Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
R.T. Haftka
University of Florida
Outline
Most structural analysis y FE codes are displacement p based In this chapter we discuss interpolation methods and elements l t based b d on di displacement l ti interpolations t l ti Stiffness matrix formulations will be presented Shortcomings and restrictions of the elements due to the interpolations used will be discussed We will review the governing equations (for solids elastic bodies) to help us understand the solution methods and accuracy
R.T. Haftka
University of Florida
R.T. Haftka
University of Florida
Stress
Stresses are distributed internal forces that result from externally applied forces
F1 B1 F2 F1 F
F3 F5 F4 F5
Note: There are two types of forces: Surface forces that act on an area of external surface and body force that acts on the volume of the body
R.T. Haftka EML5526 Finite Element Analysis University of Florida
Force is a 1st order tensor (vector) Stress is a 2nd order tensor. Why is it a tensor?
Needs two vectors to specify it it. One is the vector of the internal force and the other is the outward normal of the defining g area
R.T. Haftka
University of Florida
R.T. Haftka
University of Florida
Stress at a point
State of stress at a point in 3-D Cartesian Coordinates
yy
Matrix representation of state of stress y x
yz zy
yx
xy xx
zx xz zz
xx xy zx xy yy yz xz yz zz
Sign Conventions: Normal stresses are positive when acting outward from a surface (tension) Shear stresses are positive when they act in the +ve direction on a positive face and ve direction on a ve face
R.T. Haftka EML5526 Finite Element Analysis University of Florida
xy
/2
O
XA = x ( Al ) + xy ( Am)
Which simplifies to
xy y y
X = x l + xy m Y = y m + xy l
EML5526 Finite Element Analysis University of Florida
R.T. Haftka
X = l x + m xy + n zx
y B N
Y = l xy + m y + n yz Z = l xz + m yz + n z
Where l, m and n are the direction cosines of the normal vector of the x arbitrary plane
O C z
R.T. Haftka
University of Florida
10
Stress transformations in 3D
Stress transformation in 3D space can be defined using the directions cosines as follows. follows
11
Direction cosines in 3D
The direction cosines l.m and n between the new coordinate axes x, , y and z and the original coordinate system x, y and z are defined as follows
x x y' z l1 l2 l3
y m1 m2 m2
z n1 n2 n3
Since the transformation is orthogonal, the direction cosines must satisfy the following properties
12
y yx + d yx
xy
dy
FX
x
Summation of forces in X-direction
x + d x
dx
FY
yx
=0
University of Florida
( x + d x )dy y x dy y + ( xy + d xy )dx xy dx + FX
R.T. Haftka EML5526 Finite Element Analysis
13
x dx x xy = dy y y
x xy x + y + FX
R.T. Haftka
dxdy = 0
University of Florida
14
x xy + +X=0 x y xy x + y y +Y =0
Establishing moment equilibrium by the same method will provide the condition for symmetry p y y of the stress tensor
xy y =y yx
R.T. Haftka EML5526 Finite Element Analysis University of Florida
15
Strain
Why do we need the strain measures? Will displacement not suffice? Strain better q quantifies the deformation of the body and eliminates rigid body motion/ rotation y general g terms is a measure of Strain in very relative deformation
Relative to what?
Undeformed body : Lagrangian strain Deformed body: Eulerian strain
R.T. Haftka
University of Florida
16
l = l
P dx Q
P, P
du dv
u
P
P=P(x+u,y+v,z+w)
University of Florida
17
P P P,
du dv
du =
dv =
P dx Q
u dx x
v dx x
x =
2
dx dx dx
2 2
Q 1 u u v x = 1+ 2 + + x x x 2
2
u v 2 dx + dx + dx dx x x x = dx
x = 1+ 2
u u v + + 1 x x x
2 2
2 2 u 1 u v x = + + x 2 x x
R.T. Haftka
University of Florida
18
dx
u dx Q x
B v dx x Q
O
P
dx
dy
Q P
1
dy
= 1 + 2
x v P dy y A
u dy y
R.T. Haftka
University of Florida
19
Shear strain
With a bit of trigonometry (see for example, Allen and Haisler, Introduction to Aerospace Structural Analysis, p.60)
v u u v u u + + + x y x x y y
xy y =
R.T. Haftka
University of Florida
20
x =
u x
y =
v y
xy =
1 v
v u + x y
Engineering Strains
+ xy = x y 2
Tensor strains
Shear strain definitions are of two forms. The above form is referred to as engineering strains. The alternate form referred as tensorial strains have a factor of applied to engineering strains. To apply coordinate transformations need the tensor form.
