FINITE ELEMENT ANLYSIS OF SKELETON STRUCTURES

1. Introduction
It has been mentioned earlier that a continuum is divided in to a number of units or elements.
This is also applied to skeleton structures and it gives a number of members or elements
which are connected at a number of joints or nodes. The basic differences in the two cases are
as follows. In skeleton structures, the nodes are taken at the physical joints, supports and
terminating points where the members in between these nodes are treated as single element
which is shown with a frame structure in Fig.1 where the number of nodes is four and the
number of elements is three. In this case, it is not required to subdivide the members as the
variation of displacement and member forces within the members follow a well defined
pattern which helps to idealised these type of structure as a discrete system. Now a prismatic
folded plate structure consists of three rectangular flat plates and having a cross-section
somewhat similar to the geometry of the frame shown in Fig. 1 may be considered as an
example of an elastic continuum. In this case the plates in between two joints or edges are
subdivided in both the directions which generate a number of nodes within the flat panel.
Actually, the solution accuracy in these problems depends on the number of elements as the
variation of displacement and stresses within an element is not known and it is approximated
in terms of nodal displacements of that element. It is basically a continuous system where the
variation of displacement and stresses do not follow any definite pattern even in a regular
portion. As the purpose of dividing a structure in to a number of elements is to represent the
structure as a discrete system, this process is called discritisation. It is the first step used in
any finite element analysis. Now consider the loads which act on the structure. It is
represented as a system of forces acting at the discrete nodal points and the system of force is
known as nodal loads or nodal force vector. The next job is to establish the relationship
between these two quantities i.e., the nodal loads and nodal displacements which is explained
in the following section. In this section, all the steps or procedure followed in finite element
analysis will be explained in terms of skeleton structures [1-4]. The beginning will be made
with a very simple example and it will be gradually moved towards more and more general
cases in order to have a better way of understanding.

Fig. 1
2. Force Displacement Relationship
As an elementary case, take a spring as shown in Fig. 2 which is supported at one of its end
and subjected to a force P at its other end. The elongation of the spring or the displacement X
at the loaded point may be defined as
X = P / K or K X = P (1),

where K is the property of the spring known as spring constant. It is also defined as stiffness
of the spring which is a measure of force required to produce unit elongation or displacement
(X=1). In this example the degrees of freedom (DOF) is one and it is X.

Fig. 2
The above relation is applicable to a structure or a structural member if it can be idealised as
a single DOF system. In this context the basic type of structural members are presented here.
As a first case, a bar having cross-section area A, length l and modulus of elasticity E
under the action of axial force F as shown in Fig. 3(a) is considered. The elementary theory
of structural mechanics gives the elongation of the member as ∆=Fl/(AE) which may be
represented as

 AE 
 ∆ = F . (2)
 l 
The above equation may be compared with equation (1) which gives the stiffness of the bar
as K=AE/l.
As a second case, a beam having length l and flextural rigidity EI under the action of a
moment M at the right end as shown in Fig. 3(b) is considered. According to the elementary
structural mechanics concept the rotation of the member at its right end is θ=Ml/(4EI) which
may be represented as

 4 EI 
 θ = M . (3)
 l 
The comparison of the above equation with equation (1) gives that the force P, displacement
X and member stiffness K are M, θ and 4EI/l respectively in the present case.
As a third case, a beam having length l and flextural rigidity EI under the action of a
transverse force F at the right end as shown in Fig. 3(c) is considered. With the elementary
structural mechanics concept, the transverse displacement at the right end may be written as
δ=Fl3/(12EI) and it may be represented as
 12 EI 
 3 δ = F . (4)
 l 

2
The comparison of the above equation with equation (1) gives the stiffness K of the member
as 12EI/l3.
As a fourth case, a shaft having length l and torsional rigidity GJ under the action of a
twisting moment T at the right end as shown in Fig. 3(d) is considered. In this case, the angle
of twist φ at its right end φ=Tl/(GJ) may be expressed as

 GJ 
 φ = T . (5)
 l 
The comparison of the above equation with equation (1) gives the stiffness of the member as
GJ/l.

Fig. 3
Now take the example of two springs connected in series as shown in Fig 4. This is an
example of two DOF and they are X1 and X2 at the two nodes which are shown in Fig. 4 along
with the other quantities. The first spring, which is in the left, carries a load of (P1+P2) and its
elongation is X1 while the other spring carries a load of P2 and (X2 - X1) is its elongation. The
above equation (1) is now applied to the two springs individually and it leads to

Fig. 4
K 1 X 1 = P1 + P2 and K 2 ( X 2 − X 1 ) = P2 .

In the first equation, the second equation may be substituted to eliminate P2 and the equations
may be rearranges as follows.
( K 1 + K 2 ) X 1 − K 2 X 2 = P1 and − K 2 X 1 + K 2 X 2 = P2

In matrix form the above equations may be expressed as

3
 K1 + K 2 − K 2   X 1   P1 
 −K   =   or [K ]{X } = {P} (6),
 2 K 2   X 2   P2 

where [K ] , {X } and {P} are defined as stiffness matrix, nodal displacement vector and nodal
load vector respectively. This concept may be extended to a larger system having any number
of nodes. In the above examples, the number of nodes is equal to the number of DOF as the
DOF at the nodes is one but it will be more in most of the structural mechanics problems. In
that case, the problem size is not dictated by the number of nodes but it is dictated by the
number of DOF. Thus it is better to refer anything in terms of DOF. For structural system
having a total DOF n, the force displacement relationship may be expressed as

 K 11 K 12 ..... K 1 j ..... K 1n   X 1   P1 
K K 22 ..... K 2 j ..... K 2 n   X 2   P2 
 21    
 ..... ..... ..... ..... ..... .....  .....  .....
   =  (7).
 K i1 K i2 ..... K ij ..... K in   X i   Pi 
 ..... ..... ..... ..... ..... .....  .....  .....
    
