COMPUTATIONAL MECHANICS

New Trends and Applications

S. Idelsohn, E. O nate and E. Dvorkin (Eds.)
c CIMNE, Barcelona, Spain 1998
3D SOLID INCOMPRESSIBLE VISCOELASTIC FINITE
ELEMENT IN LARGE STRAINS FOR THE CORNEA.
REFRACTIVE SURGERY APPLICATION
Fabio A. Guarnieri

and Alberto Cardona

Centro Internacional de Metodos Computacionales en Ingeniera (CIMEC)
INTEC - Universidad Nacional del Litoral / Conicet
G uemes 3450 - 3000 Santa Fe - Argentina
e-mail: aguarni@intec.unl.edu.ar
Key Words: Computational Biomechanics, Nonlinear Viscoelasticity, Refractive Surgery,
Cornea.
Abstract. Several models
13
have been developed in order to reproduce the corneal behav-
ior in ophthalmological procedures as tonometry, radial keratotomy and photokeratectomy
with Excimer Laser. It has been found that the viscoelastic eect of a biological soft tis-
sue, as the cornea, is negligible in tonometry.
3
Nevertheless, clinical studies on humans
showed refractive changes with time in radial keratotomy. Wound healing is responsible
for the long time eect (measured in years). On the other hand, the short-time eect
(hours or days) has not been claried.
In this work a 3D viscoelastic nite element model is developed taking into account
incompressibility and large strains. An internal variable is introduced by means of a
multiplicative decomposition of the deformation gradient. The nal goal of the study is to
determine the importance of the viscoelastic eect in the Refractive Surgery.

Fellow from Conicet

Professor, UNL, and researcher from Conicet.

2
F.A. Guarnieri and A. Cardona
1 INTRODUCTION
In order to simulate the viscoelastic eect in the cornea after a refractive procedure, as
the radial keratotomy, we developed a three-dimensional incompressible viscoelastic nite
element in large strains. Bryant et al.
4
suggested that material nonlinearity is important
in modeling the Refractive Surgery. It is also important to consider a 3-D model in order
to simulate the complex geometry of the cornea with arcuate incisions. To obtain the 3-D
geometry we employed a computer-assisted videokeratographer, which optically scans the
corneal topography and produces a three-dimensional map of the corneal surface. We
used the output of this device to capture in detail the corneal shape and generate the
nite element mesh. Therefore the biomechanical models of Refractive Surgery may be
customized to each patient.
In a previous work
15
we developed a nonlinear geometrical elastic model of the human
cornea. We simulate dierent types of incisions in a static case.
Figure 1: Computer Simulation of Refractive Surgery by using Finite Elements. Module of displacements
is shown.
To develop the nonlinear viscoelastic model, we follow Le Tallecs approach
5, 6
where a
dierential form is chosen which introduces an internal variable through a multiplicative
decomposition of the deformation gradient. This model is thermodinamically consistent,
preserves incompressibility in the viscoelastic response, and is easy to solve as a com-
pressible hyperelastic model. This approach can be extended in a simple way to an
incompressible hyperelastic model in the equilibrium.
3
F.A. Guarnieri and A. Cardona
The viscoelastic incompressibility constraints the admissible solution of the internal
variable (we have ve unknowns instead of six in the deformation tensor). The nal
problem expressed in terms of displacements- internal variable is reduced to an standard
problem in displacements by eliminating the internal variable.
2 CORNEAL RHEOLOGY
The cornea is a soft tissue composite layered material made up of collagen brils and
ground substance. X-ray diraction measurements made on wet rat tail tendon (which
is composed primarily of collagen type I and ground substance) showed that 55% of the
volume of the ber is occupied by collagen molecules and the remaining 45% by water.
Such information is available for enucleated human corneal stroma, but not for stroma in
its normal hydrated state.
2
It is assumed that collagen bers Poissons ratio is 0.40 and ground substances Pois-
sons ratio is 0.49.
2
This result could change in the case of incisions made in the cornea
where the material could swell.
Hanna et al.
2
used a nite element model to simulate the corneal lamella. They cal-
culate the macroscopic constitutive properties from the experimental microscopic ones.
They found incompressibility in the Poissons ratio (0.47) in the plane perpendicular to
the cornea, and compressibility in the plane of the cornea (0.26). They found elastic
modulus values of 11 MPa and 39 KPa for corneal plane and radial plane, respectively.
Kobayashi et al.
3
determined the viscoelastic response of intact cornea subjected to
physiological intraocular pressure, measured from local deformations. The tests results of
ve paired eyes showed that in prepressurized enucleated eyes, the viscoelastic response
was insignicant while signicant viscoelastic response existed in the nonpressurized eyes.
Seiler et al.
11
studied the viscoelastic properties of human corneal strips with and with-
out Bowmans layer by relaxation measurements to determine the relative contribution
of Bowmans layer to the biomechanical properties of the cornea. At a strain of 2%, the
stress was measured to be (5.062.01)KPa and (4.721.30)KPa with and without Bow-
mans layer, respectively. They found fast relaxation time of 10.58s and slow relaxation
time of 269s for intact cornea.
Soergel et al.
12
obtained the shear compliance spectra by applying dynamic mechan-
ical shear compliance spectroscopy to dierent corneal tissues as a function of corneal
hydration and temperature in the frequency range from 0.1 mHz to 100 Hz.
In this work we will try to simulate the viscoelastic behavior of the cornea described
by these authors. We will also simulate the viscoelastic eect in the Refractive Surgery,
in particular with arcuate incisions for the correction of astigmatism.
4
F.A. Guarnieri and A. Cardona
2.1 Viscoelastic constitutive law
2.1.1 Small Strain
When a viscoelastic material is subjected to a step loading, there exists both an instan-
taneous and a long term equilibrium response. The standard linear solid is composed of
combinations of linear springs with spring constants E

