Professional Documents
Culture Documents
Chapter I
Chapter I
Chapter I
P. E. McHugh, D. O’Mahoney
1. Continuum Mechanics
Consider the stretching of the material fibre dx from the point of view of
change in magnitude. Consider, in particular, the difference of the square
of the fibre length:
dy − dx = (F ⋅ dx ) ⋅ (F ⋅ dx ) − dx ⋅ dx
2 2
( )
= dx ⋅ F T ⋅ F ⋅ dx − dx ⋅ dx (1.9)
= dx ⋅ (F )
⋅ F − I ⋅ dx
T
= 2dx ⋅ E ⋅ dx
which allows the Lagrangian Strain tensor, E, to be defined
E= 1
2 (F T
⋅F − I ) (1.10)
This is demonstartes a non-linear relationship between strain and the
displacement gradient. The Lagrangian strain rate is given by
E 2 (
& = 1 F& T ⋅ F + F T ⋅ F& ) (1.11)
4 Topics in Bio-Mechanical Engineering
Another very important strain rate measure, normally associated with the
~
current configuration, is the Rate of Deformation tensor, D , derived
from the spatial velocity gradient as
~ ~
~
~ ~
() (
D = sym L = 12 LT + L ) ~ ~T
(1.12)
Both E and D are symmetric tensors, i.e., E = E , and D = D . Under
T
Cauchy (true) Stress is given the symbol σ and describes the force per unit
area on the current configuration. It is a symmetric tensor and is related to
the traction t on a surface (internal or external) in the current configuration
and a unit normal vector to the surface n ~,
t = σ⋅n ~
(1.16)
ti = σ ij n~j
Other important stress tensors are the following:
Kirchhoff Stress (symmetric), τ = Jσ , where J = dV(x)/dV0(x), is the
Jacobian of the deformation, i.e., the ratio of local infinitesimal volumes in
the neighbourhood of a material point x, between the current configuration
(dV(x)) and the reference configuration (dV0(x))
Finite Elements in Biomedical Engineering 5
1.4 Equilibrium
The basic concept of static equilibrium is that the sum of external forces
and moments on a body is zero. It is useful to have point-wise version of
this concept, expressing equilibrium at (in the neighbourhood of) a
material point x. Such an expression can be written in terms of the
divergence of the stress, for the case of small deformations as
∂σ ij
= σ ij,j = 0
∂x j (1.17)
∇⋅σ = 0
For large deformation, we can say, in terms of quantities in the current
configuration,
∂σ ij
= σ ij,j = 0
∂y j (1.18)
∇⋅σ = 0
where σ is the Cauchy stress in this case and gradients are taken to be
spatial gradients. If the body is subject to a body force distribution, e.g., as
might come from gravity or a magnetic field, we can say
∂σ ij
+ fi = 0 (1.19)
∂y j
where f i is known as the body force density. For dynamic situations, an
inertia force density can be included
6 Topics in Bio-Mechanical Engineering
∂σ ij
+ f i − ρu&&i = 0 (1.20)
∂y j
Equilibrium at a point on the surface of body can be expressed in terms of
the relationship between stress and surface traction using equation (1.16).
For the purposes of dealing with boundary conditions (BCs) on the surface
of a body or region, the surface S can be divided into the displacement
oundary S1 and the traction boundary S2 as indicated in Fig. 2 below.
Problems in solid mechanics usually require that for a given body one can
determine the complete mechanical state for any point in time. This means
that (assuming small deformation terminology for convenience) we can
determine ε(x), σ(x), u(x) and t(x), plus rates of these quantities for rate
dependent or dynamic problems, plus other state variables (e.g., a strain
Finite Elements in Biomedical Engineering 7
hardening parameter for a plastic material) and their rates, for any time t.
To do this we are usually required to solve the following
Equilibrium equation (conditions on σ )
Constitutive equation ( σ − ε relationship)
Compatibility equation ( ε − u relationship)
Together, these form a set of partial differential equations (PDEs). We
must also consider the boundary conditions. Combining the PDE and BC,
we are presented with a Boundary Value Problem (BVP) to solve. This
concept readily extends to heat flow, diffusion problems, etc.
