République du Cameroun

Republice of Cameroun

UUniversité de University of
Yaoundé I Yaounde I

FINITE ELEMENT METHODS

par

DR LEZIN

Ecole Nationale
Supérieure Polytechnique
FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN

SOMMAIRE

1. MATHEMATICAL BACKGROUND .............................................................................. 6

1.1. Introduction ................................................................................................................ 6
1.2. Vector Analysis .......................................................................................................... 6
1.3. Matrix Theory ............................................................................................................ 6
1.4. Differential Equations ................................................................................................ 7
1.5. Cartesian Tensors ....................................................................................................... 7
1.6. Problems..................................................................................................................... 8
1.6.1. problem1............................................................................................................. 8
Solution1 ................................................................................................................ 8
1.6.2. Problem2 ............................................................................................................ 8
Solution2 ................................................................................................................ 8
1.6.3. Problem3 ............................................................................................................ 9
Solution3 ................................................................................................................ 9
1.6.4. Problem4 ............................................................................................................ 9
Solution4 ................................................................................................................ 9
1.6.5. Problem5 .......................................................................................................... 10
Solution5 .............................................................................................................. 10
1.6.6. Problem6 .......................................................................................................... 12
Solution6 .............................................................................................................. 12
2. ONE-DIMENSIONAL FINITE ELEMENT *Programming and Matrix Opération*... 14
2.1. Introduction .............................................................................................................. 14
2.2. Engineering Equations ............................................................................................. 14
2.2.1. Elasticity........................................................................................................... 14
2.2.2. Heat conduction................................................................................................ 15
2.2.3. Potential flow ................................................................................................... 15
2.2.4. Mass – transport ............................................................................................... 16
2.3. Variational function.................................................................................................. 16
2.4. Interpolation function and shape functions .............................................................. 17
2.5. Stiffness matrix ........................................................................................................ 17
2.6. Boundary conditions ................................................................................................ 17
2.7. Problems in cylindrical coordinates ......................................................................... 18
2.8. The direct method..................................................................................................... 18
2.9. problems ................................................................................................................... 19
2.9.1. Problem 1: ........................................................................................................ 19
2.9.2. Problem 2 ......................................................................................................... 20
Solution 2 ............................................................................................................. 21
2.9.3. Problem 3 ......................................................................................................... 27
Solution 3 ............................................................................................................. 27

ENSP Page 2/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
2.9.4. Problem 4 ......................................................................................................... 28
Solution 4 ............................................................................................................. 28
2.10. Homework number 2:........................................................................................... 33
2.10.1. Homework 1..................................................................................................... 33
2.10.2. Homework 2..................................................................................................... 34
3. Two – dimensional Finite Element .................................................................................. 35
3.1. Introduction .............................................................................................................. 35
3.2. Two-dimensional boundary values problems. ......................................................... 35
3.3. Theory of elasticity................................................................................................... 35
3.4. Variational functions ................................................................................................ 37
3.5. Transformations ....................................................................................................... 38
3.6. Cylindrical coordinates .......................................................................................... 39
3.7. Problems................................................................................................................... 40
3.7.1. Problem1: ......................................................................................................... 40
Solution 1 ............................................................................................................. 40
3.7.2. Problem2 .......................................................................................................... 43
Solution2 .............................................................................................................. 43
3.7.3. Problem3 .......................................................................................................... 46
Solution 3 ............................................................................................................. 46
3.7.4. Problem4 .......................................................................................................... 50
Solution4 .............................................................................................................. 50
3.8. Homework number3................................................................................................. 52
3.8.1.................................................................................................................................. 52
3.8.2.................................................................................................................................. 53
3.8.3.................................................................................................................................. 53
3.8.4.................................................................................................................................. 53
4. BEAM and FRAME FE. (Plane)...................................................................................... 54
4.1. Introduction: ............................................................................................................. 54
4.2. Governing Differential Equation.............................................................................. 54
4.3. The Displacement Method for Beam Analysis ........................................................ 54
4.4. Beam Finite Elements: ............................................................................................. 55
4.5. Matrix transformation and Frame elements ............................................................. 55
4.6. Problems................................................................................................................... 56
4.6.1. Problem 1: ........................................................................................................ 56
4.6.2. Problem 2 ......................................................................................................... 56
4.6.3. Problem 4: ........................................................................................................ 57
Solution1 .............................................................................................................. 57
Solution2 .............................................................................................................. 59
Solution4 .............................................................................................................. 64
5. ISOPARAMETRIC FE .................................................................................................... 68
5.1. INTRODUCTION.................................................................................................... 68
5.2. Numerical Integration .............................................................................................. 69

ENSP Page 3/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
5.3. Shape Function Formulas......................................................................................... 69
5.4. ISOPARAMETRIC ELEMENT .............................................................................. 70
5.4.1. Triangular Element:.......................................................................................... 70
5.4.2. Rectangular Element ........................................................................................ 70
5.5. Axisymmetric Formulations..................................................................................... 71
5.6. Modelling considerations: ........................................................................................ 72
5.6.1. Element geometries .......................................................................................... 72
5.6.2. Mesh Generation; ............................................................................................. 73
5.6.3. Load Representation:........................................................................................ 73
5.7. problems ................................................................................................................... 74
5.7.1. Problem 1 ......................................................................................................... 74
5.7.2. Problem2 .......................................................................................................... 74
5.7.3. Problem 3 ......................................................................................................... 74
5.7.4. Problem 4: ........................................................................................................ 74
Solution1 .............................................................................................................. 74
Solution2 .............................................................................................................. 76
Solution3 .............................................................................................................. 77
Solution4 .............................................................................................................. 81
6. COMPUTER IMPLEMENTATION OF 2-DIMENSION PROBLEMS ........................ 83
6.1. INTRODUCTION.................................................................................................... 83
6.2. Assembly of Element Equations .............................................................................. 84
6.3. Geometry and characteristics. .................................................................................. 85
6.4. Types of structures and DOF ................................................................................... 85
6.5. Examples: ................................................................................................................. 86
7. SELECTED TOPICS IN FEM......................................................................................... 87
7.1. Introduction .............................................................................................................. 87
7.2. Eigenvalue Problems................................................................................................ 87
7.2.1. Background: ..................................................................................................... 87
7.2.2. Heat Transfer and Bar like Problems ............................................................... 88
7.3. Three-dimension F.E................................................................................................ 89
7.4. Plate F.E. .................................................................................................................. 90
7.4.1. Relation between Force and moment to Stress. ............................................... 90
7.4.2. Deformation assumptions................................................................................ 91
7.4.3. Stress-Strain and Moment-curvature relations................................................. 92
7.5. GALERKIN APPROXIMATION ........................................................................... 93
7.6. Problems................................................................................................................... 93
7.6.1. Problems N° 1 .................................................................................................. 93
Solution1 .............................................................................................................. 93
7.6.2. Problem 2: ........................................................................................................ 94
Solution 2 ............................................................................................................. 94
7.6.3. Problem 3: ........................................................................................................ 95
Solution 3 ............................................................................................................. 95
7.6.4. Problem 4: ........................................................................................................ 97

ENSP Page 4/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
Solution 4 ............................................................................................................. 97
7.6.5. Problem 5: ........................................................................................................ 98
Solution 5 ............................................................................................................. 98
7.6.6. Problem 6 ....................................................................................................... 100
Solution 6 ........................................................................................................... 100
7.6.7. Problem 7 ....................................................................................................... 101
Solution 7: .......................................................................................................... 101
7.6.8. Examples of structures and analysed with FEM ............................................ 103

ENSP Page 5/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN

1. MATHEMATICAL BACKGROUND
1.1. Introduction
The FEM is a numerical method dealing with real world problems that in volve complicated
physics, and geometry, and boundary conditions. In a FEM, a given domain is viewed as a
collection of sub domains, and over each sub domain, the governing equation is approximated
by any traditional methods.

SHEMA ICI

A given domain is subdivided in to sub domains, called finite elements, and an approximate
solution of the problem to solve is developed over eaels element. Three fundamental steps of
the FEM might be illustrated as:
¾ Divide the whole domain in to parts.
¾ Over each part. Seek an approximation to the solution as a linear combination
of nodal values and approximation function, and drive relations among nodal
values.
¾ Assemble the parts and obtain the solution to the whole.

1.2. Vector Analysis

A vector is defined as a physical quantity that can be described by a single magnitude and a
direction that is related to a coordinate reference frame:

SHEMA ICI

A vector a in the space can be given in terms of unit vectors i, j, k directed along the x, y, z
axis:
a = axi + ayj + azk (1-1)

The vector differential operator del ∇ is defined as:

∂ ∂ ∂
∇= i+ j + k (1-2)
∂x ∂y ∂z

The ∇ operator has vector properties and is used to define vector operations as the gradient,
the divergence and the curl useful in the definition of integral vector theorems.

1.3. Matrix Theory

[A] = [aij ] 1≤ i ≤ m
¾ An m x n matrix is given as the array of numbers: .
1≤ j≤ n
¾ The transpose of this matrix is [A]T = [a ji ], and it’s inverse as [A]-1.
The determinant of [A] is detA = A = det[A ] = ...

ENSP Page 6/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
The minor of a determinant is a determinant of M ij row i column j removed. The cofactor of
an element of a determinant C ij = (−1) i + j M ij , the adjoint matrix (transpose of the cofactor
matrix) are characteristic used to compute the inverse matrix.
¾ A set of simultaneous equations can be written in matrix form as
[A]{x} = {f } (1-3)

and in solution: [A]-1[A]{x}=[I]{x}={x}=[A]-1{f}

The use of inverse for solving a set of simultaneous equations is in efficient for large sets of
equations. A method called Gaussian Elimination is faster and more efficient than the
inverse method.

1.4. Differential Equations

Finite Element analysis is a method for the numerical solutions of a differential
equation. The fundamental differential equations in this course are of the form:
d d
α ( x)A(x) φ ( x) + C(x)A(x) = 0 (1-4)
dx dx
Where α(x) = material parameter, C(x) = external
source, A(x) = cross-sectional area.

These parameters being a function of x are allowed to change from element to element.
Eq (1-4) regardless of x is written in a more elementary form as:
d 2φ
α + C = 0 (1-5)
dx 2

Or in the most general form:

d 2φ ( x) dφ (x)
α 2
−β − γφ ( x) + C = 0 (1-6)
dx dx

The one dimensional counterpart of Eq (1-6) in cylindrical coordinate is:

α d ⎡ dφ (r) ⎤ dφ (r)
⎢ r ⎥ −β − γφ (r ) + C = 0 (1-7)
r dr ⎣ dr ⎦ dr

1.5. Cartesian Tensors

Cartesian Tensor (indicial or subscript tensor) notation is the simplified version of tensor
notation that can be referenced to any curvilinear coordinate system. Since only subscripts are
required for proper representation of physical equations, rather than use x, y, z, let the
coordinates be called x1, x2, x3 i.e (x, y, z ) a (x 1 , x 2 , x 3 ) . Any vector can be given as:

ENSP Page 7/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
f = f x i + f y j + f z k a f = f1i + f 2 j + f 3 k a f i (1-8)

i having the range 1, 2, 3.

1.6. Problems
1.6.1. problem1
A position vector emanates in space from P(10,15,5) and terminates at Q(-2,5,3).
1) - use the concept of addition and subtraction of vectors to determine the components
of the position vector and compute the magnitude of the position vector.
2) – define the unit vector and compute the components of a unit vector directed from P
to Q.

Solution1

SHEMA ICI

1- rP=10i + 15j + 5k, rQ=-2i + 5j + 3k, rQ=rP + rPQ ⇒ rPQ = rQ –rP, or rPQ = -12i – 10j – 2k
[
⇒ rPQ = 12 2 + 10 2 + 2 2 ] 1
2
= 157

[
2- A unit vector U is a vector with a unit magnitude, i.e. U 2x + U 2y + U 2z ]
1
2
=1

The position vector rPQ is defined in terms of a local coordinate system using angles θx,
θy, and θz.

SHEMA ICI

Hence:
rPQ x − 12
cosθ x = = = −0,762
rPQ 15,7
− 10
cosθ y = = −0,625
15,5
cosθ z = −0,127
⇒ U = -0,762i - 0,635j - 0,127k

1.6.2. Problem2
3
Given a scalar function φ defined as: φ = N 1φ1 + N 2φ 2 + N 3φ3 = ∑ N iφi write φ in
i =1
matrix form.

Solution2

ENSP Page 8/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
Define N i ⇒ [N ] = [N 1 N 2 N 3 ] and [φ ] = [φ1 φ 2 φ3 ]. as a result:

⎧φ1 ⎫
⎪ ⎪
φ = [N ]{φ } = [N ][φ ] = [N 1 N 2
T
N 3 ]⎨φ 2 ⎬ = [N 1 N 2 N 3 ][φ i ]
T

⎪φ ⎪
⎩ 3⎭

1.6.3. Problem3
Given a function y = ax2, where a = constant. Given that [A] is a symmetric n x n matrix
and {x} n x 1 matrix, a matrix equation that is equivalent to the function y above is

y = [x] T [A] {x}. Show that dy = [A ]{x} .

d{x}

Solution3
Assume n = 2, then
⎡a a 12 ⎤ ⎧x 1 ⎫
y = [x 1 x 2 ]⎢ 11 ⎥ ⎨ ⎬ = a 11 x 1 + a 12 x 2 x 1 + a 21 x 1 x 2 + a 22 x 2
2 2

⎣a 21 a 22 ⎦ ⎩x 2 ⎭
∂y
= 2a 11 x 1 + a 12 x 2 + a 21 x 2
∂x 1
∂y
= a 12 x 1 + a 21 x 1 + 2a 22 x 2 since [A ] is symetric a 12 = a 21 , then
∂x 2
⎡a a 12 ⎤ ⎧x 1 ⎫
⎨ ⎬ = 2[A ]{x}
dy = 2 ⎢ 11
d{x} ⎣a 21 a 22 ⎥⎦ ⎩x 2 ⎭

1.6.4. Problem4
The divergence of vector function is defined as ∇ a
1-Discuss this vector operation in terms of the definition of a scalar product.
2-Write the same function using Cartesian tensor notation.

Solution4
1- From equations 1-1 and 1-2, we have:
⎛ ∂ ∂ ∂ ⎞
∇a = ⎜⎜ i + j + k ⎟⎟.(a x i + a y j + a z k )
⎝ ∂x ∂y ∂z ⎠
∂a ∂a y ∂a z
= x + +
∂x ∂y ∂z
⇒ ∇.a ≠ a.∇ because the ∇ operateur should act vector a.
2-differentiation is denoted using a cumin a. A partial differentiation of a scalar quantity
A has vector properties:

ENSP Page 9/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
∂A ∂A ∂A
∇.A = i+ j+ k.
∂x ∂y ∂z the corresponding statement in subscript tensor notation is
A,i. with i = 1,2,3

∂A ⎛ ∂A ∂A ∂A ⎞
i.e. A ,i ⇒ ⇒ ⎜⎜ or or ⎟.
∂x i ⎝ ∂x 1 ∂x 2 ∂x 3 ⎟⎠

The divergence of a vector is written ∇ a => aij, where the repeated subscript implies
the summation:
3
∂a i ∂a 1 ∂a 2 ∂a 3
a i,i ⇒ ∑ = + + .
i =1 ∂x i ∂x 1 ∂x 2 ∂x 3

1.6.5. Problem5
Use the separation variable technique (u(x, t) = U(x).T(t)) to compute the steady-state
temperature distribution for present plate. a constant temperature of T0 = 100 is
maintained along the edge y = w, and all other edges have zero temperature. The
thermal conductivity kx = ky = 1.

SHEMA ICI

Solution5

ENSP Page 10/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
The problem is defined by Laplace’s equation
∂ 2T ∂ 2T
+ = 0 (1)
∂x 2 ∂y 2
With T(0,y) = T(x, 0) = 0 and T(x, w) = T0. Assume a solution that operation the depend
variables:
T(x, y) = X(x)Y(y) (2)
d2X d2Y
And substitute into equation (1) Y 2 + X 2 = 0 . Group the functions and equal to a
dx dy
constant:
1 d2X 1 d2Y
− 2
= 2
= k 2 (3) ⇒ (2) Ordinary diff. eq.
X dx Y dy

d2X d2Y
2
+ k 2
X = 0 and 2
− k 2 Y = 0 (4)
dx dy
Both equations are solved as homogeneous linear equations with constants coefficients.
Assume X = C emx for the firs equation, we obtain the characteristic equation m = ±ki.
⎛ nπ x nπ x ⎞ ⎛ nπ ⎞
The general solution for X is X = ∑ ⎜ A n sin( ) + B n cos( ) ⎟ (5) ⎜ k n = ⎟.
⎝ L L ⎠ ⎝ L ⎠
With boundary condition, X(0) = 0, X(L) = 0 => Bn = 0 and Ansin(knL) = 0.
A similar assumption Y = Cemy, gives a solutions of second equation with the
characteristic equation m2-k2 = 0 and general solution.

⎛ nπy nπ y ⎞
Y = ∑⎜ Cn sinh( ) + Dn cosh( )⎟ (6)
⎝ L L ⎠
With boundary condition Y(0) = 0 gives Dn = 0, and
eq. (2) becomes
+∞
nπx nπy
T = ∑ E n sin( )sinh( ) En = AnCn (7).
n =1 L L

The last boundary condition T(x, w) = T0 is substituted in to equation (7), and the
orthogonality of the trigonometric functions is employed to determine En multiply both
⎛ mπ x ⎞
if observed of the equation by sin ⎜ ⎟ and integrate from 0 to L.
⎝ L ⎠

ENSP Page 11/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
mπ x ⎛ nπ w ⎞ ⎛ nπ x ⎞ ⎛ mπ x ⎞
L +∞L

∫ T0 sin ( )dx = ∫ ∑ E n sinh ⎜ ⎟ sin ⎜ ⎟ sin ⎜ ⎟dx (8)

0
L 0 n =1 ⎝ L ⎠ ⎝ L ⎠ ⎝ L ⎠
L
mπ x ⎛nπ x⎞ ⎧⎪0 m≠n
it follows that ∫ sin ( ) sin ⎜ ⎟ dx = ⎨L
0
L ⎝ L ⎠ ⎪⎩ 2 m=n

nπ w nπ x
L
2
(8) ⇒ E n sinh( ) = ∫ T0 sin( )dx. (9)
L L0 L
Evaluating En and substituting in to eq.(7) gives the analytical result of the problem.

1.6.6. Problem6
Vector transformation in 2-dimension space for a vector f from the old system (x1, x2, x3)
to the new system (x’1, x’2, x’3) is
⎧f 1' ⎫ ⎡ cosθ sinθ 0⎤ ⎧f 1 ⎫
⎪ '⎪ ⎢ ⎪ ⎪
⎨f 2 ⎬ = ⎢- sinθ cosθ 0⎥⎥ ⎨f 2 ⎬ ,
⎪ '⎪ ⎢ 0 0 1⎥⎦ ⎪⎩f 3 ⎪⎭
⎩f 3 ⎭ ⎣
3
And the corresponding subscript tensor statement is f i' = ∑ a ij f j , while the inverse
j=1

transformation from the primate coordinate to the original is system is the

3
reverse f k = ∑ a jk f j .
j=1

If the inverse transformation seen as 2nd – order transformation is seen as a different 2nd
order tensor as an extension of the 1st one, show that the stress transformation equations for a
2-dimension elasticity, derived in elementary mechanics of materials are given by equation
σ’rs = ari asj σij (1)

Solution6

ENSP Page 12/103

FINITE ELEMENT METHODS / 1. MATHEMATICAL BACKGROUND Version ou da
Dr LEZIN
The 2-dimension stress tensor can be written as

⎛ a 11 a 12 ⎞ ⎛ cosθ sinθ ⎞
⎜⎜ ⎟⇒⎜ ⎟ (2)
⎝ a 21 a 22 ⎟⎠ ⎜⎝ - sinθ cosθ ⎟⎠

Expand eq.(1) with r = s = 1 and note that I and j are summation indices from 1 to 2.
==> σ’11 = a11a11 σ11+ a11a12 σ12 + a12a11 σ21 + a12 a12 σ22
Expand eq.(1) with r = 1 and s = 2
σ’12 = a11a21 σ11+ a11a22 σ12 + a12a21 σ21 + a12 a22 σ22
Expand eq.(1) with r = 2 and s = 1
σ’21 = a21a11 σ11+ a21a12 σ12 + a22a11 σ21 + a22 a12 σ22
Expand eq.(1) with r = s = 2
σ’22 = a21a21 σ11+ a21a22 σ12 + a22a21 σ21 + a22 a22 σ22
Substitute the values of aij from eq.(2):
σ’11 = σ11cos2 θ + σ22sin2 θ + 2 σ12sin θcos θ
σ’12 = σ’21 = (σ22 - σ11)sin θcos θ + σ12 (cos2 θ - sin2 θ)
σ’22 = σ11 sin2 θ + σ22 cos2 θ -2 σ12sin θcos θ

ENSP Page 13/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN

2. ONE-DIMENSIONAL FINITE ELEMENT *Programming and

Matrix Opération*
2.1. Introduction
In a FEM, a given domain is viewed as a collection of sub domains (elements) interconnected
at discrete points (nodes), and an approximate solution of the problem to solve is developed
over each element. Element are multidimensional, and one – dimension element may have
two or more nodes with linear, parabolic or cubic shape.

SCHEMA ICI

Global Model (representing a structure)

SCHEMA ICI

Local Model (element)

Local Element
Global element
Node 1 Node 2
I 1 2
Connectivity matrix
II 2 3
III 3 4
IV 4 5
V 5 6

2.2. Engineering Equations

Equations governing engineering phenomena are usually derived from a balance equation and
constituve equation. The equations and derived from a variational principle as a basis for
deriving the corresponding FE model.

2.2.1. Elasticity
If σ = normal stress, A = area and f an axial body force, then the force in the rod (one
dimensional element) is σ (x).A(x) and the change in the force is balanced by external body
force:
d[σ (x) A(x)]
+ f(x).A(x) = 0 (2-1)
dx

From the Hook’s law, the strain ∈ in a material with Young modulus E is related to axial
displacement U as

ENSP Page 14/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
du(x)
σ ( x) = E(x) ∈ (x) and ∈ ( x) =
dx
(2-2)
du(x)
σ ( x) = E(x)
dx

(2-1) + (2-2) ==> 2nd order differential equation in terms of displacement

d ⎡ du(x) ⎤
⎢ E(x) A(x) + f(x) A(x) = 0 (2-3)
dx ⎣ dx ⎥⎦

2.2.2. Heat conduction

The balance of energy states that the change in the heat flux q is balanced by an external heat
source Q:
d[q(x) A(x)]
= Q(x) A(x) (2-4)
dx
(convection heat transfer)
A(x) = area Q<0 ==> the system is giving out energy.
The constituve equation is known as

Fourier’s law is stated as

dT(x)
q(x) = -k(x) (2-5)
dx
T = temperature, k = thermal conductivity. (2-4) + (2-
5) ==> the governing 2nd order differential equation

⎡ dT(x) ⎤
d ⎢k(x).A(x) + Q(x)A(x) = 0 (2-6)
⎣ dx ⎥⎦

E E t°
q(x) in 2 ; A(x) in ; T(x) in ; k(x) in E
tL t L3 T tLT

2.2.3. Potential flow

Potential flow can be applied problems in groundwater movement. With the assumption of
steady incompressible flow, a potential function is postulated in one dimension assuming
constant area as

ENSP Page 15/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
⎛z+p⎞
φ ( x) = -K(x) h(x) = - K(x) ⎜⎜ ⎟⎟
⎝ γ ⎠ (2-7-a)
dφ
and u(x) =
dx
u = fluid velocity, h = piezometric head, z = elevation
head, γ = specific weight of water for ground water
problems, p = pressure and K = coefficient of
permeability or hydraulic conductivity.

It follows that:
dh(x)
u(x) = -K (2-7-b)
dx

For steady incompressible flow in 1-dimension, the velocity is constant.

2.2.4. Mass – transport

Diffusion, with in the most elementary assumptions, results in a governing equation similar to
eq. (2-8). Assuming a constant area, the balance of mass for a dilute mixture can be written
as:
dC(x) dj(x)
u(x) + + K r C(x) = m (2-9)
dx dx
Where u(x) = velocity of the mixture, C(x) =
concentration of dilute species, j(x) = flux of the
species, Kr = a reaction rate, and m = external mass
source. The constituve equation is known as Fick’s
law and is given as:
dC(x)
j(x) = -D(x) (2-10)
dx
D(x) = diffusivity

(2-9) + (2-10) gives the governing equation:

dC(x) d ⎡ dC(x) ⎤
u(x) − D(x) + K r C(x) = m (2-
dx dx ⎢⎣ dx ⎥⎦
11)

The velocity u(x) (first order derivative) is assumed to be known (2-1) / (2-8) do not have 1st
order.