R.T. Haftka
University of Florida
21
Compatibility
Deformation must be such that the pieces fit together without any gaps or overlap. overlap Why is this an issue? In 2-D we require only 2 displacements u, and v to describe deformation, but have three strain quantities x,y, and xy. This implies only two of the three strain terms are independent.
x =
u x
y =
v y
xy =
v u + x y
2 2 x y 2 2 u 2 v ( xy ) = 2 + 2 y 2 + x 2 xy y x x y
R.T. Haftka
University of Florida
22
23
24
Isotropy assumes that there is no directional variation on property. Using this argument we can obtain
25
x =
x
E
x
E
The relation between the shear stress component and its corresponding shear-strain shear strain component is called the modulus of rigidity or modulus of elasticity in shear and is denoted by the letter G. E G= = G= 2(1 + )
R.T. Haftka EML5526 Finite Element Analysis University of Florida
26
x =
] ]
xy = yz zx
xy
R.T. Haftka
University of Florida
27
z = xz = yz = 0
This simplifies p the stress strain relationship p to the form as shown below.
1 x y E 1 y = y x E ( x + y ) z = E
x =
x =
E x + y 1 2 E y = y + x 1 2
] ]
xy = xy
xy
G = G xy
R.T. Haftka
University of Florida
28
Plane strain
Strains that deform the body normal to the reference plane are absent or are negligible
z = xz = yz = 0
This indicates that the stress normal to the p plane of strain is dependent on the stresses in the plane of the 1 strain z = z ( x + y ) = 0 z = ( x + y ) E Substituting z into other strain expressions we obtain
x =
1 + (1 ) x y E 1 + (1 ) y x y = E
] ]
x =
] ]
xy =
xy
G
xy = G xy
R.T. Haftka
University of Florida
29
E* * E= , = 1 * 1 *
or E * = E
1 + 2 (1 + )2
and =
1 +
R.T. Haftka
University of Florida
30
x , y , z , xy , yz , zx
u,v, w
3 6
---- x , y , z , xy , yz , zx 9
u , v, w
---- x , y , z , xy , yz , zx 6
R.T. Haftka
University of Florida
31
Principle of superposition
Effects of several forces acting together are equal to the combined effect of the forces acting separately separately. This is valid only when
The stresses and displacements are directly proportional to the load The geometry and loading of the deformed object does not differ significantly from the undeformed configuration g
R.T. Haftka EML5526 Finite Element Analysis University of Florida
32
Energy Principles
Strain Energy Density: When an elastic body is under the action of external forces, the body deforms and work ork is done b by these forces forces. The work ork done b by the forces is stored internally by the body and is called the strain energy. Let us consider the unit element of volume dxdydz with only the normal stress x acting on it it. The work done done, or work stored in the element is
z D C D C
x = x
x
A
B B A
B x
u d u + dx dydz x x x =0 =
x = x x =0
x = x
x =0
d (u )dydz
x
u dxdydz x
y
R.T. Haftka EML5526 Finite Element Analysis University of Florida
33
Strain Energy
Using Hookes law
x = x x =0
u = x = x x E x
1 x d x dxdydz = dxdydz E 2 E
2
Work done =
For shear stresses, stresses it can be similarly shown that the work done is
The h strain energy stored d in an element l dxdydz under d a general three dimensional stress system is calculated as
12 dxdydz 2G
dU =
R.T. Haftka
1 ( x x + y y + z z + xy xy + yz yz + zx zx )dxdydz 2
EML5526 Finite Element Analysis University of Florida
34
dU 0 =
1 ( x x + y y + z z + xy xy + yz yz + zx zx ) 2
dU 0 =
1 ( 11 + 2 2 + 3 3 ) 2 1 2 dU 0 = I1 2(1 + ) I 2 2E I1 = 1 + 2 + 3
I 2 = 1 2 + 2 3 + 1 3
R.T. Haftka
University of Florida
35
R.T. Haftka
University of Florida
36
R.T. Haftka
University of Florida
37
Polynomial interpolation
The polynomial function (x) is used to interpolate a field variable based on its values at n n-points points
( x) = ai xi
or
X = 1 x x
i =0
= X {a}
... x n and
{a} = a0
a1
a2
... a n