 K n1 K n2 ..... K nj ..... K nn   X n   Pn 

The physical meaning of a stiffness coefficient, say Kij (the element in the ith row and jth
column of the stiffness matrix), may be given by substituting all the nodal displacements as
zero except Xj. In this situation the ith equation will be reduced to Kij Xj = Pi and it may
further be reduced to Kij = Pi by substituting Xj = 1. Thus Kij may be defined as the force to
be given in the direction of ith DOF to get a structural deformation having unit displacement
in the direction of jth DOF while displacements in the direction of other DOF are all zero.
This concept may be utilised to get the stiffness coefficients of the entire jth column which
are the forces to be given at the different DOF to get this particular deformed shape. The
other columns of the stiffness matrix can be obtained in the similar manner by imposing the
appropriate deformation pattern of the structure.
To verify the above concept, the system consists of two springs as shown in Fig. 4
may be taken. The first column of the stiffness matrix may be obtained by imposing X1 = 1
and X2 = 0. In that case the force is to be given in the direction of 1st DOF will pull the first
spring and push the second spring by an equal amount X1 = 1. Thus the total force will be
summation of the individual forces necessary to deform the individual springs and they are K1
and K2 for the first and second springs respectively. As the direction of both the forces is
positive according to the convention followed in Fig. 4, the total force will be (K1+K2) which
is also obtained in equation (6) corresponding to K11. To have this deformed condition a force
of K2 is to be given in the negative direction of 2nd DOF which is also obtained in equation
(6) corresponding to K21. The second column of the stiffness matrix can be generated in the
similar manner by imposing the condition X1 = 0 and X2 = 1.
The stiffness matrix has certain properties which may be discussed now. First of all,
the stiffness matrix is symmetric i.e., Kij = Kji and it can be verified easily with the help of
reciprocal theorem of Maxwell and Betti. The other important property is that the diagonal

4
elements Kii are always positive. This is obvious as the force to be given in the direction of ith
DOF to get unit displacement in the same direction should be positive when the
displacements in the direction of other DOF are all zero. Usually diagonal elements are heavy
compared to off-diagonal elements as the off-diagonal elements are produced by the
deformation corresponding to other DOF. All these properties are found in the stiffness
matrix of equation (6).
Now similar to the case of single DOF, the force displacement relationship derived for
the system of springs having multiple DOF is applicable to a structure having any number of
DOF. The basic type of structural members considered in the case of single DOF system as
shown in Fig. 2 will find their use again but they are to be reinvestigated considering all the
DOF. The different types of member considered in Fig. 2 may be defined as bar or axial force
member, beam or bending member and shaft or torsion member. In all the cases, the nodes
are at the two ends of the member i.e., the number of nodes in a member is two. The DOF at
a node is one for axial force member and torsion member while it is two (deflection and
rotation) for bending member. In Fig. 5(a), a bending member is deformed in such a manner
so as to have unit displacement corresponding to the 1st DOF while displacements
corresponding to other DOF are all zero. The forces are to be given in the direction of all the
DOF to have this deformed shape of the member are also mentioned in the figure. These
quantities will be utilised to generate the stiffness matrix of a structure which will be shown
later. The above process is repeated for other DOF of the beam which are shown in Fig. 5(b-
d). In the similar manner this is shown for axial force member and torsion member in Fig.
5(e-f) and Fig. 5(g-h) respectively. As the directions of all the forces are not identical, the
sign convention should be followed according to the axis system shown the figure.
The above procedure may be applied to analyse a structure having multiple DOF. This
can be implemented in two ways, one is the direct approach and the other one is element-wise
approach, which are explained in the following sections.

3. Direct Approach of Analysis

In this approach, the force displacement relation is directly obtained from the whole structure
which gives a system of linear simultaneous equations in matrix form. The known quantities
in this equation are the stiffness matrix and the load vector while the unknown is the
displacement vector which becomes known after solving the system of equations. Once the
nodal displacement vector for the whole structure is obtained, the member force such as
bending moment, shear force, axial force or torsion within any member can be evaluated as
follows.
The member forces at the ends of any member can be determined with the help of the
basic cases given in Fig. 5 using the nodal displacements of a member. For a bending
member having its nodal displacements v1 , θ 1 , v 2 and θ 2 , the shear force at the 1st node
may be written with the help of Fig. 5(a-d) as
12 EI 6 EI 12 EI 6 EI
V1 = 3
v1 + 2 θ 1 − 3 v 2 + 2 θ 2 . (8a)
l l l l

5
 Fig. 5
The other member end forces of the bending member can be similarly written as
6 EI 4 EI 6 EI 2 EI
M1 = 2
v1 + θ 1 − 2 v2 + θ2 (8b)
l l l l
12 EI 6 EI 12 EI 6 EI
V2 = − 3
v1 − 2 θ 1 + 3 v 2 − 2 θ 2 (8c)
l l l l
6 EI 2 EI 6 EI 4 EI
M1 = 2
v1 + θ 1 − 2 v2 + θ2 . (8d)
l l l l
The member end forces of an axial force member having nodal displacements ∆ 1 and ∆ 2 can
be written with the help of Fig. 5(e-f) as

