Documente Academic
Documente Profesional
Documente Cultură
http://ocw.mit.edu
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
2.094— Finite Element Analysis of Solids and Fluids
— Fall ‘08 —
MIT OpenCourseWare
Contents
1.2.4 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
6.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7 Isoparametric elements 28
9 u/p formulation 37
1
MIT 2.094 Contents
15 Field problems 61
19 Slender structures 81
2
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
• analytical solution
• F.E. solution(s)
• holes included
• F.E. solution(s) ⇒
You will use ADINA in your projects (and homework) for structures and fluid flow.
3
MIT 2.094 1. Large displacement analysis of solids/structures
Question Is there a unique solution? Yes, for infinitesimal small displacement/strain. Not necessarily
for large displacement/strain.
For example:
Snap-through problem
4
MIT 2.094 1. Large displacement analysis of solids/structures
Not physical
t
R is in “direction” of bending moment ⇒ Not in
equilibrium.
II. Compatibility
5
MIT 2.094 1. Large displacement analysis of solids/structures
1.2.4 Notation
t t
τij i, j = 1, 2, 3 τij = t τji (1.1)
6
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
Reading:
Ch. 1, Sec.
Assume that on t Su the displacements are zero (and t Su is constant). Need to satisfy at time t: 6.1-6.2
(For i = 1, 2, 3)
t
τij,j + t fiB = 0 in t V (sum over j) (2.2)
t t t Sf
τij nj = fi on t Sf (sum over j) (2.3)
t Sf t
(e.g. fi = τi1 t n1 + t τi2 t n2 + t τi3 t n3 ) (2.4)
S
And: t τ11 t n1 + t τ12 t n2 = t f1 f
• Stress-Strain law
t
τij = function t uj
� �
(2.5)
7
MIT 2.094 2. Finite element formulation of solids and structures
� � �
t t Sf S
τij t eij d t V = t B
fi ui d t V + fi u i f d t Sf (2.6)
tV tV tS
f
where
� �
1 ∂ui ∂uj
t eij = + t (2.7)
2 ∂ t xj ∂ xi
�
with ui � = 0 (2.8)
�
t Su
2.2 Example
Derivation of (2.9)
t
τ11,1 + t f1B = 0 by (2.2) (2.10)
�t
τ11,1 + t f1B u1 = 0
�
(2.11)
∗ or Principle of Virtual Displacements
8
MIT 2.094 2. Finite element formulation of solids and structures
Hence,
�
�t
τ11,1 + t f1B u1 d t V = 0
�
(2.12)
tV
� �
t
� tS t t
τ11 u1 � t Sf − u1,1 τ d V + u1 t f1B
d t V = 0 (2.13)
� �� u� tV ���� 11 tV
S te11
u1 f tτ11 tSf
Therefore we have
� �
t t S
e
t 11 τ 11 d V = u1 t f1B d t V + u1 f t Pr t Sf (2.14)
tV tV
From (2.12) to (2.14) we simply used mathematics. Hence, if (2.2) and (2.3) are satisfied, then (2.14)
must hold. If (2.14) holds, then also (2.2) and (2.3) hold!
Namely, from (2.14)
� � �
� tS S
u1,1 tτ11 d tV = u1 tτ11 � tSf − u1 tτ11,1 d tV = u1 tf1B d tV + u1 f
tPr tSf (2.15)
tV u tV tV
or
�
�t S �
τ11,1 + tf1B d tV + u1 f tPr − tτ11 tSf = 0
� �
u1 (2.16)
tV
� ��
x t
�
Now let u1 = x 1 − tL τ11,1 + tf1B , where tL = length of bar.
Hence we must have from (2.16)
t
τ11,1 + tf1B = 0 (2.17)
9
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
Reading:
Sec. 6.1-6.2
We need to satisfy at time t:
• Equilibrium
∂ tτij t B
+ fi = 0 (i = 1, 2, 3) in t V (3.1)
∂ txj
t t Sf
τij tnj = fi (i = 1, 2, 3) on t Sf (3.2)
• Compatibility
• Stress-strain law(s)
� �
1 ∂ ui ∂uj
teij = + t (3.4)
2 ∂ t xj ∂ xi
• If (3.3) holds for any continuous virtual displacement (zero on tSu ), then (3.1) and (3.2) hold and
vice versa.
10
MIT 2.094 3. Finite element formulation for solids and structures
Major steps
III. Use divergence theorem. Obtain a boundary term of stresses times virtual displacements on t S =
t
Su ∪ t Sf .
IV. But, on t Su the ui = 0 and on t Sf we have (3.2) to satisfy.
Result: (3.3).
Example
� �
t t S
τ11 te11 d tV = ui f1 f d tSf (3.6)
tV tS
f
11
MIT 2.094 3. Finite element formulation for solids and structures
1 1
u(r) =(1 + r) u1 + (1 − r) u2 (3.7)
2 2
t 1 1
u(r) = (1 + r) u1 + (1 − r) t u2
t
(3.8)
2 2
1 1
u(r) = (1 + r) u1 + (1 − r) u2 (3.9)
2 2
� �
T t for el. (1) T
te11t τ11 d V −→ [u1 u2 ] B (1) t
τ11 d t V (3.11)
tV tV
� �� �
= t F (1)
for el. (1)
−→ [u1 u2 ] t F (1) (3.12)
⎡ ⎤
� t (1) �
F̂
= ⎣ U1 U 2 U 3⎦ (3.13)
���� ���� 0
u2 u1
where
t (1) (1)
F̂1 = t F2 (3.14)
t (1) t (1)
F̂2 = F1 (3.15)
R.H.S.
⎡ ⎤
(unknown reaction at left)
� �
U1 U2 U3 ⎣ 0 ⎦ (3.17)
t t Sf
� ��
T
� Sf · f
1
Now apply,
T � �
U = 1 0 0 (3.18)
then,
T � �
U = 0 1 0 (3.19)
then,
T � �
U = 0 0 1 (3.20)
12
MIT 2.094 3. Finite element formulation for solids and structures
This gives,
⎡ ⎤
� � � � unknown reaction
t
F̂ (1) 0 0
+ =⎣ (3.21)
⎢ ⎥
t
0 F̂ (2)
⎦
t tSf t
f1 · Sf
We write that as
t
F = tR (3.22)
t
F = fn t U1 , t U2 , t U3
� �
(3.23)
13
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
The exact solution of the mathematical model must satisfy the conditions:
• Compatibility
• Stress-strain law(s)
I. Differential formulation
II. Variational formulation (Principle of virtual displacements) (or weak formulation)
We obtained
t
F = tR (4.1)
where t F is a function of displacements/stresses/material law; and t R is a function of time.
Assume for now linear analysis: Equilibrium within 0 V and on 0 Sf , linear stress-strain law and small
displacements yields
t
F = K · tU (4.2)
We want to establish,
KU (t) = R(t) (4.3)
14
MIT 2.094 4. Finite element formulation for solids and structures
Consider
Û T =
� �
U1 V1 W1 U2 · · · WN (N nodes) (4.4)
We also say
Û T = U1
� �
U2 U3 · · · Un (n = 3N ) (4.5)
We now assume
⎡ ⎤(m)
u
u(m) = H (m) Û , u(m) = ⎣ v ⎦ (4.6a)
w
e.g. γxy = +
∂x ∂y
We also assume
15
MIT 2.094 4. Finite element formulation for solids and structures
�� T �� (m) T (m)
�(m) τ (m) dV (m) = Û fB dV (m) (4.8)
m V (m) m V (m)
m V (m)
� � (4.9)
T ��
(m) T B (m) (m)
Û H f dV
m V (m)
Finally,
� �
T ��
� (m) T (m) (m) (m)
�Û B C B dV Û =
m V (m)
� � (4.11)
T ��
(m) T (m)
fB dV (m)
�
Û
� H
m V (m)
with
T T T
ˆ B (m)
�(m) = U
(4.12)
KÛ = RB (4.13)
where K is n x n, and RB is n x 1.