, E

and dashpots with coecient

of viscosity

. It is able to model the two responses in a simple way.

E
1
E

q
1
q

Figure 2: Rheological Model.

From equilibrium we have
= E

+
m

=1
q

where is a linear strain and q

is the stress in the -dashpot. If

q

= E

) =

by dierentiation of q

we obtain the evolution equations for the internal stresses

.
q

+
q

=
d
dt
(E

)
5
F.A. Guarnieri and A. Cardona
where

is the relaxation time of the branch . The generalization to three-dimensional

non-linear elasticity is obtained by dening a stored energy function
=

() +
m

=1

(,

) =
1
2
E

2
+
m

=1
1
2
E

)
2
Thus the non-linear stress and the rate equations are:
=
d
d

() +
m

=1
q

.
q

+
q

=
d
dt
[
d
d

()]
where
q

|
t=0
= q

0
= E

0
q

|
t
= 0
2.1.2 Finite Strains. Generalized Linear Model.
We generalize the above simple model to three-dimensional situations involving isochoric
large deformations. In the nite strains case, the right Cauchy-Green tensors C, C
e
, and
C
v
measure the total deformation, the elastic part and the viscous part of the viscous
branch, respectively. Variable C
v
is an internal viscoelastic variable. In analogy with the
small strains case we assume a free energy potential of the form:
13
W(C, C
v
) = W
0
(C) + W
e
(C
e
)
where W
0
measures the stored energy of the elastic branch (long term behavior) and W
e
measures the stored energy of the viscous branch. The latter disappears progressively
during relaxation. A dissipation function is postulated in the dashpot such that it
must satisfy the Clausius-Duhem inequality (2
nd
law of thermodynamics),
= (
.
Cv
) :
.
Cv
0
If we choose = 4D
v
: D
v
, we get
(
.
Cv
) =
.
C
1
v
where is a symmetric denite-positive tensorial viscosity and D
v
is the viscous defor-
mation rate tensor.
6
F.A. Guarnieri and A. Cardona
2.1.3 Multiplicative decomposition of the deformation
Instead of an additive decomposition of the deformation, a multiplicative decomposition
is more appropiate in large strains. In crystals we were allowed to formally decompose
the deformation at microscopic level into
=
e