2.3. Variational function

The variatonal function that corresponds to the governing equations in 2-2 (except eq. (2-11))
can be written in a general form for an individual element. Let f = f(x)

ENSP Page 16/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
1 ⎡ ⎛ df ⎞ ⎤
2

J1 (f) = ∫ ⎢α ⎜ ⎟ + β f 2 − 2 γ f ⎥ dV (2-12)
V
2 ⎢⎣ ⎝ dx ⎠ ⎥⎦

For the purpose of deriving a FE model, let us represent the governing diff. eq. by pseudo
variational function corresponding to eq. (2-11) with C = C(x).

1 ⎡ ⎛ dC ⎞ ⎤
2
dC
J 2 (C) = ∫ ⎢D⎜ ⎟ +CU + K r C 2 − 2 m C⎥ dV
V
2 ⎣⎢ ⎝ dx ⎠ dx ⎦⎥
(2-13)

2.4. Interpolation function and shape functions

The discrete model is composed of one or more interpolation polynomials, and the continuous
function is divided in to finite pieces called elements. Each element is defined using an
interpolation function to describe its behaviour between its ends points (nodes) shape function
(Ni) is usually the coefficient that appears in the interpolation polynomial. The shape function
is written for each individual node of a FE and has the property that its magnitude one at that
node and zero for all other nodes in that element. (Refer to problems)

2.5. Stiffness matrix

The stiffness matrix originates from structural analysis. In FE terminology we define 2
stiffness matrices: The local stiffness matrix corresponds to an individual element. The global
stiffness matrix is the assemblage of all local stiffness matrices of the entire system.
If the element matrix is Ke = [kij], where
kij = stiffness coefficient (or influence coefficient) relating nodes [i,j]
⎡k k 12 ⎤
For a 2-node element K e = ⎢ 11 ;
⎣k 21 k 22 ⎥⎦
⎡ k 11 k 12 k 13 ⎤
For a 3-node element K e = ⎢⎢ k 22 k 23 ⎥⎥
⎢⎣ sym k 33 ⎥⎦
As example of the given model in 2-1, let a assemble the global stiff matrix (1node = 1 dim)

SCHEMA ICI

2.6. Boundary conditions

There must be two boundary conditions in order to solve for the constants of integration; and
usually they are specified on surface at each end of the one dimensional element.
¾ Essential boundary conditions are called Dirichlet boundary conditions.
For two dimensional steady – state heat condition for a rod of length L and
constant area using equation (2-6)

ENSP Page 17/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
d 2T
+ Q = 0 (2-15)
dx 2
Q = volume rate flow (fluid) = convection heat
transfer (heat flow)
T(0) = T0 and T(L) = TL.

¾ Natural boundaries conditions (Neumann boundary conditions)

correspond to the both boundary conditions (at 0 to L) specify conditions
for the 1st derivate. In heat conduction, this would be condition on the flux,
such as equation (2-15) combined with
dT(0) dT(L)
k0 = q 0 and k L = q L (2-16)
dx dx

The third type of boundary condition is called mixed and is a combination of equations (2-16)
and (2-15). In FE analysis only the Dirichlet problem is usually considered

2.7. Problems in cylindrical coordinates

Axis symmetric formulation of problems such as heat conduction results in 1-dimension
differential equation that is similar to equation 2-6.
¾ The equation for the electric potential in axis symmetric cylindrical
coordinate is:
d 2φ E dφ
E + + ρ = 0 (2-17)
dr 2 r dr

where E = the permittivity of the materiel; φ = electric potential; ρ = charge density.

E d ⎛ dφ ⎞
The area being constant r = cylinder eq. (2-17) is the same as ⎜ r ⎟ + ρ = 0 ⇒ the
r dr ⎝ dr ⎠
corresponding variational function is:

⎡
r2
⎛ dφ ⎞
2
⎤
J (φ) = ∫ ⎢π r E ⎜ ⎟ − 2π r ρφ⎥dr (2-18)
r1 ⎢
⎣ ⎝ dr ⎠ ⎥⎦

Where dV = 2 π rdr.

2.8. The direct method

The fundamental idea of the direct method is to use solutions for rod or beam problems
derived in mechanics of materials. For example axial stress in a rod subject to a constant force
P is computed as σ = P , combined with eq. (2-2), we get the deformation of a rod subject to
A
an axial force:

ENSP Page 18/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
PL
U= (2-19)
AE
U = total average deformation, L = Length. The
stiffness matrix is also derived from this expression.

2.9. problems
2.9.1. Problem 1:

SHEMA ICI

ENSP Page 19/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
Derive a one – dimensional interpolation formula for a function u = u(x) that is valid in
the range u1 through u2.
Derive the shape function Ni for a one – dimension linear FE.
Derive a linear 1-dimension interpolation formula in terms of shape functions Ni and
nodal variables and write the result as a matrix equation.
Solution 1:
a linear eq. that will approximate u(x) between u1 and u2 is in the form u = A+Bx (a), A
and B are constants ==> ui = A+Bxi ==> solving for A and B gives the interpolating
polynomial
u1x 2 − u 2 x1 u 2 − u1
A= ; B=
x 2 - x1 x 2 - x1
x2 − x x − x1
u = u(x) = u 1 + u2 (b)
x 2 - x1 x 2 - x1
2-The shape function N1 at node 1 is the coefficient of u1
The shape function N2 at node 2 is the coefficient of u2
x2 − x x − x1
⇒ N1 = ; N2 =
x 2 - x1 x 2 - x1

x = x 1 ⇒ N 1 = 1 and N 2 = 0⎫
⎬ Property given in 2-4
x = x 2 ⇒ N 1 = 0 and N 2 = 1⎭
3-using result of 1- and 2-, we define [N] = [N1 N2]
{Ui} = [U1 U2]T, and therefore u = u(x) = [N]{Ui}.

2.9.2. Problem 2
The use of equation 2-12 in the form of heat conduction gives the variational of the
function (if we make the following analogy f a T ; α a k ; β a 0 and γ a Q ) for heat
d 2T
conduction. We can show that k + Q = 0 is the governing equation for heat
dx 2
conduction
K = constant = thermal conductivity; Q = constant = convection heat transfer; T = t°
1- Use the Rayleigh – Ritz method to obtain an approximate solution for the 1 –
dimension heat conduction problem. Assume a rod of length L, constant area, and
constant heat source along the length rod. T(0) = T(L) = 0.
2- Use the Rayleigh – Ritz method and 2 - linear interpolation polynomials to solve this
heat conduction problem.

SCHEMA ICI

ENSP Page 20/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN

3-Assume a rod of elastic material fixed at both ends with constant area A and length of
3 L with uniform body force loading f. Use three linear elements of length L and
formulate the Rayleigh – Ritz solution u = u(x) using shape functions rather than
interpolation formulas.
4- for the same rod (3-) write the variational function In terms of linear shape functions.
Derive a general model ( a local stiffness matrix) for the rod using the variational
function.

Solution 2

ENSP Page 21/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
d 2T Q
1- the governing equation is given in the form 2
=− and offer 2 integration
dx R
Qx 2
gives(a) T = - + C1 x + C 2 . Substituting boundary (b) conditions and we arranging,
2k
we obtain:
Q(Lx -x 2 )
T= (c)
2k
“The Rayleigh – Ritz method states that a function is assumed in of eq. (a). The result is
a system of algebraic equations that can be solved for the unknown coefficient ci “.
n
Assuming T = ∑ c i x i-1 (d)
i =1

If T is quadratic function T = c1 + c2x + c3x2 (e)

Subtitting the bundary conditions; T(0) = 0 ==> c1 = 0;
(
T(L ) = 0 ⇒ c 2 = - c 3 L ⇒ T = c 3 x 2 - Lx )
dT (f)
= c 3 (2x - L )
dx
L
(f) in to the rearranged eq. (2-12) with the given analogy, and replacing ∫ dV with A∫ dx
V 0

gives:

[ ( )]
L
J =∫
1 2
2
) (
kc3 4 x 2 − 4 Lx + L2 − 2Qc3 x 2 − Lx Qdx or after integration on limits,
0

Qkc32 L3 Qc3 AL3

J= + (g)
6 6
The valve of c3 that makes J a is obtained in derivation with respect to c3
∂L Q
= 0 ⇒ c3 = − (h)
∂c3 2k
(h) in to (f) gives the final result:
Q(Lx -x 2 ) exact solution in this application.
T=
2k
With the given figure, we use interpolation formulas of problem 1-1.
For the element of the left side
x 2 − x1 x − x1
T = T1 + T2
L L
2 2
With boundary condition T1 = 0 at x1 =0;

ENSP Page 22/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
2 T2 x L
T= 0≤x≤ (a)
L 2
x3 − x x − x2
T = T2 + T3
L L
For the right side element, 2 2 , and with boundary condition T3
= 0 at x3 = L,

T = 2 T2
(L -x) ; L ≤ x ≤ L (b)
L 2
L⎡ 2
⎤
1 ⎛ dT ⎞
J ( T) = ∫ ⎢ k ⎜ ⎟ − QT⎥ Adx (c)
0⎢⎣ 2 ⎝ dx ⎠ ⎥⎦

(b) in to (c) gives

L
2⎡
x⎤ ⎛ 2T22 L- x⎞
L
2T22
J ( T) = ∫ ⎢k 2 − 2QT2 ⎥ Adx + ∫
L⎦
⎜ k 2 − 2QT2
⎜ L L
⎟Adx (d)
⎟
0 ⎣ L L
2
⎝ ⎠
After integration and replacement:

⎛ 2kT22 QT2 L ⎞
J (T) = ⎜⎜ - + ⎟A
⎟ (e)
⎝ L 2 ⎠

∂J (T) 4kT2 QL QL2

= − = 0 and T2 = (f)
∂T2 L 2 8k

in to eqs. (a) and (b), we have

QLx L
T= 0≤x≤
4k 2
QL( L − x) L
T= ≤x≤L
4k 2
L L 3QL2 L
At x = 2 , the solution is exact. At x = 4 , the solution is 32k , while here T( 4 ) =
QL2
16k (1st question result).
The solution process is similar to that of question 2)-, and for element I using the I =
element result in problem (1-3), we have

ENSP Page 23/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN

⎧U I = N I 1U 1 + N I 2U 2
⎪
⎨ dU I dN I 1 dN I 2 (a )
⎪⎩ dx = U 1 + U2
dx dx
Where NI1 is the linear shape function for node 1 of element I (same for NI2) using the
result for problem 1-2 we have
L−x dN I 1 1
N I1 = =−
L dx L
x dN I 2 1
NI2 = =
L dx L
Similarly for element II and III
UII = NII.2 U2 + NII.3 U3
UIII = NIII.3 U3 + NIII.4 U4

SCHEMA ICI

2L − x dN II 2 1
N II 2 = =−
L dx L
x−L dN II 3 1
N II 3 = = draw NI,I only
L dx L
3L − x dN III 3 1
N III 3 = = − (b)
L dx L
x − 2L dN III 4 1
N III 4 = =
L dx L
Substituting into eq. (2-12) + (a) and using the boundary conditions U1 = U4 = 0 with the
following analogy

ENSP Page 24/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
α ⇒ E ; f ( function) a u ; β ⇒ 0 and γ ⇒ f (bodyforce)
L
⎛ U 22 U2x ⎞ 2L
⎛ U 22 − 2U 2 U 3 + U 32 U 2 (2L - x) U 3 (x - L) ⎞
J ( U) = ∫ ⎜ E 2 − f
⎜ ⎟Adx + ∫ ⎜⎜ E
⎟ 2
− − ⎟⎟Adx
0⎝ 2L L ⎠ L ⎝ 2L L L ⎠

1
L L 3L
⎛ U 32 U 3 (3L - x ) ⎞
= ∫ EAε dx - ∫ ∆.fAdx + ∫ ⎜⎜ E 2 −
2
⎟⎟Adx
20 0 2 L⎝ 2L L ⎠
EAε 2 dx = σAεdx (réaction energy) ; U.dF = ∆.fAdx; (action energy)
integrating
U 22 − U 2 U 3 + U 3
J ( U) = EA − AfL(U 2 + U 3 ) (C)
L
Minimizing J ( U) with respect to Ui gives (U1 = U4 = 0)
∂J 2U 2 − U 3
= EA − fAL = 0
∂U 2 L
(d)
∂J 2U 3 − U 2
= EA − fAL = 0
∂U 3 L
We give (d) in matrix form as
EA ⎡ 2 − 1⎤ ⎡ U 2 ⎤ ⎡1⎤
= fAL (e)
L ⎢⎣− 1 2 ⎥⎦ ⎢⎣ U 3 ⎥⎦ ⎢1⎥
⎣⎦
fL2 fL2
Solving (e) give us U 2 = ; U3 = , and the final result
E E
fLx
UI = 0≤x≤L
E
fL2
U II = L ≤ x ≤ 2L
E
fL(3L - x)
U III = 2L ≤ x ≤ 3L
E
The exact solution is given in the form of the first question as
f3Lx - x 2
U=
2E
For x = L and x = 2L 0% error
3
For x = L 11% error
2
The variational function can be written as follows for one element of length L:
L L
1 dU dU
2 ∫0 dx dx
J (U) = E Adx - ∫ UfAdx (a)
0

ENSP Page 25/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
We use matrix definition of the linear shape function of problem 1.3 and rewrite eq. (a)
T L

A
L
{U}T ⎧⎨ dN ⎫⎬⎡ dN ⎤
E ⎢ ⎥{U}dx - A ∫ {U} [N ] fdx
T T

2 ∫0
J (U) = ⎩ dx ⎭ ⎣ dx ⎦ 0

⎡ dN 1 ⎤
[E ]⎢ ⎥{U}dx − A ∫ [U1 U 2 ]⎡⎢ 1 ⎤⎥fdx
L L
A ⎢ dx ⎥ ⎡ dN ⎤ N
= ∫ [U 1 U 2 ]⎢ ⎥ (b)
2 0 dN ⎣ dx ⎦ ⎣N 2 ⎦
⎢⎣ 2 dx ⎥⎦ 0

The minimization of the first eq. of (b) is in the form

L T L
∂J ( U) ⎡ dN ⎤ ⎡ dN ⎤
= A ∫ ⎢ ⎥ [E ]⎢ ⎥{U}dx − A ∫ [N ] fdx = 0 (c)
T

∂{U} 0 ⎣
dx ⎦ ⎣ dx ⎦ 0

Using result of question 3-

⎧ 1⎫ ⎧L - x ⎫
L ⎪⎪− L ⎪⎪ ⎡ 1 1 ⎤ ⎧U 1 ⎫
L ⎪⎪ L ⎪⎪
A∫ ⎨ ⎬[E ] − ⎨ ⎬ dx − A ∫0 ⎨ x ⎬fdx = 0
0⎪
1 ⎪ ⎢⎣ L L ⎥⎦ ⎩U 2 ⎭ ⎪ ⎪
⎪⎩ L ⎪⎭ ⎪⎩ L ⎪⎭
or
⎡ E E⎤ ⎧L - x ⎫
⎢ 2 − L⎪ ⎪
L2 ⎥ ⎧U 1 ⎫⎬dx = A ⎪⎨ L ⎪⎬fdx
L
A∫ ⎢ L ∫0 x
E E ⎥ ⎨⎩U 2 ⎭
0 ⎢− ⎥ ⎪ ⎪
⎣ L2 L2 ⎦ ⎪⎩ L ⎪⎭
and after a final integration
⎡ AE AE ⎤ ⎧ AfL ⎫
⎢ L2 − ⎪ ⎪
L2 ⎥ ⎧⎨U 1 ⎫⎬ = ⎪⎨ 2 ⎪⎬
⎢ AE ⎥
AE ⎩U 2 ⎭
⎢− 2 ⎥ ⎪ AfL ⎪
⎣ L L ⎦
2
⎪⎩ 2 ⎪⎭
or
⎧ AfL ⎫
AE ⎡ 1 − 1⎤ ⎧U 1 ⎫ ⎪⎪ 2 ⎪⎪
⎨ ⎬=⎨ ⎬
L ⎢⎣− 1 1 ⎥⎦ ⎩U 2 ⎭ ⎪ AfL ⎪
⎪⎩ 2 ⎪⎭

Or
[Ke]{U} = {fe} (element local stiffness matrix for a rod in one dimension space.
[K]{U} = {P} (stiffness equation).

SCHEMA

ENSP Page 26/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN

2.9.3. Problem 3
Compare the formulation of 1-dimension FE problem using 2 nodes linear elements
versus 3 node quadratic elements.

Solution 3
2 node linear element was discussed in problem 2.
A 3 node quadratic element has similar properties. The function must span 2 paces and
connect 3 nodes. A 3 node element can have a length of L (or 2L or any°) for
convenience of computations. Using previous formulation of the shape function we have
for element I
fI =NI1f1 + NI2f2 + NI3f3 (a)

and for element II

fII =NII3f3 + NII4f4 + NII5f5 (b)

Cubic interpolation formulas here are 4 nodes and the element will span 3 spaces.
SCHEMA

Equations (a) and (b) are given in matrix form as

ENSP Page 27/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
⎧f 1 ⎫
⎪ ⎪
f I = [N I1 N I2 N I3 ]⎨f 2 ⎬
⎪f ⎪
⎩ 3⎭
or f = [N]{fi} (c)
⎧f 3 ⎫
⎪ ⎪
f II = [N II3 N II4 N II5 ]⎨f 4 ⎬
⎪f ⎪
⎩ 5⎭

The procedure for formulating the local stiffness matrix is identical with that in all the
preceding examples. The shape function are approximated using
Interpolation formulas such as:
Fi = A + Bx + Cx2.
The method used in problem 1 can be used to derive a general shape function even if the
algebra becomes must more cumbersome.

2.9.4. Problem 4
1- Derive of stiffness matrix equation of a rod (problem 2-4) in a general3-dimension
coordinate system.
2- Apply the first question using paragraph 2-5 to find displacements at each node and
internal forces in each bar of the following truss.all members have identical areas of
cross section A and modulus E.

SCHEMA

Solution 4

ENSP Page 28/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
1- suppose a 1-dimension element IJ with respective coordinates Xi , Yi , Zi (Xj , Yj , Zj)
in the space general coordinate system. Let us define for example the displacement of
the element IJ by its projection on different axis:
U φ = U x cos α + U y cos β + U z cos γ (a)
X j − Xi
where cos α , cos β and cos γ are cosine unit vectors of segment IJ (cos α = ,...).
L
⎧U xi ⎫
⎪ ⎪
⎧U i ⎫ ⎡cos α cos β cos γ 0 0 0 ⎤ ⎪U yi ⎪
⎨ ⎬=⎢ ⎨ ⎬ (b)
⎩U j ⎭ 1
⎣ 04444 0 0 cos α cos β cos γ ⎥⎦ ⎪M ⎪
44442444444443
[Λ ]=[2×6 ] ⎪U zj ⎪
⎩ ⎭

[Λ] = the transfer matrix.

the final stiffness matrix in the general coordonate system is a 6 × 6 square matrix [K I ]
I = element number.
[K I ] = [Λ ]T [K e ][Λ ] (c)
Or in a general form (with l = cosα , m = cosβ , n = cosγ )
NB: for a 2 – dimension general coordinate system, neglect all erase rows and column
with the third cosine unit vector n.
[KI] is 6X6 for 3 – dimension and 4X4 for 2 – dimension coordinate system.
⎡l 2 lm ln - l2 - lm − ln ⎤
⎢ ⎥
⎢ m 2
mn - lm -m − mn ⎥
2

⎢ n 2
- ln - mn - n2 ⎥
AE ⎢
[K I ] = ⎥ (d)
L ⎢ (sym) l2 lm ln ⎥
⎢ ⎥
⎢ m2 mn ⎥
⎢ 2 ⎥
⎣ n ⎦
NB: last page, it is also important to find by the same procedure the expression of the
associated to nodal loads for external loading.
2- We use 3 finite elements to model the structure.

SCHEMA

ENSP Page 29/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
Global Geometric Material ¾
Element orientation
nodes properties properties
1 1 2 A; L E θ 1 = 0°
2 2 3 A; L E θ 2 = 90°
3 1 3 A; √2L E θ 3 = 45°
Element matrices using eq. (d) without terms with n.

ENSP Page 30/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN

⎡1 0 −1 0⎤
⎢ 0⎥
EA ⎢ 0 0 0
l=1, m=0 [K ] = ⎥ (a)
L ⎢− 1 0 1 0⎥
I

⎢ ⎥
⎣0 0 0 0⎦
⎡0 0 0 0⎤
⎢ 0 − 1⎥
EA ⎢0 1
l = 0, m=1 [K ] = ⎥ (a)
L ⎢0 0 0 0⎥
II

⎢ ⎥
⎣0 − 1
0 1⎦
⎡ 1 1 − 1 − 1⎤
⎢ ⎥
2 EA ⎢ 1 1 − 1 − 1⎥
m=l= [K ] = (a)
2 2 2L ⎢− 1 − 1 1 1 ⎥
III

⎢ ⎥
⎣− 1 − 1 1 1 ⎦
Assembly of the global stiff matrix from paragraph 2-5.
[K] = [ki,j] where i,j 1 ≤ i, j ≤ 6 refer to local degree of freedom

⎡K +K K +K K K K ⎤ K
I III I III I I III III
11 11 12 12 13 14 13 14

⎢ K +K I III
KI
K I
K K
III ⎥ III

⎢ 22 22 23 24 23
⎥ 24

⎢ K +K
I II
K +K
I II
K K
II
⎥ II

[K] = ⎢ 33 11 34 12 13
⎥(b
14

⎢ (sym) K +K
I
44
II
22
K K
II
23 ⎥
II
24

⎢ K +K K +K ⎥
II III II III

⎢ ⎥
33 33 34 34

⎣ K +K ⎦ II
44
III
44

Then

ENSP Page 31/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
1 2 3 4 5 6 DOF Displ
⎡1,3536 0,3536 - 1.0 0.0 - 0,3536 − 0,3536⎤ 1 u1 u1x u1
⎢ 0,3536 0.0 0.0 - 0,3536 - 0,3536⎥ 2 u2 u1y v1
⎢ ⎥
⎢ 1.0 0.0 0.0 0.0 ⎥ 3 u3 u 2x u2
[K I ] = AE ⎢ ⎥
L ⎢ (sym) 1.0 0.0 - 1.0 ⎥ 4 u4 u1y v2
⎢ 0,3536 0,3536 ⎥ 5 u5 u 3x u3
⎢ ⎥
⎣ 1,3536 ⎦ 6 u6 u1y v3
(c)
We impose
u = horizontal displ.
v = vertical displ.
¾ Boundary conditions
The specify displacement and force DOF are:
u1 = v1 = u2 = v2 = 0; F3x = P; F3y = -2P (d)
¾ Condensed equations
Since displacements u1, u2, v1, v2 are zero (pin – connection) corresponding DOF (1, 2, 3,
4) with related columns and rows can be removed from the general stiff matrix to obtain
the condensed equation derived from problem 2-4 equation (d) to obtain

EA ⎡0,3536 0,3536⎤ ⎧u 3 ⎫ ⎧P ⎫
⎨ ⎬=⎨ ⎬ (e)
L ⎢⎣0,3536 1,3536 ⎥⎦ ⎩ v 3 ⎭ ⎩- 2P ⎭
And the condensed equations for the unknown reactions are (from the first four
equations of the system.

⎡ F1x ⎤ ⎡− 0,3536 − 0,3536⎤

⎢F ⎥ ⎢ ⎥
⎢ 1y ⎥ = EA ⎢− 0,3536 − 0,3536⎥ ⎧⎨u 3 ⎫⎬ (f)
⎢F2x ⎥ L ⎢ 0.0 0.0 ⎥ ⎩ v 3 ⎭
⎢ ⎥ ⎢ ⎥
⎣F2y ⎦ ⎣ 0.0 − 1.0 ⎦
u3 and v3 is the solution of eq. (e).