Direct stiffness method:
�
K= K (m) (4.14)
m
(m)
�
RB = RB (4.15)
m
�
(m) T
K = B (m) C (m) B (m) dV (m) (4.16)
(m)
�V
(m) T (m)
RB = H (m) f B dV (m)
(4.17)
V (m)
16
MIT 2.094 4. Finite element formulation for solids and structures
E = Young’s Modulus
F.E. model
⎡⎤
U1
U = ⎣ U2 ⎦ (4.18)
U3
Element 1
⎡ ⎤
U1
u(1) (x) = x x
� �
1− 100 0 ⎣ U2 ⎦ (4.19)
� �� 100 � U
3
H (1)
17
MIT 2.094 4. Finite element formulation for solids and structures
⎤ ⎡
U1
�(1) 1 1
� �
xx (x) = − 100 0 ⎣ U2 ⎦ (4.20)
� ��100 � U
3
B (1)
Element 2
u(2) (x) = x x
� �
0 1 − 80 80 U (4.21)
� �� �
H (2)
�(2) 1 1
� �
xx (x) = 0 − 80 80 U (4.22)
� �� �
B (2)
Then,
⎤ ⎡ ⎡ ⎤
1 −1 0 0 0 0
E ⎣ 13E ⎣
K= −1 1 0 ⎦ + 0 1 −1 ⎦ (4.23)
100 240
0 0 0 0 −1 1
where,
� �
E(1) AE
≡ (4.24)
100 L
� �
E · 13 13 E
= (4.25)
3 · 80 3 80
� �� �
A∗
� �
A� < A∗ < A�
� �
η=0 η=80 (4.26)
1 < 4.333 < 9
18
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
KU = R (5.1a)
�
K= K (m) (5.1c)
m
(m)
�
R= RB
(5.1d)
m
�(m) = B U (m)
(5.1e)
U T = U1 U2 · · · Un , (n = all d.o.f. of element assemblage)
� �
(5.1f)
�
T
K (m) = B (m) C (m) B (m) dV (m) (5.1g)
(m)
�V
(m) T (m)
RB = H (m) f B dV (m) (5.1h)
V (m)
Surface loads
19
MIT 2.094 5. F.E. displacement formulation, cont’d
�
(m) T (m)
Rs(m) = HS fS dS (m) (5.6)
S (m)
Need to add contributions from all surfaces of all loaded external elements.
KU = RB + RS + Rc (5.7)
Solve for Ua :
where Ub is known!
Then use
⎛ ⎞ ⎡ ⎤⎛ ⎞
τxx 1 ν 0 �xx
E ⎣ ν
⎝ τyy ⎠ = 1 0 ⎦ ⎝ �yy ⎠ (5.11)
1 − ν2 1−ν
τxy 0 0 2 γxy
20
MIT 2.094 5. F.E. displacement formulation, cont’d
⎛ ⎞
u1
⎜
⎜ u2 ⎟
⎟
� �
⎜
⎜ u3 ⎟
⎟
u(x, y) ⎜ u4 ⎟
=H⎜ ⎟ (5.12)
v(x, y) ⎜
⎜ v1 ⎟
⎟
⎜
⎜ v2 ⎟
⎟
⎝ v3 ⎠
v4
If we can set this relation up, then clearly we can get H (1) , H (2) , H (3) , H (4) .
u(x, y) = a1 + a2 x + a3 y + a4 xy (5.14)
v(x, y) = b1 + b2 x + b3 y + b4 xy (5.15)
(e.g.) u(1, 1) = a1 + a2 + a3 + a4 = u1 .
h2 (x, y) = 14 (1 − x)(1 + y)
21
MIT 2.094 5. F.E. displacement formulation, cont’d
h3 (x, y) = 14 (1 − x)(1 − y)
h4 (x, y) = 14 (1 + x)(1 − y)
u(x, y) = h1 u1 + h2 u2 + h3 u3 + h4 u4 (5.16)
v(x, y) = h1 v1 + h2 v2 + h3 v3 + h4 v4 (5.17)
⎛ ⎞
u1
⎜
⎜ u2 ⎟
⎟
� � � �⎜
⎜ u3 ⎟
⎟
u(x, y) h1 h2 h3 h4 0 0 0 0 ⎜ u4 ⎟
= ⎜ ⎟ (5.18)
v(x, y) 0 0 0 0 h1 h2 h3 h4 ⎜ v1 ⎟
� �� �⎜
⎜ v2
⎟
⎟
H (2x8) ⎜ ⎟
⎝ v3 ⎠
v4
We also want,
⎛ ⎞
u1
⎜
⎜ u2 ⎟
⎟
⎛ ⎞ ⎡ ⎤⎜ u3 ⎟
�xx h1,x h2,x h3,x h4,x 0 0 0 0 ⎜ ⎟
⎝ �yy ⎠ = ⎣ 0
⎜ u4 ⎟
0 0 0 h1,y h2,y h3,y h4,y ⎦⎜ ⎟ (5.19)
⎜ v1 ⎟
γxy h1,y h2,y h3,y h4,y h1,x h2,x h3,x h4,x ⎜ ⎟
� �� �⎜
⎜ v2 ⎟
⎟
B (3x8) ⎝ v3 ⎠
v4
∂u
�xx = (5.20)
∂x
∂v
�yy = (5.21)
∂y
∂u ∂v
γxy = + (5.22)
∂y ∂x
22
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
6.1 Example
Reading:
Ex. 4.6 in
the text
KU = R; R = RB + R s + R c + R r (6.1)
�
� T
K= K (m) ; K (m)
= B (m) C (m) B (m) d V (m) (6.2)
m V (m)
�
� (m) (m) T
RB = RB ; RB = H (m) f B(m) d V (m) (6.3)
m V (m)
u1 u2 u3 u4 v1 v2 v3 v4
⎡↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓⎤
� � � � × × × × × ← u1
K� = ⎢ . (6.4)
�
..
⎢ ..
⎥
el. (2) ⎥ .
⎢ ⎥
⎣ ⎦
23
MIT 2.094 6. Finite element formulation, example, convergence
In practice,
⎛ ⎞
u1
⎜ .. ⎟
⎜ . ⎟
� ⎜ ⎟
� ⎜ u4 ⎟
K� = B T CB dV ; �=B⎜ (6.5)
� ⎜ ⎟
el ⎜ v1 ⎟
⎟
V
⎜ . ⎟
⎝ .. ⎠
v4
where K is 8x8 and B is 3x8.
Assume we have K (8x8) for el. (2)
U1 U2 U3 U4 U5 · · · U11 · · · U18
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
× × × × × × × × ×
⎡ ⎤
.. .. .. ← U1
..
⎢ ⎥
⎢ . . . ⎥
⎢
.. .. ..
⎥ .
..
⎢ ⎥
⎢ . . . ⎥
K = ⎢
⎢ ⎥ . (6.6)
���� � � � ⎥
assemblage
⎢
⎢ .. .. ..
⎥
⎥ ← U11
⎢ . . . ⎥ ..
⎢
. . ..
⎥ .
. .
⎢ ⎥
⎢ . . . ⎥ ..
⎢
⎣
⎥
⎦ .
← U18
× × × × × × × × ×
Consider,
� �
T
RS = H S f S d S; HS = H� (6.7)
�
S on surface
� �
h1 h2 h3 h4 0 0 0 0 ← u(x, y)
H= (6.8)
0 0 0 0 h1 h2 h3 h4 ← v(x, y)
⎛ ⎞
u1
⎜ u2 ⎟
..
⎜ ⎟
⎜ ⎟
⎜
⎜ . ⎟
⎟
u=⎜
⎜ u4 ⎟
⎟ (6.9)
⎜
⎜ v1 ⎟
⎟
⎜ .. ⎟
⎝ . ⎠
v4
24
MIT 2.094 6. Finite element formulation, example, convergence
�
HS = H� (6.10)
�
y=+1
⎡ 1 1
⎤
2 (1 + x) 2 (1 − x) 0 0 0 0 0 0
=⎣ ⎦ (6.11)
1 1
0 0 0 0 2 (1 + x) 2 (1 − x) 0 0
From (6.7);
1
⎡ ⎤
2 (1 + x) 0
� +1 1
� �
⎢
2 (1 − x) 0 ⎥ 0
RS = ⎢ 1
⎥
⎦ −p(x) (0.1) dx (6.12)
−1
⎣ 0 2 (1 + x) � �� �
1
0 2 (1 − x) thickness
⎡ ⎤
0
⎢ 0
⎥
RS = ⎢ ⎥
⎣ −p0 (0.1)
⎦ (6.13)
−p0 (0.1)
Want h1 , h2 , h3 , h4 , h5
�5
u(x, y) = i=1 hi u i .
h5 = 12 (1 − x2 )(1 + y)
25
MIT 2.094 6. Finite element formulation, example, convergence
1 1
h1 = (1 + x)(1 + y) − h5 (6.14)
4 2
1 1
h2 = (1 − x)(1 + y) − h5 (6.15)
4 2
1
h3 = (1 − x)(1 − y) (6.16)
4
1
h1 = (1 + x)(1 − y) (6.17)
4
Note:
�
hi = 1
�
We must have i hi = 1 to satisfy the rigid body mode condition.