The number of terms in the polynomial is chosen to match the number of given quantities at the nodes. With one quantity per node node, we calculate ais s using the n n-equations equations resulting from the expressions for i at each of the n-known points
( x j ) = ai x ij
i =0
R.T. Haftka
{e } = [A ]{a}
EML5526 Finite Element Analysis
{a} = [A]1 {e }
University of Florida
38
= X{a}
{e } = [A ]{a}
{a} = [A]1 {e }
= X{a}
= X{a}
{e } = [A ]{a}
= X [ A]
1
{a} = [A]1 {e }
= N {e }
N = X [ A]
University of Florida
{e }
R.T. Haftka
39
Degree of Continuity
In FEM field quantities are interpolated in piecewise fashion over each element This implies that is continuous and smooth within the element However, may not be smooth between elements m An interpolation function with C continuity provides a continuous variation of the function and up to the mm derivatives at the nodes
For example in a 1-D interpolation of f(x) C continuity indicates that f is continuous at the nodes and f f,x is not continuous. continuous 0 If the displacement u(x) is C then displacements are continuous between elements, but the strains are not (bar elements)
0
R.T. Haftka
University of Florida
40
Degree of Continuity
Function 1 is C0 continuous while 2 is C1 continuous
d dx
(x)
1 2
d1 dx
C0
d 2 dx
C1 x1 x2 x x0
C0 x1
C1 x2 x
x0
R.T. Haftka
University of Florida
41
R.T. Haftka
University of Florida
42
x1
1
x2
1
x3
N1 =
(x2 x )(x3 x ) (x2 x1 )(x3 x1 ) (x1 x )(x3 x ) (x1 x2 )(x3 x2 ) (x1 x )(x2 x ) (x1 x3 )(x2 x3 )
N1 =
1
x2 x x 2 x1 x x1 x 2 x1
x1
N2 =
x2
x3
1
N2 =
1 x1 1
2 = N 1 2 x2 2
N3 =
x1
x2
x3
1
x1 1
2
x2 2
3
x3 3
1 = N 2 3
R.T. Haftka
University of Florida
43
Nk =
44
Properties of C0 Interpolation Shape F Functions ti All shape functions Ni, along with function are polynomials of the same degree For any shape function Ni, Ni = 1 at node i ( (x=x d zero at t all ll other th nodes d j j, ( (xj xi) i) and C0 shape functions sum to one
N
k =1
R.T. Haftka
=1
University of Florida
45
C1 Interpolation p
Also called Hermitian interpolation (Hermite polynomials) Use the ordinate and slope information at the nodes to interpolate
,x2 ,x1
C0 interpolation curve
C1 interpolation curve
1 x
2 x
University of Florida
R.T. Haftka
46
R.T. Haftka
University of Florida
47
f ( x, y ) = 1 x
y a1
a2
a3
If there are two or more components (e (e.g., g u, v and w displacements) then the same interpolation function is used for all components
R.T. Haftka EML5526 Finite Element Analysis University of Florida
48
49
= U W = U + V = 0
U is strain energy, W is work done, or V is potential of the forces
= U W
R.T. Haftka
University of Florida
50
R.T. Haftka
University of Florida
51
[ k ] = [ B ] [ E ][ B ] dV
T
{re} =[ B] [ E]{0} dV[ B] {0} dV+ [ N] {F} dV+ [ N] {} dS ) Loads due to initial strain, initial stress, y forces and surface tractions body
T T T T
R.T. Haftka EML5526 Finite Element Analysis University of Florida
Constitutive C tit ti equations ti for f a linearly li l elastic l ti and d isotropic material in plane stress (i.e., z=xz=yz=0):
52
x 0 = y 0 = T , xy0 = 0
where
R.T. Haftka
University of Florida
53
R.T. Haftka
University of Florida
54
55
The node numbers sequence must go counter clockwise p field so strains are constant! Linear displacement
R.T. Haftka
University of Florida
CST ELEMENT
Decompose two-dimensional domain by a set of triangles. Each triangular element is composed by three corner nodes. Each element shares its edge g and two corner nodes with an adjacent j element Counter-clockwise or clockwise node numbering Each node has two DOFs: u and v displacements interpolation using the shape functions and nodal displacements. Displacement is linear because three nodal data are available available. Stress & strain are constant. y v3 3 v1 1 u1 v2 2 u2 x
56
u3
The goal is how to calculate unknown coefficients i and i, i = 1, 2, 3, in terms of nodal displacements.