Here, σ and t are the real stress and surface traction, respectively,
~
and δD and δv are the virtual rate of deformation and surface velocity,
respectively. The fundamental point is that the relationship is true even
when the virtual quantities (deformation rate and velocity) are not due to
the real quantities (stress and traction). The virtual velocities (or
displacements) must satisfy the kinematic (velocity or displacement)
boundary conditions. Kinematic boundary conditions are also known as
essential boundary conditions. A commonly used form of the PVW is the
one used in small deformation analysis
∫ σ : δεdV = ∫ t ⋅ δudS
V S
(1.22)
8 Topics in Bio-Mechanical Engineering
2.1 Introduction
or, σ = Dε . Very often, plane stress and plane strain assumptions are
used to simplify problems from 3D to 2D. If we assume plane strains
conditions, that is ε 33 = 0 , then equation (2.4) simplifies to:
σ 11 1 − ν ν 0 ε11
σ = E ν ε
22 (1 + ν )(1 − 2ν ) 1 −ν 0 22 (2.5)
σ 12 0 0 1− 2ν
γ
2 12
10 Topics in Bio-Mechanical Engineering
So what is the strain εij? For the general 3D case the infinitesimal
strains are given by equation (1.14). It is instructive to look at the
individual terms in this general expression. Assume that at an arbitrary
position x1, a slice of material of length ∆x1 is deformed a very small
amount ∆u1, the strain component ε11 is given by:
∂u1
ε11 = (2.7)
∂x1
∂u2 ∂u3
Similarly, ε 22 = and ε 33 = , where u2 and u3 are the
∂x2 ∂x3
deformations in the x2 and x3 directions, respectively. The shear strains
∂u1 ∂u2 ∂u2 ∂u3
are given as follows γ 12 = + , γ 23 = + and
∂x2 ∂x1 ∂x3 ∂x2
∂u1 ∂u3
γ 13 = + . For the 2D plane problems we are directly interested
∂x3 ∂x1
in ε11 , ε 22 and γ 12 , and in order to find these strain components we
need to have u1(x1, x2) and u2(x1, x2).
x2
x1
Fig. 4. Prototype triangular element in 2D.
So what do the functions Ni look like? Well given that they are in fact
an “average” we must have for all points inside ∆abc that all Ni are
valued between zero and one. Also, the sum ∑Ni must be equal to one.
Finally, each Ni(x1k,x2k), where x1k and x2k are the two coordinates of
node k, is equal to one when i is equal k, and is zero otherwise, i.e. at
each of the other nodes.
A solution to this is to use ‘area co-ordinates’ first put forward by the
Polish mathematician Courant in 1943. If we assume we have a
triangular element as shown above, the weighting functions (or shape
functions) Ni can be defined as follows:
12 Topics in Bio-Mechanical Engineering
ε = 0 − 1 0 0 0 1 1b u
22 u (2.13)
γ 12 − 1 − 1 0 1 1 0 2b
u1c
u2c
This equation illustrates that the equations emanating from the finite
element formulation are most conveniently expressed in matrix-vector
form. The Bαβ coefficients, which are spatial gradients of the weighting
functions (equation 2.10), generate a matrix B (on the rhs of equation
2.13). The nodal displacement components can be organised into a vector
Finite Elements in Biomedical Engineering 13
ue (on the far rhs of equation 2.10). On this basis, equation (2.13) can be
written simply as
ε = Bu e (2.14)
As the strain energy per unit volume is Ψ = ½ ε T D ε , the strain
energy stored in the element, Φe, of volume Ve must be given
by Φ e = Ψ V e = ½ ε T D ε V e . This leads to the definition of the
element stiffness matrix, K, by substituting for the strain (equation 2.14)
Φ e = ½ ε T D ε V e = ½ ( Bu e ) T D ( Bu e )V e
(2.15)
T
= ½ u e B T DBu e V e
Comparing this with the expression for the strain energy of a spring,
involving the stiffness of the spring K and its displacement u, (equation
2.1) we can define K as
K = B T DB V e (2.16)
and hence Φ e = ½ u e T Ku e .