�
u(x, y) = h i ui (6.18)
i
� �
u(x, y) = h i u ∗ = u∗ h i = u∗ (6.19)
i i
From (6.1),
� �
�
(m)
K U =R (6.20)
m
� �� �
(m) T (m) (m) (m)
B C B dV U =R (6.21)
m V (m)
�� T
B (m) τ (m) d V (m) = R (6.22)
V (m)
m
�
�
(m) (m) T
F = R; F = B (m) τ (m) d V (m) (6.23)
V (m)
m
Two properties
I. The sum of the F (m) ’s at any node is equal to the applied external forces.
26
MIT 2.094 6. Finite element formulation, example, convergence
�
ˆ T F (m) = U
U ˆT T
B (m) τ (m) d V (m) (6.24)
V (m)
� �� �
T
=�(m)
�
T
= �(m) τ (m) d V (m) (6.25)
V (m)
=0 (6.26)
27
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
Reading:
Sec. 5.1-5.3
B T CB dV , RB = H T f B dV . Unique correspondence (x, y) ⇔ (r, s)
� �
We want K = V V
4
�
x= hi xi (7.1)
i=1
4
�
y= hi yi (7.2)
i=1
where
1
h1 =
(1 + r)(1 + s) (7.3)
4
h2 = (1 − r)(1 + s) (7.4)
4
...
4
�
u(r, s) = hi ui (7.5)
i=1
4
�
v(r, s) = hi vi (7.6)
i=1
28
MIT 2.094 7. Isoparametric elements
� = Bû ûT =
� �
u1 u2 · · · v4 (7.7)
⎡ ∂u ⎤
∂x
⎢ ⎥
⎢ ∂v ⎥
⎥ = Bû
⎢
⎢ ∂y ⎥ (7.8)
⎣ ⎦
∂u ∂v
∂y + ∂x
⎛ ∂
⎞ ⎡ ∂x ∂y
⎤⎛ ∂
⎞
∂r ∂r ∂r ∂x
⎝ ⎠=⎣ ⎦⎝ ⎠ (7.9)
∂ ∂x ∂y ∂
∂s ∂s ∂s ∂y
� �� �
J
⎛ ∂ ⎞ ⎛ ∂
⎞
∂x ∂r
⎝ ⎠ = J −1 ⎝ ⎠ (7.10)
∂ ∂
∂y ∂s
J must be non-singular which ensures that there is unique correspondence between (x, y) and (r, s).
Hence,
� 1� 1
K= B T CB t det(J ) dr ds (7.11)
−1 −1 � �� �
dV
� 1 � 1
Also, RB = H T f B t det(J ) dr ds (7.12)
−1 −1
K∼
��
T
=t Bij CBij det(Jij ) × (weight i, j) (7.13)
i j
(i = 1, 2) (7.14)
(j = 1, 2) (weight i, j = 1 in this case) (7.15)
29
MIT 2.094 7. Isoparametric elements
9-node element
9
�
x= hi xi (7.16)
i=1
9
�
y= hi yi (7.17)
i=1
9
�
u= hi u i (7.18)
i=1
9
�
v= hi vi (7.19)
i=1
3 = 26
� � � �
0
J= 3 (7.20)
0 2
Then, we can determine the number of appropriate integration points by investigating the maximum
order of B T CB.
For a rectangular element, 3x3 Gauss integration gives exact K matrix. If the element is distorted,
a K matrix which is still accurate enough will be obtained, (if high enough integration is used).
30
MIT 2.094 7. Isoparametric elements
Find u, solution, in V, vector space (any continuous function that satisfies boundary conditions),
satisfying
�
�T C� dV = a(u, v) = (f , v) for all v, an element of V. (7.22)
V � �� � � �� �
bilinear form R(v)
Example:
Finite Element problem Find uh ∈ Vh , where Vh is F.E. vector space such that
Note:
Also,
31
MIT 2.094 7. Isoparametric elements
Property I Define: eh = u − uh .
Hence,
a(u − uh , vh ) = 0 (7.26)
a(eh , vh ) = 0 (7.27)
Property II
Proof:
�� � 0 by Prop. I
2a(u
= a(uh , uh ) + � ��h , eh ) + a(eh , eh ) (7.30)
� �� �
≥0
32
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
(A) Find
u ∈ V such that a(u, v) = (f , v) ∀v ∈ V (Mathematical model) (8.1)
a(v, v) > 0 ∀v ∈ V, � 0.
v= (8.2)
where (8.2) implies that structures are supported properly. E.g.
Properties eh = u − uh
33
MIT 2.094 8. Convergence of displacement-based FEM
� �
(C) Assume Mesh� “is contained in” Mesh�
� �
h1 h2
� �
e.g. Mesh� not contained in Mesh�
� �
h1 h2
uh minimizes! (Recall eh = u − uh )
Proof: Pick wh ∈ Vh .
�� �0
a(eh + wh , eh + wh ) = a(eh , eh ) + �
2a(e
��h , wh ) + a(wh , wh ) (8.8)
� �� �
≥0
Take wh = uh − vh .
Using property (III) and (C), we can say that we will converge monotonically, from below, to a(u, u):
34
MIT 2.094 8. Convergence of displacement-based FEM
(Ch. 4.3)
Hence,
By theory,
35
MIT 2.094 8. Convergence of displacement-based FEM
By experiment, we can evaluate log(E − Eh ) for different meshes and plot log(E − Eh ) vs. log h
or
(Small error in �V results in huge error on pressure as κ → ∞, the constant C in (8.15) can be very large
⇒ locking)
36
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
II. Compatibility
III. Stress-strain law
Solution
where
�
1 i=j
δij = Kronecker delta = (9.7)
0 � j
i=
Deviatoric strains:
�V
��ij = �ij − δij (9.8)
3
� τkk �
τij = −pδij + 2G��ij p=− (9.9)
3
(9.2) becomes
� �
�T � �
� C � dV + �V κ�V dV = R (9.10)
V V
� �
T
�� C � �� dV − �TV p dV = R (9.11)
V V
37
MIT 2.094 9. u/p formulation
p + κ�V = 0 (9.12)
�
p (p + κ�V ) dV = 0 (9.13)
�V �
p�
− p �V + dV = 0 (9.14)
V κ
For an element,
u = Hû (9.15)
�
� = BD û (9.16)
�V = BV û (9.17)
p = Hp p̂ (9.18)
p = Hp p̂ = [1]{p0 } (9.21)
p(x, y) = p0 (9.22)
�
T
�
Kuu = BD C BD dV (9.24a)
V
�
Kup =− BVT Hp dV (9.24b)
V
�
Kpu =− HpT BV dV (9.24c)
V
�
1
Kpp =− HpT Hp dV (9.24d)
V κ
38
MIT 2.094 9. u/p formulation
In practice, we use elements that use pressure interpolations per element, not continuous between
elements. For example:
Then, unless ν = 0.5 (where Kpp = 0), we can use static condensation on the pressure dof’s.
Use p̂ equations to eliminate p̂ from the û equations.
−1
� �
Kuu − Kup Kpp Kpu û = R (9.25)
p(x, y) = p0 + p1 x + p2 y (9.26)
⎡ ⎤
=�V
� � �� �
⎢ Vol qh � · vh dVol ⎥
⎢ ⎥
inf sup ⎥≥β>0 (9.27)
�qh � �vh �
���� ���� ⎣ ⎢
⎦
qh ∈Qh vh ∈Vh � �� �
for normalization
Qh : pressure space.
If “this” holds, the element is optimal for the displacement assumption used (ellipticity must also be
satisfied).
Note:
For example,
inf {1, 2, 4} = 1
sup {1, 2, 4} = 4
39
MIT 2.094 9. u/p formulation
For (entry [3,1] in matrix) assume the circled entry is the minimum (inf) of .
Case 1 βh = 0
⇒ no equation for ph |j
⇒ spurious pressure! (any pressure satisfies equation)
Case 2 βh = small = �
� · uh |i = 0 ⇒ uh |i = 0
∴�· ph |j + uh |i · α = Rh |i
� �
Rh |i
displ. = 0
⇒ ph |j = ⇒ as � is small
� pressure → large
The behavior of given mesh when bulk modulus increases: locking, large pressures. See Example
4.39 textbook.