In matrix notation
u1 u2 = u 3 1 x1 1 x2 1 x3 1 y1 y 2 2 y 3 3
58
u1 f3 b3 u2 c3 u3
59
60
v ( x, y ) = [N1 N2
N1, N2, and N3 are linear functions of x- and y-coordinates. Interpolated displacement changes linearly along the each coordinate direction.
61
{u( x, y )} = [N( x, y )]{q} [N]: 26 matrix, {q}: 61 vector. For a given point (x,y) within element, calculate [N] and multiply it with {q} to evaluate displacement at the point (x,y).
62
N xi ui = i =1 N yi v i = i =1
3
i ui 2A i =1 3 i vi 2A i =1
b c
xy
u v + = y x
ci b ui + i v i 2A i =1 i =1 2 A
63
0 c2 b2
b3 0 c3
[B] matrix t i is i a constant t t matrix t i and d depends d d only l on th the coordinates di t of f the three nodes of the triangular element. the strains will be constant over a given element
64
65
A is the area of the triangle and xij=xi- xj. (textbook has results for a coordinate system with x aligned with side 1-2
k = BT EBtA
1 (e ) T (e ) {q } [k ]66 {q( e ) } 2
Element Stiffness Matrix: [k( e ) ] = hA[B]T [C][B] Different from the truss and beam elements elements, transformation matrix [T] is not required in the two-dimensional element because [k] is constructed in the global coordinates.
The strain energy of the entire solid is simply the sum of the element strain energies
U=
U (e ) =
e=1
NE
66
along l th the x-axis i and d linear along y-axis Exact Solution: xx = 60 MPa Max deflection vmax = 0.0075 0 0075 m
xx is constant
Max v = 0.0018
xx
x
67
68
69
R.T. Haftka
University of Florida
70
Displacement field:
So, u and v are bilinear in x and y. Because of form, sides id are stiffer tiff th than di diagonals-artificial l tifi i l anisotropy! i t !
R.T. Haftka EML5526 Finite Element Analysis University of Florida
71
R.T. Haftka
University of Florida
72
Fig. Fig (a) is the correct deformation in pure bending while (b)
is the deformation of Q4 (sides remain straight). Physical y interpretation: p applied pp moment is resisted by ya spurious shear stress as well as flexural (normal) stresses.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
73
R.T. Haftka
University of Florida
74
Fig 3.6-2
See Eqn. 3.6-6 for strain-displacement matrix (= Nd=Bd). All in all all, Q4 converges properly with mesh refinement and works better than CST in most problems.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
75
R.T. Haftka
University of Florida
Stress is constant along the x x-axis axis (pure bending) linear through the height of the beam Deflection is much higher than CST element element. In fact fact, CST element is too stiff. However, stress is inaccurate.