¾
From eq. (e) we solve for u3 and v3

ENSP Page 32/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
PL PL PL
u 3 = (3 + 2 ) = 5,828 ; v 3 = −3 and from (f) F1x = P; F1y = -P; F2x = 0.0;
EA EA EA
F2y = 3P.
¾ Computation of internal forces

SCHEMA

I θe is the angle between the positive x – axis and positive xe axis, measured in the
counter clockwise direction then the transfer matrix [ Λ ] in eq. (1-b) for a 2 – dimension
coordinates system can be written as:
⎡u 1e ⎤ ⎡ cosθ e sinθ e 0 0 ⎤ ⎡u 1 ⎤
⎢ e⎥ ⎢ 0 ⎥ ⎢⎢ v 1 ⎥⎥
⎢ v 1 ⎥ =≥ ⎢− sinθ e cosθ e 0
⎥ (g)
⎢u e2 ⎥ ⎢ 0 0 cosθ e sinθ e ⎥ ⎢u 2 ⎥
⎢ e⎥ ⎢ ⎥⎢ ⎥
⎣{v2 ⎦ ⎣1404444 0 − sinθ e cosθ e ⎦ ⎣ v 2 ⎦
424444443 {
local [Λ ] global

Internal forces Fe in element can be determined from the element equation of problem
(2-4-d) as:

⎡F1e ⎤ EA ⎡ 1 − 1⎤ ⎧u 1e ⎫
⎢ e⎥ = ⎢− 1 1 ⎥ ⎨ e ⎬ (h)
F
⎣ 2⎦ L ⎣ ⎦ ⎩u 2 ⎭
u ei are from eq. (g).
Solving eq. (g) and (h) for each element, we find:

F1I = −F2I = 0; F1II = −F2II = 3P (c) ; F1III = −F2III = − 2P (T)

2.10. Homework number 2:

2.10.1. Homework 1
SCHEMA

ENSP Page 33/103

FINITE ELEMENT METHODS / 2. ONE-DIMENSIONAL FINITE ELEMENT
*Programming and Matrix Opération* Version ou date
Dr LEZIN
a-Derive the shape functions for a general tree – node quadratic element in terms of x1,
x2 and x3 as shown here.(see problem 2-9-1).
b-Refer to question a- and let x1 = -L, x2 = 0, and x3 = L.
Derived the corresponding shape functions Ni.
c-Refer to problem (2-4) and derived the local stiffness matrix for an element of length
2L with coordinates (-L, 0,L).

SCHEMA

2.10.2. Homework 2

Consider the steel column (a typical column in a multi-storey building structure) shown
here. The loads shown are due to the loads of different floors. The elasticity modulus is E
= 2,1×105MPa and cross-sectional area of the column is A = 32 cm2. Using the procedure
in this chapter, determine the vertical displacements and axial stresses in the column at
various floor – column connection points

ENSP Page 34/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN

3. Two – dimensional Finite Element

3.1. Introduction
The one dimension concept of chap 2 is extended to 2 – dim. With 2 fundamental FE: the four
node quadrilateral element and the 3 – node triangular element.

SCHEMA

The quadrilateral element is rectangular and must conform to the cartesian global coordinate
system. The triangular element is used to model 2 – dimension problems with a curved
boundary by approximating the curve with a series of straight lines.

3.2. Two-dimensional boundary values problems.

Various physical problems described in chap 2 can be extended to two dimensions and, with
the exception of eq. 2.11, can be written in the general form as:
∂ ⎡ ∂φ ( x, y ) ⎤ ∂ ⎡ ∂φ ( x, y ) ⎤
αx + αy + βφ ( x, y ) = f ( x, y )
∂x ⎢⎣ ∂x ⎥⎦ ∂y ⎢⎣ ∂y ⎥⎦
(3.1)

Where αx, αy and β are known parameters that can be a function of x and y.
The 2-dimension equation of mass transport is: similar to equation 2.11 and is:

∂C(x, y) ∂C(x, y) ∂ ⎡ ∂C(x, y)⎤ ∂ ⎡ ∂C(x, y)⎤

ux + uy − ⎢Dx + Dy + KrC(x, y) = m
∂x ∂y ∂x ⎣ ∂x ⎥⎦ ∂y ⎢⎣ ∂y ⎥⎦
(3.2)

Where ux, uy, Dx, Dy are velocities and diffusivities in x and y directions.
A special cause of eq. 3.1 when β = 0 and αx= αy gives the Poisson equation in the Cartesian
coordinates system.

3.3. Theory of elasticity

A two dimensional elasticity problem is formulated in terms of displacements, and all FE
nodes must have 2 DOF to represent a displacement in each coordinates direction.
The firs derivatives of the two components vector {u(x,y),v(x,y)} displacement correspond to
the various train – displacement relationship is given as:
⎡ ⎤
⎧ε xx ⎫ ⎢ ∂ ∂x 0 ⎥
⎪ ⎪
{ε } = ⎨ε yy ⎬ = ⎢⎢ 0 ∂ ∂y ⎥⎥ ⎡⎢ ⎤⎥ = [L]⎡⎢ ⎤⎥ (engineering
u u
⎪ ⎪ ⎢ ⎣v ⎦ ⎣v ⎦
ε ∂ ∂ ⎥
⎩ xy ⎭ ⎢ ∂y ∂x ⎥⎦
⎣
definition of strain) (3.3)

εxx, εyy, = normal strain; εxy = shew strain.

ENSP Page 35/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
The equations of equilibrium defining balance of forces on a differential element must be
satisfied when analytical solutions are attempted. These equations define a relationship among
normal stress, shear stress, and body force f:
∂ σ xx ∂ σ xy
+ + fx = 0
∂x ∂y
(3.4)
∂ σ xy ∂ σ yy
+ + fy = 0
∂x ∂y

¾ If the material is the homogeneous and isotropic with Young’s E modulus and
Poisson’s ratio ν
Being constant then a two-dimension problems usually involve either plane stress or plane
strain.

σ = D.ε (3.5)
D = matrix of elastic contents.

¾ Plane stress occurs when the thickness dimension t is small compared to the
length and width dimensions. A thin plate or disk of material loaded in its
plane is an example of plane stress. It follows that:
σ xz = σ yz = σ zz = 0
⎡ ⎤
⎢ 1 0 0 ⎥
(3.6)
[D] = E 2 ⎢ν 1 0 ⎥
1 -ν ⎢ 1 −ν ⎥
⎢0 0 ⎥
⎣ 2 ⎦

schema

Plane strain can be assumed when the length dimension is large compared with the cross
section dimensions, such as a gun barred

schema

It follows that:
σ xz = σ yz = 0; σ zz = ν (σ xx + σ yy )

ENSP Page 36/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
ε xz = ε yz = ε zz = 0
⎡ ν ⎤
⎢ 1 1 −ν
0 ⎥
E (1 − ν ) ⎢⎢ ν ⎥ (3.7.a)
[D] = 1 0 ⎥
(1 + ν )(1 − 2ν ) ⎢1 − ν ⎥
⎢ 1 − 2ν ⎥
⎢ 0 0
2(1 − ν ) ⎥⎦
⎣

The relationship between shear stress and shear strain is from eq. (3.5) and given as:
E
σ xy = Gε xy = ε xy (3.7.b)
2(1 + ν )

If the displacement in eqution (3.3) is given in term of shape functions Ni, then we define the
strain matrix corresponding to that node i as [Bi] ≠ [ε]
⎡ ∂N i ⎤
⎢ 0 ⎥
⎢ ∂x ⎥
∂N i ⎥ ⎧u ⎫
[Bi ] = ⎢⎢ 0 ; {ε } = [Bi ]⎨ ⎬ (3.7.c)
∂y ⎥ ⎩v ⎭
⎢ ∂N ∂N i ⎥
⎢ i ⎥
⎣⎢ ∂y ∂x ⎥⎦

3.4. Variational functions

Variational function for two dimension problem is obtained as an extension of the 1-
dimension function in the same way that the governing differential equation 3.1 was obtained.
The variational function corresponding to plane elasticity can be written in a general form
using Cartesian tensor notation in terms of the stress tensor and strain tensor:
potential E ext
⎛1 ⎞ 6 474 8
J (u) = ∫ ⎜ σ kj ε kj − f k u k ⎟dv − ∫ Tk u k ds
V⎝
⎜2 {⎟
bodyforce⎠ S
14442444 3
potential E int
(3.8)

This equation is analogous to the principle of minimum potential energy discussed in theory
of elasticity, with the following FE results: (e = element)

ENSP Page 37/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN

Me&u&e + Keue = Fe + Qe equilibrium

equation (3.9)
Where:

Me = element mass matrix; Ke = element stiff matrix;

Fe = element local vector; Qe = element internal force
vector.

With [Ke] given earlier;

[M ] = ∫ ρ h [N] [N]dx ; [F ] = ∫ h [N] f dx ; Q = ∫ h[N]
e T e T e T
tds
Ωe Ωe re

(3.10)
[N] = 2n×2n shape function matrix; ρ = density; h =
⎧f x ⎫
element height; f = ⎨ ⎬ = distributed force
⎩f y ⎭
parameter;

⎧t x ⎫
t = thickness ⎨ ⎬ .
⎩t y ⎭

3.5. Transformations
Matrix transformations are manipulations that modify a matrix based upon some type of
constraint or condition. From the local to the global coordinates system, matrices in the
equilibrium equations are transformed as illustrated here:
We can press the vector components of F with ε , η coordinates. In terms of components in
the x, y coordinates:
⎡Fx ⎤ ⎡ cosθ sinθ ⎤ ⎧x ⎫ ⎧Fx ⎫
⎢F ⎥ = ⎢ ⎥ ⎨ ⎬ = [T0 ]⎨ ⎬ (3.11)
⎣ y ⎦ ⎣− sinθ cosθ ⎦ ⎩ y ⎭ ⎩Fy ⎭

Schema

If we wish to express the global displacements at a given node in terms of local displacement
at a specific node, we construct the transformation of the whole (global) system as:
⎡ [I] [0] [0]⎤
[T] = ⎢⎢[0] [T0 ] [0]⎥⎥ (3.12)
⎢⎣[0] [0] [I]⎥⎦
This all displacement DOF that are not constrained are unaffected, and only global
displacement that are constrained are transformed with their specific transformation matrix
[T0]i. from the local to the global coordinate system we have:

ENSP Page 38/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
⎧∆1 ⎫ ⎡ I 0 0 ⎤ ⎧∆ ⎫
1

⎪ 2⎪ ⎢ ⎪ ⎪
⎨∆ ⎬ = ⎢0 [T0 ]T 0⎥⎥ ⎨U ec ⎬ (3.13)
⎪ 3 ⎪ ⎢0 0 I ⎥⎦ ⎪∆2 ⎪
⎩∆ ⎭ ⎣ ⎩ ⎭
Where ∆1 and ∆2 denote the constrained displacement
DOF Ue.

3.6. Cylindrical coordinates

The 3-dimension counterpart of equation (3.1) in cylindrical coordinate r,θ,z can be written in
the general form. The material constants will be assumed to be independent of the coordinate
and any spatial dependence can be included in the FE formulation.
⎡ ∂ 2φ (r,θ , z) 1 ∂φ (r,θ , z) ⎤ 1 ∂ 2φ (r,θ , z) ∂ 2φ (r,θ , z)
αr ⎢ + ⎥ + α θ 2 + α z + βφ (r,θ , z) = f (r,θ ,
⎣ ∂r 2 r ∂r ⎦ r ∂θ 2 ∂z 2
(3.14)

The 3-dimension strain-displacement equations of elasticity in cylindrical coordinate where u,

v, w are the displacement in the r, θ, z direction are: u = f(r, θ, z); v = f(r, θ, z); w = f(r, θ, z)
∂u 1 ∂v ∂w
εr = εθ = εz =
∂r r ∂θ ∂z
(engineering definition)
1 ∂U ∂v v ∂U ∂w 1 ∂w ∂v
ε rθ = + − ; ε rz = + ; εθ z = +
r ∂θ ∂r r ∂z ∂r r ∂θ ∂z
(3.15)

The 3-dimension stress-strain equations for isotropic elasticity are similar to the plane strain
equations:

ENSP Page 39/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
E
σ rr = [(1 − ν )ε r r + ν (ε θ θ + ε z z )]
(1 + ν )(1 − 2ν )
E
σθθ = [(1 − ν )ε θ θ + ν (ε r r + ε z z )]
(1 + ν )(1 − 2ν ) (3.16)
E
σ zz = [(1 − ν )ε z z + ν (ε θ θ + ε r r )]
(1 + ν )(1 − 2ν )
σ rθ = Gε r θ ; σ r z = Gε r z ; σ θ z = Gε θ z

3.7. Problems
3.7.1. Problem1:
1- A rectangular FE with dimensions axis defined in an x, y coordinates system as shown
in fig P1A. Assume the function of the form:
φ =A + Bx + Cy + Dxy
And derive the shape function for the rectangular element. Write the final result in the
form
φ = N1 φ 1 + N2 φ 2 + N3 φ 3 + N φ 4

Schema

2-A 3-node triangular element is defined in fig. P1.B in an x, y coordinate system with
nodes 1(x1, y1), 2(x2, y2) and 3(x3, y3) in the global system derive shape functions in terms
of global coordinates. Assume φ = C1 + C2x + C3y.
3- A quadrilateral element is shown in an x, y coordinates system. Temperature
distribution has been computed at each node as T1 = 100°, T2 = 60°, T3 = 50°; T4 = 90°.
Use the shape function in equation (1-) to compute the temperature at at node with
coordinate (2,2;2,5).

schema

Solution 1

ENSP Page 40/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
1-the boundary conditions for the element when we point 1 is taken as new origin are:
φ (0,0) = φ 1, φ (a,0) = φ 2 ; φ (a,b) = φ 3 ; φ (0,b) = φ 4 (a).
Eq.(a) in to the first form of φ in terms of the unknown constants are φ 1 = A ; φ 2 = A +
Ba;
φ 2 = A + Ba + Cb + Dab ; φ 4 = A + Cb. It follows that:
φ 2 − φ1 φ 4 − φ1 φ1 − φ 2 + φ 3 − φ 4
A = φ1 ; B = ;C = ;D = (b)
a b ab
Eq. (b) in to the second form of gives the expression of the shape functions by
identification.

N1 =
(a - x )(b - y ) ; N =
x (b - y )
; N3 =
xy
; N1 =
(a - x )y ; (a = x − x b = y − y ) (c).
2 2 1 1 2
ab ab ab ab
We use that the shape function Ni has a magnitude of 1 at node I and is 0 at the
remaining nodes.

Schema

ENSP Page 41/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
2- The unknown quantity is φ = C1 + C2x + C3y (a)

⎧C1 ⎫
⎪ ⎪
Or in the matrix form φ = [1 x y]⎨C 2 ⎬ (b)
1424 3
[α ] ⎪C ⎪
⎩123
3⎭
{C}

The boundary conditions are the terms of nodal point values of φ :

φ (x1, y1) = φ 1, φ (x2,y2) = φ 2 ; φ ( x3,y3) = φ 3 (c)
Eq (c) in to eq. (a) gives as a system of 3 equations with 3 unknowns
⎧φ1 ⎫ ⎡1 x 1 y1 ⎤ ⎧C1 ⎫
⎪ ⎪ ⎪ ⎪
φ i = C1 + C 2 xi + C 3 y i or ⎨φ 2 ⎬ = ⎢⎢1 x 2 y 2 ⎥⎥ ⎨C 2 ⎬
⎪φ ⎪ ⎢1 x (d)
⎩ 3⎭ ⎣ 3 y 3 ⎥⎦ ⎪⎩C 3 ⎪⎭
1≤ i ≤ 3
Eq. (d) is also given in the form
{ φ i}=[X]{C} (e)
Solving for Ci
{C} = [X]-1{ φ i} (f)
Eq. (f) in to (b) gives φ = [α][X]-1{ φ i} = [N]{ φ i} (g)
The shape functions are the product of the first 2 matrices on the right hand side of eq.
(g)
[N]= [α] [X]-1 = [N1 N2 N3] = [1 x y][X}-1 (h)
Solving (h) gives:

N1 =
[(x 2 y 3 − x 3 y 2 ) + x(y 2 − y 3 ) + y(x 3 − x 2 )]
2A
N2 =
[(x 3 y1 − x 1 y 3 ) + x(- y1 + y 3 ) + y(x 1 − x 3 )] (i)
2A
N 3 = [ x 1 y 2 − x 2 y1 + x y1 − y 2 + y - x 1 + x 2 ]
( ) ( ) ( )
2A
⎡1 x 1 y1 ⎤
1
Where A = det ⎢⎢1 x 2 y 2 ⎥⎥ = area of the triangular element (j)
2
⎢⎣1 x 3 y 3 ⎥⎦

3- Using eq. (a) of question (1-) and the functional expression

T(x, y) = N1T1 + N2T2 + N3T3 + N4T4.
with a = 2,0 ; b = 1,0 ; point (2,5 ;2,5) being with in the element at (a/4 ; b/2), the shape
function are computed as:

ENSP Page 42/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN

N1 =
(a - a )(b - b ) 3
4 2 = ;N
a
= 4
(b - b 2 ) = 1
2
ab 8 ab 8
∑N =1
N3 = 4
( )( )
a b
2 1
= ; N4 =
(
a-a b
4 2 =3
)( ) i

ab 8 ab 8
3 1 1 3
The t° is computed as T(2,5;2,5) = .100 + .60 + .50 + .90 = 85 o .
8 8 8 8

3.7.2. Problem2
1- Deduce variational function that corresponds to 2-dimensional heat transfer. Write
the variational function in terms of linear 4-node shape function and arrive at a general
matrix statement corresponding the variational function.
2- Derive a local stiffness matrix for heat transfer using shape function for a 4-node
quadrilateral element.

Solution2

ENSP Page 43/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
1- We use eq. (2.12) and extend 1-dimension to 2-dimensions:
⎡k ⎛ ∂T ⎞
2
k y ⎛ ∂T ⎞
2
⎤
J (T) = ∫ ⎢ x ⎜ ⎟ + ⎜⎜ ⎟⎟ − QT ⎥ t dx dy (a)
A⎢⎣
2 ⎝ ∂x ⎠ 2 ⎝ ∂y ⎠ ⎥⎦

Where:
t = constant thickness; kx, ky Æ thermal conductivities; Q = external source conductivity
heat transfer.
The shape function representing the t° distribution with in an element has the form of φ
in prob. (3.1), and in matrix form:
⎧T1 ⎫
⎪T ⎪
⎪ 2⎪
t = [N 1 N2 N3 N 4 ]⎨ ⎬ = [N ]{T} N i = N i ( x, y) (b)
⎪T3 ⎪
⎪⎩T4 ⎪⎭

We define operator matrices that represent the partial derivatives occurring in eq. (a)
as:

[L x ] = ⎡⎢ ∂ ⎤⎥[L y ] = ⎡⎢ ∂ ⎤⎥ (c)
⎣ ∂x ⎦ ⎣ ∂y ⎦
[Lx] = 1×1; [N] = 1×4 Î [Lx][N] is a 1×4 matrix. kx, ky also are represented as 1×1
matrices Î eq. (a) can now be written in a matrix eq. as:
⎛1 1 ⎞
J (T ) = ∫ ⎜ {T} [N ] [Lx ] [kx ][Lx ][N ]{T} + {T} [N ] [Ly] [ky][Ly][N ]{T} − {T} [N ] Q ⎟ t dx dy
T T T T T T T T

A⎝
2 2 ⎠
(d)
The final result is a 1×1 matrix and indicate a scalar quantity that can be minimized
with respect to {T} (or Ti).
⎡∂⎤
⎢ ∂x ⎥
Eq. (c) is use to define the partial st of derivatives as [L] operating on [N]. [L] = ⎢ ⎥
⎢∂ ⎥
⎢⎣ ∂y ⎥⎦
and:
⎡∂⎤ ⎡ ∂N 1 ∂N 2 ∂N 3 ∂N 4 ⎤
⎢ ⎥ ⎢
[B] = [L][N] = ⎢ ∂∂x ⎥[N1 N 4 ] = ⎢ ∂x ∂x ∂x ∂x ⎥ (e)
⎢ ⎥
N2 N3
∂N ∂N 2 ∂N 3 ∂N 4 ⎥
⎢ 1 ⎥
⎢⎣ ∂y ⎥⎦ ⎣⎢ ∂y ∂y ∂y ∂y ⎥⎦

The term (e) is refer to the [B] matrix.

In structural engineering, the B matrix may be refer to the strain matrix given in eq.
(3.7). The variational function can be written in a more compact form by defining a
thermal conductivity matrix:

ENSP Page 44/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
⎡k x 0⎤
[k ] = ⎢ to obtain:
⎣0 k y ⎥⎦

⎛1 ⎞
J (T ) = ∫ ⎜ {T} [N ] [L] [k ][L][N ]{T} − {T} [N ] Q ⎟ t dx dy
T T T T T

A⎝
2 ⎠
(f)
⎛1 ⎞
= ∫ ⎜ {T} [B] [k ][B]{T} − {T} [N ] Q ⎟ t dx dy
T T T T

A⎝
2 ⎠
Eq. (f) is an equivalent matrix statement of the variational function.
2- The variational function of eq. (f) of question 1- is minimized with respect to the
unknown variable {T}. it follows that

∫ ([B] [k ][B]{T}) t dx dy = ∫ [T]

T T
Q t dx dy (a)
A A

The first 3 matrices on the left-hand side are multiplied together to give a 4×4 local
stiffness matrix defined as:
[K ] = ∫ [B]T [k ][B] t dx dy (b)
A

For illustration, the kij term in the stiffness matrix appears as:
a b
⎛ ∂N ∂N j ∂N i ∂N j ⎞
k ij = ∫ ∫ ⎜⎜ i k x + ky ⎟ t dx dy (c)
0 0⎝
∂x ∂x ∂y ∂y ⎟⎠
The derivatives of the shape functions are evaluated from eq. (c) of problem (3.7.1.1) as:
∂N 1 b-y ∂N 1 a-x ∂N 2 b - y
=− =− =
∂x ab ∂y ab ∂x ab
∂N 2 x ∂N 3 y ∂N 3 x
=− = = (d)
∂y ab ∂x ab ∂y ab
∂N 4 y ∂N 4 a - x
=− =
∂x ab ∂y ab
Eq (d) in to eq. (c) allow as to final kij terms terms.

For example, i = j = 1 Î k 11 =
(b k2
x + a 2 k y )t
.
3ab
The local stiffness matrix is symmetric and appears as follows:

⎡2b 2 k x + 2a 2 k y − 2b 2 k x + a 2 k y − b2k x − a 2k y b 2 k x − 2a 2 k y ⎤
⎢ ⎥
2b 2 k x + 2a 2 k y b 2 k x − 2a 2 k y − b2k x − a 2k y ⎥
[ ]
K =
e t ⎢
6ab ⎢ (sym.) 2b 2 k x + 2a 2 k y − 2b 2 k x + a 2 k y ⎥
⎢ ⎥
⎣⎢ 2b 2 k x + 2a 2 k y ⎦⎥

ENSP Page 45/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
The term on the right-hand side of eq. (a) represents the distribution of the heat-source
term.
⎧ ab ⎫
⎧(a - x )(b - x )⎫ ⎪4⎪
a b ⎪x (b - y ) ⎪ ⎪ ⎪
t ⎪ ⎪ ⎪ ab ⎪
∫ ∫ ab ⎨xy ⎬Q dx dy = ⎨ ⎬tQ (f)
0 0 ⎪ ⎪ ⎪4⎪
⎪⎩ y(a - x) ⎪⎭ ⎪ ab ⎪
⎪4⎪
⎩ ⎭
The heat-source Q is distributed equally to the four nodes.

3.7.3. Problem3
1- Use the variational function of eq. (3.8) and write a corresponding function in matrix
format that can be used for plane elasticity.
2- Use result of question 1- to derive the stiffness matrix for a four-node rectangular FE
for plane elasticity in Cartesian coordinates.
3- Assume the rectangular element of figure (P1.A) in question paragraph (3.7.1.1) has
thickness t and uniform pressure loading Py (force/area), is distributed along the edge
(3.4). use the variational function of question (1-) and (2-) and determine the distribution
of the pressure to nodes 3 and 4.

Solution 3

ENSP Page 46/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
1- eq. (3-) is general statement written in Cartesian tensor notation. Consider for now
only the first term in the volume integral of eq. (3.8) .the analogous matrix is:
1
∫ 2 [σ ] [ε ]dv
T
(a)
V

Hooke’s law in matrix notation is:

{σ} = [C]{ε} (b)
And each term in eq. (b) has a form that depends upon the elasticity problem to be
solved. Matrix is defined by equations (3.6) and (3.7) (plane strain and plane stress).
Eq. (b) in to (a) gives:
1
∫ 2 [ε ] [C][ε ]dv
T
(c)
V
schema

Consider the 4-node element with corresponding displacements in x, y directions.