6
 AE AE AE AE
F1 = ∆1 − ∆ 2 and F2 = − ∆1 + ∆2 . (9)
l l l l
For a torsion member having nodal displacements φ1 and φ 2 , the member end forces can be
written with the help of Fig. 5(g-h) as
GJ GJ GJ GJ
T1 = φ1 − φ 2 and T2 = − φ1 + φ2 (10)
l l l l
Within a member, the bending moment varies linearly while other member forces remain
constant.
Now consider a simple example of a continuous beam as shown in Fig. 6 (a) to
describe the above procedure. The structure has three DOF and these are the rotations X1, X2
and X3 at the three supports (see Fig. 6(a)). For sign convention, anti-clockwise direction is
taken as positive which is similar to that followed in Fig. 5. In this case, the size of the
stiffness matrix will be 3x3 and it can be determined as follows. To generate the first column
of the matrix impose X1 =1, X2 =0 and X3 =0 which gives the deformed shape of the structure
as shown in Fig. 6(b). With the help of Fig. 5(b), the stiffness coefficients of the first column
may be written as
4( EI )1 2( EI )1
K 11 = , K 21 = and K 31 = 0 .
l1 l1

The second column of the stiffness matrix may obtained by imposing X1 =0, X2 =1 and X3 =0
which gives the structural deformation as shown in Fig. 6(c). The stiffness coefficients of this
column may be written with the help of Fig. 5(d) and Fig. 5(b) for the left and right spans
respectively as
2( EI )1 4( EI )1 4( EI ) 2 2( EI ) 2
K 12 = , K 22 = + and K 32 = .
l1 l1 l2 l2

For the last column, impose X1 =0, X2 =0 and X3 =1, and it gives the structural deformation
as shown in Fig. 6(d). With the help of Fig. 5(d) the stiffness coefficients may be written as
2( EI ) 2 4( EI ) 2
K 13 = 0 , K 23 = and K 33 = .
l2 l2

If the loads acting on the structure are the moments M1, M2 and M3 at the supports as shown
in Fig. 6(a), the situation is identical to that in Fig. 4. In that case, the force displacement
relation of the structure may be written as

 4( EI )1 2( EI )1 
 0 
 li l1   X1  M1 
 2( EI )1 4( EI )1 4( EI ) 2 2( EI ) 2     
+  X 2  = M 2  . (11)
 l l1 l2 l2     
 1 2( EI ) 2 4( EI ) 2   X 3   M 3 
 0
 l2 l 2 

7
 Fig. 6
For a specific structure, the stiffness matrix and the load vector can be evaluated following
the procedure discussed above. With the known quantities, the nodal displacement vector can
be obtained by solving the equations from the force displacement relationship. Once the
nodal displacements became known, the member forces within the elements can be evaluated
with the help of equation (8).
In this problem, the loads are taken only at the supports and they are corresponding to
the DOF which has helped to represent the loads in terms of nodal load vector easily but the
actual loading in a structure may be something else.
To have a general idea regarding the loading, the same structure is considered with a
loading which is somewhat usual in a practical problem (see Fig. 7(a)). The load is not only
acting in a direction different from that of DOF but also there is not specific point where the
load is acting in case of distributed load. In that case, the analysis is done in two steps and

8
these steps are shown in Fig. 7 (b) and Fig. 7(c). The solution of the actual problem (Fig.
7(a)) is obtained by superposing the solutions of the two problems in Fig. 7 (b) and Fig. 7(c).
In the first step, the displacements at all the DOF are made restrained by imposing clamped
boundary condition at all the nodes as shown in Fig. 7 (b). This has made the individual
members independent as the fixed or clamped boundaries do not allow to transmit the
influence of one member to the others. Thus the task in this step is to analyse of a number of
fixed beams which is an elementary problem of structural mechanics. In that case, the
information of supporting forces and moments at the member ends, known as fixed end
forces, are sufficient to analyse a fixed beam. For ready reference, the fixed end forces for
some standard loading conditions are given in Fig. 8. In the present problem, the fixed end
forces are only the fixed end moments. Using the first case of Fig. 8, the fixed end moment at
the end A of member AB may be written as FEM AB = ω l12 / 12 . Similarly the other fixed end
moment may be written as FEM BA = −ω l12 / 12 , FEM BC = W l 2 / 8 and FEM CB = − Wl 2 / 8 .
From this, it clearly indicates that the structure is subjected to moments M A = FEM AB =
ω l12 / 12 , M B = FEM BA + FEM BC = −ω l12 / 12 + Wl 2 / 8 and M C = FEM CB = −Wl 2 / 8 at A, B
and C respectively which are imposed by the supports as they are made clamped. These are
nothing but the summation of fixed end forces appropriately taken at the different nodes. As
these moments do not exist in the actual structure, the contribution of these extra loads is to
be eliminated which is performed in another step shown as in Fig. 7(c). In this step the
structure is analysed with a system of nodal loads which are equal and opposite of these extra
loads i.e., P1 = − M A , P2 = − M B and P3 = − M C (see Fig. 7(c)). The present problem (Fig.
7(c)) can be solved followed the procedure used for that in Fig. 6 (a) as they became similar.
If a concentrated force or moment acts at a node, it is simply added in the appropriate
position of the nodal load vector.

Fig. 7

9
 Fig. 8
So far the DOF is one at all the nodes but it may be more that one and it may have different
numbers in different nodes. To show this, the same problem is reinvestigated with a different
boundary and loading condition as shown in Fig. 9(a). The structure has three DOF and they
are rotation at B (X1), vertical displacement at C (X2) and rotation at C (X3) (see Fig. 9(c)). As
there are some loads acting on the members, the analysis is to be carried out in two steps as
mentioned earlier. In the 1st step, the displacements in the direction of all the DOF are
restrained by imposing clamped boundary conditions at B and C as shown in Fig. 9(b). Using
Fig. 8, the fixed end forces can be determined as FEM AB = ω l12 / 12 , FEM BA = −ω l12 / 12 ,
FEM BC = W1l 2 / 8 , FEM CB = −W1l 2 / 8 and FEFCB = W1 / 2 where FEFCB indicates fixed end
force at the end C of member BC and it will be necessary as the node at C can move in the
vertical direction. With these fixed end forces, the nodal load vector of the problem shown in
Fig 9(c) may be obtained as

 P1  − ( FEM BA + FEM BC ) ω l1 / 12 − W1l 2 / 8

2

     
 P2  =  − FEFCB − W2  =  − W1 / 2 − W2  . (12)
P   − FEMPBC   W1l 2 / 8 
 3    

10
In the above equation the concentrated load W2 is added with a negative sign in the second
row of the load vector as W2 acts in the negative direction of the second DOF.