76
N1 = ( x 1)( y 1) N 2 = x( y 1) N 3 = xy N 4 = ( x 1) y
N1 / x = ( y 1) N 2 / x = ( y 1) N 3 / x = y N 4 / x = y
N1 / y = ( x 1) N 2 / y = x N 3 / y = x N 4 / y = ( x 1)
77
N1 / y = x 1 N 2 / y = x N 3 / y = x N 4 / y = ( x 1)
1 2
Strain 4 N I xx = u I = 1 0.0002022
I =1 4
yy =
4
N I vI = 0.0002022 x + 0.0002022 x = 0 I =1 y
xy =
N I N vI + I u I = 0.0002022 0.0004044 x y I =1 x
Stress:
RECTANGULAR ELEMENT
y-normal y normal stress and shear stress are supposed to be zero zero.
yy Plot
xx is a linear function of y alone yy if a linear function of x alone xy is a linear function of x and y
xy Plot
N1 / x = ( y 1)
N1 / y = ( x 1)
N 2 / x = ( y 1) N 2 / y = x N 3 / x = y N 3 / y = x N 4 / x = y N 4 / y = ( x 1) )
79
xx
4 N NI = uI yy = I v I I =1 y I =1 x
RECTANGULAR ELEMENT
Discussions
Cant represent constant shear force problem because xx must be a linear function of x. Even E if xx can represent t linear li strain t i i in y-direction, di ti the th rectangular t l element cant represent pure bending problem accurately. Spurious shear strain makes the element too stiff.
u = 1 + 2 x + 3 y + 4 xy v = 1 + 2 x + 3 y + 4 xy
Exact
xx = 2 + 4 y yy = 3 + 4 x xy = ( 3 + 2 ) + 4 x + 4 y
4 0
Rectangular element
80
RECTANGULAR ELEMENT
Two-Layer Two Layer Model
xx = 3.48107 vmax = 0.0053
81
As element is distorted distorted, the solution is not accurate any more more.
82
Sxx is supposed to change linearly along x x-axis. axis But But, the element is unable to represent linear change of stress along x-axis. Why? Exact solution: v = 0.005 m and xx = 6e7 Pa.
83
u ( x, y ) = a0 + a1 x + a2 y + a3 x + a4 xy
2
+ a5 y 2 + a6 x 2 y + a7 xy 2
Strain
u ( x, y ) = a1 + 2a3 x + a4 y + 2a6 xy + a7 y 2 x
Can this element accurately represent pure bending and constant shear force problem?
84
85
Sxx
Syy
86
Bubble modes
Strain xx can vary linear along x-dir. Shear strain xy can vanish for pure bending Nodeless DOFs, a1, a2, a3, and a4, are condensed in the element level ( (total 12 DOFs) )
87
88
Modeling the previous bending problem with Q6 elements gives the following stresses:
R.T. Haftka
University of Florida
89
3.7-1
90
The edges x=a deform into a parabola (i.e., quadratic displacement in y) (same for y=b)
R.T. Haftka
University of Florida
91
represent many strain states exactly. For example, p states of constant strain, bending g strain, etc.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
92
Mechanical loads: concentrated loads surface traction loads, traction, body forces.
R.T. Haftka
University of Florida
93
R.T. Haftka
University of Florida
94
95
R.T. Haftka
University of Florida
96
In (b), a Q4 element and two LSTs share the top midnode so that the nodal loads from Q4 and the right LST are combined.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
97
LST has no vertex loads and vertex loads of Q8 are upwards! The resultant in all cases is W, the weight of the element.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
98
. the beam is extended. Rotational dof at A, B and C are In (b) associated with the beam elements only only. A plane element with drilling dof would also work but is not recommended.
R.T. Haftka EML5526 Finite Element Analysis University of Florida
99
R.T. Haftka
University of Florida
10 0
Stress calculation
Combining g Hookes law with strain-displacement p equation q
= E ( Bd 0 ) e.g. x E y = 2 1 xy 0 1 T 1 0 Bd T 0 0 0 (1 ) / 2
R.T. Haftka
University of Florida
10 1
There are interpolation and extrapolation techniques that we will study y later
R.T. Haftka EML5526 Finite Element Analysis University of Florida
10 2
R.T. Haftka
University of Florida
10 3
R.T. Haftka
University of Florida