The matrix of nodal displacement is {U} = [u1 v1 u2 v2 u3 v3 u4 v4]T
The element displacements are (d) defined in terms of {U} using shape function s as:
u e ( x, y) = N 1 u 1 + N 2 u 2 + N 3 u 3 + N 4 u 4
4 (e)
v e ( x, y) = ∑ N i v i
i =1

Or in the matrix form

⎧u e ⎫
⎨ ⎬ = [N ]{U} (f)
⎩v e ⎭
Where [N] depends up on the numbering sequence defining nodal displacement in the
element using the expression of {ε} in eq. (3.3) with the expression of (f)
{ε} = [L][N]{U} (g)
We have the strains at any x, y locations in terms of nodal displacements. The matrix
[L][N] defines the [B] matrix given eq. (3.7).
Eq. (g) in to (c) is the final form for the first term of the variational function:
1 1
∫ 2 {U} [N] [L] [C][L][N]{U}dv = ∫ 2 {U} [B] [C][B]{U}dv
T T T T T
(h)
V V

¾ The second term of volume integral of eq. (3.8) represents the body force
and similar to heat-source term of problem (3.7.2.1). It becomes:
1
∫ 2 [U] [N]
T T
fdv (i)
V

ENSP Page 47/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
¾ The surface integral on eq. (3.8) represents a natural boundary condition
and is called a surface traction, a force boundary condition, or a distributed
pressure boundary condition. If {f} = [Tx Ty]T, then the surface integral in
matrix form is
∫ {U} [N] {T}ds
T T
(j)
V

2- The variational function in question 1- is minimized with respect to the displacement

vector as:
∂J
∫ [N]T [L]T [C][L][N]{U}dv − ∫ [N]T fdv − ∫ [N]T {T}ds = 0 (a)
∂{U} V V S

Using the figure in question (1-), eq. (e) of (1-) is the correct definition of {U}, and the
corresponding shape function matrix is:

[N ] = ⎡⎢
N1 0 N2 0 N3 0 N4 0 ⎤
(b)
⎣0 N1 0 N2 0 N3 0 N 4 ⎥⎦

The operator matrix [L] is defined by eq. (3.3), [B] matrix as the product [L][N] is a 3×8
matrix:
⎡∂N 1 ∂N 2 ∂N 3 ∂N 4 ⎤
⎢ 0 0 0 0 ⎥
∂x ∂x ∂x ∂x
[B] = ⎢⎢ 0 ∂N1 ∂y 0 ∂N 2
∂y
0
∂N 3
∂y
0 ∂N 4 ⎥
∂y ⎥
(c)
⎢∂N ∂N 1 ∂N 2 ∂N 2 ∂N 3 ∂N 3 ∂N 4 ∂N 4 ⎥
⎢ 1 ∂y ∂x ∂y ∂x ∂y ∂x ∂y ∂x ⎥⎦
⎣
The volume integral is changed to an area integral by assuming a constant thickness in
the z-direction. The stiffness matrix computed is:
[K ] = ∫ [B]T [C][B]t dA (d)
A

We use eq. (paragraph3.7.2.2.c) to compute the terms kij of the stiffness matrix with the
Cij matrix given by equations (3.6) and (3.7) for plane elasticity.
For example:
⎛ ∂N
a b
∂N ∂N ∂N 1 ⎞
K 11 = ∫ ∫ ⎜⎜ 1 C11 1 + 1 C 33 ⎟ t dx dy
0 0⎝
∂x ∂x ∂y ∂y ⎟⎠
⎛
a b
(b - y)
2
(a - x) ⎞
2
= ∫ ∫ ⎜ C11 2 2 + C 33 2 2 ⎟⎟ t dx dy
⎜
0 0⎝ a b a b ⎠
⎛C b C a⎞
= ⎜ 11 + 33 ⎟ t
⎝ 3a 3b ⎠
The stiffness matrix therefore is an 8×8 system matrix.
¾ The second term of eq. (a) is similar to the heat source term of eq. (f) of
paragraph (3.7.2.2).

ENSP Page 48/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
3- The surface integral is seen in the next question.
the surface integral has been minimized as eq. (a) of question (2-), and is:

∫ [N] {T}ds
T
(a)
S

Substituting eq. (b) of question (2-) in to (a-) above we obtain a general expression with
⎧Tx ⎫
T ⎨ ⎬ as:
⎩Ty ⎭
⎡c11b C33a C12 +C33 c11b C33a C12 −C33 c11b C33a -C12 −C33 c11b C33a C12 −C33 ⎤
⎢ 3a + 3b 4
− +
3a 6b 4
− −
6a 6b 4
−
6a 3b 4 ⎥
⎢ c33b C22a -C12 +C33 c33b C22a -C12 −C33 c33b C22a C12 −C33 c33b C22a ⎥
⎢ + − + − − − ⎥
⎢ 3a 3b 4 3a 6b 4 6a 6b 4 6a 3b ⎥
⎢ c11b C33a -C12 −C33 c11b C33a C12 −C33 c b C a C12 +C33 ⎥
⎢ + − − 11 − 33 ⎥
3a 3b 4 6a 3b 4 6a 6b 4
⎢ c33b C22a -C12 +C33 c33b C22a C12 +C33 c33b C22a⎥
⎢ + − − − ⎥
[ ]
K =t⎢
e

⎢
3a 3b 4 6a 3b
c11b C33a C12 +C33
4 6a 6b ⎥(f)
c11b C33a C12 −C33 ⎥
⎢ + − + ⎥
3a 3b 4 3a 6b 4
⎢ c33b C22a -C12 +C33 c33b C22a ⎥
⎢ (Sym) + − ⎥
⎢ 3a 3b 4 3a 6b ⎥
⎢ c11b C33a -C12 −C33 ⎥
⎢ + ⎥
3a 3b 4
⎢ c33b C22a ⎥
⎢ + ⎥
⎣ 3a 3b ⎦
to problem of paragraph (3.73.2)

ENSP Page 49/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN

⎡ N1 0⎤ ⎧ N 1Tx ⎫
⎢0 ⎪N T ⎪
⎢ N 1 ⎥⎥ ⎪ 1 y⎪
⎢N 2 0⎥ ⎪ N 2 Tx ⎪
⎢ ⎥ ⎪ ⎪
0 N 2 ⎥ ⎧Tx ⎫ ⎪ N 2 Ty ⎪
∫S ⎢⎢ N 3 ⎨ ⎬ds = ∫ ⎨
0 ⎥ ⎩Ty ⎭ S⎪
N T
⎬ds (b)
⎢ ⎥
3 x ⎪
⎢0 N3 ⎥ ⎪ N 3 Ty ⎪
⎢N ⎪ ⎪
0⎥ ⎪ N 4 x⎪
T
⎢ 4 ⎥
⎢⎣ 0 N 4 ⎥⎦ ⎪ ⎪
⎩ N 4 Ty ⎭
The element edge connecting nodes 3-4 corresponds to the edge y = b for local
coordinate as eq. (c) of paragraph (3.7.1.1), and substituting y = b Î N1 = N2 = 0; N3 =
x/a ; N4 = (a-x)/a. the surface traction are replaced with Tx = Px = 0 and Ty = Py. ds is
replaced by tdx; t = thickness. Integration is along x axis from 0 to a. after substitution
in eq. (b) and integration, the final load vector is

[
t 0 0 0 0 0 Py a
2
0 Py a
2
]
T
(c)

This result indicates that one-half of the pressure loading should be distributed to each
node.

3.7.4. Problem4
The plate shown is 15×25cm and 5mm thick and loaded with a tension of Px =
120N/mm2. Compute the nodal displacements, the strains and stresses using a one-
element using a one-element analysis. Assume E = 2,1×1011Pa; and solve the problem for
υ = 0,3 .

Solution4
The plate should be analyzed as plane stress. For the eight possible node displacements,
u1 = v1 = v2 = u2 = 0 are the boundary conditions.
The result obtained in problem of paragraph (3.73.3) indicates that the pressure loading
Px should be distributed to nodes 2 and 3as 1200N/mm2×15mm×5×1/2=4,5kN.
¾ The local stiffness matrix is given by eq. (e) of paragraph (3.7.3.2)
becomes the global stiffness matrix for this one-element model. The
material constants Cij are computed using the plane stress constitutive
equations of eq. (3.6):

ENSP Page 50/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
C11 = C 22 = E
(1 - υ ) = 2,3077 × 10
11
2 Pa

= υE
(1 - υ ) = 0,6923 × 10 Pa
11
C12 2

=E = 0,8077 × 10 Pa
2(1 + υ )
11
C 33

With zero-displacements in boundary conditions.

The reduced “global” stiff matrix equation is:

⎡ C11 b C 33 a C12 − C 33 C12 + C 33 ⎤

⎢ 3a + 3b
C11 b C 33 a
− ⎥ ⎧u2⎫ ⎧4,5KN⎫
⎢
6a 3b
C11 b C 33 a
4
C12 + C 33
4
C12 − C 33 ⎥ ⎪u ⎪ ⎪ ⎪
⎢ + ⎥ ⎪ 3⎪ ⎪4,5KN⎪
t⎢ 3a 3b 4 4 ⎥ ⎨ ⎬ ⎨ ⎬× t
⎢ C 33 b C 22 a C 33 b C 22 a ⎥ v
⎪ 3⎪= ⎪ 0
⎢ (sym ) 3a
+
3b 3a
+
6b ⎥
⎪
⎢ C 33 b C 22 a ⎥ ⎪⎩v4⎪⎭ ⎪⎩0 ⎪⎭
⎢ + ⎥
⎣ 3a 3b ⎦
Or
⎡0,91026 − 0,21795 − 0,02885 0,3750 ⎤ ⎧u 1 ⎫ ⎧45000⎫
⎢ ⎪ ⎪
⎢ 0,91026 0,3750 − 0,02885⎥⎥ ⎪u 3 ⎪ ⎪⎪45000⎪⎪
10 ×
11
⎨ ⎬=⎨ ⎬
⎢ (sym) 1,44359 0,35384 ⎥ ⎪v 3 ⎪ ⎪0 ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪⎭
⎣ 1,44359 ⎦ ⎩v 4 ⎭ ⎩0
u2 = u3 = 7,193×10-4mm; v3 = v4 = -1,385×10-4mm
¾ The strains are computed using eq. (c) of problem of paragraph (3.7.3.2)
and eq. (3.7).

ENSP Page 51/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
∂N 1 − 15 + y ∂N 1 - 25 + x
= ; = ;
∂x 375 ∂y 375
∂N 2 15 − y ∂N 2 − x
= ; = ;
∂x 375 ∂y 375
With x =a/2; y = b/2,
∂N 3 y ∂N 3 x
= ; = ;
∂x 375 ∂y 375
∂N 4 − y ∂N 4 25 − x
= ; = ;
∂x 375 ∂y 375
∂N1 −1 ∂N1 −1
= ; =
∂x 50 ∂y 30
∂N 2 1 ∂N 2 −1
= ; =
∂x 50 ∂y 30
∂N 3 1 ∂N 3 1
= ; =
∂x 50 ∂y 30
∂N 4 −1 ∂N 4 1
= ; =
∂x 50 ∂y 30
⎡0 ⎤
⎢0 ⎥
⎡ −1 1 1 −1 ⎤⎢ ⎥
⎢ 50 0 50 0 50 0
50
0 ⎥ ⎢ 7,193 × 10−4 ⎥
⎡ε xx ⎤ ⎢ ⎥⎢ ⎥ ⎡ 2,877 ⎤
⎢ ⎥ −1 −1 1 1 ⎥ ⎢0 ⎥ ⎢
ε = ⎢ε yy ⎥ = ⎢ 0 0 0 0 = −0,923⎥⎥ × 10−5
⎢ε ⎥ ⎢
⎢ 30 30 30 30 ⎥ 7,193 × 10 ⎥ ⎢
⎢ −4

⎥⎢ ⎥ ⎢0 ⎥⎦
⎣ xy ⎦ ⎢ −1 −1 −1 1 1 1 1 −1 ⎥ ⎢ −1,385 × 10−4 ⎥ ⎣
⎢⎣ 30 50 30 50 30 50 30 50 ⎥⎦ ⎢ 0 ⎥
⎢ ⎥
⎢ −1,385 × 10−4 ⎥
⎣ ⎦

The stresses are computing using eq.(3.5) to find that

⎧σ xx ⎫ ⎡ ⎤ ⎡6 ⎤
⎪ ⎪ ⎢ ⎥ ×10−4 = ⎢ 0,13⎥ MPa
⎨σ yy ⎬ = D x ⎢ ⎥ ⎢ ⎥
⎪ ⎪ ⎢⎣ ⎥⎦ ⎢⎣ 0 ⎥⎦
⎩σ xy ⎭

3.8. Homework number3

3.8.1.
Derive a local stiffness matrix for heat transfer using the three-node triangular element
as it was done for a four-node quadrilateral element.

schema

ENSP Page 52/103

FINITE ELEMENT METHODS / 3. Two – dimensional Finite Element Version ou da
Dr LEZIN
3.8.2.
Derive the shape functions for the nine-node rectangular element shown.
a= x3- x1; b= y3 – y1.

Schema

The results of homework (2-1) can be used for one side of the rectangular element. If Nix
an Niy are such results in specific direction, then the shape function here is Ni = Nix×Niy.

3.8.3.
An eight-node rectangular element has node numbers in sequence with corner
numbered 1, 3, 5, 7 and midside nodes numbered 2, 4, 6, 8 with corner nodes appearing
first in the solution vector. Derive the transformation matrix that will renumber the
element to conform with element numbers in problem (II-), with corner nodes still
appearing first in the solution vector (neglecting the ninth node).

3.8.4.
The plate shown is a 60×100×1(cm3) loaded with a tension of Px = 12kN/mm3.Using a 2-
quadrilateral element model, compute nodal displacements and strains and stresses in
each element. Assume E = 2,1×1011Pa ; υ = 0,3.

ENSP Page 53/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN

4. BEAM and FRAME FE. (Plane)

4.1. Introduction:
A bar element in a pane is called a truss element. A plane truss contains 2-DOF
per node (u, v), axial (u) and transverse (v) displacements. By definition, a bar element can
carry only axial loads and deform axially, where as beams can take transverse loads and
bending moments about an axis perpendicular to the plane of the member . a superposition of
beam an bar FE give a frame FE as it is shown. Members of a frame structure are connected
by rigid connection ⇒ axial and transverse forces and bending moments are developed in the
members. ⇒ A frame element has 6 DOF in the plane and 12 DOF in the space coordinate
system.
(Schema)

4.2. Governing Differential Equation

The governing differential equation for beam deflection in 2-dimension is derived
in any textbook on Structural Mechanics. Assuming small deflection theory, the governing
equation is written in term of deflection of the beam v(x) as:
d 4 v( x)
EI = ω ( x) (4.1)
dx 4

E and I can also be function of x, but are assumed to

be constant.

The assumption that constitute elementary beam the lead to a linear theory for beam analysis.
Each derivative of the deflection has a particular meaning:
v(x) = deflection
dv( x)
= θ ( x) = slope
dx

d 2 v( x)
EI = M ( x) = bending moment (4.2)
dx 2

d 3 v( x) dM ( x)
EI 3
= = V ( x) = transverse shear force
dx dx

d 4 v( x) dV ( x)
EI 4
= = ω ( x) = distributed load
dx dx

The differential equations (4.2) are based upon a classical beam sign convention necessary for
successful development of the derivation of the beam FE.

4.3. The Displacement Method for Beam Analysis

The displacement (or stiffness) method, for 2-dimension beam, is used to solve for
the displacements and rotations (slopes) at each end of the beam. The idea is then to write an
equation that can be solved to obtain the magnitude of the force or moment reaction in the
terms of end displacements and rotations. First, assume that any beam to be idealized, is fixed
at both ends (fixed-fixed beam); and is subjected to an external applied loading. The reactions

ENSP Page 54/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
of such a beam can be computed, but are given in table 4.1. This procedure will be seen in
future problems.

4.4. Beam Finite Elements:

The analysis of 2-dimension beams using FE formulation is identical to matrix
analysis of structures. The derivation of the stiffness matrix is based upon defining shape
function that satisfied the governing differential equation and fixed-fixed boundary
conditions.
A cubic displacement can be assumed in the form:

v( x) = a 0 + a1 x + a 2 x 2 + a3 x 3 (4.3)

a i are computed to satisfy boundary conditions.

A general variational function was defined by equation 3.8, and can be used to define the
proper potential energy function for deriving beam FE. Hermite’s interpolation formula is
used for deriving the shopped function that can be written as:

n n
dv i
Ni = ∑ U ( x)v + ∑ W ( x) dx
i =0
i i
i =0
i (4.4)

dv i
vi , are deflection and slope at xi. The polynomial is of degree 2n+1, and for a cubic
dx
equation n = 1. Ui(x) and Wi(x) are polynomials with properties:

⎡ dL(xi ) ⎤
(x − xi )⎥ [ Li (x)]
2
Ui (x) = ⎢1− 2
⎣ dx ⎦ (4-5)
Wi (x) = (x − xi )[ Li (x)]
2

Li(x) are ith Lagrange coefficient polynomial of degree

n.

4.5. Matrix transformation and Frame elements

Matrix transformations that can described a boundary condition on
plane elasticity problems has been discussed in §3.5. Since each element has its own specific
orientation, the transformation from local coordinate to global coordinate of the element
stiffness matrix and force vector are given as:

ENSP Page 55/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN

[K ] = [T ] [K ] [T ]
e
G
T e
L
(4.6)
{F } = [T ] {F }
e
G
T e
L

e = element
L = local coordinate system
G = global coordinate system.

As a particular case of equation 3.12, since the slope coordinates are invariant with axis
rotations α, will be:

⎡ cos α sin α 0 ⎤
⎢− sin α cos α 0 0 ⎥
⎢ ⎥
[T ]
e
⎢ 0
=⎢
0 1
cos α sin α 0⎥
⎥
⎥ (4.7)
⎢
⎢ 0 − sin α cos α 0⎥
⎢ ⎥
⎣⎢ 0 0 1⎦⎥
coordinate DOF from §4.1

4.6. Problems

4.6.1. Problem 1:
1°) The beam shown (with θ1 > 0 ) is important in the derivation
schema
of the equations for the stiffness matrix for beam FE it is also
It is also useful when interpreting results of the final beam analysis.
schema
Use the differential equations defined by equation (4.1) to solve for shear and moment
reactions for this fixed-fixed beam.
2°) The following beam (v1>0) schema
Is important for the derivation of the stiffness matrix for beam FE. Use the differential
equation (4.1) to solve for shear and moment reaction for this fixed-fixed beam
3°) This beam can be used to identify the end displacements an rotations for a beam
subjected to the ends forces and bending moments shown. schema
Construct a set of four equations that relate each force at moment action to the
displacements and rotations
schema

4.6.2. Problem 2
schema

ENSP Page 56/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
This two-span beam is fixed at both ends and supported between the end with a simple
support that allows rotation. Compute the rotation at the simple support at reactions at
all supports. Construct the corresponding shear and moment diagrams.

4.6.3. Problem 3
1) Use Hermite’s interpolation formula to derive cubic shape functions for the
transverse deflection of beam
2°) Derive the local FE stiffness matrix for a beam with combined transverse loading
and axial force.
3°) Derive the corresponding stiffness matrix for a beam oriented in a local ε , η
coordinate system and referenced to the global coordinate x, x.
schema

4.6.4. Problem 4
The given frame in fixed at node 1 and free to translate in the x-direction atr node 3.
compute the displacements and reaction at all nodes.

Solution1
1°) In the case equation (4.1) will be
EIv ′′′' − w = 0 or v ′′′' = 0 ( a)
The beam is statically indeterminate (hypperstatique) of degree of indeterminacy 2 ⇒ we
need 2 boundary conditions (b.c) to solve the differential equations. Integrating (a) ⇒
v′′′ = c ,

From equation EIv ′′′ = v( x) and at x = 0 , V (0) = V = R1 and equation (b) will be:

R1
v ′′′ = (c)
EI
x
after integrating of (c) v ′′ = R1 + c2 (d)
EI
The boundary condition using our structure is EIv ′′(0) = − M and substituting in equation (d)
and solving for c2 gives:

ENSP Page 57/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
x M1
v ′′ = R1 − (e)
EI EI
We assume as a joint sign convention far beam FE that ( M ) Q↑+ moments rotating
+
counter clock wise are positive and shear forces acting on the joint upward in the (+y)
direction are positive.
Integrating equation (e) twice gives equations for the slope θ(x) and for the deflection
v(x):
R1 2 M 1 x
v′ = x − + c3
2 EI EI
(f)
R1 3 M 1 x 2
v= x − + c 3 x + c4
6 EI 2 EI
at x = 0, v' (0) = θ1 , v(o) = 0
The boundary condition into (f ) gives:
6 EI 4 EI
R1 = 2
θ1 and M 1 = θ1 (g)
L L
There are again 2 additional equations in static ( ∑ f y = 0 and ∑ M z = 0 ) that can be use
to find R2 and M2. Otherwise, we can use equation (4.2) with x =L into equations (c) and
(e), and find that
6 EI 2 EI
R2 = − 2
θ1 and M 2 = θ1 (h)
L L
2°) Beginning again with equation (4.1) and integrating twice results in
v ′′′' = 0
v ′′′ =c 1 (a)
v ′′ =c 1 x + c 2 (b)
The appropriate boundaries conditions are EIv ′′′(o) = R1 and EIv ′′(o) = − M 1 into (a) and
(d), after solving for ci gives
x M1
v′′ = R1 − (c)
EI EI
With a other integration
R1 2 M 1 x
v′ = x − + c3 (d)
2 EI EI
R M x2
v = 1 x3 − 1 + c 3 x + c4 (e)
6 EI 2 EI
The boundary conditions are v’(0) = 0 and v(0) = v1 in (e) and (d) help us to find c3 = 0
and c4 = v1. v’(L) = 0 and v(L) = 0 are used with equations (d) and (e) to obtain 2
equations that can be solved for M1 and R1
12 EI 6 EI
R1 = 3
v1 and M 1 = 2 v1 (f)
L L

ENSP Page 58/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
Static equations are used to compute R2 and M2
12 EI 6 EI
R2 = 3
v1 and M 2 = 2 v1 (g)
L L

3°) It is desirable, first, to write an equation that relates R1 to each of the displacement
and rotations of the given figure.
R1 = f (v1 ,θ1 , v2 ,θ 2 ) (a)
Let us assume, before assigning boundary conditions to the beam, that both supports are
fixed. The function f(v1) was evaluate in question 2°) and the result is used there.
Similarly R1 is affected by f(θ1) computed in question 1°). f(v2) and f(θ2) are computed in
the same way.
⇒ The function of equation (a) using the principle of superposition is written as:

R1 = R1v1 + R1θ1 + R1v2 + R1θ 2

12 EI 6 EI 12 EI 6 EI
= 3
v1 + 2 θ1 − 3 v 2 + 2 θ 2 (b)
L L L L
Similarly, M 1 = f (v1 ,θ1 , v 2 ,θ 2 ) and the functions are again evaluated using the result of
question 1°) and 2°). To find that:
6 EI 4 EI 6 EI 2 EI
M1 = 2
v1 + θ1 − 2 v2 + θ2 (c)
L L L L
The reactions are evaluated at the right end of the beam in a similar manner:
12 EI 6 EI 12 EI 6 EI
R2 = − 3
v1 − 2 θ1 + 3 v2 − 2 θ 2 (d)
L L L L
6 EI 2 EI 6 EI 4 EI
M 2 = 2 v1 + 2 θ1 − 3 v2 − 2 θ 2 (e)
L L L L
A more compact form of equation (b) and (e) is given in the matrix form as:

⎧ R1 ⎫ ⎡ 12 6 L − 12 6 L ⎤ ⎧ v1 ⎫
⎪M ⎪ ⎢ 6 L 4 L2 − 6 L 2 L2 ⎥ ⎪θ ⎪
⎪ 1 ⎪ EI ⎢ ⎥ ⎪⎨ 1 ⎪⎬
⎨ ⎬= 3 (f)
⎪ R2 ⎪ L ⎢− 12 − 6 L 12 − 6 L ⎥ ⎪v 2 ⎪
⎪⎩M 2 ⎪⎭ ⎢ ⎥
⎣ 6L 2L
2
− 6 L 4 L2 ⎦ ⎪⎩θ 2 ⎪⎭

or {f } = [K e ]{V } (g)
[Ke] = beam elements stiffness matrix in local coordinates
{f} = force vector in local coordinate
{V} = displacement vector in local coordinates.

Solution2

ENSP Page 59/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
Two local elements are combined to form a global stiffness matrix according to
the procedure explained in equation (2.14). The two beams have a common rotation and
displacement at support 2.
According to the displacement method, we assume that the beam is fixed–fixed
and compute equivalent joint given in table 4.1. Apply boundary conditions to the fixed-
fixed beam to make it appear as the simple beam that is being analysed. The given
boundaries are deleted if they are zero. Solve for the remained displacements and
replace them in the stiffness equation to find reaction support. The local stiffness
matrices are constructed first from equation (f) of problem §4.6.1.3. The equivalent
point loads for the applied loads are obtained from table 4.1.
Beam 1-2
⎧ − ωL 2 ⎫ ⎡ 12 6 L − 12 6 L ⎤ ⎧ v1 ⎫
⎪− ωL2 12⎪ ⎢ 6 L 4 L2 − 6 L 2 L2 ⎥ ⎪θ ⎪
⎪ ⎪ EI ⎢ ⎥ ⎪⎨ 1 ⎪⎬
⎨ ⎬= 3 (a)
⎪ − ωL 2 ⎪ L ⎢− 12 − 6 L 12 − 6 L ⎥ ⎪v 2 ⎪
⎪⎩ ωL2 12 ⎪⎭ ⎢ ⎥
⎣ 6L 2L
2
− 6 L 4 L2 ⎦ ⎪⎩θ 2 ⎪⎭

Beam 2-3
⎧0⎫ ⎡ 12 6( 2 L ) − 12 6( 2 L ) ⎤ ⎧ v 2 ⎫
⎪0⎪ ⎢6(2 L) 4(2 L) 2 − 6(2 L) 2(2 L) 2 ⎥ ⎪θ ⎪
⎪ ⎪ EI ⎢ ⎥ ⎪⎨ 2 ⎪⎬
⎨ ⎬= (b)
⎪0⎪ (2 L) ⎢ − 12 − 6(2 L) − 6( 2 L ) ⎥ ⎪ v 3 ⎪
3
12
⎪⎩0⎪⎭ ⎢ 2 ⎥
⎣6(2 L) 2(2 L ) − 6(2 L) 4(2 L) ⎦ ⎪⎩θ 3 ⎪⎭
2

The [Ke] are combined through v2 and θ2 to give a 6 × 6 global stiffness matrix [k].

