Biomedical Engineering Reference
In-Depth Information
Stationarity of
L
in relation to the displacement field u gives
∂L
u
σ ( u )
ε
σ ( u )
ε
=
ε ( u )
· ε ( ξ )
+
ε ( ξ ) · ε ( λ )d
, ξ
d
b
b
b
b
i σ ( u )
+
ε ( ξ ) · ε ( λ )
d
i
ε
σ ( u )
ε
+
ε ( ξ ) · ε ( λ )
d
=
0
ξ V
(2.16)
s
s
where the symmetry of the material tensor σ ( u )
ε
= C ρ in
b has been used. Since
for both
h the tangent matrix is the linear elasticity tensor, the condition
can be rewritten as
b and
σ ( u )
ε
T
b C ρ (
ρ
) ε ( λ ) · ε ( ξ )
+
ε ( λ ) · ε ( ξ )
d
d
b
i
i
s
b C ρ (
+
s C
ε ( λ ) · ε ( ξ )
d
=−
ρ
) ε ( u )
· ε ( ξ )
d
ξ V
(2.17)
s
b
This equation is called the adjoint problem, and it provides the adjoint solution λ .
Note that, in contrast to the state Eq. (2.18), this problem is linear in λ , allowing a
quite simple calculation.
Finally, the Lagrangian variation in relation to the design variable
ρ
takes the
form
∂L
∂ρ
C ρ
∂ρ
C ρ
∂ρ
1
2
,
β
=
ε ( u )
· ε ( u )d
+
ε ( u )
· ε ( λ )
d
b
b
b
b
σ ( u )
∂ρ
+
· ε ( λ )
d
+ α
d
=
0
(2.18)
i
b
Using conventional finite elements, Eqs (2.14) and (2.17) are easily discretized
leading to the following algebraic equations:
F int = K b U + F i ( U )
+ K s U = F ext
(2.19)
K b
+ K s
+ K i ( U ) T
= F adj ( U )
(2.20)
where K b , K s ,and F ext denote the conventional stiffness matrices and external
forces, F i ( U ) denotes the nonlinear internal forces at the interface, K i ( U ) denotes
the tangent matrix at the equilibrium configuration U ,and F adj ( U ) denotes the
adjoint forces from Eq. (2.17). Given N and B , the arrays of shape functions and
their derivatives, respectively, these matrices are computed as
B T
C ρ B d
s B T
s
K b
=
K s
=
C
B d
(2.21)
b
i B T σ ( U )
B T
F i ( U )
=
σ
( U )d
K i
=
B d
(2.22)
ε
b
B T
C ρ ( ρ ) ε ( u )d b
F adj ( U ) =−
(2.23)
b
Search WWH ::




Custom Search