction

Exercise: Get familiar with the program

 Change boundary conditions
 Change topology
 Change material and sections
 try with different material and sections in the same analysis
 Change the discretization (element coordinates)
 should all be on the x-axis, i.e. y- and z-coordinate equal zero.
 Compare with analytical solution, slide 25.
 Determine the temperature variation of the following problem.
Discretize into 5 elements, for the area use the height at the middle
of the element (thickness=1.0 m)

x = [0.15 0.45 0.75 1.05 1.35]

h = [0.87 0.66 0.53 0.46 0.47]

57
 Finite Element Method
Introduction, 1D heat conduction

Thank you for your attention

58