40
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
We developed Reading:
� Ch. 6
t
τij teij d tV = tR (10.1)
tV
� �
1 ∂ui ∂uj
teij = + t (10.2)
2 ∂ txj ∂ xi
�
t
τij δt eij d tV = tR (10.3)
tV
� �
1 ∂(δui ) ∂(δuj )
δt eij = + (≡ t eij ) (10.4)
2 ∂ t xj ∂ t xi
In FEA:
t
F = tR (10.5)
In linear analysis
t
F = K t U ⇒ KU = R (10.6)
In general nonlinear analysis, we need to iterate. Assume the solution is known “at time t”
t 0
x= x + tu (10.7)
41
MIT 2.094 10. F.E. large deformation/general nonlinear analysis
F ∼
= tK ΔU , t
K = tangent stiffness matrix at time t (10.11)
From (10.8),
t t+Δt
K ΔU = R − tF (10.12)
t+Δt
We use this to obtain an approximation to U . We obtain a more accurate solution for U (i.e. F)
using
t+Δt
K (i−1) ΔU (i) = t+Δt
R − t+ΔtF (i−1) (10.13)
t+Δt (i) t+Δt (i−1) (i)
U = U + ΔU (10.14)
Also,
t+Δt
F (0) = tF (10.15)
t+Δt (0) t
K = K (10.16)
t+Δt
U (0) = tU (10.17)
Note:
��
2
�a�2 = (ai )
i
�
(i)
ΔU =U
i=1,2,3...
Process
42
MIT 2.094 10. F.E. large deformation/general nonlinear analysis
So, perform this process for i = 1, 2, 3, . . . , n, where n is the total number of degrees of freedom.
Pictorially,
.. ..
⎡ ⎤
⎢ . . ⎥
t
K = ⎢ ... ..
⎢ ⎥
⎣ . ··· ⎥
⎦
.. ..
. .
The strain measure energy-conjugate to the 2nd P-K stress 0tSij is the Green-Lagrange strain 0t�ij
Then,
�
t
0 Sij δ 0t�ij d 0V = tR (10.21)
0V
Also,
�
t+Δt
0 Sij δ t+Δt0�ij d 0V = t+Δt
R (10.22)
0V
Example
43
MIT 2.094 10. F.E. large deformation/general nonlinear analysis
t
F = tR (10.23)
t+Δt t+Δt
F = R (10.24)
i = 2,
t+Δt
K (1) ΔU (2) = t+Δt
R − t+ΔtF (1) (10.27)
t+Δt (2) t+Δt (1) (2)
U = U + ΔU (10.28)
44
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
d tx1
⎡ ⎤
d tx = ⎣ d tx2 ⎦ (11.3)
d tx3
⎡ 0 ⎤
d x1
d 0x = ⎣ d 0x2 ⎦ (11.4)
d 0x3
Implies that
d tx = 0tX d 0x (11.5)
t
0C = 0tX T 0tX (11.6)
Some applications
45
MIT 2.094 11. Deformation, strain and stress tensors
� 0 T t T��t 0
�
� t �2 d x 0X 0X d x
λ = , from (11.5) (11.9)
d 0s · d 0s
Express
d 0x = d 0s 0n
� �
(11.10)
0 0
n = unit vector into direction of d x (11.11)
� t �2 0 T t 0
⇒ λ = n 0 C n (11.12)
� 0 T t 0 � 12
∴ tλ = n 0C n (11.13)
Also,
�T � t � � t � � t �
d tx̂ · d x = d ŝ d s cos tθ,
�
(a · b = �a��b� cos θ) (11.14)
From (11.5),
� ��
ˆT
�
d 0x̂T 0tX t
0X d 0x � �
t ˆ
cos tθ = 0X ≡ 0tX (11.15)
d tsˆ d ts
d ŝ n̂ 0 C n d 0s
0 0 T t 0
= (11.16)
d tsˆ · d ts
0
n̂T 0tC 0n
∴ cos tθ = (11.17)
tλ̂ tλ
Also,
0
t ρ
ρ= (see Ex. 6.5) (11.18)
det 0tX
Example Reading:
Ex. 6.6 in
the text
46
MIT 2.094 11. Deformation, strain and stress tensors
1
h1 = (1 + 0x1 )(1 + 0x2 ) (11.19)
4
..
.
t
xi = 0xi + tui (11.20)
4
�
= hk txki , (i = 1, 2) (11.21)
k=1
where txki are the nodal point coordinates at time t ( tx11 = 2, tx12 = 1.5)
Then we obtain
5 + 0x2 1 + 0x1
� �
t 1
0X = 1 1 (11.22)
4 2 (1 + 0x2 ) 0
2 (9 + x1 )
At 0x1 = 0, 0x2 = 0,
� �
t
� 1 5 1
0 X �0 = (11.23)
�
xi = 0x2 =0 4 12 9
2
t 1 �t T t � 1 �t �
0� = 0X 0X − I = 0C − I (11.24)
2 2
�0
xi + tui
�
∂ txi ∂ ∂ tui
= = δ ij + (11.25)
∂ 0xj ∂ 0xj ∂ 0xj
We find that
t 1 �t t t t
�
47
MIT 2.094 11. Deformation, strain and stress tensors
� √ ��
3 1 4
�
t 2 −
√2 3 0
0X = 1 3 0 3 (11.30)
2 2 2
Then,
t
� �2
0C = 0tX T 0tX = 0tU (11.31)
t 1 �� t �2 �
0� = 0U −I (11.32)
2
This shows, by an example, that the components of the Green-Lagrange strain are independent of a
rigid-body rotation.
48
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
We discussed:
� �
t ∂ txi � t �−1 t
0X = ⇒ d tx = 0tX d 0x, d 0x = 0X dx (12.1)
∂ 0xj
t
0C = 0tX T 0tX (12.2)
� �
0
� t �−1 ∂ xi
d 0x = 0t X d tx where 0t X = 0X = (12.3)
∂ txj
Note in FEA
� ⎫
0
xi = hk 0xki ⎪
⎪
⎬
k
� for an element (12.6)
t
ui = hk tuki ⎪
⎪
⎭
t
xi = 0xi + tui → for any particle (12.7)
49
MIT 2.094 12. Total Lagrangian formulation
E.g., k = 4
Then
� �
t
0 Sij δ 0t�ij d 0V = t
τij δ t
eij d tV = tR (12.12)
0V tV
t+Δt
0S = 0tS + 0 S (12.13)
t+Δt
0 Sij = 0tSij + 0 Sij
(12.14)
t+Δt
t
0� = 0� + 0� (12.15)
t+Δt t
0�ij = 0�ij + 0�ij (12.16)
Assume the solution is kown at time t, calculate the solution at time t + Δt. Hence, we apply (12.12)
at time t + Δt:
�
t+Δt t+Δt 0 t+Δt
0 Sij δ 0�ij d V = R (12.17)
0V
Look at δ 0t�ij :
1 �t
δ 0t�ij = δ t t t
�
0ui,j + 0uj,i + 0uk,i 0uk,j (12.18a)
2� �
t 1 ∂δui ∂δuj ∂δuk ∂ tuk ∂ tuk ∂δuk
δ 0�ij = + 0 + 0 · 0 + 0 · 0 (12.18b)
2 ∂ 0xj ∂ xi ∂ xi ∂ xj ∂ xi ∂ xj
1
δ 0t�ij = δ 0ui,j + δ 0uj,i + δ 0uk,i 0tuk,j + 0tuk,i δ 0uk,j
� �
(12.18c)
2
We have
t+Δt
0�ij − 0t�ij = 0�ij (12.19)
0�ij = 0eij + 0 ηij (12.20)
50
MIT 2.094 12. Total Lagrangian formulation
where 0eij is the linear incremental strain, 0 ηij is the nonlinear incremental strain, and
⎛ ⎞
1⎜ t t
0eij = ⎝ u + u + u u + u u ⎠ (12.21)
⎟
2 0 i,j 0 j,i �0 k,i 0 k,j �� 0 k,i 0 k,j�
initial displ. effect
1
0 ηij = u u (12.22)
2 0 k,i 0 k,j
where
∂uk t+Δt
0uk,j = , uk = uk − tuk (12.23)
∂ 0xj
Note
From (12.17):
�
�t
δ 0eij + δ 0 ηij d 0V
�� �
0 Sij + 0 Sij
0V
�
�t t
� 0
= 0 Sij δ 0eij + 0 Sij δ 0eij + 0 Sij δ 0 ηij + 0 Sij δ 0 ηij d V (12.25)
0V
t+Δt
= R (12.26)
Linearization
⎛ t
⎞
t
0 KL U 0 KN L U
� �
⎜� �� � �t �� �⎟ 0 t+Δt t 0
⎜0 Sij δ 0eij + 0 Sij δ 0 ηij ⎟ d V =
⎝ ⎠ R− 0 Sij δ 0eij d V (12.27)
0V 0V
� �� �
t
0F
We use,
51
MIT 2.094 12. Total Lagrangian formulation
Left hand side as before but using (k − 1) and right hand side is
�
t+Δt t+Δt (k−1) 0
= t+ΔtR − 0 Sij δ 0�ij d V (12.30)
0V
gives
t+Δt t+Δt (k−1)
R− 0F (12.31)
52
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
�
t+Δt
τij δ t+Δteij d t+ΔtV = t+Δt
R (13.1)
t+ΔtV
�
t+Δt t+Δt 0 t+Δt
0 Sij δ 0�ij δ V = R (13.2)
0V
t+Δt
0 Sij = 0tSij + 0 Sij (13.3)
t+Δt t
0�ij = 0�ij + 0�ij (13.4)
0�ij = 0eij + 0 ηij (13.5)
where 0tSij and 0t�ij are known, but 0 Sij and 0�ij are not.
