Generation of shape functions for straight

beam elements
Charles E. Augarde
Originally published in Computers and Structures, 68 (1998) 555-560
Corrections highlighted.

Abstract

Straight beam finite elements with greater than two nodes are used for edge
stiffening in plane stress analyses and elsewhere. It is often necessary to
match the number of nodes on the edge stiffener to the number on a whole
plane stress element side. Beam elements employ shape functions which are
recognised to be level one Hermitian polynomials. An alternative to the
commonly adopted method for determining these shape functions is given
in this note, using a formula widely reported in mathematical texts which
has hitherto not been applied to this task in the finite element literature.
The procedure derives shape functions for beams entirely from the set of
Lagrangian interpolating polynomials. Examples are given for the derivation
of functions for a three and four-noded beam elements.

keywords: finite elements, beams, Hermitian interpolation, shape func-

tions

1
Introduction

Analysis of structures using the finite element method is well-established.

Many formulations exist for complex elements but simple elements remain
popular since they are usually well-tested and easy to implement into an
analysis program. Two-dimensional plane stress analysis, for thin structures
subject to in-plane loading, may employ continuum elements, such as the
fifteen-node triangle, having a large number of nodes along a side . Where
edge stiffening is required, beam elements can be connected to continua edge
nodes. There is then a requirement for formulations of beam elements having
more than two nodes.

Conventional two-dimensional beam elements have two degrees of freedom

at each node: one lateral displacement and one rotation. Unless the struc-
ture is loaded entirely laterally, axial stiffness must also be incorporated, by
an additional degree of freedom at each node. With this amendment, beam
elements are usually referred to as frame elements. The axial effects are un-
coupled for straight beams and the determination of suitable shape functions
is straightforward [1].

The author has recently contributed to the development of a complex three-

dimensional finite element model at the Department of Engineering Science,
Oxford University to study the damage accruing to surface structures from
adjacent tunnelling [2]. Modelling was also undertaken in two-dimensions
(to validate the complex model) where tunnel linings were represented with
simple beam elements. The procedure outlined in this note was used in
the implementation of these beam elements into an existing in-house finite
element code (OXFEM).

In this note we are concerned with the generation of the shape functions which
interpolate the lateral displacements along beam elements having more than
two nodes. Bernoulli-Euler beam theory is assumed where transverse shear
deformation is zero. While most finite element texts describe the simple two-
noded beam [1,4] few explain how more complex elements may be formulated
[3].

2
Standard procedure

The beam element of Figure 1a is defined by two nodes, a distance l apart

along the element local x−axis. A common convention, and that adopted
here, is to make the nodal coordinates non-dimensional that is
x
x̄ = (1)
l
so that the two-noded beam has nodes at x̄ = 0, +1.

Using shape functions, lateral displacement w(x̄) of the two-noded beam

element of Figure 1a is
w(x̄) = N d (2)
where

N = { N1 N2 N3 N4 } & dT = { w1 θ1 w2 θ2 } (3)

Nj , (j = 1, 4) are the bending shape functions and wi , θi , (i = 1, 2) are the

displacements and rotations at the nodes. Determination of the former, using
the method to be found in many finite element texts [4,5], proceeds by first
writing w(x̄) as an n-termed polynomial with unknown coefficients, n being
the number of degrees of freedom in the element

w(x̄) = Xα (4)

where,

X = { 1 x̄ x̄2 x̄3 } & αT = { α1 α2 l α3 l2 α4 l3 } (5)

Taking equation (4) and its first derivative with respect to x̄, four further
equations can be formed, one for each degree of freedom:

d = Aα (6)

where  
1 0 0 0
 0 1 0 0 
A=


 (7)
 1 1 1 1 
0 1 2 3

3
Solving for α in Equation (6) and substituting into Equations (4) and (2)
gives
N = XA−1 (8)

The three-node beam of Figure 1b, of overall length 2l, has nodes at x̄ =
−1, 0, +1. The same procedure as above yields

N = { N1 N2 N3 N4 N5 N6 }, (9)
dT = { w1 θ1 w2 θ2 w3 θ3 }, (10)
X = { 1 x̄ x̄2 x̄3 x̄4 x̄5 } (11)
αT = { α1 α2 l α3 l2 α4 l3 α5 l4 α6 l5 } (12)

and to solve for N requires the inversion of

 
1 −1 1 −1 1 −1

 0 1 −2 3 −4 5 
 
 1 0 0 0 0 0 
A=

 (13)
 0 1 0 0 0 0 
 
 1 1 1 1 1 1 
0 1 2 3 4 5

Hermitian interpolation

The shape functions in Equation (2) are Hermitian polynomials since the
displacement w(x) is interpolated from nodal rotations as well as nodal dis-
placements. This contrasts with Lagrangian interpolation, used for contin-
uum elements’ shape functions and for the axial effects in frame elements.
Considering small displacements, the nodal rotations are the first deriva-
tives of the unknown real displacement function at the nodes thus fulfilling
the definition of Hermitian interpolation. This property allows an alternative
procedure to be used to determine the shape functions to that outlined above