ENSP Page 60/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
⎡ 12 6L − 12 6L 0 0 ⎤
⎢ 6l 4L 2
− 6L 2L 2
0 0 ⎥⎥ v1
⎧ − ωL 2 ⎫ ⎢
⎪− ωL2 12⎪ 12 12 L 12 12 L
⎢− 12 − 6 L 12 + − 6L + − ⎥ θ ⎧v ⎫
⎪ ⎪ ⎢ 8 8 8 8 ⎥ 1 ⎪ 1⎪
⎪⎪ − ωL 2 ⎪⎪ EI ⎢ 12 L 16 L2 12 L 8 L2 ⎥ v 2 ⎪θ1 ⎪
⎨ ⎬ = 3 ⎢ 6 L 2 L2
− 6 L + 4 L2
+ ⎨ ⎬ (c)
⎪ ωL 12 ⎪ L ⎢ 8 ⎥⎥ θ 2 ⎪v 2 ⎪
2
8 8 8
⎪ 0 ⎪ ⎢ 0 12 12 L 12 − 12 L ⎥ v ⎪θ ⎪
0 − − 3 ⎩ 2⎭
⎪ ⎪ ⎢ 8 8 8 8 ⎥
⎪⎩ 0 ⎪⎭ ⎢ θ
12 L 8L2 12 L 16 L2 ⎥ 3
⎢ 0 0 − − ⎥
⎣ 8 8 8 8 ⎦
The boundary conditions for the given beam structure are v1 = v2 = v3 = θ1 = θ 3 . The
solution reduces to one equation and one unknown θ2.
EI ωL2 ωL3
( 4 L2
+ 2 L2
)θ 2
= or θ 2 = (d )
L3 12 72 EI
The result for each beam element must be computed using the element local stiffness
matrix and found unknown and boundary conditions values:
Beam 1-2:

⎧ VA ⎫ ⎧ 0 ⎫ ⎧ ωL 2 ⎫
⎪M ⎪ ⎪ 0 ⎪ ⎪ ⎪
⎪ ⎪ ωL 12 ⎪
2
⎪ A ⎪ EI e
⎨ [
⎬ = 3 K 1− 2 ]
⎪
⎨ 0 ⎬+⎨ ⎬ (e)
⎪ VB ⎪ L ⎪ ωL3 ⎪ ⎪ ωL 2 ⎪
⎪⎩M B ⎪⎭ ⎪ ⎪ ⎪− ωL2 12⎪⎭
⎩ 72 EI ⎭ ⎩
And multiplication gives:
7ωL 5ωL ωL2 − ωL2
VA = RA = ; VB1 = ; MA = ; MB = ; ( RB = VB1 + VB 2 )
12 12 9 36
Similarly, the stiffness matrix of equation b) is used to computed shears and moments
for beam 2-3:

⎧VB 2 ⎫ ⎧ 0 ⎫ ⎧0⎫
⎪M ⎪ ⎪ ωL3 ⎪ ⎪ ⎪
⎪ B⎪
⎨ [
⎬ = K 2 −3
e
]
⎪ ⎪ ⎪0⎪
⎨ 72 EI ⎬ + ⎨ ⎬ (f)
⎪ VC ⎪ ⎪ 0 ⎪ ⎪0⎪
⎪⎩M C ⎪⎭ ⎪ 0 ⎪ ⎪⎩0⎪⎭
⎩ ⎭
ωL ωL ωL2 ωL2
Or VB 2 = ; Vc = RC = − ; MB = ; MB = ; RB = VB1 = VB 2 = .....
48 48 36 72

The stress caused by bending moment is derived in strength materials (RDM) as

M(x).y
σx = (b)
I
Using Hooke’s law, eq.(b) and eq.(4.2) we obtain

ENSP Page 61/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
σ (x) M(x).y d 2 v(x)
ε (x)= = =y (c)
E EI dx 2
And (a) can be written as
y 2 ⎛ d 2 v(x) ⎞
L L
J (v ) = ∫ ∫ ⎜ ⎟EdAdx-∫ ω vdx
A 0
2 ⎝ dx 2 ⎠ 0

∫ y dA=I we get
2
Knowing that
A

L L
1
J (v) = ∫ EI(v′′) 2 dx − ∫ ω vdx (d)
20 0

The 2nd derivate in eq. (d) is defined in terms of shape function Ni and joint displacement
given by eq. (g) of question (1- ). Then we have v′′ = [ N ′′]{v} in to eq. (d) gives the
potential energy in matrix form:
L L
1
∫ {v} [ N ′′] [ EI][ N ′′]{v} dx − ∫ {v} [ N ′′] ω dx (e)
T T T T
J (v ) =
20 0

Minimizing J(v) with respect to {v}, we obtain the matrix eq. that defines the local beam
finite element:
L L
J (v) = ∫ [ N ′′] [ EI ][ N ′′]{v} dx − ∫ [ N ] ω dx = 0 (f)
T T

0 0

d 2 Ni
Substituting the values of Ni from question (1-) (eq. (e) to (f)) and results in the
dx 2
matrix equation

⎡ 6 12x ⎤ ⎧ x2 x3 ⎫
⎢ L2 + L3 ⎥ ⎪1-3 2 + 2 3 ⎪
⎢ ⎥ ⎪ L2 L ⎪
⎧ v ⎫ ⎪ x3 ⎪
L −
⎢ 4 + 6x ⎥ ⎪
1
⎪ x-2
x
+
⎢ L L3 ⎥ ⎪θ1 ⎪ ⎪⎪ L2 L3 ⎪⎪ ω dx (g)
∫0 ⎢ 6 12x ⎥ [ EI ][ N ′′ ] ⎨
v
⎬ dx= ⎨ 2 3 ⎬
⎪ ⎪ ⎪ x x
⎢ − ⎥ 2
3 −2 3 ⎪
⎢ L2 L3 ⎥ ⎪⎩θ 2 ⎪⎭ ⎪ L2 L ⎪
⎢ 2 6x ⎥ ⎪ 3 2 ⎪
⎢− + 2 ⎥ ⎪x − x ⎪
⎣ L L ⎦ ⎩⎪ L3 L2 ⎭⎪
Performing the matrix multiplications integrating will give the stiffness matrix which
can be compared with eq. (f) of problem of paragraph (4.6.1.3). the matrix of equivalent
joint loading corresponds to the uniformly loaded fixed-fixed beam of table 4.1.

ENSP Page 62/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
⎧ω L ⎫
⎧ v ⎫ ⎪ 2 ⎪
⎡ 12 6L -12 6L ⎤ 1
⎢ 6L 4L2 -6L 2L2 ⎥ ⎪θ ⎪ ⎪ω L ⎪
2
EI ⎢ ⎪ ⎪
⎥⎨ 1⎬= ⎨⎪ 2 ⎪ (h)
⎬
L3 ⎢ -12 -6L 12 -6L ⎥ ⎪v2 ⎪ ⎪ω L
⎢ 2⎥⎪ 2 ⎪
⎣ 6L 2L -6L 4L ⎦ ⎩θ 2 ⎭
2 ⎪ ⎪ ⎪
⎪− ω L ⎪
2

⎩ 2⎭
schema

The stiffness matrix for axial force acting on a rod (or a beam) was derived as an
example of elementary elasticity in problem of paragraph (2.9.2.4) eq.(d). Replacing
body forces {fe} by axial forces N1 and N2, we can write:

AE ⎡ 1 −1⎤ ⎧u1 ⎫ ⎧ N1 ⎫
⎨ ⎬ = ⎨ ⎬ (a)
L ⎢⎣ −1 1 ⎥⎦ ⎩u 2 ⎭ ⎩ N 2 ⎭
The complete local stiffness matrix is obtained by combining eq. (a) above and eq. (f) of
problem of (paragraph 4.6.1.3) and using the procedure in paragraph 2.5.
⎧ N1 ⎫ ⎡ C1 0 0 -C1 0 0 ⎤ ⎧ u1 ⎫
⎪R ⎪ ⎢ ⎪ ⎪
⎪ 1 ⎪ ⎢ 0 12C 2 6C 2 L 0 -12C 2 6C 2 L ⎥⎥ ⎪v1 ⎪
⎪⎪ M 1 ⎪⎪ ⎢ 0 6C 2 L 4C 2 L 0 -6C 2 L 2C 2 L2 ⎥ ⎪⎪θ1 ⎪⎪
2

⎨ ⎬ ⎢ = ⎥ ⎨ ⎬ (b)
N
⎪ ⎪ ⎢2 -C1 0 0 C1 0 0 ⎥ ⎪u 2 ⎪
⎪ R2 ⎪ ⎢ 0 -12C 2 -6C 2 L 0 12C 2 -6C 2 L ⎥ ⎪v2 ⎪
⎪ ⎪ ⎢ ⎥⎪ ⎪
2
0 -6C 2 L 4C 2 L2 ⎦⎥ ⎩⎪θ 2 ⎪⎭
⎣⎢ 0 6C 2 L 2C 2 L 244444444
⎩⎪ M 2 ⎭⎪ 144444444 3
⎡Ke ⎤
⎣ ⎦

AE AE
Where C1 = and C2 = 3 . The axial ui transverse vi deformations are uncoupled for
L L
beams when the local axis of the beam coincides with the global axis.
3- The beam is shown with local and global coordinates system. Upon rotation of axis we
can use the transformation equation given in equation (4.6) directly, using the local
matrix for the beam with combined axial force as derived equation (b) of question 2.

ENSP Page 63/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
⎡⎣ K e ⎤⎦ = [ T ] [T ]
T
⎡⎣ K e ⎤⎦ (a)
G
123L
(b) of question 2.

or
⎡⎣ K e ⎤⎦ = [ T ] ⎡⎣ K e ⎤⎦ [ T ]
T
xy Gη

⎡ 2 12I 2 12I 6I 12I 12I 6I ⎤

⎢ Ac + L2 s (A-
L2
)cs -
L
s −(Ac 2 + 2 s 2 )
L
-(A- 2 )cs
L
- s⎥
L
⎢ ⎥
⎢ 12I 6I 12I 12I 2 6I ⎥
As 2 + 2 c 2 c -(A- 2 )cs −(As + 2 c )
2
c
⎢ L L L L L ⎥
⎢ 6I 6I ⎥
E⎢ 4I s - c 2I ⎥
⎡⎣ K e ⎤⎦ = ⎢ L L ⎥
xy L⎢
12I 12I 6I ⎥
⎢ Ac 2 + 2 s 2 (A- 2 )cs s⎥
⎢ L L L ⎥
⎢ 12I 6I ⎥
⎢ ( sym.) Ac 2 + 2 s 2 - c⎥
⎢ L L ⎥
⎣⎢ 4I ⎦⎥
c = cosα
s = sinα
A = cross-sectional area
I = moment of inertia.

Solution4
schema

First the stiffness matrix and corresponding displacement matrix are evaluated for both
elements:
Element1: x,y and ξ, η coincide Î α = 0 Î c = cos0° = 1; s = sin0° = 0.
Also for both elements:
12I 12 × 6130 × 10−8
2
= 2
= 73,56 × 10−7 m 2
L 10
6I 6 × 6130 × 10−8
= = 36, 78 × 10−6 m3
L 10
E 2,1 × 1011
= = 2,1 × 1010 Pa
L 10 m

ENSP Page 64/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
Substituting eq.(b) of problem of paragraph (4.6.3.3) gives the stiffness matrix:
A = 7640000×10-10
⎡ 76400000 0 0 −76400000 0 0 ⎤ ⎛ u1 ⎞
⎢ ⎜ ⎟
⎢ 0 73560 367800 0 −73560 367800 ⎥⎥ ⎜ v1 ⎟
⎢ 0 367800 2452000 0 −367800 1226000 ⎥ ⎜θ1 ⎟
2,1 ⎢ ⎥ ⎜ ⎟ (a)
⎢ −76400000 0 0 764000 0 0 ⎥ ⎜ u2 ⎟
⎢ 0 −73560 −367800 0 73560 −367800 ⎥ ⎜ v2 ⎟
⎢ ⎥⎜ ⎟
0 367800 1226000
⎣14444444444444 0 −367800 2452000 ⎦ ⎜⎝θ 2 ⎟⎠
4244444444444444 3
⎡ K1-2
e ⎤
⎣ ⎦

Element2

schema

The element axis is assumed to be directed from node2 to node3; by inspection the angle
between the local axis and the global axis x axis is α = 270°. Then c = cos270° = 0,
s = sin270° =-1.
And in to eq. (b) of paragraph (2.6.3.3). We get the stiff
⎡ 73560 0 367800 -73560 0 367800 ⎤ ⎛ u 2 ⎞
⎢ 0 764000 0 0 −76400000 0 ⎥⎜v ⎟
⎢ ⎥⎜ 2 ⎟
⎢ 367800 0 2452000 −367800 0 1226000 ⎥ ⎜θ 2 ⎟
2,1 ⎢ ⎥ ⎜ ⎟ (b)
⎢ −73560 0 −367800 73560 0 −36700 ⎥ ⎜ u 3 ⎟
⎢ 0 −764000 0 0 764000 0 ⎥ ⎜ v3 ⎟
⎢ ⎥ ⎜⎜ ⎟⎟
⎣ 0 −764000 1226000 − 367800 0 2452000 θ
1444444444444 424444444444444 3⎦ ⎝ 3 ⎠
⎡ K e2-3 ⎤
⎣ ⎦
The global stiff matrix is obtained by combing equations (a) and
(b)
⎡ 764000 0 0 −764000 0 0 0 0 ⎤ ⎛ u1 ⎞
0
⎢ 0 73560 367800 0 −73560 367800 0 0 ⎥⎜v ⎟
0
⎢ ⎥⎜ 1 ⎟
⎢ 0 367800 2452000 0 −367800 226000 0 0 0 ⎥ ⎜θ1 ⎟
⎢ ⎥⎜ ⎟
⎢−76400000 0 0 837560 0 367800 −73560 0 367800 ⎥ ⎜ u2 ⎟
2,1⎢ 0 −73560 −367800 0 76473560 −367800 0 −764000 0 ⎥ ⎜ v2 ⎟
⎢ ⎥⎜ ⎟
⎢ 0 367800 1226000 367800 −367800 4904000 −367800 0 12260000⎥ ⎜θ2 ⎟
⎢ 0 0 0 −73560 0 −367800 73560 0 −367800 ⎥ ⎜⎜ u3 ⎟⎟
⎢ ⎥
⎢ 0 0 0 0 −764000 0 0 764000 0 ⎥ ⎜ v3 ⎟
⎢ ⎜ ⎟
⎣ 0 0 0 367800 0 1226000 −367800 0 2452000 ⎥⎦ ⎝θ3 ⎠

ENSP Page 65/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN
⎧0 ⎫
⎪ ⎪
⎡u1 ⎤ ⎪− wL ⎪ ⎧0 ⎫
⎢ ⎥ ⎪ 2 ⎪ ⎪
⎪
⎢ v1 ⎥ ⎪− wL2 ⎪ ⎪−60000 ⎪
⎢θ1 ⎥ ⎪ 12 ⎪ ⎪−105 ⎪
⎢ ⎥ ⎪0 ⎪ ⎪0 ⎪
⎢ u2 ⎥ ⎪ ⎪ ⎪ ⎪
⎢v ⎥ = ⎪⎨− wL ⎪⎬ = ⎪⎨−60000 ⎪⎬
⎢ 2 ⎥ ⎪ 2
⎪ ⎪ ⎪
⎢θ 2 ⎥ ⎪ wL ⎪ ⎪+105 ⎪
2
⎢u ⎥ ⎪ −
12 ⎪ ⎪0 ⎪
⎢ 3 ⎥ ⎪0 ⎪ ⎪ ⎪
⎢ v3 ⎥ ⎪ ⎪ ⎪ 0 ⎪
⎢ ⎥ ⎪0 ⎪ ⎪ ⎪⎭
⎣θ3 ⎦ ⎩ 0
⎪0 ⎪
⎪⎩ ⎪⎭
(Tab.4.1) ( N , m) (c)
Since
u1 = v1 = θ1 = v3 = 0, these displacement are remouved to obta in ed the reduced stiffness equation as
⎡ 76473650 0 367800 −73560 367800 ⎤ ⎧u2 ⎫ ⎧0 ⎫
⎢ ⎥ ⎪v ⎪ ⎪−60000 ⎪
⎢ 0 76473560 −367800 0 0 ⎥ ⎪⎪ 2 ⎪⎪ ⎪⎪ ⎪
⎪
2,1 × ⎢ 367800 −367800 4904000 −367800 1226000 ⎥ ⎨θ 2 ⎬ = ⎨+10 5
⎬ (d )
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ −73560 0 −367800 73560 −367800 ⎥ u3
⎪ ⎪ ⎪
0
⎪
⎢⎣ 367800 0 1226000 −367800 2452000 ⎥⎦ ⎪⎩θ3 ⎪⎭ ⎪⎩0 ⎪⎭

Solving for displacements in (d) gives (N,m, rad)

⎧u2 ⎫ ⎧−2,917 × 10−12

⎪ ⎪ ⎪
⎪⎪v2 ⎪⎪ ⎪⎪−0, 0002804
⎨θ 2 ⎬ = ⎨0, 01938 (m)rad
⎪u ⎪ ⎪0,1938
⎪ 3⎪ ⎪
⎪⎩θ3 ⎪⎭ ⎪⎩0, 01938

Substituting the displacement into equation each element matrix equation with the
added joint loading from table 4.1, we obtain the corresponding
Element 1-2
⎧u1 ⎫ ⎧0 ⎫ ⎧0 ⎫ ⎧ N1 ⎫ ⎧0 ⎫
⎪v ⎪ ⎪15012,9 ⎪ ⎪60000 ⎪ ⎪V ⎪ ⎪ ⎪
⎪1⎪ ⎪ ⎪ ⎪ ⎪ ⎪ 1 ⎪ ⎪75012,9 ⎪
⎪⎪θ ⎪⎪ ⎪⎪50106 ⎪⎪ ⎪⎪10 ⎪⎪ ⎪⎪ M 1 ⎪⎪ ⎪150106 ⎪
5
⎡⎣ K1e− 2 ⎦⎤ = ⎨ 1 ⎬ = ⎨ ⎬+⎨ ⎬=⎨ ⎬=⎨ ⎬
⎪u2 ⎪ ⎪0 ⎪ ⎪0 ⎪ ⎪ N 2 ⎪ ⎪0 ⎪
⎪v2 ⎪ ⎪−15012,9 ⎪ ⎪60000 ⎪ ⎪V2 ⎪ ⎪44987,1 ⎪
⎪ ⎪ ⎪ 5 ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎪⎩θ 2 ⎪⎭ ⎩⎪10 ⎪⎭ ⎩⎪−105 ⎭⎪ ⎪⎩ M 2 ⎪⎭ ⎩0 ⎭

ENSP Page 66/103

FINITE ELEMENT METHODS / 4. BEAM and FRAME FE. (Plane) Version ou da
Dr LEZIN

Element 2-3
⎧u2 ⎫ ⎧0 ⎫ ⎧0 ⎫ ⎧ N 2 ⎫ ⎧0 ⎫
⎪v ⎪ ⎪ ⎪ ⎪ ⎪ ⎪V ⎪ ⎪ ⎪
⎪ 2 ⎪ ⎪44982 ⎪ ⎪0 ⎪ ⎪ 2 ⎪ ⎪−44982 ⎪
⎪⎪θ ⎪⎪ ⎪0 ⎪ ⎪0 ⎪ ⎪⎪ M 2 ⎪⎪ ⎪0 ⎪
⎡⎣ K 2e−3 ⎦⎤ = ⎨ 2 ⎬ = ⎨ ⎬+⎨ ⎬ = ⎨ ⎬ = ⎨ ⎬
⎪u3 ⎪ ⎪0 ⎪ ⎪0 ⎪ ⎪ N 3 ⎪ ⎪0 ⎪
⎪v3 ⎪ ⎪449892 ⎪ ⎪0 ⎪ ⎪V3 ⎪ ⎪44982 ⎪
⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪
⎩⎪θ3 ⎭⎪ ⎩0 ⎭ ⎩0 ⎭ ⎩⎪ M 3 ⎭⎪ ⎩0 ⎭
The results of this last equation must be transformed back to the local system for final
interpretation since the z [local and global] coordinate system are differentiable
equation (4.7) is used as
[V2−3 ]η ,ε = [T ]{V }xy

⎧ N 2 ⎫ ⎡ 0 −1 0 0 0 0 ⎤ ⎧0 ⎫ ⎧0 ⎫
⎪V ⎪ ⎢
⎪ 2 ⎪ ⎢1 0 0 0 0 0 ⎥ ⎪−44982 ⎪ ⎪44982 ⎪⎪
⎥ ⎪ ⎪ ⎪
⎪⎪ M 2 ⎪⎪ ⎢0 0 1 0 0 0 ⎥ ⎪0 ⎪ ⎪0 ⎪
⎨ ⎬=⎢ ⎥⎨ ⎬=⎨ ⎬
⎪ N3 ⎪ ⎢0 0 0 0 −1 0 ⎥ ⎪ 0 ⎪ ⎪0 ⎪
⎪V3 ⎪ ⎢0 0 0 1 ⎥ ⎪
0 0 44982 ⎪ ⎪ −44982 ⎪
⎪ ⎪ ⎢ ⎥⎪ ⎪ ⎪ ⎪
⎩⎪ M 3 ⎭⎪ ⎣0 0 0 0 0 0 ⎦ ⎩0 ⎭ ⎩0 ⎭
The final results are in a F.B.D (free body diagram) for each element
schema

ENSP Page 67/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN

5. ISOPARAMETRIC FE
5.1. INTRODUCTION
FEM always requires the transformation of the problem of coordinate system from local to
global, or to normalized coordinate system.
schema
The transformation between x and ε an be represented by the linear “stretch” transformation
x = a + bε
1
and b = ( xb − x a ) = 1 L
2 2
1 1
a = ( xb + x a ) = x a + L
2 2
1
x(ε ) = x a + L(1 + ε )
Hence we have 2 (5.1)

The local coordinate ε as normalized is useful for its simplicity. Convenience in construction
and computations.
The interpolation (or shape function) has the following property as it was seen in, chapter 2:
⎧1 if i = j
N i (ε ij ) = ⎨
⎩0 if i ≠ j (5.2)

Where ε ij is the ε coordinate of the node in the element. For an element with in nodes,
ψ i (i = 1,2,..., n ) are polynomials of degree n-1.
For each linear element derived in CHAP.2 we find that
schema
1 1
N1 = (1 − ε ) N 2 = (1 + ε )
2 2
1
2
( ) 1
N 1 = − ε (1 − ε ); N 2 = 1 − ε 2 ; N 3 = ε (1 + ε )
2
−9
N3 =
27
16
( )
⎛1 ⎞
1 − ε 2 ⎜ + ε ⎟; N 4 = (1 + ε )⎛⎜ 1 − ε 2 ⎞⎟
⎝3 ⎠ 16 ⎝9 ⎠
−9
N1
16
(
(1 − ε )⎛⎜ 1 − ε 2 ⎞⎟; N 2 = 27 1 − ε 2 )⎛⎜ 13 − ε ⎞⎟
⎝9 ⎠ 16 ⎝ ⎠
(5.3)

It is natural to think of an approximation of the geometry (x) on the same way we have an
approximation of the dependent variable (say- displacement or heat). For example
x = f (ε ) and u = g (ε ) . F and g be different functions. With different degree of approximation
η ij

ENSP Page 68/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
m n
x = ∑ N i .xi u(x ) = ∑ N j u j
i =1 j =1 (5.4)

Depending on the degree of approximation used for coordinate transformation and that used
for the dependent variable, the FE formulations are classified into 3 categories:
¾ Sub parametric formulations: m<n (Euler-Bernoulli beam element)
¾ Isoperimetric formulations m=n => (i=j)
¾ Super parametric formulations m>n (not in practice)

5.2. Numerical Integration

The Gauss-quadrature has became the accepted numerical integration schema in the majority
of FE applications.Quadrature numerical integration it general, an integral is
evaluated approximated as
n
I = ∫ f ( x )dx = ∑ wk f ( x k )
b

a
k =1 (5.5)

x k = sampling points (Gauss points); wk = weights.