(d)

(e)

(f)

Fig. 9

The next job is to determine the stiffness matrix of the structure (Fig. 9(a) or 9(c)) and it can
be obtained by imposing deformation of the structure as shown in Fig. 9(d), (e) and (f) to
generate its 1st, 2nd and 3rd column respectively. Using Fig. 5, the stiffness matrix of the
structure may be written as

11
  4( EI )1 4( EI ) 2 6( EI ) 2 2( EI ) 2 
 + − 
 l1 l2 l 22 l2 
[K ] =  − 2 2 6 ( EI ) 12( EI ) 2
−
6( EI ) 2 
(13)
l2 l 23 l 22 
 
 2 ( EI )2 6( EI ) 2 4( EI ) 2 
−
 l2 l 22 l2 

In the previous example, all the members are oriented in a particular direction but they may
have different orientations in most of the cases. In this example, a frame as shown in Fig.
10(a) is considered where the members are oriented in different directions. In this structure,
the DOF are four and they are horizontal displacement (X1), vertical displacement (X2) and
rotation (X3) at joint B and rotation (X4) at joint C as shown in Fig. 10(c). The structure is to
be analysis in two steps due to the nature of loading as discussed earlier. In the first step
clamped boundary conditions will be imposed at joints B and C (see Fig. 10(b)) to restrain the
displacements in the direction of all the DOF. With the help of Fig. 8, the necessary fixed end
forces can be determined as FEFBAx = −W / 2 , FEM BA = −Wl1 / 8 , FEFBCy = ω l 2 / 2 , FEM BC
= ω l 22 / 12 and FEM CB = −ω l 22 / 12 where the superscript in FEF indicates its direction. For
2nd step of the problem as shown in Fig. 10(c), the nodal load vector can be obtained with
these fixed end forces as

 P1   − FEFBAx   W /2 
P    
− ω l2 / 2 
 2  − FEFBCy
  
   = =
  . (14)
P
  
3 − ( FEM BA + FEM )
BC  Wl 1 / 8 − ω l 2
2 / 12 
 P4   − FEM CB   ω l 22 / 12 

With the help of Fig. 5, the stiffness matrix of the structure can be generated by imposing
deformation of the structure as shown in Fig. 10(d-g) as

 12( EI )1 ( AE ) 2   6( EI )1  
 +  0   0 
 l1
3
l 2  2
 l1  
  ( AE )1 12( EI ) 2   6( EI ) 2   6( EI ) 2 
 0  +     
l 23  2 2
[K ] = 
  l1  l2   l 2  . (15)
 6( EI )1   6( EI ) 2   4( EI )1 4( EI ) 2   2( EI ) 2 
  2
  2
  +   
 l
 1   2 l  l1 l2   2 l
  6( EI ) 2   2( EI ) 2   4( EI ) 2 
 0  2
    
  l2   l2   l 2 

12
 Fig. 10

The direct approach is preferable if the any problem is relatively small and tried to solve it
manually. But it is clear from the above discussion that the major involvement in this process
lies with the generation of the stiffness matrix and the solution of the system of simultaneous
equations. Thus this approach is useful only in a problem having very few DOF. For a
structure having higher DOF, it is preferable to use the element-wise approach which is
discussed in the following section.

13
4. Element-wise Approach of Analysis
In a practical structure, the number of DOF is usually quite large which generates a large
number of coupled equations. This huge system of equations can not be solved manually and
it requires the help of computers where the work can be executed easily. Apart from solution
of these equations, generation of the stiffness matrix for the whole structure is another major
task. The element-wise approach gives a systematic approach to carry out this work using
computers. In this approach, the stiffness matrix and nodal load vector of the individual
members or elements are computed separately and they are combined together or assembled
to form the overall stiffness matrix and load vector of the whole structure. Before it is applied
to a structure, there are certain points to be discussed. The members of the structure shown in
Fig. 5 or 6 have equal number of nodes and equal number of effective DOF at the nodes but
this is not for the members of the structure shown in Fig. 9 or 10. This has happened due to
different types of support or boundary conditions at the different nodes. Thus the order or size
of the stiffness matrix of the individual elements will be different for the structure shown in
Fig. 9 or 10. This will not only make the assembly process cumbersome but also create other
problems. To avoid this difficulty and to have an equal order of element stiffness matrix,
element load vector and element displacement vector, all the possible DOF depending on the
type of structure are considered at both the nodes. Thus the DOF at all the nodes will be
vertical deflection and rotation (two) in the beam problem shown in Fig. 9 while it will be
horizontal deflection, vertical deflection and rotation (three) in the frame problem shown in
Fig. 10. Though the boundary conditions are violated at the element level by including these
undesirable DOF, but it will be taken care of by imposing the boundary condition on the
overall system of equations or stiffness matrix which will eliminate the effect of these extra
DOF. The other problem is concerned with the member orientation as in the structure shown
in Fig. 10. The direction of axial displacement of the two members is different and this is also
true for transverse displacement. In this problem the axial displacement of one member is
incidentally oriented in the direction of transverse displacement of the other member as the
members are at right angle. Moreover, the members are parallel to the axis system which has
made the problem easier. The situation will be much more complex when the members will
be inclined to the axis system. In order to overcome this problem, the stiffness matrix, nodal
load vector and nodal displacement vector of a member is expressed with reference to an axis
system oriented along the member axis which is known as local axis system for that member.
The stiffness matrix expressed in local axis system of a member is defined as local stiffness
matrix and it will have same expression for all the members which is really a big advantage.
This is also true for local load vector for a particular type of loading on the member. Though
the local axis system of the different members may have different orientations but they can be
correlated with a common axis system, known as structural axis system or global axis system,
through coordinate transformation. Utilising this correlation between the two axes systems,
the local stiffness matrix and load vector of a member can be converted in to global stiffness
matrix and global load vector of the member i.e., stiffness matrix and nodal load vector of the
member are expressed with reference to global axis system. Now the global stiffness matrix