1� t t
�
0eij = 0ui,j + 0uj,i + 0uk,i 0uk,j + 0uk,j 0uk,i (13.6)
2
1� �
0 ηij = 0uk,i 0uk,j (13.7)
2
53
MIT 2.094 13. Total Lagrangian formulation, cont’d
�
t t T t 0
0 KL = 0 BL 0 C 0 BL d V (13.10)
0V
�
t t T t t 0
0 KN L = 0 BN L 0 S 0 BN L d V (13.11)
0V ����
matrix
�
t t T t 0
0F = 0 BL 0 Ŝ d V (13.12)
0V ����
vector
t+Δt
U (i) = t+Δt
U (i−1) + ΔU (i) (13.14)
∂u1 ∂ tu ∂uk
0e11 = 0 + 0 k · 0 (13.15)
∂ x1 ∂ x1 ∂ x1
� �
1 ∂uk ∂uk
0 η11 = · (13.16)
2 ∂ 0x1 ∂ 0x1
We are after
u11
⎛
⎞
⎜ u12 ⎟ t
0e11 = 0tBL ⎜
⎝ u21 ⎠ = 0 BL û
⎟ (13.17)
u22
2
�
ui = hk uki (13.18)
k=1
�
t
ui = hk tuki (13.19)
k=1
54
MIT 2.094 13. Total Lagrangian formulation, cont’d
1 � �
0e11 =0 −1 0 1 0 û
L
⎛ ⎞
⎜ ⎟
⎜0 ⎟
⎜ L + ΔL ⎟ 1 � �
− ⎟ 0L −1
+⎜ cos θ 1 ⎟· 0 1 0 û
⎜ 0L
⎜� �� �⎟
⎝ ⎠
∂ tu1
∂ 0x1
⎛ ⎞
⎜ ⎟
⎜0 ⎟
⎜ L + ΔL ⎟ 1 � �
+⎜
⎜
0L
sin θ⎟⎟ · 0L 0 −1 0 1 û (13.20d)
⎜� �� � ⎟
⎝ ⎠
∂ tu2
∂ 0x1
=0tBL û (13.20e)
Hence,
0
L + ΔL � �
0e11 = 2
− cos θ − sin θ cos θ sin θ û (13.20f)
( 0L)
where the boxed quantity above equals 0tBL . In small strain but large rotation analysis we assume
ΔL � 0L,
1 � �
0e11 = 0L
− cos θ − sin θ cos θ sin θ û (13.20g)
� �
1 ∂u1 ∂u1 ∂u2 ∂u2
0 η11 = 0 0 + 0 (13.21a)
2 ∂ x1 ∂ x1 ∂ x1 ∂ 0x1
� �
1 ∂δu1 ∂u1 ∂u1 ∂δu1 ∂δu2 ∂u2 ∂u2 ∂δu2
δ 0 η11 = + 0 + 0 + 0 (13.21b)
2 ∂ 0x1 ∂ 0x1 ∂ x1 ∂ 0x1 ∂ x1 ∂ 0x1 ∂ x1 ∂ 0x1
� �
∂δu1 ∂u1 ∂δu2 ∂u2
= + 0 (13.21c)
∂ 0x1 ∂ 0x1 ∂ x1 ∂ 0x1
� � tS �� ∂u1
�
t
�
∂δu1 ∂δu2 0 11 0 ∂ 0x1
0 S11 δ 0 η11 = ∂ 0x1 ∂ 0x1 0 t ∂u2 (13.21d)
0 S11 ∂ 0x1
� �� �
t
0S
� ∂u1
� � �
∂ 0x1 1 −1 0 1 0
∂u2 = û (13.21e)
∂ 0x1
0L 0 −1 0 1
� �� �
BN L
55
MIT 2.094 13. Total Lagrangian formulation, cont’d
0C =E (13.22)
t
0 Ŝ = 0tS11 (13.23)
t EA
0K =0L
cos2 θ − cos2 θ
⎡ ⎤
cos θ sin θ − cos θ sin θ
⎢ sin2 θ − sin θ cos θ − sin2 θ ⎥
=⎢ ⎥
⎣ cos2 θ sin θ cos θ ⎦
sym sin2 θ
� �� �
t
0 KL (13.24)
⎡ ⎤
1 0 −1 0
t
P ⎢ 0 1 0 −1 ⎥
+0 ⎣ ⎢ ⎥
L −1 0 1 0 ⎦
0 −1 0 1
� �� �
t
0 KN L
When θ = 0, 0tKL doesn’t give stiffness corresponding to u22 , but 0tKN L does.
56
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
�
t+Δt
τij δ t+Δteij d t+ΔtV = t+Δt
R (14.1)
t+ΔtV
�
t+Δt t+Δt 0 t+Δt
0 Sij δ 0�ij δ V = R (14.2)
0V
⏐
� linearization
⏐
� � �
0 t 0 t+Δt t 0
0 Cijrs 0ers δ 0eij δ V + 0 Sij δ 0 ηij δ V = R− 0 Sij δ 0eij δ V (14.3)
0V 0V 0V
Note:
δ 0eij = δ 0t�ij
F.E. discretization
� � �
t+Δt
0
xi = hk 0xki t
xi = hk txki xi = hk t+Δtxki (14.4a)
k k k
� � �
t
ui = hk tuki t+Δt
ui = hk t+Δtuik ui = hk uik (14.4b)
k k k
57
MIT 2.094 14. Total Lagrangian formulation, cont’d
Truss
ΔL
0L � 1 small strain assumption:
t E 0A
0K = 0L
cos2 θ − cos2 θ
⎡ ⎤
cos θ sin θ − cos θ sin θ
2
⎢ cos θ sin θ sin θ − sin θ cos θ − sin2 θ ⎥
=⎢ ⎥
⎣ − cos2 θ − cos θ sin θ cos2 θ sin θ cos θ ⎦
2 (14.6)
− cos θ sin θ − sin θ sin θ cos θ sin2 θ
⎡ ⎤
1 0 −1 0
t
P ⎢ 0 1 0 −1 ⎥
+ ⎢
0L ⎣ −1
⎥
0 1 0 ⎦
0 −1 0 1
� � � �� �
u1 cos θ sin θ u1
= (14.7)
v1 − sin θ cos θ v1
t E 0A
0K = 0L
(14.8)
⎡ ⎤ ⎡ ⎤
1 0 −1 0 1 0 −1 0
⎢ 0 0 0 0 ⎥ t
P ⎢ 0 1 0 −1 ⎥
= ⎢
⎣ −1
⎥+ 0L
⎢
⎣ −1
⎥ (14.9)
0 1 0 ⎦ 0 1 0 ⎦
0 0 0 0 0 −1 0 1
58
MIT 2.094 14. Total Lagrangian formulation, cont’d
t
P
Q 0L = tP · Δ ⇒ Q= 0L
· Δ (14.10)
t
P
where the boxed term is the stiffness. In axial direction, 0L is not very important because usually
0 t t
E A P P
0L � 0L . But, in vertical direction, 0L is important.
⎡⎤
− cos θ
t t ⎢ − sin θ ⎥
⎢ ⎥
0 F = P ⎣ cos θ ⎦ (14.11)
sin θ
t t 1 �� �2 � �2 �
�
0 11 = u
0 1,1 + u u
0 1,1 0 1,1 + u u
0 2,1 0 2,1 + u
0 1,1 + u
0 2,1 (14.12)
� �� � 2� �� �
0e11
0 η11
0�22 = · · · (14.13)
0�12 = · · · (14.14)
(Axisymmetric)
0�33 = ? (14.15)
t1 �� t � 2 �
0� =
0U −I (14.16)
2⎡ ⎤
× × 0
⎣ × × 0 ⎦
t 2 0 0 ×
0U = (14.17)
↑
t 2
( λ)
� �
t d ts 2π 0x1 + tu1
λ= 0 =
d s 2π 0x1
t
(14.18)
u
=1+ 0 1
x1
�� �2 �
t
t 1 u1
0�33 = 1+ 0 − 1
2 x1
�2 (14.19)
t
1 tu1
�
u
= 0 1 +
x1 2 0x1
59
MIT 2.094 14. Total Lagrangian formulation, cont’d
t
�t �2
t+Δt u + u1 1 u 1 + u1
0�33 = 10 + 0x (14.20)
x1 2 1
t
� �2
t+Δt u1 u 1 u1 1 u1
0�33 = 0�33 − 0t�33 = 0x + 0x · 0x + 2 0x (14.21)
1 1 1 1
• F.E. solution ∼ uh
Find �u − uh � and �τ − τh �.