v
Therefore the deformation gradient and right Cauchy-Green strain tensor may be ex-
pressed in the form
F = F
e
F
v
C = F
T
v
C
e
F
v
J = J
e
J
v
where J = det(F) > 0. This idea is taken from plasticity where F = F
e
F
p
10
.
Then, by applying the rst principle of thermodynamics in an isothermic process, we
have
Work Free energy = Dissipated heat
In a reference conguration we have
(
.
Cv
) : dC
v
=
1
2
S : dC dW = (
1
2
S
W
C
) : dC
W
C
v
: dC
v
where S is the symmetric Piola-Kirchho stress tensor. Since the state variables C, C
v
are supposed to be independent, we obtain
S = 2
W(C, C
v
)
C
(
.
Cv
) =
W(C, C
v
)
C
v
det(C
v
) = 1
where the rst equation is a standard hyperelastic constitutive law with C
v
as a consti-
tutive parameter; the second equation is a rst order dierential equation in time where
the variable C
v
introduces the time dependence in the model. The latter equation is
the incompressibility relation of the viscous part ( experimental data suggest that time-
dependent change in the free energy is only due to isochoric deformations).
2.2 Hyperelastic Material Model
We choose a simple hyperelastic model as:
W(C, C
v
) = W
0
(E) + W
e
(C
e
)
W
0
(E) =
1
2
(tr E)
2
+ E : E
7
F.A. Guarnieri and A. Cardona
W
e
(C
e
) = W
1
(tr C
e
) + W
2
(det C
e
)
= W
1
(A : C) + W
2
(det C)
= k
1
(A : C) k
1
ln(det C)
since C
e
=
_
C
1
v
C
_
C
1
v
=

AC

A, where A = C
1
v
is the internal variable, E is
the Green strain, and are the Lame constants, k
1
is a constant, and W
0
and W
e
is the equilibrium and non-equilibrium free energy function, repectively. Then, the 2
nd
Piola-Kirchho stress may be written:,
S = 2E + tr E1 + 2
C
W
e
(C
e
)
S=2E+trE1+2k
1
A-2k
1
C
1
and choosing an adequate dissipation form ((
.
Cv
) =
.
C
1
v
) we obtain the dissipative
constitutive law, and multiplying by C
v
we have

.
C
1
v
+
W(C, C
v
)
C
v
=

.
A
1

W(C, A)
A
=

.
A
1
k
1
C = 0
det(A) = 1
2.3 Viscous Incompressiblity Constraint
Instead of using the mixed form of the incompressibility constraint det(C
v
) = 1, we follow
the proposal of LeTallec et al.
5, 6
who dened a constrained space of the internal variable
such that
C
v
A = {A L
2
(), det(A) = 1}
The incompressibility condition can be rewritten by developing the expression of the
determinant in the third line:
a
31
cof
31
A + a
32
cof
32
A + a
33
cof
33
A = 1
Since C
v
is positive denite, the diagonal cofactors are dierent from zero. Then we can
write
a
33
=
1 a
31
cof
31
Aa
32
cof
32
A
cof
33
A
8
F.A. Guarnieri and A. Cardona
3 APPROXIMATION IN TIME
The backward Euler scheme is chosen since, although not second order accurate as the
midpoint rule, it requires less computer memory and has very nice sti stability and long
term convergence properties.
5
Let t > 0 be the time interval; for each iteration n 0, we have to solve
_

h
F
n+1
h
_
2
W
C
(C
n+1
h
, A
n+1
h
)
_
:
h
dV =
_

(t
p

h
) dV

(A
n+1
)
1
(A
n
)
1
t

W(C
n+1
, A
n+1
)
A
= 0 (1)
with A(, t
0
) = A
0
= given value.
After discretization in time, we eliminate the viscoelastic variable. At each time step,
we calculate A
n+1
by solving the dissipation equation (1).
4 MATRIX RELATIONS
4.1 Elimination of the Internal Variable
In the dissipation equation we have ve unknowns, the components of symmetric tensor
C
v
. The sixth one, a
33
is given by the above equation. We have ve equations since the
six components of the dissipation equation are not independent because of the incom-
pressibility condition. This is performed at each integration point. If
A(Y ) =
_
_
Y
1
Y
3
Y
5
Y
3
Y
2
Y
4
Y
5
Y
4
Y
6
_
_
with
a
33
= Y
6
=
1 + Y
1
Y
2
4
2Y
3
Y
5
Y
4
+ Y
2
Y
2
5
Y
1
Y
2
Y
2
3
Finally, we have to solve the A = G(C) by a Newton iteration method
Y = arg min
ZR
5
_
W(C
n+1
, A(Z)) +
(A
n
)
1
t
: A(Z)
_
where the rst and second derivatives are
F
t
(Y ) =
_
W(C
n+1
, A(Y ))
A
+
(A
n
)
1
t
_
:
A
Y
= 0
=
_
k
1
C +
(A
n
)
1
t
_
:
A
Y
9
F.A. Guarnieri and A. Cardona
K =
F
t
Z
=