14
and global load vector of all the members can be conveniently assembled to form the overall
stiffness matrix and overall load vector.
The evaluation of nodal load vector of an element in its local axis system is very easy.
It is nothing but the negative values of the fixed end forces for the loading which acts on the
element. As the concentrated loads and moments acting at the different nodes are not
incorporated in the element load vector, they are simply added at their corresponding
locations of the overall load vector once it is obtained after assembly of all the element load
vectors.
In the following sections, some important aspects such as local stiffness matrix
coordinate transformation, assembly of element stiffness matrix and load vector, and
boundary conditions are discussed.

4.1. Local stiffness matrix

In this section, the local stiffness matrix of an element of different types of skeleton structures
is presented. In all the cases, the number of nodes in an element is two and they are at the
element ends.
First of all, the truss or bar element is considered which has one DOF at each node in
the local axis system and this is valid for plane truss as well as space truss. Thus the DOF of a
truss element are two and the size of its local stiffness matrix is 2×2. With the help of Fig.
5(e-f), the stiffness matrix may be expressed as

− AE / l 
[K ] = 
AE / l
. (16)
− AE / l AE / l 

The shaft element is similar to the bar element since they have same number of DOF which is
angle of twist in the present element whereas it was axial displacement in the bar element.
The stiffness matrix of the shaft element may be written with the help of Fig. 5(g-h) as

− GJ / l 
[K ] = 
GJ / l
. (17)
− GJ / l GJ / l 

In the beam element, the DOF at each node are usually two and they are transverse
displacement in the plane of bending and bending rotation. Thus the size of its local stiffness
matrix is 4×4 and it may be expressed with the help of Fig. 5(a-d) as

 12 EI / l 3 6 EI / l 2 − 12 EI / l 3 6 EI / l 2 
 
2
− 6 EI / l 2
[K ] =  6 EI / l 3 4 EI / l 2 EI / l 
. (18)
− 12 EI / l − 6 EI / l 2 12 EI / l 3 − 6 EI / l 2 
 
 6 EI / l
2
2 EI / l − 6 EI / l 2 4 EI / l 

15
The DOF at each node of the 2D frame element as shown in Fig. 11 is three and they are
axial displacement, transverse displacement in the plane of bending and bending rotation.
Thus it is a combination of bar element and beam element. The stiffness matrix may be
obtained by combing equations (16) and (18) and it is

 AE / l 0 0 − AE / l 0 0 
 0 12 EI / l 3
6 EI / l 2
0 − 12 EI / l 3
6 EI / l 
2

 0 6 EI / l 2
4 EI / l 0 − 6 EI / l 2 2 EI / l 
[K ] =  . (19)
− AE / l 0 0 AE / l 0 0 
 0 − 12 EI / l 3 − 6 EI / l 2 0 12 EI / l 3 − 6 EI / l 2 
 
 0 6 EI / l 2 2 EI / l 0 − 6 EI / l 2 4 EI / l 

Fig. 11

For the grillage or grid element as shown in Fig. 12 the DOF at each node are also three but
they are angle of twist, transverse displacement in the plane of bending and bending rotation.
Using Fig. 12, the stiffness may be written as

 GJ / l 0 0 − GJ / l 0 0 
 0 4 EIY / l − 6 EIY / l 2 0 2 EIY / l 6 EIY / l 2 
 
 0 − 6 EIY / l 2 12 EIY / l 3 0 − 6 EIY / l 2 − 12 EIY / l 3  .
[K ] =  
(20)
− GJ / l 0 0 GJ / l 0 0 
 0 2 EIY / l − 6 EIY / l 2 0 4 EIY / l 6 EIY / l 2 
 
 0 6 EIY / l 2 − 12 EIY / l 3 0 6 EIY / l 2 12 EIY / l 3 

In a space frame element as shown in Fig. 13, the DOF at each node are six and they are axial
deformation, transverse displacement in both the plane of bending, angle of twist and bending
rotation for the biaxial bending i.e., three translations and three rotations. Thus it is a
combination of 2D frame element and grillage element. The stiffness matrix may be obtained
by combing equations (19) and (20) and it is

[K ] [K12 ]
[K ] =  11  (21)
[K 21 ] [K 22 ]
where

16
 Fig. 12

 AE / l 0 0 0 0 0 
 0 12 EI / l 3 0 0 0 6 EI / l 
2

 0 0 12 EI Y / l 3
0 − 6 EI Y / l 2 0 
[K11 ] =  
 0 0 0 GJ / l 0 0 
 0 0 − 6 EI Y / l 2
0 4 EI Y / l 0 
 
 0 6 EI / l 2 0 0 0 4 EI / l 

− AE / l 0 0 0 0 0 
 0 − 12 EI / l 3
0 0 0 6 EI / l 
2

 0 0 − 12 EI Y / l 3
0 − 6 EI Y / l 2 0 
[K12 ] =  
 0 0 0 − GJ / l 0 0 
 0 0 6 EI Y / l 2
0 2 EI Y / l 0 
 