The development of the weights and sampling points is based upon the Legendre polynomial,
giving way to the Gauss-Legendre quadrature, which requires limits from -1 to +1:
+1 n
I = ∫ f (r )dr = ∑ wk f (rk )
−1
k =1 (5.6)

The Gauss-Legendre quadrature is easily extended to 2 or dimensions.

5.3. Shape Function Formulas

Interpolation formulas and shape function for higher order finite elements are often derived
by inspection.
Lagrange polynomial Formula.
The Lagrange polynomial formula is an interpolation formula that is useful for generating
shape function and is
n
x − xm (x − x0 )(x − x1 )....(x − xn )
Lk = ∏ =
m =0 x k − x m ( x k − x0 )( x k − x1 )...( x k − x n )
k =m

(5.7)

When x = xk ⎯
⎯→ Lk = 1
x = x m (with m ≠ k ) Lk = 0 ⇒ Lk has the properties described in equation (5.2). the Lk can
be used to construct an interpolation formula or shape function corresponding to any line
element, and the line element is easily extended to higher dimensions.
Triangular Shape Function Formulas:
Shape function can be derived using the general coordinate seen in chapter 3, with the 3-node
triangular element definite relative to the linear sides of the triangle. The shape function in
terms of general coordinate using the formula

ENSP Page 69/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
n
Fù (L1 , L2 , L3 )
Nk = ∏
m =1 Fm L1 , L2 , L3 (5.8)
n=is the order of the triangle and is equal to 1 less
than the number of nodes a long a side. The function
Fm is obtained from the equations of n lines that pass
through all the nodes except the one of interest. The
denominator is the valve of Fm when evaluated at the
coordinates of node k. (see problems)

5.4. ISOPARAMETRIC ELEMENT

5.4.1. Triangular Element:
The linear (3-node) triangular element was developed earlier. Higher-order triangular
elements can be developed with the help of the Pascal triangle, with contains the terms of
polynomials of various degrees in 2-coordinate systemas shown.
Degree of the
Number of terms
Pascal Triangle complete Element with nodes
in the polynomial
polynomial
1 0 1
x y
x 2
xy y2 1 3

x3 x2 y xy 2 y3 schema
4 3 2 2 3 4
2 6
x x y x y xy y
3 10
x 5 x 4 y x 3 y 2 x 2 y 3 xy 4 y 5 4 15
5 21
Equations (5.4) remain valid here in the general or normalized coordinate system
(x, y )[or (ε ,η )] , and can be experessed in matrix for as:
1≤ i ≤ n
⎧u ⎫
φ1 = ⎨ 1 ⎬
⎩v1 ⎭
⎧φ x ⎫
⎨ ⎬ = [N ]{φi }
n node number (5.9)
⎩φ y ⎭
φ being any functional value (u, v, x, y, …).

As an example let see these applications:

Schema

5.4.2. Rectangular Element

Analogous to the triangular elements, the Lagrange rectangular elements can be developed
from a rectangular array as shown here. In general, a first order Langrage rectangular element
has n modes with.

ENSP Page 70/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
n = ( p + 1)
2
(4.12)

Rectangular array of element Lagrange element

x x2 x3 x4 x5 .........................

y xy x 2 y x 3 y x 4 y x 5 y..............

y2 xy 2 x2 y2 x3 y 2 x4 y 2 x 5 y 2 ................
schema
y3 x y3 x2 y3 x3 y3 x4 y3 x 5 y 3 ......................

y4 x y4 x2 y4 x3 y 4 x4 y4 x 5 y 4 ........................

...................and so on.........................

As an example let us consider a 8-nodal 16-nodes elements quadrilateral element in general

coordinate and in normalized coordinate system as shown
Schema
With equations (5.4) being valid, shape function N i are given as

Np =
1
(1 + εε p )(1 + ηη p ) 1≤ p ≤ 4
4
Np =
1
2
( )
1 − η p2 ε 2 − ε p2η 2 (1 + εε p + ηη p ) 5≤ p≤8
((5.13)

5.5. Axisymmetric Formulations.

Isoparametric finite elements for axisymmetric problems are derived in the same manner as
all other 2-dimensional formulations. A primary difference is that the volume integration must
be replaced with dV = 2πrdrdz (see 3.6). as discussed in chapter 3 the strain-displacement
∂
equations to be modeled are obtained from equations 3.15 assuming v = = 0:
∂θ
∂u u
ε rr = ε θθ =
∂r r
⎧ε rr ⎫
⎪ε ⎪ ∂w ∂u ∂u
ε zz = ε rz = +
{ε } = ⎪⎨ θθ ⎪⎬ ∂r ∂r ∂z
⎪ε zz ⎪
⎪⎩ε rz ⎪⎭
(5.14)

The equations of equilibrium are independent of the tangential coordinate θ and are:

ENSP Page 71/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
∂σ rr ∂σ rz ∂σ rr − ∂σ θθ
+ + =0 (5.15)
∂r ∂z ∂r
∂σ rz ∂σ zz ∂σ rz
+ + =0 (5.16)
∂r ∂z ∂r

The matrix of material constants is obtained form equation (3.16) as (ref.eqs.(3.6),(3.7)

⎡1 − υ υ υ 0 ⎤
⎢ υ 1 −υ υ 0 ⎥
E ⎢ ⎥
[C ] =
(1 + υ )(1 − 2υ ) ⎢⎢ υ υ 1 −υ 0 ⎥
⎥
⎢ 0 0 0
(1 − 2υ ) ⎥
⎣ 2⎦
(5.17)

And the stiffness matrix can be written as

[ K ] = 2π rarg A [ B ] [C ][ B ]
T

(5.18)

The Laplace operator for cylindrical coordinates ( r ,θ , z ) is written as

∂2 1 δ 1 ∂2 ∂2
∇2 = + + +
∂r 2 r δ r r 2 ∂θ 2 ∂z 2 (5.19)

The dependence open the θ coordinate is deleted in the axisymmetric formulation

5.6. Modelling considerations:

Numerical simulation of a physical process requires (a) a mathematical model that describes
the process and (b) a numerical method to analyze the mathematical model. In development of
mathematical model we often make a set of a assumptions about the process to derive the
mathematical relationships governing the system. Valid assumptions can be made only if we
have a qualitative understanding of how the system works. In FE modeling we give guide
lines concerning element geometries, mesh refinements, and load representations.

5.6.1. Element geometries

Numerical evaluation of integral over actual elements involves a coordinate transformation
from the actual element to the master element. The transformation is accepted if and only if
every point in the actual element is mapped uniquely into point in the master element and vice
versa. This requirement can be expressed as
[ ]
J e = det J e > 0 (5.20)

Everywhere in the element

[J ]= Jacobian matrix of the element ≅ ratio of an area element in the actual element to the
e

corresponding area element in the master element.

ENSP Page 72/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
dA ≅ dxdy = J e dεdη (5.21)

¾ If J e =0, a non-zero area element in the actual element is mapped into a

zero-area in the master element => unacceptable => errors.
¾ If J e <0, a right-handed coordinate system => instability in the
computation
¾ To ensure J e >0, certain geometric shapes of real elements must be
avoided. (in general a 10 o ≤ α ≤ 160 o
(schema)

5.6.2. Mesh Generation;

Generation of a finite element mesh for a given problem should follow the guidelines listed
below:
¾ The mesh should represent the geometry of the computational domain and
local representation accurately
¾ The mesh should be such that large gradients in the solution are adequately
represented.
¾ The mesh should not certain elements with unacceptable geometries
(schema)
¾ Range of acceptable location
¾ Combining different order elements.
¾ Using transition element with different number of nodes on different sides
¾ Impose a condition that constraint the midside node to have the same valve
as that experienced at the node by the lower order element.
Generally, local mesh refinements should be such that very small elements are not placed
adjacent to very large one, and the transition (from big to small) must be gradual.

5.6.3. Load Representation:

The accuracy of the result depends in the element and mesh used to represent the distributed
boundary (or domain of loading).
The use of linear element versus quadratic element might change the actual load distribution
as seen
A solid plate in contact with a circular disc generates a reactive force that can be represented
either as a point load, or as a locally distributed force.
In (a) we might not know the area of distributed load in each case.

ENSP Page 73/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
5.7. problems
5.7.1. Problem 1
1- Use Gauss-Legendre numerical integration to integrate the body force term for the
one-dimensional linear FE derived in problem 2.9.2.4)
2- Use Gauss-Legendre quadratures with n=2 to numerically integrate
b b
∫∫
a a
xydxdy

3- Use Lagrangian interpolation formula for deriving one-dimension, three-node shape

functions for the element illustrated. Specialize the shape function for an element of
length L with a node at its center.

5.7.2. Problem2
1- Derived the interpolation function for a four-node isoparametric quadrilateral
element.
(schema)
2- A four-element model of a plane area is shown here. Use the interpolation function for
question 1°) and for element III, show that coordinate location (x=7.0, y=6.0)
corresponds to point (1,1) in the normalized space. Also, for ε = 0,5 and η = −0,5 ,
determine the corresponding point in the generate coordinate system.

5.7.3. Problem 3
(schema)
1- For the isoparametric master and actual elements shown, discuss the transformation
that relates partial derivatives in the general x,y coordinate to the normalized
ε , η coordinate
2- The fundamental theory for the development of a local stiffness matrix for heat
transformation is given by problem 3.7.2 as
[K ] = ∫ [B ]T [K ][B ]tdxdy (a)
A

Give and discuss the formulation of the [B] matrix for a 4-mode quadrilateral
isoparametric element
3- A quadrilateral element is shown. Evaluate the stiffness matrix for heat transfer using
the definition given in question 2°)
Assume the thermal conductivity as kx=ky=1, and a unit thickness for element

5.7.4. Problem 4:
Consider the 3-element mesh of quadrilaterals show with the given node numbering.
Investigate the effect of node numbering and element convexity on the transformation
from the master element to each of the three elements use the definition in 5.6.1

Solution1

ENSP Page 74/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
1- The matrix equation to be evaluated is written for reference as
⎧( x − L)⎫
⎪
L ⎪
A∫ ⎨ x ⎬ fdx (a)
0
⎪⎩ L ⎪⎭

Using a 2 point integration formula (n=2 from table 6.1), we get the exact solution

The argument of the integral must be change to represent limit -1 to +1, or x =

(rL ) − L)
2
dr
with dx = L
2
And the integral to be evaluated is
⎧ (1 − r ) ⎫
+1⎪
⎪ ⎪⎪
AfL ∫ ⎨ 4 ⎬dr (b)
−1 (r + 1)
⎪ ⎪
⎩⎪ 4 ⎭⎪
Substituting the sampling point and weight valves for n=2 gives
⎛ AfL ⎞ AfL
⎜ ⎟{[1 − (0,577735)] + [1 − (− 0,577735)](1)} =
⎝ 4 ⎠ 2
Similarly, the second term gives also the exact answer that was found in (2.9.2.4)
2- The integration arguments are modified using
(r + 1) dr b ds
x= , dx = a , and y = ( s + 1) , dy = b
2 2 2 2
And the integration to be integrated is
+1 +1 a b ab dr = wi ⎫
∫−1 ∫−1 2 (r + 1) 2 (s + 1) 4 drds ⎬weights
ds = w j ⎭

Or in the discrete form (n=2)

a 2b 2
+ 1)wi (s j + 1)w j
2 2

16
∑∑ (r
i =1 j =1
i

a 2b 2
[{(− 0,577735 + 1)w1 [(− 0,577735 + 1)w1 + (0,577735 + 1)w2 ]}
16
+{(0,577735 + 1)w2 [(− 0,577735 + 1)w1 + (0,577735 + 1)w2 ]}] ( from table 6.1,w1 = w2 = 1)
2 2
a b
=
4
What is also the exact result?
3- Substitute into equation 5-7:
(x − xi )(x − x k ) (x − xi )(x − xk ) (x − xi )(x − xk )
N i = Li = Lj = Lk = N k =
(xi − x j )(xi − xk ) (x j − xi )(x j − x k ) (x k − xi )(x k − x j )

ENSP Page 75/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
L
Let xi = 0 ; x j = ; x k = L with i = 1, j = 2, k = 3
2

⇒ Ni =
(x − L / 2)(x − L ) N2 =
x( x − L )
N3 =
x(x − L / 2 )
(− L / 2)(− L ) (L / 2)(L / 2 − L ) L (L − L / 2 )

Solution2
1- We use the Lagrange polynomial formula and the given figure. The interpolation
function along lines ε = −1 and η = −1 . The interpolation function is identical to the
shape function and the notation for the shape function is used.

N 1 = L1ε L1η =
ε −1
×
η −1
=
(1 − ε )(1 − η ) the remaining
−1−1 −1−1 4
Function are derived in a similar manner

N 2 = L2ε L2η =
ε +1 η −1
× =
(1 + ε )(1 − η )
1+1 −1−1 4

N 3 = L3ε L3η =
ε +1 η +1
× =
(1 + ε )(1 + η )
1+1 1+1 4

N 4 = L4ε L4η =
ε −1
×
η +1
=
(1 − ε )(1 + η )
−1−1 1+1 4
2- The interpolation function given in equation 5.9 can be written as follows using the
notation of question 1°)
4 4
x = ∑ N i xi y = ∑ N i yi (a)
i =1 i =1

Or
1
x= [(1 − ε )(1 − η )x1 + (1 + ε )(1 − η )x2 + (1 + ε )(1 + η )x3 + (1 − ε )(1 + η )x4 ] (b)
4
1
y = [(1 − ε )(1 − η ) y1 + (1 + ε )(1 − η ) y 2 + (1 + ε )(1 + η ) y 3 + (1 − ε )(1 + η ) y 4 ] (c)
4
¾ Substituting the general coordinate values for element III
1
x = [(1 − ε )(1 − η ) × 2 + (1 + ε )(1 − η ) × 5,5 + (1 + ε )(1 + η ) × 7 + (1 − ε )(1 + η ) × 4] (e)
4
1
y = [(1 − ε )(1 − η ) × 3 + (1 + ε )(1 − η ) × 3 + (1 + ε )(1 + η ) × 6 + (1 − ε )(1 + η ) × 6] (d )
4
Substituting the values of ε = 1 and η = 1 into equation (d) and (e) and computing the
corresponding x and y values
We find
1
x = [0 + 0 + (1 + 1)(1 + 1) × 7 + 0] = 7
4
1
y = [0 + 0 + (1 + 1)(1 + 1) × 6 + 0] = 6
4

ENSP Page 76/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
¾ The same
x(0,5; − 0,5) = 5,0313
y (0,5;−0,5) = 3,75

Solution3

ENSP Page 77/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
1°) Shape function N i are given in 5-7-2
N i = N i (ε ,η ) 1≤ i ≤ 4 (a)
(N i = shape function )
The x, y coordinate are defined in terms of ε , η coordinate as
x = x( N i ) = x(ε ,η ) y = y ( N i ) = y (ε ,η ), ( N i = int erpolation )
The derivatives if the shape functions can be written as follows, using the chain rule:
∂N i ∂N i ∂x ∂N i ∂y
= +
∂ε ∂x ∂ε ∂y ∂ε
(c)
∂N i ∂N i ∂x ∂N i ∂y
= +
∂η ∂x ∂η ∂y ∂η
And in matrix form as
⎡ ∂N i ⎤ ⎛ ∂x ∂y ⎞ ⎡ ∂N i ⎤
⎢ ∂ε ⎥ ⎜ ⎟⎢ ⎥
⎢ ⎥ = ⎜ ∂ε ∂ε ⎟ ⎢ ∂x ⎥ (d)
⎢ ∂N i ⎥ ⎜ ∂x ∂y ⎟ ⎢ ∂N i ⎥
⎢⎣ ∂η ⎥⎦ ⎜⎝ ∂η ∂η ⎟⎠ ⎢⎣ ∂y ⎥⎦

We define the Jacobian matrix as

⎛ ∂x ∂y ⎞
⎜ ⎟
J = ⎜ ∂∂εx ∂ε ⎟
∂y ⎟
⎜
⎜ ∂η ∂η ⎟⎠
⎝
Multiplying equation (d) by J-1, gives the forms of the equation that can be used to
compute derivatives of the shape function in x, y system
⎡ ∂N i ⎤ ⎡ ∂N i ⎤ ⎛ ∂x ∂y ⎞
−1
⎡ ∂N i ⎤
⎢ ∂x ⎥ ⎢ ∂ε ⎥ ⎜ ⎟ ⎢ ∂ε ⎥
⎢ ⎥ = J −1 ⎢ ⎥ = ⎜ ∂ε ∂ε ⎟ ⎢ ⎥ (f)
∂N
⎢ i⎥ ⎢ i ⎥ ⎜ ∂x
∂ N ∂y ⎟ ⎢ ∂N i ⎥
⎢⎣ ∂y ⎥⎦ ⎢⎣ ∂η ⎥⎦ ⎜⎝ ∂η ∂η ⎟⎠ ⎢⎣ ∂η ⎥⎦

Substituting equations (a) of problem 5-7-2 2) gives the final form

−1
⎡ ∂N i ⎤ ⎛ ∂x ∂y ⎞ ⎡ ∂N ⎤
⎢ ∂x ⎥ ⎜ ∑ ∂ε xi ∑ ∂ε yi ⎟ ⎢ ∂ε i ⎥
⎢ ⎥=⎜ i i ⎟ ⎢ ⎥ (g)
∂N
⎢ i⎥ ⎜ ∂x ∂y ⎟ ⎢ ∂N i ⎥
⎢⎣ ∂y ⎥⎦ ⎜⎝ ∑ ∂η i i ∑ ∂η yi ⎟ ⎢ ∂η ⎥
x
i ⎠ ⎣ ⎦
The definition of J can be extended to any interpolation function as

ENSP Page 78/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
⎡ ⎛ ∂N i ⎞⎤
⎢∑ ⎜ ∂ε ⎟⎥
⎝ ⎠⎥
J =⎢
⎢ ⎛ ∂N i
[xi − yi ] ( h)
⎞⎥
⎢∑ ⎜⎜ ⎟⎟⎥
⎣⎢ ⎝ ∂η ⎠⎦⎥
[2 × n] [n × 2]
For a 4-mode isoparametric Element, the Jacobian matrix in a 2-dimensio system will be

⎡ x1 y1 ⎤
1 ⎡− (1 − η ) (1 − η ) (1 + η ) − (1 + η )⎤ ⎢⎢ x 2 y 2 ⎥⎥
J= (i )
4 ⎢⎣ − (1 − ε ) − (1 + ε ) (1 + ε ) (1 − ε ) ⎥⎦ ⎢ x3 y3 ⎥
⎢ ⎥
⎣ x4 y4 ⎦
2- Equation of problem 3-7-2 gives the [B] matrix as [B]=[L][N] and that corresponds to
left hands side of equation (g) of question 1), that can be expanded for i=4 as
⎡ ∂N 1 ∂N 2 ∂N 3 ∂N 4 ⎤ ⎡ ∂N 1 ∂N 2 ∂N 3 ∂N 4 ⎤
−1
⎢ ∂x ∂x ∂x ∂x ⎥ = ⎡ J 11 J 12 ⎤ ⎢ ∂ε ∂ε ∂ε ∂ε ⎥
⎢ ∂N ∂N 4 ⎥ ⎢ J ⎢ ∂N 4 ⎥
(b)
⎢ 1
∂N 2 ∂N 3
⎥ ⎣ 21 J 22 ⎥⎦ ⎢ ∂N 1 ∂N 2 ∂N 3
⎥
⎢⎣ ∂y ∂y ∂y ∂y ⎥⎦ ⎢⎣ ∂η ∂η ∂η ∂η ⎥⎦
The components J ij of [J] are given by equation (i) of question 1). The derivative of the
shape function above with respect to ε and η are the same with those the equations (i)
and (h) of question 1°), and these are the same with those of equations (i) and (h) of
question 1°), and these are 2 matrices are always identical for a “4-mode isoparametric
element with one DOF per mode”.
(b) above represents shape function
(i), (h) in 1°) represents interpolation function for geometry transformation
3- A 2 × 2 Gaussian quadrature (numerical integration) is use as the master element is
showing. The [B] matrix given in question 2 most be evaluated for each Gauss point, and
each time the [B] matrix is evaluated, the contribution to the stiffness matrix is
computed as [B]T[K][B], and the final stiffness matrix is the sum of the four
contribution.
Computation is shown in detail, bat are usually alone by computer for each gauss point
of each element of the structure
Comment if [Ke] is 8 × 8 with 4 points, and 1000 elements => 4000 [Ke]
The matrix that defines the element in the n coordinate system is evaluated using data as
⎡2 1⎤
⎢5 2⎥⎥
⎢ (a)
⎢4 6⎥
⎢ ⎥
⎣1 4⎦
Gauss point 1
The location of the Gauss point is obtained form table S.I and shown in the given figure.
Let

ENSP Page 79/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
ε i =1 = −0,57735 η j =1 = −0,57735 using equation (i) of 1°) to evaluate J ij of the
Jacobian matrix
1
J 11 = [− (1 − η )x1 + (1 − η )x2 + (1 + η )x3 − (1 + η )x4 ]
4
1
= [− (1 + 0,57735).2 + (1 + 0,57735).5 + (1 − 0,57735).4 − (1 − 0,57735).1] = 1,50
4
1
J 12 = [− (1 + 0,57735).1 + (1 + 0,57735).2 + (1 − 0,57735).6 − (1 − 0,57735).4] = 0,60566
4
1
J 21 = [− (1 + 0,57735).2 − (1 − 0,57735).5 + (1 − 0,57735).4 + (1 + 0,57735).1] = −0,5
4
1
J 22 = [− (1 + 0,57735).1 − (1 − 0,57735).2 + (1 − 0,57735).6 + (1 + 0,57735).4] = 1,60566
4
⎡ 1,5 0,60566⎤ ⎡0,59221 − 0,22338⎤
⇒J =⎢ ⎥ J −1 = ⎢ ⎥ (b)
⎣− 0,5 1,60566 ⎦ ⎣ 0,18411 0,55324 ⎦
Form equation 5.21
dA = dxdy det J dεdη = det J × wi =1 × w j =1 = det J = 2,71133 ( wi = w j = 1)

0,59221 − 0,22338⎤ ⎡− 1,57735 1,57735 0,42265 − 0,42265⎤

[B]1 = ⎡⎢ ⎥⎢ ⎥
⎣ 0,18411 0,55324 ⎦ ⎣− 1,57735 − 0,42265 0,42265 1,57735 ⎦
⎡− 0,18544 0,25713 0,03897 − 0,15066⎤
=⎢ ⎥ (c )
⎣ − 0,2988 0,01426 0,07794 0,19868 ⎦
The contribution to the stiffness matrix for Gauss point 1 can be written as
[K ]1 = [B ]1 [K
T
][B ]1 dA (see b) of 3-7-2.2)

⎡0,28676 − 0,11265 − 0,07684 − 0,09728⎤

⎢ 0,17982 0,03018 − 0,09735⎥⎥
And [K ]1 = ⎢ (e )
⎢ 0,02059 0,02606 ⎥
⎢ ⎥
⎣ 0,16857 ⎦
The procedure is the same for the remaining Gauss points and it is easier to show that
for
Gauss point 2 ε i =1 = −0,57735 η j = 2 = 0,57735

⎡ 1,5 0,89434⎤ −1 ⎡0,59227 − 0,31318⎤

J =⎢ ⎥ J = ⎢ 0,17509 0,52527 ⎥
⎣− 0,5 1,60566 ⎦ ⎣ ⎦
0,09250 0,18863 − 0,15216⎤
[B]2 = ⎡⎢
0,06409
⎥
⎣− 0,22564 − 0,03700 0,12455 0,13809 ⎦
⎡0,15711 0,04077 − 0,04573 − 0,15216⎤
⎢ 0,02834 0,03667 − 0,10578⎥⎥
[K ]2 =⎢ (f)
⎢ 0,14591 − 0,13685⎥
⎢ ⎥
⎣ 0,39479 ⎦

ENSP Page 80/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
Gauss point 3 ε i = 2 = 0,57735 η j =1 = −0,57735

⎡ 1,5 0,60566⎤ ⎡0,60246 − 0,19262⎤

J =⎢ ⎥ J −1 = ⎢ ⎥
⎣− 0,5 1,89434 ⎦ ⎣ 0,15902 0,47705 ⎦
− 0,21722 0,31353 − 0,01230 − 0,08401⎤
[B]3 = ⎡⎢
⎣ − 0,11311 − 0,12541 0,20492 0,03360 ⎥⎦
⎡0,18859 − 0,16954 − 0,06448 0,45428 ⎤
⎢ 0,35855 − 0,09293 − 0,09607 ⎥⎥
[K ]3 = ⎢ (g)
⎢ 0,13251 0,02490 ⎥
⎢ ⎥
⎣ 0,02574 ⎦
Gauss point 4 ε i = 2 = 0,57735 η j = 2 = 0,57735

