Finite Element Types

and
Analysis of 3D Problems
Jayadeep U. B.
M.E.D., NIT Calicut
Ref.: 1. Finite Elements and Approximation, Zienkiewicz, O. C.,
and Morgan, K., John Wiley & Sons.
2. Zienkiewicz, O. C., The Finite Element Method, Tata
McGraw Hill Edition.

Department of Mechanical Engineering, National Institute of Technology Calicut

Introduction
Till now, we have been mostly considering the linear shape
functions. In 1D, this gave us linear elements; while in 2D,
this resulted in linear triangles and bi-linear rectangles.
However, while introducing the concept of shape functions in
1D, the possibility of using higher order shape functions as a
means for improving the accuracy of approximation, without
refining the mesh was discussed.
We derived the quadratic shape functions in 1D and also, the
possibility of higher order shape functions was discussed in
2D, for achieving greater accuracy.
This section is intended to be a general discussion on higher
order elements and an introduction to finite element analysis
of 3D problems.

Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 1D

In 1D, the shape function of any order can be expressed by
the Lagrange Interpolation functions. In general, the shape
function for node l using a Lagrange interpolation function
of order p is given as:

e
l

x x0 x x1 L x xl 1 x xl 1 L x x p

xl x0 xl x1 L xl xl 1 xl xl 1 L xl x p

From this, using the natural coordinates, we can obtain the

shape functions:
1 & N e 1
Linear: N 0e
1
2
2
1
1 1
1

e
e
e
Qudratic: N 0
, N1
& N2
2
2
2
Cubic and higher order shape functions can be similarly
obtained.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 1D

contd.

This type of formulation is most commonly used in FEM, and

is called the Lagrangian Family of Elements.
However, there is a major disadvantage for this formulation, if
we are planning to improve the accuracy by successively
increasing the order of shape functions.
The shape functions of any given node are different in each
case and therefore, we will not able to re-use any part of the
stiffness matrix calculated for a lower order shape function.
The issue can be (partly) rectified using the shape functions
called Hierarchical shape functions.
The basic idea here is to increase the order of polynomials
using additional terms, without changing the original shape
function, and so stiffness matrix of a lower order element
becomes part of new elements. Details left for self-study.
However, the more preferred method is mesh refinement.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 2D

A similar procedure can be adopted in 2D, especially for the
rectangular type Lagrangian elements.
When two perpendicular natural coordinate directions are
defined as discussed earlier, any order shape functions can be
obtained as products of linear shape functions. For example:
1 1

e
Bi-Linear: N 0
4
1 1
e
Bi-Quadratic: N 0

2
2

Cubic and higher order shape functions can be similarly

obtained.
The bilinear element has a linear variation for shape functions
along the edges, while it has a quadratic variation along a line
connecting the corners. This can be observed by substituting
= in the above equation.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 2D

contd.

This can be expressed in the form of Pascals triangle, giving

the terms represented by the shape functions:
1

2
3

2
2

However, the approximation is exact only up to the level of

the completed polynomials. In other words, there could be
errors of the order of 2 in case of a bi-linear element.

Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 2D

contd.

A similar analysis on bi-quadratic element gives:

1

2
3
4

2
3

2
2

2 2

3
3

In other words, there could be errors of the order of 3, even

though there 4th order terms in the shape functions.
The shape functions of 4th order is unnecessary as we have
only one term of the 4th order polynomial available.
This leads to the formulation of the Serendipity family of
elements, where this higher order terms are not present.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Higher Order Shape Functions 2D

contd.

The eight-noded serendipity element and the completed

polynomial is given in the figure below:

2
3
4

2
3

2
2

2 2

3
3

Similarly there can be 12 noded and even higher order

serendipity elements.
Further details on this family of elements is left for self-study.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Introduction to 3D FEM
Examples of 3-D Problems Heat Transfer Problem:
The steady-state 3D heat transfer equation is:

Q k k k Q 0
x x
y y
z
z
2

The major differences between the 3D and 2D heat transfer

problem are due to geometry:
The domain is a volume (3D space) as against an area in
case of 2D.
The boundary is a surface, instead of a curve.
Any heat transfer problem is actually a 3D problem. Hence,
any problem can be analyzed as 3D, but simplifications are
possible to 1D or 2D problems in many cases.

Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Introduction to 3D FEM contd.

Examples of 3-D Problems Stress Analysis:
There are no simplifications involved in a general 3D
formulation, and hence, there are no further classifications in
3D problems.
The equilibrium equations in 3D are:

xy y yz
x xy xz

X 0,

Y 0 &
x
y
z
x
y
z
xz yz z

Z 0
x
y
z
There could be three independent displacements at any point,
and hence there are 3 d.o.f. per node.
The stress and strain vectors have six elements, and hence the
strain-displacement and stress-strain relations gets modified
accordingly.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

General Comments on FEM in 3D

The finite element analysis in 3D is computationally more
complicated due to larger number of unknowns and equations.
This is especially true in the vector field problems, like stress
analysis, since the number of unknowns per node also is
higher than 1D and 2D problems.
However, most of the derivations used in 2D could be easily
extended, after making proper modifications, to 3D.
Also, the 3D problems could be conceptually simpler, since
we do not make much assumptions in 3D formulation.
However, due to computation reasons, it is preferable to
perform a 1D or 2D analysis, if it is justifiable.

Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Finite Elements in 3D
The extension of a linear triangle in 2D to 3D is the Fournoded Tetrahedron Element.
As in case of triangular elements, if used in case of stress
analysis, we get a constant strains/stresses everywhere in a
given element, since the displacement has linear variation.
The standard tetrahedron element in natural coordinates is
shown in figure below:
3
The weights and sampling
(0,0,1)
points for Gauss Quadrature
method can be easily
obtained.

(0,1,0)
(1,0,0)

Higher order elements, with quadratic or even higher order

variation for field variable, can be formulated as in 1D & 2D.
Lecture - 01

Department of Mechanical Engineering, National Institute of Technology Calicut

Finite Elements in 3D
The extension of a bi-linear rectangle is the Eight-noded Brick
Element.
For better accuracy, we can have the higher order elements of
Lagrangian family, or the higher order Serendipity elements.
The standard brick element in natural coordinates is shown in
figure below:
(-1, -1, 1)
(1, -1, 1)
(1, -1, -1)

(-1, 1, 1)

The weights and sampling

points for Gauss Quadrature

method can be easily

obtained by extending the
(-1, 1, -1)
same from 1D.

(1, 1, -1)

Utilizing the parametric formulation, any linear boundary can

be exactly matched by this element. For curved boundaries,
either linear approximation or higher order mapping is used.
Lecture - 01