 0 − 6 EI / l 2
0 0 0 2 EI / l 

17
 − AE / l 0 0 0 0 0 
 0 − 12 EI / l 3
0 0 0 − 6 EI / l 
2

 0 0 − 12 EI Y / l 3
0 6 EI Y / l 2 0 
[K 21 ] =  
 0 0 0 − GJ / l 0 0 
 0 0 − 6 EI Y / l 2
0 2 EI Y / l 0 
 
 0 2 EI / l 
2
6 EI / l 0 0 0

 AE / l 0 0 0 0 0 
 0 12 EI / l 3
0 0 0 − 6 EI / l  2

 0 0 12 EI Y / l 3
0 6 EI Y / l 2 0 
[K 22 ] =  .
 0 0 0 GJ / l 0 0 
 0 0 6 EI Y / l 2
0 4 EI Y / l 0 
 
 0 − 6 EI / l 2 0 0 0 4 EI / l 

Fig. 13

4.2. Coordinate Transformation

The local stiffness matrix and load vector of the individual elements can not be assembled as
the nodal displacements and nodal forces are represented in their own local axes having
different orientations. They are transformed in to global stiffness matrix and global load
vector where all the quantities are represented in a common axis system which help to
assemble the stiffness matrix and load vector of the different elements to form the overall
stiffness matrix and load vector of the structure. Before the transformation is applied to
element stiffness matrix and load vector, it is explained with the position vector or
coordinates of a point in the two dimensional system as shown in Fig. 14. The x-y coordinate
system may be considered as the global axis system, while x′-y′ is the local axis system which
makes an angle α with the global axis system. With the simple geometric relationships, the
coordinates of any point P (see Fig. 14) with reference to the above axes system may be
correlated as x ′ = x cos α + y sin α and y ′ = − x sin α + y cos α . In matrix form the above
equation may be written as

 x ′   cos α sin α   x   x ′   cos α cos β   x 

 =    or   =    or {X ′} = [T ]{X } (22)
 y ′ − sin α cos α   y   y ′ − cos β cos α   y 

18
 Fig. 14

where β = 900-α and the matrix [T] is defined as the transformation matrix which contains
the direction cosines of x′ in its first row and that of y′ in the second row. This is similarly
applicable to a three dimensional system and it may be expressed as

 x ′   l1 m1 n1   x 
    
 y ′ = l 2 m2 n2   y  or {X ′} = [T ]{X } (23)
 z ′  l m3 n3   z 
  3

where l1, m , n1 are the direction cosines of x′ and it is l2, m2, n2 for y′ and l3, m3, n3 for z′.
The transformation of the coordinates or position vector is applicable to nodal displacement
vector {X} and nodal force vector {P} as all these quantities are of similar type and they as
follows.
{X L } = [T ]{X G } and {PL } = [T ]{PG } (24)

where the subscripts L and G used with any quantity indicates that the quantity is represented
with reference to local axis system and global axis system respectively. As the transformation
matrix is orthogonal, its inverse is equal to its transpose. Using this property, equation (24)
may be written in reverse way which is as follows

{X G } = [T ]T {X L } and {PG } = [T ]T {PL }. (25)

For the transformation of element stiffness matrix, take the force displacement relationship of
an element in its local axis system which may be written as
[K L ]{X L } = {PL }
As the transformation matrix is orthogonal [T ][T ]T is an identity matrix [I ] which can be
post-multiplied with [K L ] in the above matrix without affecting the equation and it becomes

19
[K L ][T ][T ]T {X L } = {PL }
Now [T ]T is pre-multiplied in both the sides of the above equation and it becomes

[T ] T [K L ][T ][T ]T {X L } = [T ] T {PL }

With the help of equation (25), the above equation may be written as

[T ] T [K L ][T ] {X G } = {PG }
In the above equation, [T ] T [K L ][T ] is nothing but the global stiffness matrix [K G ] i.e.,

[K G ] = [T ] T [K L ][T ] (26)

Now the transformation matrix of the different types of structural elements is presented.
The transformation matrix of a 2D frame element or grillage element which is in x-y
plane and makes an angle α with global x axis may be written as

[ λ ] [ 0] 
[T ] =   (27)
[0] [λ ]

 cos α sin α 0
where [λ ] = − sin α cos α 0 .
 0 0 1

The inclination of the member may be expressed in terms of coordinates of its two nodes
( x1 , y1 ) and ( x 2 , y 2 ) which represents the above matrix [λ ] as

 Lx / L L y / L 0
[λ ] = − L y / L Lx / L 0
 0 0 1

where L x = x 2 − x1 , L y = y 2 − y1 and L = L2x + L2y .

In the above case, the transformation matrix is obtained easily as the position of local axis
system can be obtained by rotating the global axis system with respect to its z axis only. In
space frame, the problem is much more complex as the position of the local axis system is
obtained by rotating the global axis system in three steps. The first rotation may be given
with respect to z by an angle α, the second rotation may be given with respect y in its new
position by an angle β and the last rotation is with respect to x in its latest position by an
angle γ. The β may be taken negative which will help to express it in terms of nodal
coordinates. With these quantities the transformation matrix may be written as

20
 [λ ] [ 0] [ 0] [ 0] 
 [ 0] [λ ] [0] [0] 
[T ] =  (28)
 [ 0] [ 0] [ λ ] [ 0] 
 
 [ 0] [0] [0] [λ ]

[ ][ ]
where [λ ] = λγ λ β [λα ] and they are expressed as follows.