2
W
AZ
:
A
Y
+
_
W(C
n+1
, A(Y ))
A
+
(A
n
)
1
t
_
:

2
A
Y Z
=
_
k
1
C +
(A
n
)
1
t
_
:

2
A
Y Z
where K is the symmetric linearized operator. This is equivalent to solve the dissipation
equation (1).
4.2 Tangent Viscoelastic Matrix
After A is computed we obtain the Tangent Viscoelastic Matrix by linearization of the
stress relation using
C =
S
E
=

2
W
0
E
2
+

2
W
e
E
2
= 2I+1 1+4
_

2
A:C
W
1
AA +
A:C
W
1
A
C
_
+
4
_

2
det C
W
2
I
2
3
C
1
C
1
+
det C
W
2
_
I
3
C
C
1
+ I
3
C
1
C
__
= 2I+1 1+4k
1
A
C
4k
1
I
C
1
where I
IJKL
=
C
IJ
C
KL
=
1
2
(
IK

JL
+
IL

JK
) and I
C
1
IJKL
=
C
1
IJ
C
KL
=
1
2
(C
1
IK
C
1
JL
+
C
1
IL
C
1
JK
)
To obtain
A
C
we use
A
E
= 2
G
C
(C)
and
Y
n+1
Y
n
= K
1
(Y
n
)F
t
(Y
n
)
Y
n+1
C
= K
1
(Y
n
)
F
t
(Y
n
)
C
4k
1
A
C
= 4k
1
A
Y
Y
C
= BK
1
B
t
Finally,
C=2I+1 14k
1
I
C
1BK
1
B
t
5 FINITE ELEMENT IMPLEMENTATION
We will use the conventional Galerkin approximation for the numerical simulation of the
constitutive model with moderate changes in material volume.
10
F.A. Guarnieri and A. Cardona
5.1 Equilibrium equations. Variational formulation
We consider the functional dened by the expression
() :=
_
B
W(X, C()) dV
_
B
(t
p
) dA
We obtain the weak form of the equilibrium equations in a xed reference conguration
B, through the rst variation of , by using the Gateaux derivative at

in the direction

0
. By neglecting the body forces, and considering t
p
pressure external forces in
t
B, we
have
D

0
=
_
B

C
W(X, C()) :
F
C
d
d

=0
F() dV
_
tB
(t
p

0
) dA
=
_
B
FS : GRAD
0
dV
_
tB
(t
p

0
) dA
Here,
0
are admissible displacement variations satisfying essential boundary conditions
W =
_

0
: B R
3
|
0
=
0
|
tB
= 0
_
5.2 Incremental Total Lagrangian Formulation
Although we use for simplicity lineal material behavior in the equations, we are able to
use several nonlinear hyperelastic models. Then the equation of motion is non-linear due
to the material and geometric non-linearities presented in the problems, and cannot be
solved directly. An incremental iterative technique can be used, wherein an approximate
solution is obtained by linearizing the non-linear equations by incremental decomposition.
Let the residual be
G(,
0
) :=
_
B
FS : GRAD(
0
) dV G
ext
(
0
) = 0
Assuming that the solution at a conguration
0
is known, we seek the solution at some
increment
0
+ u.
The linearization of G is obtained through the second variation of
L

0
G() := G(
0
) + DG(
0
) u
By using the Gateaux derivative at

in the direction u and neglecting the contribution

to the external forces
DG() u =
d
d

=0
_
B
FS : GRAD(
0
) dV
11
F.A. Guarnieri and A. Cardona
=
_
B
GRAD(u) S : GRAD(
0
) dV
+
_
B
(F
T
GRAD(u))
T
sym
: C : (F
T
GRAD(
0
))
sym
dV
5.3 Discretization
Let W
h
W be an isoparametric conforming nite element approximation of the space
of admissible variations dened in the standard way as
W
h
= {
h
W|
h
e
=
nen

