A Short Note on the FE-Formulation of

Transient Elasticity Problems

- Viscoelasticity and Elastodynamics
VSM013

FREDRIK LARSSON

Department of Applied Mechanics

CHALMERS UNIVERSITY OF TECHNOLOGY
Göteborg, Sweden 2006
Preface

This short introduction on transient problems in elasticity is part of the course Finite
Element Method - Applications, VSM013 at Chalmers University of Technology. The
text is based on, and intended as a complement to, the literature Introduction to the
Finite Element Method by N.S. Ottosen and H. Petersson and Notes on Transient Heat
Conduction and Similar Problems by J. Eriksson.
The text was developed for the course in the fall 2006.

Göteborg in November 2006

Fredrik Larsson

iii
iv

iv
Contents

1 Viscoelasticity - The Kelvin Model 1

1.1 Quasi-static equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Constitutive relation for the Kelvin model . . . . . . . . . . . . . . . . . . 3
1.3 Semi-discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Essential boundary conditions in space . . . . . . . . . . . . . . . . 9
1.3.2 Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Discrete solution in time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Dynamics 13
2.1 Strong form of the equation of motion . . . . . . . . . . . . . . . . . . . . 13
2.2 Weak form in space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Semi-discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 System of first order ODE’s . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Exercises 21

v
vi CONTENTS

vi
Chapter 1

Viscoelasticity - The Kelvin Model

In many engineering applications, viscous effects in materials need be considered. Ex-

amples are quasi-static cases, such as creep and relaxation, or the dynamic case, where
damping is crucial to obtain realistic results. In this chapter, we shall consider the simplest
possible constitutive model for a visco-elastic material and formulate the pertinent tran-
sient weak and finite element forms. For a more in-depth study on viscoelastic material
models, cf. Runesson (8).

1.1 Quasi-static equilibrium

We now consider a quasi-static mechanical problem of determining the displacements

u(x, y, z, t) = [ux uy uz ]T inside a body V (in 3D) during the time interval 0 < t < T . The
notion quasi-static refers to neglecting the effects of inertia due to accelerations, i.e., we
assume that static equilibrium is fulfilled at any given time. Following the derivations in
Ottosen and Petersson (7), we state the strong form of equilibrium for each specific time
as follows:

T
˜ σ = b(t) in V, t ∈ (0, T ),
−∇ (1.1)
t = h(t) on Sh , t ∈ (0, T ),
u = g(t) on Sg , t ∈ (0, T ),
u = u0 in V, t = 0.

1
2 1 Viscoelasticity - The Kelvin Model

Note that the only difference from the static linear elasticity case is that we require the
equations to hold during the specified time interval, where loading, boundary conditions
as well as displacements and stresses may vary, and that we introduced known initial
values of the displacement for the time t = 0. Just as for the case of transient heat flow,
as described in Eriksson (4), we note that the initial condition can be stated explicitly as

u(x, y, z, 0) = u0 (x, y, z), (1.2)

where u0 (x, y, z) is a known function in space. In the stated equations, we recall the
definitions of the stress vector and the (matrix) differential operator
   
∂
σxx 0 0
   ∂x ∂ 
 σyy   0 ∂y 0 
   
 σ   0 0 ∂ 
 zz  ˜ = ∂z 
σ= , ∇ . (1.3)

 σxy  ∂ ∂
 ∂y ∂x 0 
   
 σ   ∂ 0 ∂ 
 xz   ∂z ∂x 
∂ ∂
σyz 0 ∂z ∂y
Finally, in order to apply the natural boundary conditions on Sh , we recall the definition
of the traction as
t = Sn, (1.4)
where S is the stress tensor (stress matrix),
 
σxx σxy σxz
S =  σxy σyy σyz  , (1.5)
 

σxz σyz σzz

and n is the outwards pointing normal vector of unit length.
Repeating the steps in Ottosen and Petersson (7), the weak form of the problem can be
stated as
Z  T Z Z Z
˜
∇v σ dV = T
v t dS + T
v h(t) dS + v T b(t) dV
V Sg Sh V
for t ∈ (0, T ), (1.6)
u = g(t) on Sg , t ∈ (o, T )
u = u0 in V, t = 0,

where v is the arbitrary weight functions. Here, the natural boundary conditions h(t)
are introduced in the integral equation while the essential boundary conditions g(t) are

2
1.2 Constitutive relation for the Kelvin model 3

µ ǫ

Figure 1.1: A rhoelogical model of Kelvin visocelasticity in uniaxial stress, where a damper
with viscosity µ is coupled in parallel to a spring with stiffness E.

kept as an additional condition together with the initial condition described by u0 . Note,
again, that the equations must be satisfied at any time instance t.