1 0 0  cos β 0 − sin β   cos α sin α 0

[ ] 
λγ = 0 cos γ [ ]
sin γ  , λγ =  0
 1 0  and [λα ] = − sin α
 cos α 0
0 − sin γ cos γ   sin β 0 cos β   0 0 1

In the above equation α and β can be expressed in terms of nodal coordinates ( x1 , y1 , z1 )

and ( x 2 , y 2 , z 2 ) as cos α = L x / L2x + L2y , sin α = L y / L2x + L2y , cos β = L2x + L2y / L and

sin β = − Lz / L where L x = x 2 − x1 , L y = y 2 − y1 , L z = z 2 − z1 and L = L2x + L2y + L2y . The

rotation about its own axis γ is to be supplied as input as it can not be obtained from the nodal
coordinates.
In all the cases mentioned above, the transformation matrix is square as the number of
DOF at a node in local axis system and global axis system is equal but this is not valid in the
case of a truss. In that case the DOF at a node in local axis system is always one while that in
global axis system is two in a plane truss and three in a space truss. Thus the transformation
matrix will be rectangular in a truss problem. In a plane truss, the displacement u′ at a node of
an element in its local axis system can be expressed in terms of u, v of the corresponding
node with the help of equation (22) where the first equation of this pair of equations will be
utilised. This will be applicable for both the nodes which will make the transformation matrix
as

Lx / L L y / L 0 0 
[T ] = 
L x / L L y / L 
. (29)
 0 0

In the similar manner the transformation matrix of a space truss element may be written as

 Lx / L L y / L Lz / L 0 0 0 
[T ] = 
L x / L L y / L L z / L 
. (30)
 0 0 0

4.3. Assembly of element stiffness matrix and load vector

Once the stiffness matrix and load vector of the members are expressed in terms of global
axis system, they are combined or assembled to form the overall stiffness matrix and load
vector of the structure. The procedure followed to do this work may be explained with a
structure having total number of elements et, total number of nodes nt and degrees of freedom
at each node dn. If the number of nodes of an element is ne, which is usually two in a skeleton

21
structure, the number of DOF of an element is de=dn ne and it is the size of element stiffness
matrix and element load vector. It is dt=dn nt for the total structure which is the size of the
overall system. In an element, the numbering of its nodes are 1,2,.... ne with reference to that
element which may be referred as element level or local level. But the numbering of these
nodes is something else with reference to the structure which may be referred as global level.
These numbers in global level are not consecutive and they vary from element to element.
Thus it is very important to know the correlation between these two sets of node numbering,
which is supplied by a matrix known as nodal connection matrix, say [nod]. This is a
rectangular matrix having number of rows equal to the number of elements et and number of
columns equal to the number of nodes of a element ne. In this matrix the elements of any row,
say ith row, contain the global numbering of the nodes of the ith element. The column wise
arrangement of these numbering is guided by their local level numbering. As an example, an
element at ith row and jth column of this matrix may be considered. If it contains a number
m, it will be global number of the jth node (in element level) of the ith element. In this
process, the DOF are arranged in proper order to have a systematic representation of the
system so that the process can be implemented in a convenient manner. For this purpose, the
numbering of the DOF of a node is consecutively made and once it is over, the next node (the
node having next higher node number) is considered for the numbering of its DOF. Again,
the way the DOF are arranged at a node should be maintained at all other nodes which can be
described with an example of a beam. If the deflection is taken as the 1st DOF and the
rotation as the 2nd DOF at a node, it is to be followed at all other nodes. Based on the above
discussion, the correlation between the local numbering and global numbering can be
established as follows
il = (in − 1)d n + i d and ig = (nod ie ,in − 1)d n + id (31)

where ie is the number of an element, i n is the number a node of that element in local level,
id is the number of a DOF of that node which can vary from 1 to dn, il is the DOF at the
element level and i g is the DOF at the global level. Utilising the above equation (31), the
element stiffness matrix and element load vector can be assembled to form the overall
stiffness matrix and overall load vector.
It has already been mentioned that the stiffness matrix is always symmetric.
Moreover, it is a sparse matrix where nonzero elements are found only at the neighborhood of
the diagonal elements which gives a banded nature of the matrix as shown in Fig. 15. The
reason behind this is very simple which can be explained. It has already been mentioned that
a column, say jth column of the stiffness matrix, can be generated by imposing unit
displacement corresponding to the jth DOF while displacement corresponding to all other
DOF are restrained. The elements of this column contain the respective forces required at the
different DOF to have this deformed shape of the structure. If jth DOF is associated with the
node n, it influences only the DOF of those nodes which are directly connected to the node n
by an element. The influence is not transmitted further as all the nodes except node n are
imposed clamped boundary condition. Thus the DOF, which are only influenced in this

22
process, will contribute some nonzero values at the respective positions of this column.
Usually a node is connected to its neighboring nodes and it may be expected that the
difference between their numbers is not much. In that case the nonzero quantities are not
spread much from the diagonal position and it gives this typical banded pattern as mentioned
earlier. Thus the banded portion along the diagonal containing all the nonzero elements is the
effective portion of the overall stiffness matrix. Utilising symmetry, half of the banded
portion (usually the portion above the diagonal) is stored in a rectangular array as shown in
Fig. 15 which saves a lot of storage space and it is quite useful since the size of this matrix
becomes very large. This appears to be a shifting of the individual rows horizontally where
the diagonal elements of the actual stiffness matrix occupy the first column of the rectangular
array. The band width i.e., the number of columns of the rectangular array should be
sufficient enough to include all the nonzero elements of the overall stiffness matrix. This is
guided by the element where the difference in its node numbers is maximum compared to
that of other elements as the dispersion of nonzero elements in the rows or column is
maximum. If m and n are the corresponding node numbers of that element, the band width
may be calculated as ( m − n − 1)d n . In a general case, the number of nodes may be more that
two in an element. In such cases the difference between the highest and lowest node number
of the elements is taken. Furthermore, the banded storage helps to reduce the computing time
to solve the overall system of equations.

