Finite Element Analysis For Beams: Beam Kinematics
Finite Element Analysis For Beams: Beam Kinematics
Finite Element Analysis For Beams: Beam Kinematics
So far we have looked at finite elements with interpolation functions that satisfied C0 continuity.
However, beams are structural elements that require a different treatment. This topic though
strictly speaking not essential for study in our introductory course, is important in the sense that it
tells us that finite element analysis requires different approaches and is not always amenable
through just one method.
Shell elements are more involved than beam elements. However, the idea behind beams and shell
elements is fundamentally very similar.
Beam Kinematics
Nomenclature for the beam. The midplane for the beam is called as the neutral axis. The deforma-
tion of the Euler-Bernoulli beam happen as per this figure
4. Now noting that using the basics of Euler-Bernoulli theory, where the internal moment
d2 u
m=EI d x2
, we get the following equation
Boundary Conditions
Differential equation is:
Boundary Conditions:
This governing equation is a fourth order differential equation and the boundary conditions for this
equation are:
Note:
FEA_Beams.nb ��� 3
1. the quantity n is external normal to the beam along the length of the beam. For example, in
the Figure of the beam above with the loading, at the le� end of the beam n = -1 and at the right
end of the beam n = +1. This is not the same n as shown in the Figure above regarding nomencla-
ture of the beam.
-
2. Our sign convention is that m is positive anticlockwise. Hence the inclusion of the normal n
as discussed above ensures that moment is correctly defined as in the sign-convention figure.
-
3. Also the shear force s is positive upwards, hence by multiplying with n we ensure that we
obey the sign convention for shear force as described in the sign-convention figure.
At any end of the beam you could have boundary conditions in the following common combinations
Like in the case of elasticity at any point we cannot have energy conjugates: m and θ or
s and uy being specified simultaneously. Mathematically, this means:
where like in the previous cases, we note that the weight w should vanish on the essential bound-
aries Γu to give
We now do another integration parts on the term involving first derivatives of both m and w
4 ��� FEA_Beams.nb
dw
In the above equation, we note that dx
should vanish on the essential boundary Γθ -- here we are
encountering something completely new than anything before that.
Combining all the relations above together, we get the following final expression for the weak form
This is actually the statement of principle of virtual work. The le� hand side of the equation tells us
the internal virtual work done and the right side is the external virtual work. In this case w = δuy
can be thought of to be the virtual displacement of the beam. Now, by noting that:
d2 u
m = E I d x2y
the weak for will look as
As before we have symmetry in the way w and uy appear in the weak form derivatives.
We now have to carefully look into the structure of U and U0 , respectively, the spaces corresponding
to uy and w respectively. Since we have second derivative of both these functions appearing in the
integral we can no longer have C0 functions approximating them. We need C1 functions, i.e. func-
tions without jumps and kinks. At a more physical level, the presence of kink will make the deforma-
tions incompatible (something that was desirable in linear elasticity). Since the integrals need to
finite, C1 functions that are also integrable are called as H2 functions.
Here f and m are the internal shear force and the moments (very similar to structural mechanics.)
The Hermite polynomials for interpolation are
where
This ensures that the continuities at the nodes both for the displacement uy and the slope duy /dx
are indeed satisfied.
l +1
Also note that any integral of the form: ∫Ωe f(x) dx = 2 ∫-1 f(ξ) dξ.
When we use the Galerkin scheme, the same interpolation functions are used for uy and w.
Now, we check the properties of the shape functions . For simplicity (without loosing generality) we
choose the value of the length of the beam as l = 1.
Discrete equations
As before we can the residual (or the reactions) as
where fΩe and fΓe are the element and body forces, respectively. Everything else now is similar to
what we have been doing before.
If the stiffness E I is constant over the element, the element stiffness is given as
It can be seen that uniform load results in both nodal moments and vertical nodal forces, corre-
sponding to fixed end moments for a uniformly loaded beam.
It itself such beam elements are particularly different that what we do in structural mechanics.
However, they provide some understanding of what one does when encountered with higher order
differential equations.
Also, as we see below, these ideas could be extended for finite element analysis of shells.
Example problem:
Consider a beam problem shown in Figure below. The beam ANS is clamped at the le� side and is
free at the right side. Spatial dimensions are in meters forces in N and distributed loading p in N
m-1 . The beam bending stiffness is EI = 104 N m2 . The natural boundary conditions at x = 12 m are
- -
s = -20 N and m = 20 N m.
The beam is subdivided into two finite elements as shown in this figure.
8 ��� FEA_Beams.nb
quite obviously because of the properties of the shape functions and their derivatives.