⎡ 1,5 0,89434⎤ ⎡0,60246 − 0,27945⎤

J =⎢ ⎥ J −1 = ⎢ ⎥
⎣− 0,5 1,89434 ⎦ ⎣ 0,15204 0,45611 ⎦
− 0,03213 0,16810 0,11991 − 0,25588⎤
[B]4 = ⎡⎢ ⎥
⎣− 0,06426 − 0,16380 0,23982 − 0,01176⎦
⎡0,01697 0,01687 − 0,06334 0,02952 ⎤
⎢ 0,18116 − 0,06289 − 0,13512⎥⎥
[K ]4 =⎢ ( h)
⎢ 0,23642 − 011018 ⎥
⎢ ⎥
⎣ 0,21577 ⎦
The stiffness matrix is the sum of equations (e) → (h)

⎡0,04945 − 0,22456 − 0,25040 − 0,17449⎤

4 ⎢ 0,74787 − 0,08897 − 043433 ⎥⎥
[K ] = ∑ [K ]i = ⎢ (i )
i =1 ⎢ 0,53543 − 0,19606⎥
⎢ ⎥
⎣ 0,80488 ⎦

Solution4

ENSP Page 81/103

FINITE ELEMENT METHODS / 5. ISOPARAMETRIC FE Version ou da
Dr LEZIN
The master element is having that N i are given in 5.7.2.1) and elements of the Jacobian
J ij can be obtained from equation (g) of 5.7.3.1), we e valued the Jacobian for each of
the elements;
Element 1: we have x1 = x 4 = 0; x 2 = x3 = 2; y1 = y 2 = 0; y 3 = 3 and y 4 = 5. The
transformation and the Jacobian are given by
1
x = 2 N 2 + 2 N 3 = 1 + ε ; y = 3N 3 + 5 N 4 = (1 + η )(2 − ε ) (a)
2
∂x ∂y 1
1 − (1 + η ) 1
det J = ∂∂εx ∂∂εy = 2 = (4 − ε ) > 0 (b)
1 2
0 2− ε
∂η ∂η 2
Cleary, the determinant of Jacobian is linear in ε , and for all values of ε in − 1 ≤ ε ≤ 1
2y
is positive. And the transformation is possible as 1 + ε = x and 1 + η =
5− x
Element N°2: (student of time) x1 = x 4 = 4; x 2 = 3; x3 = 5; y1 = 0; y 2 = 2; y 3 = y 4 = 3. The
transformation of det J are given as
1 1 1 1
x = 3 + ε + η + + ηε y = 2 + ε + η − εη (c )
2 2 2 2
1 1
(eq 5.9) 1+ η (1 − η ) 3
det J 2 2 = (1 + η − ε ) (d )
1
(1 + ε ) 1 − 1 ε 4
2 2
Zero for all η , ε of element, for example is zero along the line ε − η − 1 = 0 , and is
negative in the shaded area in the master element. The same area is outside the actual
element 2. thus, elements, with any interior angle greater than -________ should not be
used in any FE mesh.
Element N°3: x1 = 2, x 2 = 0, x3 = x 4 = 5; y1 = y 4 = 3 and y 2 = y 3 = 5
The transformation and the det J become:
1 1
x = 3 − ε + 2η + εη , y = 4+ε (e )
2 2
1
− (1 − η ) 1
⎛ 1 ⎞
det J = 2 = −⎜ 2 + ε ⎟ < 0 (f)
1 ⎝ 2 ⎠
2+ ε 0
2
The negative det J indicates that a right-hand coordinate system is mapped into a left-
hand one. Must be avoided

ENSP Page 82/103

FINITE ELEMENT METHODS / 6. COMPUTER IMPLEMENTATION OF 2-
DIMENSION PROBLEMS Version ou date
Dr LEZIN

6. COMPUTER IMPLEMENTATION OF 2-DIMENSION

PROBLEMS

6.1. INTRODUCTION
In 2-dimensions, the element calculations are more involved than one dimension, owing to the
following considerations:
¾ Various geometric shapes of elements.
¾ Single as well as multivariable problems.
¾ Integration performed over areas as opposed to along lines (for 1-
dim.elements).
¾ Mixed-order integration use in certain formulation (shear deformable plates
and penalty function formulation of viscous in incompressible fluids)
A brief description of any software can be represented (only for explanation) by the flow chart
of the 2-dimension computer given below:
SHEMAS

Where:
BOUNDARY: subroutine to impose specified boundary condition on the primary and
secondary variables.
CONCTVTY: subroutine to describe connectivity conditions.
DATAECHO: subroutine to echo the input data to the program.
EGNBNDRY: subroutine to impose specified homogeneous b.c. on the primary variables in
eigenvalue problems.
EGNSOLVR: subroutine to determine eingenvalues and eingenvectors.
ELKMFRCT: subroutine to compute the element [K],[M], and [F] matrices for various
problems in rectangular elements.
ELKMFTRI: subroutine to compute the element [K],[M], and [F] matrices for various
problems in triangular element.
EQNSOLVR: subroutine to solve banded, symmetric systems of algebraic equations using
the Gauss elimination method.
INVERSE: subroutine to invert a 3 × 3 matrix explicitly.
JACOBI: subroutine to compute the jacobian matrix.
MESH2DG: subroutine to generate mesh for non-rectangular elements.
MESH2DR: subroutine to generate mesh for rectangular elements.
POSTPROC: subroutine to post-compute the solution, gradient of the solution and stresses.
QUADRTRI: subroutine to generate the quadrature points and weights for triangular
elements.
SHAPERCT: subroutine to compute the shape (interpolation) functions for linear and
quadratic rectangular elements.

ENSP Page 83/103

FINITE ELEMENT METHODS / 6. COMPUTER IMPLEMENTATION OF 2-
DIMENSION PROBLEMS Version ou date
Dr LEZIN
SHAPETRI: subroutine to compute the shape (interpolation) functions for linear and
quadratic triangular elements.
TEMPORAL: subroutine to compute the equivalent coefficient matrices and column
vectors for equations when time-dependent analysis is carried.

6.2. Assembly of Element Equations

The assembly of element equations should be carried out as soon as the element matrices are
computed, rather than waiting till the element the element coefficient of all the element are
computed. The latter requires storage of characteristics of each element. In the former case,
we can perform the assembly in the same loop in which a subroutine is called to calculate and
to store in the stiff equation (in the banded form) an element matrix.
When element matrices are symmetric, the assembled global matrix is also symmetric; with
many zeros away from the main diagonal. Therefore, it is sufficient to store only the upper
half-band of the global matrix. Let Mi be the number of matrix elements between the diagonal
element and the last nonzero element in the i-th row, after which a N element in that row; the
half band which is the maximum of (Mi+1)× NDF,(where NDF=number of DOF per node).
NHBW = max[( M i + 1) × NDF ] (6.1)
1≤ i ≤ n

n = number of the rows in the matrix.

If 2 global nodes do not belong to the same element, then the corresponding entries in the
global matrix are zeros (node I, J ∉ an element).
KIJ = 0 (6.2)

This property enables us to determine the half-band width NHBW of the assembled matrix.
NHBW = max
1≤ N ≤ NEM
{NOD( N , I ) − NOD( N , J ) + 1}× NDF (6.3)
1≤ I , J ≤ NPE

Where NEM = Number of element in the mesh

NPE = Number of nodes per elements
NDF = Number of DOF per node.

For example, for 1-dimension problems with elements connected in series, the maximum
difference between nodes of an element is equal to NPE − 1 hence
NHBW = [( NPE − 1) + 1]× NDF = NPE × NDF
≤ NEQ = nber of equations in the matrix equation

(6.4)
e
The logic for assembling the element matrices K ij
into the upper-banded form of the global
coefficient K ij
is that the assembly can be skipped whenever J < I and J > NHBW. The main

ENSP Page 84/103

FINITE ELEMENT METHODS / 6. COMPUTER IMPLEMENTATION OF 2-
DIMENSION PROBLEMS Version ou date
Dr LEZIN
diagonal, I=J, of the assembled square matrix (fall storage form), becomes the first column of
the assembled banded matrix (banded storage form) as seen here:
⎡ a11 a12 . . a1k 0 0 0 0 ⎤ ⎡ a11 a12 . . . a1k ⎤
⎢a ⎥ ⎢ a
⎢ 21 a 22 . . a2k a 2 k +1 . 0 0 ⎥ ⎢ 22 a 23 . . . a 2 k +1 ⎥⎥
⎢ a31 a32 a33 . a3k a3k +1 a3k + 2 0 0 ⎥ ⎢ a33 a34 . . . a3k + 2 ⎥
⎢ ⎥ ⎢ ⎥
⎢ . ⎥ ⎢ . . . . . . ⎥
[A] ⎢
=⎢ . ⎥
⎥ ⇒ [A] = ⎢ .
⎢ . . . . . ⎥
⎥
( sym ) ⎢ ⎥ ⎢ . . . . . . ⎥
⎢ . ⎥ ⎢ .
⎢ . ⎥ . . . . . ⎥
⎢ ⎥
⎢ ⎥ ⎢a n −1,n −1 a n −1,n . . . 0 ⎥
⎢ . ⎥ ⎢ a 0 . . . 0 ⎥⎦
⎢0 . 0 0 a nk a nk +1 a nk + 2 . a nn ⎥⎦ ⎣ nn1
⎣

6.3. Geometry and characteristics.

In any structure, there will inevitably be a large number of node and elements for which it is
necessary to input the data regarding the coordinates of nodes of each element. This can be a
very onerous because task leading to the possibility of errors. However, because most of the
data will be fairly regular, it is possible to input data only at the extremities of a regular set of
data and to let the computer generate the intermediate data.
DESSIN

NODSTR = Node at the start of line

NODEND = Node at the end of the line
NELMEND = element at the end of a line-group of elements.
NELMSTR = element at the start of a line-group of elements.
NSTEP = number of steps to reach from NODSTR to NODEND.
NODINC = Node number increment
CORINC (I) = the array sloving coordinates increment for successive
intermediate nodes.
Evidently, because the intermediate nodes are equally spaced and the node numbers are
equally incremented, it is possible and easy to generate node numbers and coordinates of
intermediate nodes.

6.4. Types of structures and DOF

ENSP Page 85/103

FINITE ELEMENT METHODS / 6. COMPUTER IMPLEMENTATION OF 2-
DIMENSION PROBLEMS Version ou date
Dr LEZIN
6.5. Examples:

Examples 1:

For the 2-D element shown, find the half bandwidth NHBW.

DESSIN a

⎡11 7 3 2 1 6 9 10⎤ element I

⎢ ⎥
LNODS ( NELEM , NPE ) = ⎢11 12 13 8 5 4 3 7 ⎥ element II
⎢⎣ 9 14 16 15 13 12 11 10⎥⎦ element III

élémentI : NHBW I = [(11 − 1) + 1]× 2 = 22 ⎫

⎪
élémentII : NHBW II = [(13 − 1) + 1]× 2 = 22 ⎬ ⇒ NHBW = 14
élémentIII : NHBW III = [(16 − 10) + 1]× 2 = 14⎪⎭

DESSIN b

NHBW =? (For a 2-D Truss).

With your own numbering?
NDH (NDOF) = 1

DESSIN c1 DESSIN c2

(personal)
NDOF = 2: use 8 nodes isoparemetric element with at least 12 elements.

DESSIN d

Examples 2:
Write a program to generate intermediate nodes and their respective coordinates for
any plane structure.
Apply this program for this particular structure.

DESSIN d

ENSP Page 86/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN

7. SELECTED TOPICS IN FEM.

7.1. Introduction
The intent of previous chapters has been to outline and discuss a particular topic
and to demonstrate that FE theory is merely an extension of numerical methods to solve
different engineering problems. This chapter is an introduction to a variety of problems of
practical interest to the user in many fields of study research. The FEM of modelling coupled
with time-dependent analysis has an impact on the analysis of time-dependent problems.
Advanced topics are introduced and a variety of applications are illustrated.

7.2. Eigenvalue Problems

7.2.1. Background:
An eigenvalue problem is defined to be one in which we seek the values of the parameter λ
such that the equation
A(U) = λ B(U) (7.1)

is satisfied for nontrivial value of U. There A and B denote either matrix operators or
differential operator, and values of λ for which equation (7.1) is satisfied are called an
d 2U d2
eigenvector (or eigenfunction); for example, the equation − = λU ( x) with A= − ,B=1
dx 2 dx 2
which arises in connection with natural axial vibration of a bar or the transverse vibration of a
cable, constitutes an eigenvalue problem. λ denotes the square of vibration w.
In structural eng, the eigenvalues denote either natural frequencies or bucking loads. In fluid
mechanics and heat transfer, they denote amplitudes of the Fourier components making up the
solution. Differential eigenvalue problems are reduced to algebraic eigenvalue problems (in
example: [A]{X } = λ [B]{X } ) by means of the FE approximation, and the methods of solution of
the algebraic eigenvalue problems are then use to solve for the eigenvalues and eigenvectors.
∂Φ( x, t ) ∂ ⎡ ∂Φ ( x, t ) ⎤
α ( x) − ⎢ β ( x) ∂x ⎥ + γ ( x)Φ( x, t ) = δ ( x, t ) (7.2)
∂t ∂x ⎣ ⎦

Where Φ( x, t ) [or u ( x, t ) , v( x, t ) ,…] is an eigenfunction,

α (x) , β (x) , γ (x) and δ ( x, t ) are functions that pertain to a
particular application.

For example if a tension in a cable is assumed to be constant for small deflection theory, then
we can get the following governing:
d 2 v( x)
T − k ( x)v( x) = − f ( x) (7.3)
dx 2

Neglecting the foundation nidulus k and external loading f, the dynamic problem is obtain by
summing the forces in y-dir and including inertial forces (with ∑ Fy = ρa y , ρ = mass or
density, ay = acceleration). The final differential equation has the form:

ENSP Page 87/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
∂v ∂ 2v
T = ρ (7.4)
∂x 2 ∂t 2

that is the hyperbolic differential equation that governs the dynamic motion of the cable. v(x,t)
is called wave propagation that often requires an homogenous solution through the separation
of variables technique [ u ( x, t ) = u ( x, t ) t + u h ( x, t ) with u h ( x, t ) = U ( x)T (t ) ]

7.2.2. Heat Transfer and Bar like Problems

Let us consider the problem of solving the equation
d ⎡ dU ⎤
− ⎢ a( x) + c( x)U ( x) = λc0 ( x)U ( x) (7.5)
dx ⎣ dx ⎥⎦

for λ and U(x). Here a, c and c0 are known quantities that depend on the physical problem, λ =
eigenvalue, U = eigenfunction. Special cases are given as:
Heat transfer: a = kA; c = Pβ; c0 = ρcA
Bars: a = EA; c = 0; c0 = ρA (7.6)

Over a typical element Ωe we seek a finite element approximation of U in form

n
U e ( x) = ∑U ej N j ( x) (7.7)
j =1

The weak form of equation (7.5) is:

xb
⎛ dw dU ⎞
0 = ∫⎜a + awU ( x) − λc0 wU ⎟dx − Qne w( xa ) − Qne w( xb )
xa ⎝
dx dx ⎠
(7.8)

Where w is the weight function and Q1e and Qne are the secondary variables at nodes 1 and
nodes n respective. (Assuming Qie = 0 when i ≠ 1 and I ≠ n)

⎡ dU ⎤ ⎡ dU ⎤
Q1e = − ⎢a Q e
= − ⎢⎣a dx ⎥⎦ (7.9)
⎣ dx ⎥⎦ xa
n
xb

Substitution of the Fe approximation in the weak form gives the FE model of the eigenvalue
equation (7.5):
[K ]{U }− λ [M ]{U }= {Q }
e e e e e
(7.10)

Where

ENSP Page 88/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
xb
⎡ dN i dN j ⎤
K = ∫ ⎢a ( x)
e
ij + c( x) N i N j ⎥ dx
xa ⎣
dxdx ⎦
xb
(7.11)
M ije = ∫ [c
xa
0 ( x) N i N j dx ]

7.3. Three-dimension F.E.

Three-dimension F.E. is an extension of the 2-dimension isoperimetric elements of chapter 5,

as well as numerical integration using the Gauss-Legendre quadrature to evaluate volume
elements. Equation (5.6) will simply be:
+1 +1 +1
I= ∫ ∫ ∫ f (ξ ,η , ρ )dξdηdρ
−1 −1 −1
or
m n l
I = ∑∑∑ wi w j wk f (ξ i ,η j , ρ k ) in discrete form
i =1 j =1 k =1
(7.12).

Where ξ ,η , ρ are the coordinates of the 3-dimension

normalized element (usually m = n = l).

Interpolated function can be derived as described for 2-dimension element and they have the same
properties.
n

∑N
i =1
i
e
( x, y, z ) = 1 (7.13)

and the geometry of the elements can be described by

the transformation equations
m
x = ∑ xi N i (ξ ,η , ρ )
i =1
m
y = ∑ y i N i (ξ ,η , ρ ) (7.14)
i =1
m
z = ∑ z i N i (ξ ,η , ρ )
i =1

DESSIN

The definition of the jacobian matrix and the numerical quadrature rules described earlier are
extended to the 3-dimension case. In example, equation (§5.7.3.1h) becomes

ENSP Page 89/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
⎡ x1 y1 z1 ⎤
⎢x y2 z 2 ⎥⎥
⎡ ∂x ∂y ∂z ⎤ ⎡ ∂N 1 ∂N 2 ∂N m ⎤ ⎢ 2
⎢ . . . . ⎢ . ⎥
⎢ ∂ξ ∂ξ ∂ξ ⎥⎥ ⎢ ∂ξ ∂ξ ∂ξ ⎥⎥ ⎢ . .
⎥
⎢
∂x ∂y ∂z ⎥ ⎢ ∂N 1 ∂N 2 ∂N m ⎥ ⎢ . . . ⎥
J =⎢ = . . . . = . (7.15)
⎢ ∂η ∂η ∂η ⎥ ⎢ ∂η ∂η ∂η ⎥ ⎢ . . . ⎥
⎢ ∂x ∂y ∂z ⎥ ⎢ ∂N 1 ∂N 2 ∂N 3 ⎥ ⎢ . . . ⎥
⎥
⎢ ⎥ ⎢ . . . . ⎥ ⎢
⎢⎣ ∂ρ ∂ρ ∂ρ ⎥⎦ ⎢⎣ ∂ρ ∂ρ ∂ρ ⎥⎦ ⎢ . . . ⎥
⎢ ⎥
⎢⎣ x m ym z m ⎥⎦

And therefore

⎡ ∂N i ⎤ ⎡ ∂N i ⎤
⎢ ∂x ⎥ ⎢ ∂ξ ⎥
⎢ ∂N ⎥ ⎢ ⎥
−1 ⎢ ∂N i ⎥
⎢ i ⎥ = [J ] (7.16)
⎢ ∂y ⎥ ⎢ ∂η ⎥
⎢ ∂N i ⎥ ⎢ ∂N ⎥
⎢ ⎥ ⎢ i⎥
⎣ ∂z ⎦ ⎢⎣ ∂ρ ⎥⎦

Expression for interpolation (shape) function for linear quadrature elements can be found in
any book.

7.4. Plate F.E.

The term “plate” refers to solid bodies that are bounded by two parallel planes whose lateral
dimensions are compared with the separation between them (thickness) as it is shown down
1
[ h ≤ (a or b) ]. Geometrically, plate problems are similar to the plane stress problems,
10
except that plates are also subjected to transverse loads (loads perpendicular to the plane of
the plate) that cause bending about axes in the plane of the plate. Because of the smallness of
the thickness dimension , it is often not necessary to model plate using 3-dimension elasticity
theory, but using simple 2-dimension theories that can be seen else where. Only a brief
development of the theory of plates is given here, and more definitive treatment is given in
specialized books.

7.4.1. Relation between Force and moment to Stress.

Dessin

Internal stresses in a plate produce bending moments M and shears Q as illustrated

above. Moment and shears are defined as acting per unit length of plate. These internal
actions are defined as follows:

ENSP Page 90/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
h h h
2 2 2
M x = ∫ σ xx zdz M y = ∫ σ yy zdz M xy = ∫ σ xy zdz
h h h
− − −
2 2 2
h h
(7.17)
2 2
Qx = ∫ σ xz dz Qy = ∫ σ yz dz
h h
− −
2 2

The predominant stress are obtained by integrating equation (7.6) as

Mxz Myz M xy z
σ xx = σ yy = σ xy = (7.18)
h 3 12 3
h 12 h 3 12

The remaining stresses are also obtained by integrating (7.6), even if they are quite small and
are maximum at the midplane of the plate, z = 0.
3Q y ⎛ 4 z 2 ⎞ 3Q x ⎛ 4z 2 ⎞
σ xz = ⎜1 − 2 ⎟⎟ σ yz = ⎜⎜1 − 2 ⎟⎟ (7.19)
2h ⎜⎝ h ⎠ 2h ⎝ h ⎠

7.4.2. Deformation assumptions

Classical thin-plate theory is somewhat similar to beam theory and is often referred to as the
Kirchhoff theory of plates. The basic assumption is that points on the midsurface of the plate
(z = 0) can displace in the z-direction. A straight line normal to the midsurface can displace
only in the z-direction. A straight line normal to the midsurface cannot displace in the x or y
direction. However, at distance z from the midsurface displace u and v, in the x and y
direction, can develop as the plate bends. These displace are illustrated and given below.

DESSIN

∂w ∂w
u = −Z v = −Z (7.20)
∂x ∂y

The corresponding strains are:

∂u ∂ 2u ∂v ∂2w ∂u ∂v ∂2w
ε xx = = −z 2 ε yy = = −z 2 ε xy = + = −2 z
∂x ∂x ∂y ∂y ∂y ∂x ∂x∂y

The strain-deformation relations in equations (7.20 - 7.21) do not allows for transverse shear
∂w
deformation. ( ε yz, zx ). ε yz = −θ y = 0
∂y

ENSP Page 91/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
A more complete assumption follows the same idea, except that straight lines normal to the
midsurface before bending remain straight but not normal to the midsurface after bending as
shown down. The angles between the z-axis and a line originally normal to the midsurface (x
or y-axis) are defined as θx or θy respectively. These assumptions are referred to as the
Mindlin theory of plates.
DESSIN

Displacements are given as

u = − zθ x v = − zθ y (7.22)

And the strains become

∂θ x ∂θ y ∂θ x ∂θ y
ε xx = − z ε yy = − z ε xy = − z ( + )
∂x ∂y ∂y ∂x
(7.23)
∂w ∂w
ε yz =( −θy) ε xz =( −θx )
∂y ∂x

This theory accounts for transverse shear deformation and is applicable for moderately thick
plates. ⇒ Mindlin (thick-plate) theory reduces to Kirkchhoff (thin-plate) theory with the
∂w ∂w
assumption that θ x = and θ y = .
∂x ∂y

7.4.3. Stress-Strain and Moment-curvature relations.

The stress-strain relations correspond to a linearly elastic material and following the matrix
notation of previous chapters may be written.
{σ } = [E ]{ε } (7.24)

Homogeneous and isotropic material properties are assumed, and stresses (equations 7.18 and
7.19) are related to the strains as:
⎧σ xx ⎫ ⎡ E (1 − ν ²) Eν (1 − ν ²) 0 0 0 ⎤ ⎧ε xx ⎫
⎪σ ⎪ ⎢ Eν (1 − ν ²) E (1 − ν ²) 0 0 0 ⎥ ⎪ε ⎪
⎪⎪ yy ⎪⎪ ⎢ ⎥ ⎪⎪ yy ⎪⎪
σ
⎨ xy ⎬ = ⎢ 0 0 G 0 0 ⎥ ⎨ε xy ⎬ (7.25)
⎪σ ⎪ ⎢ ⎥
⎢ 0 0 0 G 0 ⎥ ⎪ε yz ⎪
⎪ ⎪ yz
⎪ ⎪
⎪⎩σ zx ⎪⎭ ⎢⎣ 0 0 0 0 G ⎥⎦ ⎪⎩ε zx ⎪⎭

G = E 2(1 + ν )

ν = coefficient de poisson

Moment-curvature relations for the Mindlin theory are obtained by combining equations 7.18,
7.19 with 7.23 and substituting into equation (7.25)

ENSP Page 92/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
⎧Mx ⎫ ⎡ D νD 0 0 0 ⎤⎧ ∂θ x ∂x ⎫
⎪M ⎪ ⎢νD D 0 0 ⎥
0 ⎥⎪ ⎪ ∂θ y ∂y ⎪
⎪⎪ y ⎪⎪ ⎢ ⎪ ⎪⎪
M
⎨ xy ⎬ = − ⎢ 0 0 D (1 − ν ) 2 0 0 ⎥ ∂θ
⎨ x ∂y + ∂θ y ∂x ⎬ (7.26)
⎪Q ⎪ ⎢ ⎥⎪
⎢0 0 0 Gh 0 ⎥ θ y − ∂w ∂y ⎪
⎪ y ⎪ ⎪ ⎪
⎪⎩ Q x ⎪⎭ ⎢⎣ 0 0 0 0 Gh⎥⎦ ⎪⎩ θ x − ∂w ∂x ⎪⎭

{M } = [DM ]{K }
where D = Eh 3 [12(1 -ν ²)] is the flexural
[D M ] = material matrix and {k} = curvature matrix

7.5. GALERKIN APPROXIMATION

The GALERKIN method of approximate analysis is classified as a “method of weighted
residuals”. The analysis is based upon assuming an approximation, the differential equation
will not be satisfied and there will be an error in the solution. The error (residual) is then
optimized with respect to some parameter, and the optimisation procedure is called a
weighted residual method. A possible assumed solution TR of the differential equation of the
heat conduction in chapter 2 is:
TR = a 0 + a1 + a 2 x 2 + a3 x 3 + ...... (7.27)

ai = unknown constants. The assumed solution must satisfy the boundary condition, and it
follows that equation (7.27) must have at least one or more unknown constant than there are
boundary condition. The exact solution being T, the error or residual R is
R=T-TR (7.28)

The method of weighted residual requires that the unknowns of TR be computed using the
criterion:

Ω
∫ w ( x) R( x, a )dV = 0
i i (7.29)

Where there is a one-to-one correspondence between each wi(x) and R(x, ai); and Ω represent
the domain. In FE analysis, the assumed function ai (7.27) are the shape of the function.