Fig. 15

4.4. Boundary conditions

In most of the structures, some forms of supports are found. The purpose of providing a
support at a point is to resist the movement of the structure at that point in some specified
direction depending on the type of support. Thus some of the DOF of the nodes at the point
of supports are zero and this is satisfied by imposing appropriate boundary conditions. In
direct approach of analysis discussed in section 3, it is not required to impose such conditions
since these DOF are not consider in the analysis and they do not appear in the overall system
of equations. But they are not eliminated in the case of element-wise approach of analysis
since all the possible DOF are taken at all the nodes irrespective of boundary nodes where the
incorporation of boundary conditions becomes essential. This can be explained with a DOF,
say ith DOF (Xi=0) of a structure, where the overall system of equations may be written as

23
 K 11 K 12 ..... K 1i ..... K 1n   X 1   P1 
K K 22 ..... K 2i ..... K 2 n   X 2   P2 
 21
 ..... ..... ..... ..... ..... .....  .....  .....
    =  . (32)
 K i1 K i2 ..... K ii ..... K in   X i   Pi 
 ..... ..... ..... ..... ..... .....  .....  .....
    
 K n1 K n2 ..... K ni ..... K nn   X n   Pn 

This can be achieved by simply eliminating the ith row and ith column and its
implementation is not difficult if the overall stiffness matrix is stored in a square array as
shown in the above equation (32). But the implementation is not that easy if the matrix is
stored in the banded form or in some other manner. In that case, instead of eliminating the
row and column, the numerical value of the diagonal element K ii is made very high by
simply adding a large number compared to the actual value of K ii as

 K 11 K 12 ..... K 1i ..... K 1n   X 1   P1 
K K 22 ..... K 2i ..... K 2 n   X 2   P2 
 21
 ..... ..... ..... ..... ..... .....  .....  .....
   =  (33)
 K i1 K i2 ..... K ii (1 + γ ) ..... K in   X i   Pi 
 ..... ..... ..... ..... ..... .....  .....  .....
    
 K n1 K n2 ..... K ni ..... K nn   X n   Pn 

where γ is of the order of 1010 or something more. This makes Xi approximately zero which
can be shown taking the ith equation and dividing it by K ii (1 + γ ) as

K i1 Ki2 K in Pi
X1 + X 2 + ........ + X i + ...... Xn = (34).
K ii (1 + γ ) K ii (1 + γ ) K ii (1 + γ ) K ii (1 + γ )

As the denominator K ii (1 + γ ) is very large compared to any one of the numerators in the
above equation, all the contributing terms become approximately zero and it leads to Xi≈0.
Apart from the mathematical interpretation, a physical interpretation may also be given in this
case. It is something like replacing or idealising the support by a spring having a very high
stiffness value. Thus a yielding support having a finite stiffness value can also be taken care
of by this approach where the appropriate stiffness value of the support is to be taken. It may
be generalised further to incorporate the case of a support undergone a settlement by an
amount ∆s by a simple modification of the corresponding component Pi of the load vector as

24
 K 11 K 12 ..... K 1i ..... K 1n   X 1   P1 
K K 22 ..... K 2i  
..... K 2 n   X 2    P2 
 21 
 ..... ..... ..... ..... ..... .....  .....   ..... 
   =  (35).
 K i1 K i2 ..... K ii (1 + γ ) ..... K in   X i   K ii (1 + γ )∆ s 
 ..... ..... ..... ..... ..... .....  .....   ..... 
    
 K n1 K n2 ..... K ni ..... K nn   X n   Pn 

The boundary conditions of the structures having some form of supports are discussed above
but there are some structures such as aircraft, ship, submarine and some similar mobile
structures where no such supports are found to exist. In these structures the loads are
distributed in such a manner that they balance each other and the equilibrium of the whole
structure is maintained. Though there is no physical support in these structures but it becomes
necessary to incorporate a minimum number of boundary conditions to solve the overall
system of equations. This may be explained with a small example of a bar subjected to equal
and opposite forces at its two ends as shown in Fig. 16.

X1 X2
P1 = -P P2 = P
1 A, E, l 2

Fig. 16

Under the action of this system of forces, the bar will be in equilibrium and it will undergo an
elongation of ∆ = Pl / AE . The difficulty will be realised by taking the force displacement
relationship of the bar, which is

 AE / l − AE / l   X 1   P1 
− AE / l   =  . (36)
 AE / l   X 2   P2 

The above equation can not be solved as the stiffness matrix is singular i.e., the determinant
of the matrix is zero. It became so as the equations are not linearly independent. Apart from
the mathematical reasoning, the problem may be explained physically considering the nodal
displacement vector of the bar, which may be {X }T = [0 ∆], {X }T = [− ∆ 0] , {X } = ,
T

[− ∆ / 2 ∆ / 2] or {X } = [c c + ∆ ] where c may have any value. It indicates that the above

T

system of equations may have any number of solutions and the nodal displacements are not
independent quantities which may be related by X 2 − X 1 = ∆ . To get a unique solution of
this problem, one of the nodal displacements is to be assigned a specific value and this is
nothing but the incorporation of boundary condition at a node. This will not affect the strain
of the bar as the strain is measured by the relative change in position or distance between the
particles which is only concerned with ∆. In this case, c is that part of the displacement which

25
has no effect on strain and it is defined as the rigid body translation. Similar to this, we find
rigid body rotation in some other problems.

References
1. Gere, J.M. and Weaver Jr., W. (1965) Analysis of Framed Structures, D. Van Nostrand.
2. Coates, R.C., Coutie, M.G. and Kong, F.K. (1980) Structural Analysis, 2nd Ed. ELBS,
Hong Kong.
3. Mukhopadhyay, M and Sheikh, A.H. (2004) Matrix and Finite Element Analyses of
Structures, Ane Books, New Delhi.

26