FEA_Beams.nb ��� 9
We also have a point force acting at the center of element-1. Since the nodal forces transmitted
due to a body force of form p(x) = F0 δ(x - x0 ) is:
fΩe = ∫Ωe NeT F0 δ(x - x0 ) dx = F0 NeT (x0 ) = F0 NeT (ξ0 ),
where ξ0 corresponds to the point x0 through the equation:
ξ = 2lx - 1.
Note, that here we do not have the additional factor of 2l because δ(x - x0 ) has units of 1
length
(units
of 1x . However, ξ does not have any units.
For element 1: In this case a distributed force of p(x) = -1 acts over the beam an a point force
-10 N acts at x = 2l which implies that ξ = 0. So the total body force on this element is:
For element 2: The point force in element 2, acts on the first node where ξ = -1 giving
The direct assembly of the force matrix gives the global force vector as:
While accounting for the essential boundary conditions (Γu and Γθ ) as the partitioned degrees of
freedom, the stiffness matrix for the system is:
As usual solving for the displacements and the unknown reactions at the essential boundary point
we get
10 ��� FEA_Beams.nb
How does this answer (displacement, shear force and bending moment) compare with the exact
answer? The displacement is quite reasonable as compared with the exact solution. Consider the
origin of the shape functions for the beam.
�������� u /. DSolve[
{EI u ''''[x] ⩵ 0, u[0] ⩵ u1, u '[0] == 0, u '[l] ⩵ 0, u[l] ⩵ 0}, u, x][[1]];
%[x] // FullSimplify;
l
N1 = % /. x → (1 + ξ) // FullSimplify
2
u /. DSolve[
{EI u ''''[x] ⩵ 0, u[0] ⩵ 0, u '[0] == θ1, u '[l] ⩵ 0, u[l] ⩵ 0}, u, x][[1]];
%[x] // FullSimplify;
l
N2 = % /. x → (1 + ξ) // FullSimplify
2
u /. DSolve[
{EI u ''''[x] ⩵ 0, u[0] ⩵ 0, u '[0] == 0, u '[l] ⩵ 0, u[l] ⩵ u2}, u, x][[1]];
%[x] // FullSimplify;
l
N3 = % /. x → (1 + ξ) // FullSimplify
2
u /. DSolve[
{EI u ''''[x] ⩵ 0, u[0] ⩵ 0, u '[0] == 0, u '[l] ⩵ θ2, u[l] ⩵ 0}, u, x][[1]];
%[x] // FullSimplify;
l
N4 = % /. x → (1 + ξ) // FullSimplify
2
1
�������� u1 (- 1 + ξ)2 2 + ξ
4
FEA_Beams.nb ��� 11
1
�������� l θ1 (- 1 + ξ)2 (1 + ξ)
8
1
�������� - u2 - 2 + ξ (1 + ξ)2
4
1
�������� l θ2 (- 1 + ξ) (1 + ξ)2
8
These shape functions are the same as what we have used. Thus shape functions that we have used
earlier infact corresponding to solving the basic beam equation with boundary displacements at
nodes 1 and 2 u1 , θ1 , u2 , θ2 . As a result any problem with loading only at the nodes will be exactly
solved using the FEA formulation discussed. However, bending moment will be approximation
because with the FEA shape functions bending moments will be linear in x and will have kinks or
jumps when we move from one element to the other if there are external force or bending
moment, respectively, at the common node. The worst, however, will be shear force since wher-
ever we have a point force on the beam, we will end up creating jump in the shear force. Hence, it is
a good idea to use a new node wherever point forces are present. The actual comparison between
the current FEA solution with two elements and the exact solution is shown below.
12 ��� FEA_Beams.nb
What next?
The polynomial space of C1 elements that was used above of Hermite polynomials is not available in
FEniCS. An alternative procedure which could be implemented would be to utilize C0 elements, but
put a constraint on the kinks between the elements using the so called Discontinous Galerkin
elements
(https://fenicsproject.org/docs/dolfin/1.6.0/python/demo/documented/biharmonic/python/docum
entation.html).
The simple Euler-Bernoulli formulation can be extended to deep beams where even shear can be
dominant. A most commonly used model for this is what is called as Timoshenko beam theory
(https://en.wikipedia.org/wiki/Timoshenko_beam_theory). The corresponding finite element
formulation can given rise to what is called as shear locking phenomena, whose effect can be
so�ened using the so called reduced integration formulation
(http://14.139.134.16/cmmacs/pdf/ch06.pdf).
There are many, many other interesting features and abnormalities that can arise in finite element
formulation which are topics for a more advance course. Further, the natural transition from beam
elements is to shear elements. We will not go into such details in our course. Instead, I will provide
these as some of the topics on which you will do a mini-course project and do a 5 minute presenta-
tion.