Bringing the spring analogy further, the displacement of the spring
can be determined from an inversion of Hooke’s equation (2.1),
u = F / K = FK −1 . Similarly, if a vector of nodal forces, Fe, is known,
and when K can be inverted (see section 2.3 below), the nodal
displacements can be determined through u e = K −1Fe . This is the
situation that normally pertains in solid mechanics problems to be solved
by the finite element method; one seeks the unknown displacements of a
body under known applied loads.
As regards the convenience of matrix notation for finite element
formulations, we note that equations (2.8) can be neatly expressed in the
following form using the shape function matrix N
u N 0 Nb 0 Nc 0
u = Nu e , u = 1 , N = a
N c
(2.17)
u2 0 Na 0 Nb 0
14 Topics in Bio-Mechanical Engineering
The volume is given by area (0.5) times a unit thickness (1). This gives:
182 78 − 130 − 52 − 52 26
78 182 − 26 − 52 − 52 − 130
− 130 − 26 130 0 0 26
K = (2.18)
− 52 − 52 0 52 52 0
− 52 − 52 0 52 52 0
26 − 130 26 0 0 130
The entries in K have the units of N/m. Some properties of K are
worth noting. The first is that K is symmetric (although this may not be
generally the case). To see why we can argue as follows:
K = B T DB
⇒ K T = [B T DB] T
⇒ K T = B T D T B TT
⇒ K T = B TDTB
Q D = DT ⇒ K = K T
The second is that the matrix is not of full rank, actually it has rank
N-3. In other words it is singular, and so does not have a regular inverse.
To have full rank, K in equation (2.18) must have six independent rows
(or columns). However, firstly note that row 4 (R4) and R5 are the same
so R4 - R5 = 0. Also, R1 + R3 + R5 = 0, and finally R2 + (R5 + R6) =
0. Performing these elementary row operations we get K’ as:
Finite Elements in Biomedical Engineering 15
0 0 0 0 0 0
0 0 0 0 0 0
− 130 − 26 130 0 0 26
K' =
0 0 0 0 0 0
− 52 − 52 0 52 52 0
− 26 − 130 26 0 0 130
The significance of this is that at least three displacement boundary
conditions must be specified in order to obtain a unique solution to
Hooke’s equation Fe = Kue. These three displacement boundary
conditions relate to the three rigid body modes of deformation in 2D.
These must be eliminated before the finite element solution for the
deformation of the body can be determined. Lastly we see from equation
(2.18) that all row and column sums are equal to zero. This is because K
is formed from a derivative, i.e. K= dFe/due. All differential operators
have this property.
x2
x1
The first problem is that area co-ordinates which we used for triangular
elements are not useful here. Instead we use the following weighting (or
shape) functions:
1
N a = ( 1 - x1 )( 1 - x2 )
4
1
N b = ( 1 + x1 )( 1 - x2 )
4 (2.19)
1
N c = ( 1 + x1 )( 1 + x2 )
4
1
N d = ( 1 - x1 )( 1 + x2 )
4
It can be easily shown that these Ni meet the same requirements as those
for the triangular element. Again we wish to find an expression for the
strain within the element in terms of the nodal displacements, e.g. we
wish to have
∑
ε11 ( x1 , x2 ) = B1i ( x1 , x2 )u1i (2.20)
Starting with the shape functions, as before, we can say
∑ ∑
u1 ( x1 , x2 ) = N i ( x1 , x2 ) u1i and u2 ( x1 , x2 ) = N i ( x1 , x2 ) u2i (2.21)
By substituting these definitions into the definition of strain we generate
a result similar to that given in equation (2.13) but where this time the B
matrix has the form:
−1+ x2 0 1− x2 0 1+ x2 0 −1− x2 0
1
B= 0 −1+ x1 0 −1− x1 0 1+ x1 0 1− x1
4
−1+ x1 −1+ x2 −1− x1 1− x2 1+ x1 1+ x2 1− x1 −1− x2
(2.22)
It is worth noting that the matrix B is not a constant. Instead it varies
according to the position within the element. Therefore, the strain varies
across the element in a linear fashion. So, the product BTDB gives a
matrix which varies across the element. So K cannot be found directly.