References
[1] T. Sussman and K. J. Bathe. “Studies of Finite Element Procedures - on Mesh Selection.” Computers
& Structures, 21:257–264, 1985.
[2] T. Sussman and K. J. Bathe. “Studies of Finite Element Procedures - Stress Band Plots and the
Evaluation of Finite Element Meshes.” Journal of Engineering Computations, 3:178–191, 1986.
60
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
S = Sθ ∪ Sq
∂θ
qy = −k (15.1)
∂y
∂θ
qz = −k (15.2)
∂z
III. Compatibility: temperatures need to be continuous and satisfy the boundary conditions.
61
MIT 2.094 15. Field problems
unit cross-section
dV = dx · (1) (15.4)
q |x − q |x+dx + q B dx = 0 (15.5)
� �
∂qx
q|x − q|x + dx + q B dx = 0 (15.6)
∂x
� �
∂ ∂θ
− −k � + qB �
dx
� dx
�=0 (15.7)
∂x ∂x
� �
∂ ∂θ
k = −q B (15.8)
∂x ∂x
� � � � �
∂ ∂θ
θ k + · · · + q B dV = 0 (15.11)
V ∂x ∂x
62
MIT 2.094 15. Field problems
⎛ ∂θ ⎞
∂x
⎜ ⎟
� ∂θ
⎜ ⎟
θ =⎜
⎜ ∂y
⎟
⎟ (15.13)
⎝ ⎠
∂θ
∂z
⎡ ⎤
k 0 0
k=⎣ 0 k 0 ⎦ (15.14)
0 0 k
Radiation
� � �4 �
4
q S = κ∗ (θr ) − θS (15.16)
� � �2 � � r
2
= κ∗ (θr ) + θS θ + θS θr − θS
�� �
(15.17)
= κ θr − θS
� �
(15.18)
where κ = κ(θS ) and θr is given temperature of source. At time t + Δt
� � �
� T t+Δt t+Δt � t+Δt B S
θ k θ dV = θ q dV + θ t+Δtq S dSq (15.19)
V V Sq
t+Δt
Let θ = tθ + θ (15.20)
t+Δt (i) t+Δt (i−1) (i)
or θ = θ + Δθ (15.21)
t+Δt (0) t
with θ = θ (15.22)
From (15.19)
�
�T (i)
θ t+Δtk(i−1) Δθ � dV
V
� �
�T t+Δt � (i−1)
= θ t+Δtq B dV − θ t+Δtk(i−1) θ dV (15.23)
V V
� � � ��
S (i−1) (i)
+ θ t+Δth(i−1) t+Δtθ e − t+Δtθ S + ΔθS dSq
Sq
(i)
where the ΔθS term would be moved to the left-hand side.
We considered the convection conditions
�
S
θ t+Δth t+Δtθ e − t+Δtθ S dSq
� �
(15.24)
Sq
63
MIT 2.094 15. Field problems
F.E. discretization
t+Δt
θ = H1x4 · t+Δtθ̂4x1 for 4-node 2D planar element (15.25)
t+Δt �
θ2x1 = B2x4 · t+Δtθ̂4x1 (15.26)
t+Δt S S t+Δt
θ =H · θ̂ (15.27)
For (15.23)
⎛ ⎞
� �
�T (i) gives
θ t+Δtk(i−1) Δθ � dV =⇒ ⎝ ���� B T t+Δtk(i−1) ���� θ̂ (i)
B dV ⎠ Δ ���� (15.28)
V V � �� �
4x2 2x2 2x4 4x1
� �
θ t+Δtq B dV ⇒ H T t+Δtq B dV (15.29)
V V
� �� �
� T t+Δt (i−1) t+Δt � (i−1)
θ k θ dV ⇒ B T t+Δtk(i−1) BdV t+Δt (i−1)
θ̂ (15.30)
V V � �� �
known
�
S T t+Δt (i−1)
� � ��
t+Δt e t+Δt S (i−1) (i)
θ h θ − θ + ΔθS dSq =⇒
Sq
(15.31)
⎛ ⎛ ⎞⎞
�
T
S t+Δt (i−1)
H
� �� � h H S ⎝ t+Δtθ̂ e − ⎝ t+Δtθ̂ (i−1) +Δ ����
���� θ̂ (i) ⎠⎠ dSq
Sq � �� � � �� �
4x1 1x4 4x1 4x1 4x1
∂φ ∂φ
vx = vy = (15.35)
∂x ∂y
∂2φ ∂2φ
⇒ + 2 = 0 in V (15.36)
∂x2 ∂y
64
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
� ∂θ �� �
θ� is prescribed, k � = qS � (16.2)
� �
Sθ ∂n Sq Sq
Sθ ∪ Sq = S Sθ ∩ Sq = ∅ (16.3)
65
MIT 2.094 16. F.E. analysis of Navier-Stokes fluids
� �
∂
ρcp v θ|x − ρcp v θ|x + (ρcp vθ)dx + conduction + etc (16.5)
∂x
In general 3D, we have an additional term for the left hand side of (16.1):
So (16.1) becomes
Navier-Stokes equations
• Differential form
66
MIT 2.094 16. F.E. analysis of Navier-Stokes fluids
• Variational form
� � � �
S S
v i ρvi,j vj dV + eij τij dV = v i fiB dV + v i f fi f dSf (16.13)
V V V Sf
�
p� · vdV = 0 (16.14)
V
• F.E. solution
We interpolate (x1 , x2 , x3 ), vi , v i , θ, θ, p, p. Good elements are
×: linear pressure
◦: biquadratic velocities
(Q2 , P1 ), 9/3 element
9/4c element
So in general,
Example:
For Sf e.g.
∂vt
τnn = 0, = 0; (16.15)
∂n
67
MIT 2.094 16. F.E. analysis of Navier-Stokes fluids
∂vn
and ∂t is solved for. Actually, we frequently just set p = 0.
vL vL k
Re = Pe = α= (16.16)
ν α ρcp
• Differential equations
exp Pe
� �
θ − θL L x −1
⇒ = (16.20)
θR − θL exp (Pe) − 1
68
MIT 2.094 16. F.E. analysis of Navier-Stokes fluids
• F.E. discretization
θ�� = Peθ� (16.21)
� 1 � 1
� �
θ θ dx + Pe θθ� dx = 0 + { effect of boundary conditions = 0 here} (16.22)
0 0
1 Pe
2 (θi+1 − 2θi + θi−1 ) = (θi+1 − θi−1 ) (16.23)
(h∗ ) 2h∗
vL
Pe = (16.24)
α
Define
h vh
Pee = Pe · = (16.25)
L α
Pee Pee
� � � �
−1 − θi−1 + 2θi + − 1 θi+1 = 0 (16.26)
2 2
what is happening when Pee is large? Assume two 2-node elements only.
θi−1 = 0 (16.27)
θi+1 = 1 (16.28)
Pee
� �
1
θi = 1− (16.29)
2 2
69
MIT 2.094 16. F.E. analysis of Navier-Stokes fluids
Pee
� �
1
θi = 1− (16.30)
2 2
70
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
71
MIT 2.094 17. Incompressible fluid flow and heat transfer, cont’d
Continuity
�
pvi,i dV = 0 (17.2)
V
P.V. temperature
� � � �
S
θρcp θ,i vi dV + θ,i kθ,i dV = θq B dV + θ q S dS (17.3)
V V V Sq
F.E. solution
�
xi = hk xki (17.4)
�
vi = hk vik (17.5)
�
θ= hk θk (17.6)
�
p= h̃k pk (17.7)
⎛ ⎞
v
⇒ F (u) = R
u = ⎝ p ⎠ nodal variables (17.8)
θ
dθ d2 θ
ρcp v =k 2 (17.9)
dx dx
dθ d2 θ
Pe = 2 (17.10)
dx dx
72
MIT 2.094 17. Incompressible fluid flow and heat transfer, cont’d
vL k
Pe = , α= (17.11)
α ρcp
θ∗ is non-dimensional
exp PeLx − 1
� �
θ − θL
= (17.12)
θR − θ L exp (Pe) − 1
h
h∗ =
L
� �
ξ ξ
θ(ξ) = 1 − ∗ θi−1 + ∗ θi (17.14)
h h
For node i:
Pee Pee
73
MIT 2.094 17. Incompressible fluid flow and heat transfer, cont’d
1 − (Pee /2)
θi = (17.19)
2
Physically unrealistic solution when Pee > 2. For this not to happen, we should refine the mesh—a very
fine mesh would be required. We use “upwinding”
dθ �� θi − θi−1
� = (17.20)
dx i h∗
The result is
Unfortunately it is not that accurate. To obtain better accuracy in the interpolation for θ, use the
function
� x�
exp Pe L −1
(17.23)
exp (Pe) − 1
This implies flow-condition based interpolation. We use such interpolation functions—see references.