A=1
N
A
()
A
,
A
R
3
}
where := {(1, 1) (1, 1) (1, 1)} and N
A
: R the isoparametric shape
function, and
X
h
e
=
nen

A=1
N
A
()X
A
is the isoparametric map.
Within the isoparametric context we set

h
e
= X
h
e
+U
h
e
=
nen

A=1
N
A
()(X
A
+U
A
)
F
h
e
=
nen

A=1
(X
A
+U
A
) GRAD(N
A
)
(F
T
GRAD())
sym
=
nen

A=1
B
A
L

A
GRAD() =
nen

A=1
B
A
NL

A
Introducing these terms in the variational formulation and through the linearization of G
about
0
, on the element level, we have a system of linear algebraic equations of the form
(K
M
(
0
) +K
G
(
0
))u =F
ext
(
0
) F
int
(
0
)
where
K
M
=
_
B
T
L
: C : B
L
d
12
F.A. Guarnieri and A. Cardona
K
G
=
_
B
T
NL
S : B
NL
d
F
ext
=
_
N
T
t
p
d
F
int
=
_
B
T
L
S d
5.4 Algorithm
1. For t=0 : u = Y = 1; Solve Initial Elastic problem by Newton Method.
2. For t=0
+
: u = u
0
+, Y = 1;
3. compute Y by a Newton Method such that: A
n+1
A
n
= K
1
(Y
n
)F
t
(Y
n
);
4. updates S by adding
We
E
5. updates local stiness matrix C by adding