7.6. Problems
7.6.1. Problems N° 1
The classical vibrating string problem is described by equation 7.4. Assume free harmonic
motion for the transverse motion of the string and derive the homogeneous governing
differential equation

Solution1

ENSP Page 93/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
we assume a solution in the form v( x, t ) = V ( x)T (t ) (a) into equation 7.4, separating
the variables:
2
1 d ²V ( x) 1 d ²T (t ) ⎛ w⎞
= = −⎜ ⎟ (b)
v dx ² c ²T dt ² ⎝c⎠
Where w = frequency of the vibration, c = const = f(T,ρ). The solution is governed by
two ordinary differential equations.
2
d ²V ⎛ w ⎞ d ²T
+⎜ ⎟ V = 0 and + w2 = 0 (c)
dx ² ⎝ c ⎠ dt ²
The solution of the second equation is in the form
T = A sin wt + b cos wt (d) and it follows that equation (a) can be replaced
with the assumption
v( x, t ) = V ( x)e −iwt = 0 (d) , into equation (7.4) gives the homogeneous form:

⎡ d ²V ρ ⎤ −iwt ⎡ d ²V ρ ⎤
⎢⎣− dx ² − T w²V ⎥⎦ e = 0 (e) or ⎢ + w²V ⎥ e −iwt = 0 (f)
⎣ dx ² T ⎦

7.6.2. Problem 2:
Do the following for the problem of small deflection of a cable subject to a uniform load
and acted upon an elastic foundation (see §7.21).
Write the variational function
Assume linear shape function and derive the local stiffness matrix for an element of
length L following the method in problem 2.9.2.4

Solution 2

ENSP Page 94/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
a)- Equation (7.3) is the governing differential that we use in the varational function of
equation 2.12) with f(x) = v(x), α = T, β = k and γ = f :

1 ⎧⎪ ⎡ dvx ⎤ ⎫⎪
2

J (v) = ∫ ⎨T ⎢ + k [v ( x ) ]2
− 2 fv ( x ) ⎬dv (a)
V
2 ⎪⎩ ⎣ dx ⎥⎦ ⎪⎭

b)- Equation (a) is written as a matrix equation in terms of the shape.function for an
element of length L with constant area as
L T L L
J (v ) =
A
{v}T ⎡⎢ dN ⎤⎥ [T ]⎡⎢ dN ⎤⎥{v}dx + A {v}T [N ]T [k ][N ]{v}dx − A {v}T [N ]T fdx
∫ ∫ ∫ (b)
20 ⎣ dx ⎦ ⎣ dx ⎦ 20 0

We minimize the function of equation (b) with respect to {v} as :

L T L L
∂J (v) ⎡ dN ⎤
= ⎢∫ ⎥ [T ]⎡⎢ dN ⎤⎥{v}dx + ∫ [N ] [k ][N ]{v}dx − ∫ [N ]
T T
fdx = 0 (c)
∂{v} 0 ⎣ dx ⎦ ⎣ dx ⎦ 0 0

The area term has been divided since a cable would not normal have a variable area.
The final element results for the first and 3rd term are given in §2.9.2.4 with AE replaced
by T. the second term involves the foundation modulus, and the shape function, and the
matrix can be written as:
L L
⎧( L − x) L ⎫ ⎡ L − x x ⎤ ⎧ v1 ⎫ ⎡ ( L − x)² L ² x( L − x)² L ² ⎤ ⎧ v1 ⎫
∫ ⎨ ⎬[k ]⎢ ⎨ ⎬ dx =∫ k ⎢ x( L − x)² L ² ⎨ ⎬dx
0 ⎩ x L ⎭ ⎣ L L ⎥⎦ ⎩v2 ⎭ 0 ⎣
x ² L ² ⎥⎦ ⎩v2 ⎭

⎡ kL 3 kL 6⎤
Integrating and combining term gives the local stiffness matrix ⎢ ⎥ , and the
⎣kL 6 kL 3⎦
complete local stiffness matrix for the small deflection of the cable is:
⎡ T L − T L ⎤ ⎧ v1 ⎫ ⎡ kL 3 kL 6⎤ ⎧ v1 ⎫ ⎧ fL 2⎫
⎢− T L T L ⎥ ⎨v ⎬ + ⎢kL 6 kL ⎨ ⎬=⎨ ⎬
3⎥⎦ ⎩v 2 ⎭ ⎩ fL 2⎭
(d)
⎣ ⎦⎩ 2 ⎭ ⎣

7.6.3. Problem 3:
Derive a four-node element for 2-dimension steady-state heat conduction using the
GALERKIN method.

Solution 3

ENSP Page 95/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
the governing differential equation for steady-state heat conduction is (see equation
2.15):
∂ ²T ∂ ²T
+ ky
kx −Q = 0 (a) The interpolation function gives the
∂x ² ∂y ²
approximation of TR as:
4
TR = ∑ N i ( x, y )Ti = [N ]{T } (b) (Ti = constant).
i =1

Equation (b) into equation (a) gives the residual as a function of x, y and {T}, and the
∂ ²TR ∂ ²TR
result is written as k x + ky − Q = R ( x, y : {T }) (c) an aid in b.
∂x ² ∂y ²
Visualizing the process of forming the residual, (b) into (a) is given in matrix format (see
problem §3.7.2.1):
⎡∂⎤
⎡∂ ∂ ⎤ ⎡k x 0 ⎤ ⎢ ∂y ⎥
⎢ ⎢ ⎥[N N2 N3 N 4 ]{T } − [Q ] = [R( x, y : {T })] (d)
⎢ ∂x
⎣ ∂y ⎥⎦ ⎣ 0 k y ⎥⎦ ⎢ ∂ ⎥ 1
⎣⎢ ∂y ⎦⎥
The weighted residual explained above is formulated as a matrix equation with wi
replaced by [N]and the integral taken over the volume resulting in zero. The first term
of equation (d) is:

⎡ ⎛ ∂² N i ∂ ² N i ⎞⎤
∫ [N ] ⎢⎣∑ ⎜⎜⎝ k ⎟⎥{T }dV = 0
T
+ ky the second term of equation (d) is
∂y ² ⎟⎠⎦
x
∂x ²
V (e)
∫ [N ]
T
QdV
V

The first term of equation (e) after matrix multiplication can be transformed using the
divergence theorem (Green-Gauss theorem) to give:
⎛ ∂ ² N1 ∂ ² N1 ⎞ ⎛ ∂N ∂N ∂N ∂N ⎞ ⎛ ∂ ² N1 ∂ ² N1 ⎞
∫ ⎜⎜⎝ N k
V
1 x
∂x ²
+ N1 k y
∂y ² ⎠ V⎝
∂x ∫
⎟⎟dV = - ⎜⎜ 1 k x 1 + 1 k y 1 ⎟⎟dV + ⎜⎜ N1k x
∂x ∂y ∂y ⎠ S⎝
∂x ² ∫
n x + N1 k y
∂y ²
n y ⎟⎟dS (f)
⎠
the heat source term is identical to the one in equation §3722.f. substituting equation (b)
into equation (e) can be written as:
∂ ²[N ]
kx {T } + k y ∂ ²[N ]{T } − Q = R( x, y : {T }) (g)
∂x ² ∂y ²
∂ ²[N ]
Or in a more abbreviated form as [k ] {T } − Q = R( x, y : {T }) (h) and (e) will be
∂xi2
∂ ²[N ]
⎛ ⎞
written as ∫ ⎜⎜ [N ]T [k ] 2 {T } − [N ]T Q ⎟⎟dV = 0 (i)
V ⎝ ∂xi ⎠
The shape function matrix and an operator matrix are

ENSP Page 96/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
⎡∂ 2 ∂x ² 0 ⎤
[N ] = ⎡⎢
N1 N2 N3 N4 ⎤
⎥ and [L ] = ⎢ 2 ⎥ ( j)
⎣ N1 N2 N3 N4 ⎦ ⎣ 0 ∂ ∂y ² ⎦

The use of equations (j) and (h) will give the equivalent of equation (d).It follows from
equation (f) that the final result is:
⎛ ∂[N ]T
[k ] ∂[N ] {T } + [N ]T Q ⎟⎟dV = ⎜⎜ [N ]T [k ] ∂[N ] {T }ni ⎟⎟dS
⎞ ⎛ ⎞
∫⎜
⎜ ∂x
V⎝ i ∂xi ⎠ S⎝
∫ ∂xi ⎠
(k)

An analysis must interpret the specific form of [N], xi, and {T] that is to be used since
they must correspond to the element and coordinate system being used.

7.6.4. Problem 4:
DESSIN

The rectangular vibrating membrane is the 2-dimension counterpart of the vibrating

string and is described by an equation of the form:
∂²w ∂²w ∂²w
T +T =ρ (a) Where w = deflection, T = tension in the
∂x ² ∂y ² ∂t ²
membrane, and ρ is the density.
Formulate the FE model for the vibrating membrane and obtain a solution using the
model illustrated.

Solution 4

ENSP Page 97/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
Assuming a solution of equation (a) in the form w(x,y,t) = W(x,y)e-iwt will lead to an
analysis of the equation
∂ ²W ∂ ²W ρω 2
+ + λW = 0 λ= (b)
∂x ² ∂y ² T

Applying the GALERKIN approximation for equation (b) will lead to the result:
∂[N ] ∂[N ]
T
− λ [N ] [N ] = 0
T
(c)
∂xi ∂xi

The membrane is square and can be modelled using four squares elements, and the
same result would be computed using symmetry. Adopting kx = ky = 1 and result of
problem (§3.7.2.2) with b.c W1 = W3 = W4 = 0, we get only one equation (for node 2).
1
1 ⎡ a² a² ⎤ a² 1 24T 4,899 ⎛ T ⎞ 2
2 + 2 − λ = 0 , and it follows that ω 2
= or ω = ⎜ ⎟ (the
6(a 2 ) ⎢⎣ 4
2
4 ⎥⎦ 4 9 a² ρ a ⎜⎝ ρ ⎟⎠
1
4,443 ⎛ T ⎞ 2
exact solution is ω = ⎜ ⎟ )
a ⎜⎝ ρ ⎟⎠

7.6.5. Problem 5:
The differential equation that describes the transverse motion of a beam structure is
similar to equation (4.1) and was derived by TIMOSHENKO as:
∂ 4v ∂ 2v
EI = − ρ (a) ρ = mass density per unit length.
∂x 4 ∂t 2
Assume free harmonic motion and derive the corresponding local beam F.E.
Apply the results on a simply supported beam element at both ends

Solution 5

ENSP Page 98/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
1)- The beam F.E. that models the left hand side off equation (a) was derived and
analysed I chapter 4. It remains to develop a model for the dynamic term on the right-
hand side of the equation .the use of the GALERKIN method to derive the FE model is
adopted following equations (§4.6.3.1.g) as a trial solution:
v( x, t ) = [ N ( x)]{v(t )} (b) where [N(x)] = [N 1 N2 N3 N4 ]
and {v(t)} = [v1 v4 ] .
T
v2 v3

Substitute the trial solution into the governing equation:

∂ 4v ∂ 2v
EI {v} + ρ [ N ] = R(x, t : {v}) (c)
∂x 4 ∂t 2
Multiplying by the weight function and integrating over the length
L L
∂ 4v ∂ 2v
∫ EI [ N ] T
{v}dx + ρ [ N ∫
] T
[ N ] dt = 0 (d)
0
∂x 4 0
∂t 2

Integrate the first term by parts twice:

L L
⎡ ∂ 2 [ N ]T ∂ 2 [ N ] ∂ 2v ⎤ ⎡ ∂[ N ] ∂ 2 [ N ] ∂ 3v ⎤
∫⎢ EI {v} + ρ [ N ]T
[ N ] ⎥ dx = EI ⎢ {v} − [ N ] {v}⎥
∂x 2 ∂x 2 ∂t 2 ⎦ ⎣ ∂x ∂x
2
0⎣
∂x 3 ⎦ 0

The terms to the right of the equal sign represent the boundary condition on the
deflection and slope (moment and shear) of chapter 4. The free vibration problem
without transverse loading is governed by the differential equation within the volume
integral. Assume the solution as v( x, t ) = V ( x)e −iωt and substitute into the di equation to
obtain the corresponding eigenvalue problem:
L
⎡ ∂ 2 [ N ]T ∂ 2 [ N ] ⎤
∫ ⎢ EI {v} + ρω 2
[ N ] T
[ N ]{v}⎥ dx = 0 (e)
0 ⎣
∂x 2 ∂x 2 ⎦
The first term of equation (e) is the stiff matrix of chapter 4; the second term of equation
(e) defines the mass matrix:
L
[m] = ∫ [ρ[ N ]T [ N ]]dx (f)
0

Shape functions Ni of chapter 4 are used to evaluate the mass matrix as follows:
T
⎧ 2 x 3 − 3 x 2 L + L3 ⎫ ⎧ 2 x 3 − 3x 2 L + L3 ⎫ ⎡ 156 22 L 54 − 13L ⎤
⎪ ⎪ ⎪ 3 3⎪ ⎢ 22 L
L
ρ ⎪ x 3 L − 3 x 2 L2 + xL3 ⎪ 1 2 2
⎪ x L − 3x L + xL ⎪ ρL ⎢ 4L 2
− 13L − 3L2 ⎥⎥
[m] = ∫ ⎨ ⎬ ⎨ ⎬ dx = (g)
L3 ⎪ − 2 x 3 + 3 x 2 L ⎪ L3 3
⎪ − 2 x + 3x L ⎪
2
420 ⎢ 54 − 12 L 156 − 22 L ⎥
0
⎪ x 3 L − x 2 L2 ⎪ ⎪ x L−x L ⎪ ⎢ 2 2 ⎥
⎣− 13L − 3L − 22 L 4 L ⎦
3 2 2
⎩ ⎭ ⎩ ⎭
In summary, the eigenvalue problem for harmonic motion of a beam can be written as:
[K ]{v} − ω 2 [m]{v} = 0 (h)

2) - With matrices [K] and [M] known, b.c. are v(0) = v(L) ⇒ deleting the first and third
rows and columns of both matrices we get:

ENSP Page 99/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
EI ⎡4 L2 2 L2 ⎤ ω 2 ρL ⎡ 4 L2 − 3L2 ⎤
⎢ 2 ⎥− ⎢ ⎥=0 (i)
L3 ⎣2 L 4 L2 ⎦ 420 ⎣− 3L2 4 L2 ⎦

Solving, we get
1
10,954 ⎛ EI ⎞ 2
ω1 = ⎜ ⎟ ∆ω 1 = 11%
L2 ⎜⎝ ρ ⎟⎠
1
( j)
50,120 ⎛ EI ⎞2
ω2 = ⎜ ⎟ ∆ω 2 = 27%
L2 ⎜⎝ ρ ⎟⎠
1
n 2π 2 ⎛ EI ⎞ 2
exact ω n = 2 ⎜⎜ ⎟⎟
L ⎝ ρ ⎠
explain ∆ω i ?

7.6.6. Problem 6
The differential equation that describes the buckling of a long column is similar to
equation (4.1) and is derive in elementary mechanics of materials as:
d 4v d 2v
EI + ρ =0 (a) where ρ = concentrix axial force applied at the end of the column .
dx 4 dx 2
Derive the local finite element that can be used to study the buckling of a column with
concentric axial loading.

Dessin

Solution 6
The derivation follows that of problem §75.5.1 where the beam FE that models the first
term of equation (a) was derive using the GALERKIN method. This technique can be
used to derive the FE model for the second term of equation (a). Assume the trial
solution as in problem §7.5.5.1
v( x, t ) = [ N ( x)]{v(t )} (b) where [N(x)] = [N 1 N2 N3 N4 ]
and {v(t)} = [v1 v4 ] .
T
v2 v3

Substitute the trial solution into the governing equation, and multiply the weight
function and integrate over the length:
d 4 [N ] T d [N ]
L L 2

∫
T
EI [ N ] {v}dx + ∫ ρ [ N ] dx = 0 (c)
0
dx 4 0
dx 2

Integrate the first term by part twice and the second term once. As in problem §7.5.5.1, there
will be boundary condition for node (point) loading that represent shear moment, and axial
force. Neglect the possible point loading and concentrate on the eigenvalue problem
L
⎡ d 2 [ N ]T d 2 [ N ] d [ N ]T d [ N ] ⎤
governing column buckling: ∫ ⎢ EI {v} + ρ {v}⎥ dx = 0 (d)
0 ⎣ dx 2 dx 2 dx dx ⎦

ENSP Page 100/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
The first term of equation (d) is the stiffness matrix of problem §7.5.5.1. The second
term must be evaluated using the derivatives of the shape function given in equation (c) -
(f) of problem 4.6.3.1. The matrices inside the integral are multiplied together as follows:
⎧ 6 x 2 − 6 xL ⎫
⎪ 2 3⎪
ρ ⎪3x L − 4 xL + L ⎪ 1
[ ]
2
2 2 2 3 T
3 ⎨ 2 ⎬ 3 6 x − 6 xL 3x L − 4 xL + L − 6 x 2 + 6 xL 3 x 2 L − 2 xL2 and after
L ⎪ − 6 x + 6 xL ⎪ L
⎪ 3x 2 L − 2 xL2 ⎪
⎩ ⎭
integrating and substituting limits, the matrix represents the effect of axial force on
bending stiffness becomes:
⎡ 36 3L − 36 3L ⎤
⎢ 3L 4 L2 − 3L L2 ⎥
[D] = ρ ⎢⎢ ⎥ (e)
30 L − 36 − 3L 36 − 3L ⎥
⎢ ⎥
⎣ 3L L2 − 3L 4 L2 ⎦

In summary, the eigenvalue problem for the critical buckling load for a column can be
written:
[K ]{v} − λ2 [D]{v} = 0
ρ (f)
Where λ2 =
EI

7.6.7. Problem 7
Derive a load stiffness matrix for plate bending using the Mindlin theory and equation
(7.26) as the basic relationship than that can be used to evaluate the strain energy that
corresponds to plate bending. The use of the strain energy to derive plat elements follows the
concepts developed in chapter 4 for deriving the beam FE.

Solution 7:
The FE that is derived has been named the “hétérosis element” by HUGHS (1987)
and the discussion here follows that of Cook et al (1989). The development is general and
applies to any element; the element unknowns are the plate rotation and transverse
deflection, and are defined in terms of nodal DOF for any element using a suitable shape
function. Assume the followings relation between element unknowns and nodal
unknowns:

⎧ ω e ⎫ ⎡ N1 0 0 N2 0 0 . . repeat . . . Nn 0 0 ⎤⎧ ω ⎫
⎪ ⎪ ⎢ ⎪ ⎪
⎨θ xe ⎬ = ⎢ 0 N1 0 0 N2 0 . . for . . . 0 Nn 0 ⎥⎥ ⎨θ x ⎬ (a)
⎪θ ⎪ ⎢ 0 0 N1 0 0 N2 . . elementmethod . . . 0 0 N n ⎥⎦ ⎪⎩θ y ⎪⎭
⎩ ye ⎭ ⎣
{U } = [N ]× {d } where

ENSP Page 101/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
{U} are the element unknowns
[N] are element shape function and {d} are the nodal unknowns :
{d } = [ω1 θ x1 θ y1 ω 2 θ x 2 θ y 2 ... ... ω n θ xn θ yn ]T (b)

In general, the strain energy of deformation is for the plate can be written
h
2
1
∫ ∫ {δ } [E ]{δ }dzdA
T
U= (c) A = midsurface area of the plate.
2 A h
−
2
{δ} = is given in equation (7.25)
The integration is through the thickness of the plate, and equations §7.4 are used to
convert the strain energy into matrices corresponding equations 7.26 as follows:
1
U= {K 0 }T [DM ]{K 0 }dA
∫ (d)
2A

There are 3 DOF at each node ω, θx, θy, that are defined by equation (a). the curvature
are defined in term of nodal unknowns using an operator matrix
{K 0 } = [L ][N ]{d } = [B ]{d } (e)
Where [L] is the following 5 × 3 matrix:
⎡ ∂ ⎤
⎢ 0 ∂x
0⎥
⎢ ∂⎥
⎢ 0 0 ⎥
⎢ ∂y ⎥
⎢ ∂ ∂⎥
L=⎢ 0 ⎥ ( f ) . The strain energy can be defined as:
⎢ ∂y ∂x ⎥
⎢− ∂ 0 1⎥
⎢ ∂y ⎥
⎢ ∂ ⎥
⎢− 1 0⎥
⎣ ∂x ⎦
1 1
∫ {d }[N ] [L] [D ][L][N ]{d }dA = 2 ∫ {d } [B] [D ][B]{d }dA
T T T T
U= M M (g)
2 A A

ENSP Page 102/103

FINITE ELEMENT METHODS / 7. SELECTED TOPICS IN FEM. Version ou da
Dr LEZIN
Equation (g) leads to the formulation of stiffness matrix as:

[K ] = 1 ∫ [B ]T [DM ][B]dA (h)

2A

The derivation is the general one, and equation (h) can be used to formulate a FE using
any desired shape function. The stiffness matrix can now apply the heterosis element
formulation.
The heterosis element shown as a 9-node quadrilateral isoperimetric plate element. The
corner and midside nodes have 3 DOF (θx, θy, v) where as the center node has only 2-
DOF (θx, θy).
The element has 26 DOF. The Lagrange shape function is used to model the rotation θx
and θy, and the standard 8-node element is used to model the transverse plate deflection
ω. It follows that the element stiffness matrix (equation (f)) is split into two stiffness
matrices, one to model bending and one to and one to model transverse sheat:
[K ] = [K M ] + [K S ] (i)

Where:
[K b ] = ∫ [Bb ]T [Db ][Bb ]dA
A
( j ) Splitting the stiffness matrix is best visualized
[K s ] = ∫ [BS ]T [DS ][BS ]dA
A

by examining the result [B] = [L] [N] defined by equation (e). Let P denote the mine
node Lagrange shape function (rotation) and N represent the standard 8-node shape
function (deflection). Note that it is possible to write matrix equation similar to equation
(a) to represent the combined shape function P and N. The heterosis [B] matrix is
constructed
as:
⎡ 0 ∂R ∂x 0 0 ∂P2 ∂x 0 . . . . ∂P9 ∂x 0 ⎤
⎢ 0 0 ∂P1 ∂y 0 0 ∂P2 ∂y . . . . 0 ∂P9 ∂y ⎥⎥ [3 × 26]
⎡[Bb ]⎤ ⎢
[B] = ⎢ ⎥ = ⎢ 0 ∂P1 ∂y ∂P1 ∂x 0 ∂P2 ∂y ∂P2 ∂x . . . . ∂P9 ∂y ∂P9 ∂x ⎥ (k)
⎣[B S ]⎦ ⎢− ∂N ∂y 0 P1 ∂R ∂y 0 P2 . . . . 0 P9 ⎥
⎥
⎢ 1
⎢ − ∂N 1 ∂x
⎣ P1 0 ∂R ∂x P2 0 . . . . P9 0 ⎥⎦ [2 × 26]
⎡ D νD 0 ⎤
[D S ] = ⎡⎢
⎢ ⎥ Gh 0 ⎤
[Db ] = ⎢ D 0 material matrix from eq 7.26
⎥
⎣ 0 Gh⎥⎦
⎢⎣ sym (1 − ν ) D 2⎥⎦

7.6.8. Examples of structures and analysed with FEM

In the local coordinate, there is no θz rotation, but there is a corresponding projection θz in the
general coordinate system from local θx and θy.
The modelisation can also be done with any 2-D triangular and rectangular element, with
augmented (or not) node in the middle in a different plane.

dessin

ENSP Page 103/103