References
[1] K.J. Bathe and H. Zhang. “A Flow-Condition-Based Interpolation Finite Element Procedure for
Incompressible Fluid Flows.” Computers & Structures, 80:1267–1277, 2002.
[2] H. Kohno and K.J. Bathe. “A Flow-Condition-Based Interpolation Finite Element Procedure for
Triangular Grids.” International Journal for Numerical Methods in Fluids, 51:673–699, 2006.
74
MIT 2.094 17. Incompressible fluid flow and heat transfer, cont’d
Use (17.26) in the momentum and energy equations and use force equilibrium and compatibility at
the FSI boundary to set up the governing F.E. equations.
References
[1] K.J. Bathe, H. Zhang and M.H. Wang. “Finite Element Analysis of Incompressible and Compressible
Fluid Flows with Free Surfaces and Structural Interactions.” Computers & Structures, 56:193–213,
1995.
[2] K.J. Bathe, H. Zhang and S. Ji. “Finite Element Analysis of Fluid Flows Fully Coupled with
Structural Interactions.” Computers & Structures, 72:1–16, 1999.
[3] K.J. Bathe and H. Zhang. “Finite Element Developments for General Fluid Flows with Structural
Interactions.” International Journal for Numerical Methods in Engineering, 60:213–232, 2004.
75
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
In structures, Reading:
Sec. 8.4
F (u, p) = R. (18.1)
In heat transfer,
F (θ) = Q (18.2)
In fluid flow,
F (v, p, θ) = R (18.3)
In structures/solids
� �� t (m) T t ˆ (m) 0 (m)
F = F (m) = 0 BL 0S d V (18.4)
0V (m)
m m
Elastic materials
76
MIT 2.094 18. Solution of F.E. equations
Material law
t
0 S11
˜
0t�11
=E (18.5)
In isotropic elasticity:
E (1 − ν)
Ẽ = , (ν = 0.3) (18.6)
(1 + ν) (1 − 2ν)
�� �2 � �� �2 �
0
t 1 �� t � 2 � 1 L + tu 1 t
u
0� = U − I ⇒ 0t�11 = −1 = 1+ 0 −1 (18.7)
2 0 2 0L 2 L
0
t ρ0
0 S11 = X tτ 0X T
tρ t 11 11 t 11
(18.8)
with
0
0 L 0 0
t X11 = 0L
, ρ L = tρ tL (18.9)
+ tu
t
� 0 �2 0
L L L
⇒ 0tS11 = 0L tL
t
τ11 = t tτ11 (18.10)
L
�� �2 �
0 t
L 1 u
∴ t tτ11 = Ẽ · 1+ 0 −1 (18.11)
L 2 L
�� �2 ��
t t
�
t ẼA t u u
⇒ τ11 A = P = 1+ 0 −1 1 +
0 (18.12)
2 L L
This is because of the material-law assumption (18.5) (okay for small strains . . . )
Hyperelasticity
t
0W = f (Green-Lagrange strains, material constants) (18.13)
� �
t 1 ∂0tW ∂0tW
0 Sij = + t (18.14)
2 ∂ 0t�ij ∂ 0�ji
� t
1 ∂ 0 Sij ∂ 0tSij
�
C
0 ijrs = + (18.15)
2 ∂ 0t�rs ∂ 0t�sr
77
MIT 2.094 18. Solution of F.E. equations
Plasticity
• yield criterion
• flow rule
• hardening rule
� t
t t−Δt
τ = τ + dτ (18.16)
t−Δt
where t+Δt
U (i−1) is the value we just calculated and an approximation to U ∗ .
Assume t+ΔtR is independent of the displacements.
� � ∂ t+ΔtF ��
t+Δt t+Δt (i−1)
0= R− F − � · ΔU (i) (18.19)
∂U � t+ΔtU (i−1)
We obtain
t+Δt
K (i−1) ΔU (i) = t+Δt
R − t+ΔtF (i−1) (18.20)
∂ t+ΔtF ��
� � ��
t+Δt (i−1) ∂F ��
K = = (18.21)
∂U � t+ΔtU (i−1) ∂U � t+ΔtU (i−1)
Physically
� �
t+Δt (i−1)
Δ F1
t+Δt (i−1)
K11 = (18.22)
Δu
78
MIT 2.094 18. Solution of F.E. equations
t
i = 1; K Δu(1) = t+Δt
R − tF (18.23)
t+Δt (1) (2) t+Δt t+Δt (1)
i = 2; K Δu = R− F (18.24)
Convergence Use
But, if incremental displacements are small in every iteration, need to also use
t
�1 (18.28)
Li
79
MIT 2.094 18. Solution of F.E. equations
Beam
t 1
e.g. L = 100
(4-node el.)
(9-node el.)
→ We do not have a shear (better)
→ But, still for thin structures, it has problems
like ill-conditioning.
⇒ We need to use beam elements. For curved structures also spurious membrane strain can be
present.
80
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
t t 1 1
Beam analysis, L � 1 (e.g. L = 100 , 1000 , · · · ) Reading:
Sec. 5.4,
6.5
(plane stress)
� L �
2 0
J= (19.1)
0 2t
1
h2 = (1 − r) (1 + s) (19.2)
4
1
h3 = (1 − r) (1 − s) (19.3)
4
− 41 (1 + s) L2 − 14 (1 − s) L2
⎡ ⎤
0 0
B∗ = ⎢
⎢
⎥ etc
⎥ (19.7)
1
⎢
⎢ 0 4 (1 − r) 2t 0 − 14 (1 − r) 2t ⎥
⎥
⎣ ⎦
1
4 (1 − r) 2t − 14 (1 + s) L2 − 14 (1 − r) 2t − 14 (1 − s) L2
81
MIT 2.094 19. Slender structures
u2 v2 θ2
∂u
⎛ ⎞
− L1 t
⎡ ⎤
0 ∂x
2L s ⎜ ⎟
0
⎢ ⎥ ⎜ ⎟
Bbeam = ⎢
⎢ ···
⎥ ⎜
⎥∼⎜
⎟ (19.8)
0
� 0
� 0
� ∂v
� ⎟
∂y
⎢ ⎥ ⎜ ⎟
⎣ ⎦ ⎜ � ⎟
− L1 − 21 (1 − r)
⎝ ⎠
0 ∂u ∂v
∂y + ∂x
1
v(r) = (1 − r)v2
(19.9)
2
1 st
v(−1) = v2 (19.11)
st
u(−1) = − θ2 + u2 (19.12)
2
Kinematics is
1
u(r) = (1 − r)u2 (19.13)
2
results into �xx
∂u 2 1
→ �xx = · =− (19.14)
∂r L L
st
u(r, s) = − (1 − r)θ2 (19.15)
4
results into �xx , γxy
st
→ �xx = (19.16)
2L
∂u 2 1
γxy = · = − (1 − r) (19.17)
∂s t 2
1
v(r) = (1 − r)v2 (19.18)
2
results into γxy
1
→ γxy = − (19.19)
L
82
MIT 2.094 19. Slender structures
1 1
− v2 − (1 − r)θ2 = 0 (19.20)
L 2
for all r! ⇒ Impossible (except for v2 = θ2 = 0) ⇒ So, the element has a spurious shear strain!
dw
γ= −β (19.21)
dx
� �
1 3
I= bt (19.22)
12
As = kA = kbt (19.24)
To calculate k Reading:
� �2 p. 400
� �
1 2 1 V
(τa ) dA = dAs (19.25)
A 2G AS 2G A s
83
MIT 2.094 19. Slender structures
Now interpolate
w(r) = h1 w1 + h2 w2 (19.28)
β(r) = h1 θ1 + h2 θ2 (19.29)
1+r
w= w1 (19.30)
2
1+r
β= θ1 (19.31)
2
Shearing strain
w1 1+r
γ= − θ1 (19.32)
L 2
Shear strain is not zero all along the beam. But, at r = 0, we can have the shear strain = 0.
w1 θ1
− can be zero (19.33)
L 2
Namely,
w1 θ1 2
− = 0 for θ1 = w1 (19.34)
L 2 L
84
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
The fiber moves up and rotates and its length does not change.