The generation of Hermitian (or Hermite) polynomials from Lagrangian in-

terpolation polynomials is described in many mathematical texts [6,7,8]. De-
spite the clear understanding that bending shape functions are equivalent to
Hermitian polynomials the technique described below has not, to the author’s
knowledge, been linked to the generation of those shape functions.

4
One-dimensional interpolation is required for straight beam elements. The
single dimension is along the element centreline, defined as the x-axis. One-
dimensional Hermitian interpolation for an unknown, w(x) proceeds as [6,7]

nnod
" Ã ! Ã !#
X
r r dw r dr w
w(x) = H0i wi + H1i + . . . Hri (14)
i=1 dx i
dxr i

where Hjir is a Hermite polynomial of level r, relating to node i and to deriva-

tive order j of w. The sum is over the number of nodes, nnod where values
of w and its derivatives are available. The level of the polynomial indicates
the highest order derivative used in the interpolation.

Comparison of Equations (2) and (14) reveals that the bending shape func-
tions are level one Hermitian polynomials as follows:
1 1 1 1
N1 = H01 N2 = H11 N3 = H02 N4 = H12 (15)

Level one Hermitian polynomials are derived from Lagrangian polynomials

by the following formulae:
1
H0i = [1 − 2(x − xi )L0i (xi )][Li (x)]2 (16)
1
H1i = (x − xi )[Li (x)]2 (17)

where Li (x) is the one-dimensional Langrangian polynomial of degree (nnod−

1) calculated at node i, given by
nnod
Y x − xj
Li (x) = (18)
j=1,j6=i xi − xj

and L0i (x) is its first derivative with respect to x. A polynomial of order
(nnod × 2 − 1) is required to interpolate over nnod points, each contributing
two values. Inspection of Equations (16) and (17) shows that the Hermite
polynomials are of the correct order for interpolation.

Re-writing Equations (16) and (17) in terms of the non-dimensional coordi-

nate, x̄
1
H0i = [1 − 2l(x̄ − x̄i )L0i (x̄i )][Li (x̄)]2 (19)
1
H1i = l(x̄ − x̄i )[Li (x̄)]2 (20)

5
The advantage of a derivation based on the Lagrangian polynomials and their
first derivatives is that these are already likely to be present in a program
code. The former are required for continuum elements and the derivatives are
required for isoparametric elements. The use of this procedure also provides
a systematic approach to allow simpler coding.

Examples

Three-node beam

A three-node element with six bending degrees of freedom and total length
2l is shown in Figure 1b. This element has nodes at x̄ = (−1, 0, 1). The axial
degrees of freedom are omitted from this element as in the derivations above.
Shape functions for these degrees of freedom are the Lagrangian polynomials
of order 2. From the preceding section, it is clear that these are also required
for the derivation of the bending shape functions.

The Lagrangian polynomials are

(x − x2 )(x − x3 )
L1 (x) = (21)
(x1 − x2 )(x1 − x3 )
(x − x1 )(x − x3 )
L2 (x) = (22)
(x2 − x1 )(x2 − x3 )
(x − x1 )(x − x2 )
L3 (x) = (23)
(x3 − x1 )(x3 − x2 )
Rewriting in terms of the non-dimensional coordinate x̄ and substituting for
values of x̄ at nodes (i.e. (x̄1 , x̄2 , x̄3 ) = (−1, 0, 1) ), gives
x̄
L1 (x̄) = (x̄ − 1) (24)
2
L2 (x̄) = 1 − x̄2 (25)
x̄
L3 (x̄) = (x̄ + 1) (26)
2
The derivatives are
1
L01 (x̄) = (2x̄ − 1) (27)
2l
6
 2
L02 (x̄) = − (x̄) (28)
l
0 1
L3 (x̄) = (2x̄ + 1) (29)
2l

The bending shape functions are equivalent to the following Hermite poly-
nomials:
1 1 1
N1 = H01 N2 = H11 N3 = H02 (30)
1 1 1
N4 = H12 N5 = H03 N6 = H13 (31)
Note that the first four are not the same as those in Equation (15) since each
set is based on different order Lagrangian polynomials.