In order to find an overall value for K we assume that for a small
element of volume dV (= dx1 x dx2 x 1), at a position (x1,x2), the elemental
stiffness matrix is dK, so we have:
Finite Elements in Biomedical Engineering 17
d K = B T
DB dV
= ∫ (2.23)
T
K B DB dV
1 1
= ∫ ∫
T
K B DB dx 1 dx 2
−1 −1
In order to evaluate the last integral (equation 2.23) one could simply use
Maple or Mathematica to get:
113 33 − 74 −7 − 56 − 33 17 7
113 7 17 − 33 − 56 −7− 74
113 − 33 17 − 7 − 56 33
113 7 − 74 33 − 56
K= (2.24)
113 33 − 74 − 7
SYMM 113 7 17
113 − 33
113
However, this is obviously not practical when using a programming
language such as Fortran or Pascal. Instead Gaussian integration is
commonly used. In this approximation the integrand is evaluated at
special points on [−1,1]x[−1,1] called Gauss points. In this case the
integrand is simply a matrix, containing functions as entries, rather than
constants. On [−1,1] x [−1,1] the four Gauss points ξi(x1,x2) are at
[0.577, ± 0.577] and [−0.577, ± 0.577] . The integral is then evaluated
as a weighted sum:
4
(2.25)
K = ∑ H i B T ( ξ i ) DB ( ξ i )
i =1
In this case the weights Hi are all equal to 1, although this is not
generally the case. A more general treatment of this method can be found
in most numerical methods text books.
18 Topics in Bio-Mechanical Engineering
To see how the finite element method works in practice, let us look at the
following example. Consider the structure shown below, which is hinged
at a and is only allowed to move in the x1 direction at b. We wish to find
all remaining displacements and reaction forces.
2.7 Generalisations
Fig. 7. Stress-strain curve for an elastic-plastic material. Based on elastic unloading, the
strain can be decomposed into elastic and plastic parts.
that has the potential to deal with both linear and non-linear problems. A
very good reference on finite element formulations of the “conventional
form” is Zienkiewicz and Taylor (8). Consider a BVP, assuming small
deformation kinematics, posed in terms of the PVW, where t is the surface
traction vector and the δ refers to virtual quantities
∫ σ : δεdV = ∫ t ⋅ δudS
V S
(3.1)
Using the definition of the infinitesimal strain tensor and inserting the
finite element interpolation as illustrated in equations (2.14 and 2.17) we
can say
ε = 12 (∇u + ∇u T ) = B e u e and δε = B e δu e (3.4)
Here Be is the matrix of shape function gradients for element “e” in the
finite element mesh. For a material with a non-linear constitutive response
we can say
σ = σ (ε) → σ = σ (u e ) (3.5)
Substitution into the PVW and summation over all the elements “e” in a
finite element mesh, each with a volume Ve and a surface Se, yields
∑ ∫ δu
e Ve
T
e B Te σ (u e )dV = ∑ ∫ δu Te N Te tdS
e Se
(3.6)
If we drop the summation notation and the subscripts for convenience and
eliminate or “cancel” the arbitrary virtual displacements we get
∫B σ (u)dV = ∫ N T tdS
T
(3.7)
V S
Assume that we have solved for state at time t and we wish to update state
to t+∆t. The ut+∆t are considered as main solution variables and we must
solve the following non-linear set of equations for ut+∆t
( )
G u t + ∆t = 0 (3.9)
The Newton-Raphson (NR) method can be applied to determine the ut+∆t
by iteration. Let’s first look at a 1D analogy. Suppose we wish to solve f(x)
= 0 by NR. If we have a guess at the root xi, a better guess xi+1 given by the
NR formula
−1
df
xi+1 = xi − ⋅ f ( xi ) (3.10)
dx xi
The method is applied iteratively, i.e., xi+1 is substituted for xi and the NR
formula is reapplied. This process is continued until a convergence
criterion is satisfied
Finite Elements in Biomedical Engineering 25
u ti ++1∆t = u ti + ∆t −
(
∂G u ti + ∆t ) −1
t + ∆t
( )
G ui (3.12)
∂u
If we reorganise this in terms of the change in incremental displacement
we get
∂G u ti + ∆t
δu i +1 = u ti ++1∆t − u ti + ∆t = −
( )
t + ∆t
−1
( )