Two-node element:
Three-node element:
85
MIT 2.094 20. Beams, plates, and shells
dw
= J −1 Hw,r û (20.8)
dx � �� �
Bw
dβ
= J −1 Hβ,r u
ˆ (20.9)
dx � �� �
Bβ
Hence we obtain
� � 1 � 1 �
T
EI BβT Bβ det(J )dr + (Ak)G (Bw − Hβ ) (Bw − Hβ ) det(J )dr û
−1 −1
� 1
= HwT p det(J )dr (20.10)
−1
Kû = R (20.11)
K is a result of the term inside the bracket in (20.10) and R is a result of the right hand side.
For the 2-node element,
w1 = θ1 = 0 (20.12)
w2 , θ2 = ? (20.13)
w2 1+r
γ= − θ2 (20.14)
L 2
We cannot make γ equal to zero for every r (page 404, textbook). Because of this, we need to use
about 200 elements to get an error of 10%. (Not good!)
Recall almost or fully incompressible analysis: Principle of virtual displacements:
� �
�T � �
� C � dV + �v (κ�v )dV = R (20.15)
V V
u/p formulation
� �
T
�� C � �� dV − �v pdV = R (20.16)
V V
� �p �
p + �v dV = 0 (20.17)
V κ
But now we needed to select wisely the interpolations of u and p. We needed to satisfy the inf-sup
condition
�
qh � · vh dVol
inf sup Vol ≥β>0 (20.18)
���� ���� �qh ��vh �
qh ∈Qh vh ∈Vh
86
MIT 2.094 20. Beams, plates, and shells
4/1 element:
We can show mathematically that this element does not satisfy inf-sup condition. But, we can also
show it by giving an example of this element which violates the inf-sup condition.
v1 = �, v2 = 0 ⇒ � · vh for both elements is positive and the same. Now, if I choose pressures as
above
�
qh �vh dVol = 0, hence (20.18) is not satisfied! (20.19)
Vol
9/3 element
satisfies inf-sup
9/4-c
satisfies inf-sup
� L
γ AS γ − γ AS dx = 0
� �
(20.21)
0
where
dw
γ= − β, from displacement interpolation (20.22)
dx
87
MIT 2.094 20. Beams, plates, and shells
� L � � � L
dw
−β �γ AS
�dx = γ AS �
γ AS
�dx (20.24)
0 dx 0
� +1 � �
1+r L
⇒− θ2 · dr + w2 = γ AS · L (20.25)
−1 2 2
w2 − L2 θ2
⇒ γ AS = (20.26)
L
γ AS (shear strain) is equal to the displacement-based shear strain at the middle of the beam.
L
γ AS = 0 hence w2 = θ2 (20.27)
2
� L
�
β β � dx = M β � x=L
�
⇒ EI (20.28)
0
�� � �
2
1
⇒ EI · L θ2 = M (20.29)
L
ML M L2
⇒ θ2 = , w2 = (20.30)
EI 2EI
(exact solutions)
88
MIT 2.094 20. Beams, plates, and shells
Plates Reading:
Fig. 5.25,
p. 421
⎧
⎨ w = w(x, y) is the transverse displacement of the mid-surface
⎪
v = −zβy (x, y) (20.31)
⎪
u = −zβx (x, y)
⎩
For any particle in the plate with coordinates (x, y, z), the expressions in (20.31) hold!
We use
q
�
w= hi wi (20.32)
i=1
q
�
βx = − hi θyi (20.33)
i=1
�q
βy = + hi θxi (20.34)
i=1
where q equals the number of nodes. Then the element locks in the same way as the displacement-based
beam element.
89
2.094 — Finite Element Analysis of Solids and Fluids Fall ‘08
A = area of mid-surface
p = load per unit area on mid-surface
w = w(x, y) (21.1)
∂u ∂βx
�xx = = −z (21.5)
∂x ∂x
∂v ∂βy
�yy = = −z (21.6)
∂y ∂y
� �
∂u ∂v ∂βx ∂βy
γxy = + = −z + (21.7)
∂y ∂x ∂y ∂x
∂βx
⎛ ⎞
⎛ ⎞ ∂x
�xx
⎜ ⎟
⎜ ⎟
∂βy
⎝ �yy ⎠ = −z ⎜
⎜ ∂y
⎟
⎟ (21.8)
γxy
⎜ ⎟
⎝ ⎠
∂βx ∂βy
∂y + ∂x
� �� �
κ
90
MIT 2.094 21. Plates and shells
∂w ∂u ∂w
γxz = + = − βx (21.9)
∂x ∂z ∂x
∂w ∂v ∂w
γyz = + = − βy (21.10)
∂y ∂z ∂y
⎛ ⎞ ⎡ ⎤⎛ ⎞
τxx 1 ν 0 �xx
E ⎣ ν ⎦ ⎝ �yy ⎠ = C · �
⎝ τyy ⎠ = 1 0 (21.11)
1 − ν2 1−ν
τxy 0 0 2 γxy
(plane stress)
� � � �
τxz E γxz
= = Gγ (21.12)
τyz 2(1 + ν) γyz
⎡ ⎤⎛ ⎞
� � + 2t 1 ν 0 �xx
� � E ⎣
�xx �yy γ xy ν 1 0 ⎦ ⎝ �yy ⎠dzdA+
A − 2t 1 − ν2 1−ν
0 0 2 γxy (21.13)
� � + 2t � �� � �
� � 1 0 γxz
k γ xz γ yz G dzdA = wpdA
A − 2t 0 1 γyz A
⎛ ⎞
w1 ⎛ ⎞
u1
⎜ θx1 ⎟
⎜ v ⎟
⇒ Kb ⎜ θy1 ⎟ , also Kpl. str. ⎝ 1 ⎠ (21.14)
⎜ ⎟
⎝
.
⎠ ...
..
91
MIT 2.094 21. Plates and shells
�
u= hi u i (21.16)
�
v= hi vi (21.17)
�
w= hi wi (21.18)
�
βx = − hi θyi (21.19)
�
βy = hi θxi (21.20)
From (21.13)
⎡ ⎤
� 3 1 ν 0 � � �
Et 1 0
κT ⎣ ν 1 0 ⎦ κdA + γ T Gt · k γdA (21.21)
A 12 (1 − ν 2 ) 1−ν A 0 1
0 0 2
This displacement-based element “locks in shear”. We need to change the transverse shear interpola
tions.
1 A 1 C
γyz = (1 − r)γyz + (1 + r)γyz (21.22)
2 2
where
� ��
A ∂w �
γyz = − βy �� (21.23)
∂y evaluated at A
92
MIT 2.094 21. Plates and shells
Aside: Why not just neglect transverse shears, as in Kirchhoff plate theory?
∂w
• If we do, γxz = βx = 0 ⇒ βx = ∂w
∂x − ∂x
� 2 �
• Therefore we have ∂∂xw2 , · · · in strains, so we need continuity also for
� ∂w
�
∂x
· · ·
•
1 1
r(1 + r)w1 − r(1 − r)w2 + 1 − r2 w3
� �
w=
2 2
w2 and w3 never affect w1 (∴ w|r=1 = w1 ).
But,
∂w 1 1
= (1 + 2r)w1 − (1 − 2r)w2 − 2rw3
∂r 2 2
�
w2 and w3 affect ∂w
∂r 1 . ⇒ This results in difficulties to develop a good
�
element based on Kirchhoff theory.
With Reissner-Mindlin theory, we independently interpolate rotations such that this problem does
not arise.
For flat structures, we can superimpose the plate bending and plane stress element stiffness. For
shells, curved structures, we need to develop/use curved elements, see references.
References
[1] E. Dvorkin and K.J. Bathe. “A Continuum Mechanics Based Four-Node Shell Element for General
Nonlinear Analysis.” Engineering Computations, 1:77–88, 1984.
[2] K.J. Bathe and E. Dvorkin. “A Four-Node Plate Bending Element Based on Mindlin/Reissner Plate
Theory and a Mixed Interpolation.” International Journal for Numerical Methods in Engineering,
21:367–383, 1985.
[3] K.J. Bathe, A. Iosilevich and D. Chapelle. “An Evaluation of the MITC Shell Elements.” Computers
& Structures, 75:1–30, 2000.
[4] D. Chapelle and K.J. Bathe. The Finite Element Analysis of Shells – Fundamentals. Springer,
2003.
93