CHAPTER 3 {
THE FINITE ELEMENT
CONCEPT16
‘3 THE FINITE ELEMENT CONCEPT
In this section the finite clement concept is introduced by considering a simple one-
HE a
Ny = SD 2 HD yy (628)
ax) E2-H
‘Te compute the displacement at the midpoint ofthe element, where x = L/4, we have
~2xk
axe
and
ly, = 3812
thu, =
‘The above equation indicates thatthe displacement at the midpoint of the element is
‘merely equal tothe average ofthe displacements at its ends, which swhat we would expect
intuitively
‘Similar calealations may be performed for element two, where
= 1/2) .
erp ~
In computing the above shape functions, iti crucial to note that the local node numbers
tone and two of element two correspond to the global node numbers two and three.
Accordingly, the displacement at any pint in element two may now be written as
= Ny + Noid = NU, +NUs
‘To compute the displacement a the midpoint of the element, where
SL/A , we have
wake
Nya 2-2x8
ted
£72and
edu, +h
undu, +10,
iL?
16E
‘The finite clement solution for the displacement at various point in the rod is shown in
Tble A,
x 0 Lis rR 3Li4 L
: 0 ama a2 ie ae
16E 8 16E 26
‘Table 3.1 Finite clement solution fortwo linear elements
‘To compute the strains at any point inside an element, we differentiate the
displacements. In element one for example, equation (3.10) gives
Ay My
eta
bre
and hence
[Note that the strainis constant over each element. This is because the strain isa frstorder
differential of the linearlyvarying displacement. Sinilai for element two, we obtain,
eg ew eg ae My, 4
= Gb id + Gd = G+ GU, G2)
with
an,
ect G30)
aN, 2
ae? 63
and hence‘Comparing the strains between element one and element two, we see that they are
different. This implies that the strains are discontinuous between elements. The
displacements, on the other hand, which are the primary variables, are continuous
throughout the mesh. Generally speaking, the jump in strain between adjacent elements
{is reduced as the number of elements is increased,
(Once the strains have been found, the stresses may be computed using the stress-strain
(constitutive) law. For uniaxial elastic conditions, the axial stress is related to the axial
strain via Young's modulus F according to
on Ee 6.32)
Since the strains are constant across each element, so are the stresses. The stresses may
jump between adjacent elements. For element one we have
op = Ex SEL. EL
EX gg
‘Similarly, for element two
we xX ML
Om BX GEG
‘To assess the performance of our finite element mode, it i instructive to compare the
results from the two-element mesh with the exact solution, ‘The accuracy of the nodal
displacements and element stresses is of particular interest,
3.4 Comparison of Finite Element Solution with Exact Solution
‘The equilibrium equation of the rod is given by (3.1). Substituting (3.2) and (3.32), the
governing differential equation becomes
ait ye
£ab +0
"This may be integrated to give
where cis an integration constant, ‘The boundary conditions at the free end of the rod,
where x = L, are oy = Eilujdx = du/dx = 0. These define the integration constant as
= EL/E. Thus the exact solution for the strain distribution ia the rod is
ag t= Fu -2 833)
Using (6.32), thestrosss are given by
ox Bes = RL 63
Integrating (3.33) again, we obtain the displacements asoo iu -¥) G35)
where the integration constants er since u = 0 when x = 0
‘The exact displacements are compared with the two-element finite element solution in
‘Table 32 and Figure 35.
x ° us 12 sue L
7 5 ae | ame w@
16 ae 2
ans 0 pray ma
t RE a
%exor 0 =14 o ~7 0
‘Table 3.2; Errors infinite element displacements
For this problem, the finite element displacements match the exact displacements at the
nodes, butnot atpointsinside the elements, In general, thenodal displacements from finite
clement analysis do not match the exact displacements, and itis necessary to refine the
‘mesh by trial and error until the displacement errors are small. Tt also interesting to note
that the displacements are never greater than the exact displacements. This is always tric
for displacement finite clement analysis of linea elastic materials where it can be shown
that the finite clement stifiness matrix is invariably too “stiff”, The fact that the nodal
isplacements approach the exact displacements from below is very useful in finding
suitable mesh by tral and error,
06 —_____
05:
oa.
aE 931 finite clement
mm
0 035 050 075 100
‘Figure 3.5: Comparison of finite element solution with exact solution0
‘The errors n the stresses are generally much larger than the errors inthe displacements,
as shown in Table 3.3 and Figure 26.
element 1 clement 2
x 0 La Le 12 | 3u4 L
o ama | aera | axes | mua | xu | xrys
@aar| %E | ae | mp | me | ma | o
Yeenor | 25 0 50 50 0 =
‘Table 3.3: Errors in finite element stresses
Unlike the displacements, the computed stresses are most accurate atthe centre of each
clement, This is because the finite element procedure attempts to model the exact stress
field in a least squares sense. Because the stres is constant within each clement, and no
continuity conditions are imposed, itis inevitable that stress jumps, such as those shown in
Figure 3.6, occur between elements, This introduces ambiguity ifthe stress is sampled at
the nodes. In Table 33, for example, node two basa stress of 3X1/4in element one bat
stress of XL/4 in element two. Quite cleat, as the number of clementsis increased, the
size of the stress jumps, and the corresponding ambiguity, will iminish. In displacement
finite element analysis, its usual to sample the stestes at points inside the elenment and
the displacements atthe nodes. The precise location ofthe points fr sampling the stresses
dependson the typeof element and willbe discussed in subsequent letures. For our simple
problem, however, the finite element stresses match the exact stresses ifthe former are
‘sampled at the centre of each element
finite element
0 035050075 1.00
i
Figure 3.6: Comparison of finite element solution with exact solution