G ui (3.13)
∂u
The partial derivative term on the right hand side of equation (3.13) is the
Jacobian of the governing equations. We can identify this as the Tangent
Stiffness matrix, K. Using this we can say
(
δu i +1 = −K u ti + ∆t ) G (u )
−1 t + ∆t
i (3.14)
and hence by inversion
( ) (
K u ti + ∆t δu i +1 = −G u ti + ∆t ) (3.15)
The matrix equation (3.15) represents a set of linear algebraic equations in
the δu i +1 . It has the same form as for a linear problem (Ku = F). It must be
solved, for each iteration, for the change in incremental displacements. The
matrix K and vector G (the residual force vector) are different for each
iteration. At any particular iteration, the current increment in
displacements is given by
i
u ti + ∆t − u t = ∆u i = ∑ δu k (3.16)
k =1
The convergence criterion can be expressed in terms of G as follows
( )
G u ti ++1∆t < Tolerance (3.17)
It is important to note that the method requires accurate evaluation of
( ) ( )
K u ti + ∆t and G u ti + ∆t for each iteration i. G requires accurate evaluation
of the stress σ at the current estimate of ut+∆t. This can be seen from the
form of G,
26 Topics in Bio-Mechanical Engineering
( )
G u ti + ∆t = ∫ B T σ (u ti + ∆t )dV − ∫ N T t t + ∆t dS (3.18)
V S
which in turn means that a stress update algorithm is required. Next, look
at structure of K.
( )
K uti+∆t =
( )
∂G uti+∆t
= ∫ BT
∂σ
dV = ∫ BT
∂σ ∂ε
dV = ∫ BT
∂σ
BdV
∂u V
∂u (uti+∆t ) V
∂ε (uti+∆t ) ∂u V
∂ε (uti+∆t )
(3.19)
The partial derivative in the right-most term in equation (3.19) is the
Jacobian of the constitutive law and can be identified as the Tangent
matrix, Dtan.
∂σ
D tan =
∂ε (uti + ∆t )
( )
⇒ K u ti + ∆t = ∫ B T Dtan BdV (3.20)
V
The tangent stiffness matrix K has the same form as for linear problems,
i.e., the classic “BTDB” structure. However here K is different for each
iteration. This can lead to problems since the tangent matrix can be
difficult to evaluate for very non-linear materials.
The NR method is accurate and displays rapid convergence. However
sometimes it is modified, e.g.,
• use of a constant K, from the first iteration in the increment (or even
the first increment in the loading history – initial stress method) which
means we avoid having to recalculate K for each iteration
• use of a symmetric K if the problem generates an unsymmetric K,
otherwise we must use unsymmetric solvers
• more complex versions of the algorithm: BFGS method (Broyden-
Fletcher-Goldfarb-Shanno).
These and other methods can work well. They sometimes result in slower
convergence, but they can actually increase the radius of convergence. It
is important to note that it is acceptable to modify K but we always must
calculate G correctly and since G depends directly on the stress this
means that we must have a stress update algorithm that is accurate:
Finite Elements in Biomedical Engineering 27
u ti + ∆t = u t + ∆u i
G (u ti + ∆t ) = ∫ B T σ (u ti + ∆t )dV − ∫ N T t t + ∆t dS (3.21)
V S
σ (u ti + ∆t ) = σ (u t + ∆u i ) = σ (u t ) + ∆σ i
From equation (3.21) we must determine ∆σ i accurately, given ∆ui. This
can be quite difficult for complex non-linear constitutive laws.
accuracy. It is conditionally stable and the time step must be kept within
stability limit. Algorithms have been developed to evaluate this limit.
Explicit methods are becoming more popular since there are no iteration
and no convergence difficulties. They are suited to large deformation
problems and highly non-linear problems that might arise due to material
constitutive law complexity and surface contact, for example. The main
down side it the necessity to use a very small time step size that can slow
down computation. Time step size can be increased using mass scaling,
i.e., artificially factoring-up the mass. However this has to be carefully
monitored to avoid the introduction of artificially high kinetic energy
effects.