From (24) and (27) and with substitution for x̄1 we obtain
· µ ¶¸ · ¸2
1 3 x̄ 5 1 3
H01 = 1 − 2l(x̄ + 1) − (x̄ − 1) = x̄2 − x̄3 − x̄4 + x̄5 (32)
2l 2 4 2 4
Similarly,
· ¸2
1 x̄ lh 2 i
H11 = l(x̄ + 1) (x̄ − 1) = x̄ − x̄3 − x̄4 + x̄5 (33)
2 4
h i2
1
H02 = [1] (1 − x̄2 ) = 1 − 2x̄2 + x̄4 (34)
h i
1
H12 = lx̄ [−x̄(x̄ + 1)]2 = l x̄ − 2x̄3 + x̄5 (35)
· µ ¶¸ · ¸2
1 3 x̄ 5 1 3 5
H03 = 1 − 2l(x̄ − 1) (x̄ + 1) = x̄2 + x̄3 − x̄4 − x̄
(36)
2l 2 4 2 4
· ¸2
1 x̄ lh 2 i
H13 = l(x̄ − 1) (x̄ + 1) = −x̄ − x̄3 + x̄4 + x̄5 (37)
2 4

The suitability of the above as shape functions can be demonstrated by plot-

ting each polynomial (Figure 2). The polynomials related to displacements
have value 1 at the associated node and zero elsewhere. Those related to
derivatives have slopes of 1 at the associated node.

Four-node beam

Figure 1c shows a beam with four nodes having the same overall length as
the three-node beam. The non-dimensional coordinates (x̄) of the nodes are

7
−1, −1/3, +1/3 and +1. Eight shape functions are required, derived from
the following Lagrangian polynomials and derivatives:
· ¸
9 1 x̄
L1 (x̄) = − + + x̄2 − x̄3 (38)
16 " 9 9 #
27 1 x̄2 3
L2 (x̄) = − x̄ − + x̄ (39)
16 3 3
" #
27 1 x̄2
L3 (x̄) = + x̄ − − x̄3 (40)
16 3 3
· ¸
9 1 x̄ 2 3
L4 (x̄) = − − + x̄ + x̄ (41)
16 9 9
· ¸
9 1
L01 (x̄) = + 2x̄ − 3x̄ 2
(42)
16l 9
· ¸
27 2
L02 (x̄) = −1 − x̄ + 3x̄ 2
(43)
16l 3
· ¸
27 2
L03 (x̄) = 1 − x̄ − 3x̄2 (44)
16l 3
· ¸
9 1
L04 (x̄) = − + 2x̄ + 3x̄ 2
(45)
16l 9
The first two shape functions, relating to lateral displacement and rotation
at node 1, can then be derived as
1 h i
N1 = 13 − 15x̄ − 243x̄2 + 281x̄3 + 1215x̄4 − 1413x̄5 − 729x̄6 + 891x̄ 7
(46)
512
l h i
N2 = 1 − x̄ − 19x̄2 + 19x̄3 + 99x̄4 − 99x̄5 − 81x̄6 + 81x̄7 (47)
256
Plotting these two functions (Figures 3a & b) shows that they also satisfy
the basic requirement of shape functions as outlined above.

Conclusion

This work presents an alternative derivation of bending shape functions for

simple beam elements, for implementation of many-noded straight beam el-
ements within a finite element analysis code. While the elements described
are simple the theory will be of interest to developers of other C1 continuous
elements such as rectangular plates.

8
Acknowledgements

This work was carried out as part of an EPSRC funded project examining
numerical modelling of tunnelling at the Department of Engineering Science,
University of Oxford, UK under the guidance of Dr. H.J. Burd. The author
would like to acknowledge the help and contribution of Dr. Burd to this
work.

References

1. Astley, R.J., Finite Elements in Solids and Structures, Chapman & Hall,
London, 1992.

2. Augarde, C.E., Numerical modelling of tunnelling processes for assessment

of damage to buildings, D.Phil thesis, University of Oxford, 1997.

3. Dawe, D.J., Matrix and Finite Element Displacement Analysis of Struc-

tures, Clarendon Press, Oxford, 1984.

4. Cook, R.D., Concepts and Applications of Finite Element Analysis, 2nd

edn., John Wiley & Sons, Chichester, 1981.

5. Mohr, G.A., Finite Elements for Solids, Fluids and Optimisation, Oxford
University Press, Oxford, 1992.

6. Jacques, I. and Judd, C., Numerical Analysis, Chapman & Hall, London,
1987.

7. Morris, J.L., Computational Methods in Elementary Numerical Analysis,

John Wiley & Sons, Chichester, 1983.

8. Spanier, J. and Oldham, K.B., An Atlas of Functions, Hemisphere, Lon-

don, 1987.