2
We
E
2
to

2
W
0
E
2
6. test r, u;stop on convergence ;
This code was written in C++, by using the Object Oriented Library of Finite Elements
of the Interactive Program Oofelie.
16
6 RESULTS
As a rst test of this viscoelastic approach, we study the deformation of a rectangular
strip in pure traction with
W
e
(C
e
) = W
1
(tr C
e
) + W
2
(det C
e
)
= k
1
tr C
e
k
1
ln(det C
e
)
We observed a time history of the displacements observed at the free edge of the strip
(Figure 3). We observe a transition in time from a sti instantaneous response to a softer
long term behavior. The adimensional numerical values were k
1
= 1e5, = 0.0, E =
1e6, f = 100, = 5e4.
The next calculation studies the history of the deformation of a human eye after surgery.
The nite element mesh consists of 2 layers of 23 rings of 20 elements. The boundary is
xed. The posterior layer is loaded by a constant pressure equivalent to the intraocular
pressure. The refractive surgery consists of 2 arcuate incisions of 60 degrees of arc in a
radious of 3 mm from the apex (Figure ??). In the Figure 5, we show the time history of
the displacement of the incision gape, and in the Figure 6, the time history of the edge
13
F.A. Guarnieri and A. Cardona
1 2 3 4 5 6 7 8 9 10
6
6.5
7
7.5
8
8.5
x 10
-6
Figure 3: Time history of the displacements at the free edge of a strip.
Figure 4: Corneal mesh with 2 arcuate incisions.
displacement of the incision. This is correlated with a greater eect of the surgery in the
time. In Figure 7, we show the time history of the apex z-displacement.
14
F.A. Guarnieri and A. Cardona
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
0 2 4 6 8 10 12
Figure 5: Time history of the displacement of the incision gape.
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
0 2 4 6 8 10 12
Figure 6: Time history of the edge displacement of the incision.
We observed that in any case the cornea increases the depth chamber, leading to a
myopic shift.
The keratometric curvature (max. curvature vs min. curvature at 3 mm from the apex)
changed with the time. The time history of attest axis (greater radious of curvature)
is showed in the Figure 8, and the time history of the steepest axis (lower radious of
curvature) in the Figure 9.
We observed a atenning of the overall radious of curvature with the time, as we expect
from clinical results.
15
F.A. Guarnieri and A. Cardona
0 2 4 6 8 10 12
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
Figure 7: Time history of the apex z-displacement.
0 2 4 6 8 10 12
730
740
750
760
770
780
790
800
810
Figure 8: Time history of the attest axis.
7 CONCLUSIONS
A viscoelastic nite element as proposed by Le Tallec has been implemented and applied
to the solution of large strains in the cornea after surgery.
16
F.A. Guarnieri and A. Cardona
0 2 4 6 8 10 12
660
680
700
720
740
760
780
800
Figure 9: Time history of the steepest axis.
The implementation was performed in an object oriented program for nite element
analysis. Tasks as introduction of new constitutive laws have been greatly facilitated in
this environment.
We observed the creep of the cornea in several ways, as the displacement of the incision
edge, the incision gape, the elevation of the apex, and the keratometric curvature.
The overall attening in the steady case is in agreement with the fact that, the deepest
eect of the surgery is the instataneous response after surgery, and after some time (days,
months...) the cornea tends to recover a little its old shape.
This is a preliminary result. More cases have to be simulated. An open question is
about the viscoelastic parameters in refractive surgery.
REFERENCES
[1] Buzard K.A., Introduction to Biomechanics of the Cornea, Refractive and Corneal
Surgery, 1992; 8, 2: 127.
[2] Hanna K. D., Jouve F. E., Waring G. O., Ciarlet P. G.,Computer Simulation of
Arcuate Keratotomy for Astigmatism, Refractive and Corneal Surgery, 1992; 8, 2:
152.
[3] Kobayashi A.S., Staberg L.G., Schlegel W.A., Viscoelastic Properties of Human
Cornea, Experimental Mechanics, 1973;13, 497:503.
[4] Bryant M. R., McDonnell P.J., Constitutive Laws for Biomechanical Modeling of
Refractive Surgery, J. of Biomechanical Engineering, 1996; 118, 473:481.
17
F.A. Guarnieri and A. Cardona
[5] Le Tallec P., Rahier C., Kaiss A., Three-dimensional incompressible viscoelasticity
in large strains: Formulation and numerical approximation, Comput. Methods Appl.
Mech. and Engrg., 1993; 109, 233:258.
[6] Le Tallec P., Rahier C., Numerical Models of Steady Rolling for Non-Linear Vis-
coelastic Structures in Finite Deformations, Internat. J. Numer. Methods Engrg.,
1994; 37, 1159:1186.
[7] Simo J.C., On a Fully Three-Dimensional nite-strain viscoelastic damage model:
Formulation and Computational aspects, Comput. Methods Appl. Mech. and Engrg.,
1987; 60, 153:173.
[8] Christensen, Theory of Viscoelasticity: An Introduction, (Academic Press, New York,
1971).
[9] Gurtin M.E., An Introduction to Continuum Mechanics, (Academic Press, New York,
1981).
[10] Lubliner J., Plasticity Theory, Macmillan Publishing Company, New York, 1990.
[11] Seiler T., Matallana M., Sendler S., Bende T., Does Bowmans Layer Determine the
Biomechanical Properties of the Cornea?, Refractive and Corneal Surgery, 1992; 8,
139: 142.
[12] Soergel F., Jean B., Seiler T., Bende T.,M ucke S., Pechhold W., Pels L., Dynamic
Mechanical Spectroscopy of the Cornea for Measurement of its Viscoelastic Properties
in Vitro, German J. Ophthalmol.,1995; 4, 151-156.
[13] Lubliner J., A Model of Rubber Viscoelasticity, Mechanics Research Communications,
1985;12, 93:99.
[14] Berardi G., Jaeger M., and Martin R., Carpentier C., Modelling of a thermo-
viscoelastic coupling for large deformations through nite element analysis, Int. J.
Heat Mass Transfer, 1996;39, 3911-3924.
[15] Guarnieri F. A., Cardona A., Computer Simulation of Refractive Surgery by Using
a Finite Element Model, World Congress on Medical Physics and Biomedical Engi-
neering, Rio de Janeiro, Brazil, vol.39a part 2 p. 647, August 1994.
[16] Cardona A., Klapka I., Geradin M., OOFELIE: An Object Oriented Finite Elements
Program Led by Interactive Execution, presented to: Computational Structures
Technology, 1994, Atenas, Grecia.
18