For small elastic strains and incompressible plasticity, the Jacobian of the
mapping J is approximately equal to 1, and hence the Cauchy stress is
approximately equal to the Kirchhoff stress. Referring the PVW back to
the reference configuration, V0 and S0, we can say
∫ δD σdV = ∫ δv tℜdS
T T
(3.25)
V0 S0
Here, ℜ is the infinitesimal surface area ratio between the current and the
reference configurations. Introducing the finite element interpolation as
before we can say u = N e u e , δu = N e δu e , v = N e v e and δv = N e δv e .
~
The rate of deformation tensor D is linearly related to ve at any instant in
time through the matrix of shape function spatial gradients, Be, as follows
Finite Elements in Biomedical Engineering 31
∂v ∂v
( )
T
~ ~ ~
D= 1
L + LT = 12 + = Be ve
(3.26)
∂y ∂y
2
In contrast to the small deformation case, here the shape functions and
hence the shape function gradients can be functions of the deformation y
and the displacements u. This means that Be can be different at each
instant in time. The linear relationship is only true instantaneously and is
not true in a time integrated form, i.e., the strain and displacement are not
linearly related. In an incremental form over a time step ∆t the linear
relationship is approximately true and we can say the following
∆ε = D∆t = B e v e ∆t = B e ∆u e (3.27)
In this case, ∆ε is the increment in logarithmic strain. Substituting into the
PVW, equation (3.25), summing over all the elements “e” in a finite
element mesh and dropping the summation notation and the subscripts for
convenience we get
∫ δv B(u) T σ (u)dV = ∫ δv T N T tℜdS
T
(3.28)
V0 S0
( )
K u ti + ∆t =
(
∂G u ti + ∆t
=∫
∂BT) σdV + ∫ BT
∂σ
dV
∂u V0
∂u (u t + ∆t ) V0
∂u (u t + ∆t
i )
i
(3.31)
∂t ∂ℜ
− ∫N T
ℜdS − ∫ N T t dS
S0
∂u (u ti + ∆t ) S0
∂u (u ti + ∆t )
However, as before,
∂σ ∂σ ∂ε
= = Dtan B (3.32)
∂u (uti + ∆t ) ∂ε ∂u (uti + ∆t )
so that
( ) ∫ ∂∂Bu ∂t
T
K u ti + ∆t = σdV + ∫ BT D tan BdV − ∫ N T ℜdS
V0 (u )
t + ∆t V0 S0
∂u (u ti + ∆t )
i
∂ℜ
− ∫ NT t dS
S0
∂u (u ti + ∆t )
(3.33)
Here we see that finite deformation kinematics have introduced
considerable complexities. In the first instance B will be different at each
point in time in the evaluation of G, from equation (3.30). More
significantly however, K is quite complicated, being composed of the
familiar term involving the material tangent matrix (second term on right
in equation 3.33) and other terms coming from the variation in B, t and
ℜ, due to large deformations. A finite element programme written to
implement this method must keep track of all these variations to generate
an accurate G and an accurate K. As part of this, the stress update
algorithm is specifically required to generate an accurate σ and an
accurate Dtan. K will in general be an un-symmetric matrix, requiring the
use of un-symmetric solvers when implemented in a computer
programme.
Finite Elements in Biomedical Engineering 33
4.1 Introduction
As can clearly be seen from the previous sections the finite element
method can be developed to deal with a range of boundary value
problems, both linear to non-linear. The presentation above is concerned
with solid mechanics only, however the principles and method are also
applicable to problems involving temperature/heat flow, diffusion,
chemical reactions, electromagnetism and, or course, biological
processes. The applications of computational methods in the
biomechanics and biomedical engineering fields have grown enormously
in recent years. Many advances have been made in applying them to the
analysis and design of medical devices, and this endeavour, on the
whole, has proved extremely successful – indeed finite element analysis
is a required part of the US FDA certification process for certain classes
of medical devices. More challenging is the application of computational
methods to tissue and to the representation of biological processes; this is
a very exciting and rapidly developing area at the forefront of modern
engineering science. In this new realm, there is tremendous scope for the
enhancement of existing computational methods and for the development
of new methods, based on alternative paradigms, that more completely
capture the physical nature of the problems to be addressed. In this
context, “cellular automata” approaches, and the combination of such
approaches with continuum mechanics, may prove extremely useful in
the future.
In the following we present two examples of the finite element
method, (i) modelling deformation of trabecular bone and (ii) modelling
failure of cardiovascular stent struts. Both examples can be classified
under the heading of micromechanics, where computational mechanics is
applied to physical problems at the micron size scale. Micromechanics
has been very successfully used in the past to deal with the mechanics of
multiphase materials and advanced metal-ceramic composites, and is
now finding increased application in the biomedical area.
Micromechanical models are usually based on the concept of the periodic
or repeating “unit cell”, whereby a macroscale material or system is
34 Topics in Bio-Mechanical Engineering
Trabecular bone (TB) makes up the inner part of a typical skeletal bone,
consisting of a highly porous network of tiny plates and strands called
trabeculae. Accurate three-dimensional finite element models are the key
to fully understanding its mechanical properties and how it functions in-
vivo. The challenging shape and size scale necessitates micro-CT scans to
be used as starting point for the geometry of such finite element models.
Firstly, a 3D STL model is created from a set of TB µ-CT scans using
imaging processing software (such as MIMICS). This mesh can then be
improved using specialised finite element pre-processing software. The
finite element analysis can then be performed using a conventional solver.
Finite Elements in Biomedical Engineering 35
Fig. 8. FE model of canine trabecular bone (3) – contour plot of von-Mises stress, computed
using ABAQUS/Standard.
36 Topics in Bio-Mechanical Engineering
The largest models currently run are based on a 3.5mm cube unit cell
created from canine femur bone µCT scans (see Fig. 8 (3)). The model
consists of 1.8 million 4-noded tetrahedral elements. Linear elastic
material behaviour has been assumed. Simple boundary conditions have
been applied here; the bottom surface is constrained and the top surface is
subjected to a uniform compressive displacement. A non-uniform stress
distribution in the microstructure is clearly evident. Solving takes in the
order of 4 hrs wall time on single processor of SGI Origin 3800. Many
elements have very poor aspect ratio leading to excessive element
distortion under loading. The considerable variation in element size and
shape required excessive CPU time because the stiffness matrix for each
element has to be computed.
Voxel mesh generation is also the currently used for this analysis of
trabecular bone. In this case an eight node brick element is placed at each
position where the CT scan detects bone. This results in a highly structured
mesh with a jagged irregular surface. However, the resulting system of
equations can be solved quickly as each element will have the same
stiffness matrix.
Micromechanics has been used to study size effects in small scale metallic
medical devices, in particular vascular stents for the treatment of coronary
heart disease (7). Stents, when manufactured, typically take the form of
networks of thin stainless steel struts. The structure undergoes
considerable plastic deformation when the stent is being deployed in a
blocked artery. It is expanded into position in the artery by the inflation of
a balloon upon which it is mounted. Once deployed, the balloon is deflated
and removed. Since stent struts are quite small, on the order of 75 µm x 75
µm in cross-section, and may only have a few grains across the thickness
(on the order of 10), the issue of size effects on material properties arises.
In particular, Murphy et al. (5) studied the influence of strut size on
ductility, finding that thin strut tensile ductility was only on the order of
50% of the ductility of bulk material. In Savage et al. (7), crystal plasticity
polycrystalline models were used to study this effect in 2D generalised
Finite Elements in Biomedical Engineering 37
Fig. 9. Example of 2D stent strut model with hexagonal grains. Reprinted from Savage et
al. (7) © Biomedical Engineering Society.
38 Topics in Bio-Mechanical Engineering
Fig. 10. Contour plot of accumulated plastic strain for a stent strut model with nine grains
across the thickness. Reprinted from Savage et al. (7) © Biomedical Engineering Society.
60
Failure Strain [%]
Computational
50
Experimental
40
30
20
10
0
0 100 200 300 400 500 600
5. Concluding Remarks
Acknowledgements
References