1.2 Constitutive relation for the Kelvin model

We shall now introduce a constitutive relation to relate the stress in (1.1,1.6) to the sought
displacements u. To this end, we first recall the expression of strains used in (7),
˜
ǫ = ∇u (1.7)
where the differential operator reappears and we recall the definition of the strain vector,
 
ǫxx
 
 ǫyy 
 
 ǫ 
 zz 
ǫ= . (1.8)
 γxy 
 
 γ 
 xz 
γyz

In order to model transient effects, we shall also assume that the strain rate,
∂
ǫ̇ = ǫ, (1.9)
∂t
affects the stress. Since the coordinate system (x, y, z) is assumed to be constant in time,
we may interchange the order of differentiation, and hence
 
∂ ˜  ˜ ∂ ˜ u̇.
ǫ̇ = ∇u = ∇ u ǫ=∇ (1.10)
∂t ∂t

3
4 1 Viscoelasticity - The Kelvin Model

We shall now adopt the simplest possible model where the stress depends linearly on
the strain as well as the strain rate. This model is known as the Kelvin model for
viscoelasticity, cf. e.g. Runesson (8). Here, we assume that an elastic stress is generated
by the strains,
σ el = Dǫ, (1.11)
while the strain rate induces a viscous stress

σ v = Gǫ̇. (1.12)

Note that D is exactly the constitutive matrix as introduced in (7), while G is a new
constitutive matrix describing the viscous effects. Finally, the Kelvin model is defined
by the assumption that these stresses can be added, whereby we obtain the constitutive
relation
σ = σ el + σ v = Dǫ + Gǫ̇. (1.13)
The additive split of these stresses can be illustrated by a rheological model as shown in
Figure 1.2, where a viscous damper is coupled in parallel to a spring that represents the
elastic effects.
Following the same arguments as for the constitutive matrix D, as presented in (7), we
can study the properties of G with respect to material symmetry properties. Perhaps the
most straight-forward fashion to construct G is to model it as proportional to the elastic
constitutive matrix,
µ
G = D, (1.14)
E
where D is the elastic constitutive matrix, E is Young’s modulus of elasticity, and µ is
a reference viscosity for the uniaxial case. This simple assumption would yield a visco-
elastic model determined by one single parameter µ (in addition to the elastic parameters,
e.g. E and ν for the isotropic case).

Example 1.1 Creep test

One material behavior that can be analyzed using a viscoelastic material model is that
of creep. Creep is the long-term evolution of displacements with time under maintained
mechanical loading. A creep test is usually performed by applying a constant stress and
studying the evolution of strain with time.
For simplicity, we shall consider the 1-dimensional counterpart of (1.13) for uniaxial
stress,
σ = Eǫ + µǫ̇, (1.15)

4
1.2 Constitutive relation for the Kelvin model 5

0.8

0.6
E
σ̄
ǫ
0.4

0.2

0
0 1 2 3
E
µ
t

Figure 1.2: Illustration of a uniaxial creep test for a Kelvin visocelasticity model under-
going the constant stress σ̄. The strain saturates to the elastic response with time.

where σ and ǫ are the scalar measures of stress and strain, and E and µ are the material
parameters. The creep test can now be modeled by applying a constant stress, σ̄, and
solving for the strain from the differential equation

µǫ̇ + Eǫ = σ̄ for t ∈ (0, T ), (1.16)

ǫ(0) = 0, (1.17)

where we assume an unstrained initial case. The result is illustrated in Figure 1.1. We
see that the long term effect of this model is that of a saturated strain level, corresponding
to a purely elastic case. In a case of material failure due to creep, such a saturation
would not take place. It is thus important to realize that the Kelvin model fails to model
creep-failure.

Example 1.2 Rate-dependent response

The application of a viscoelastic model allows for modeling rate-dependent response. We
thus allow for the case that an increase of the rate of loading yields an increase of the
response. This is illustrated in Figure 1.2 where the uniaxial stress-strain relation is
indicated for different loading rates. Note that in the limit case of slow loading, ǫ̇ → 0,
the elastic response is retained.

5
6 1 Viscoelasticity - The Kelvin Model

ǫ̇2 ǫ̇1 ǫ̇0 → 0

0.8

0.6
E
1
σ

0.4

0.2

0
0 0.5 1
ǫ

Figure 1.3: Illustration of the rate-dependent response of the Kelvin visocelasticity model
undergoing different constant strain rates ǫ̇2 > ǫ̇1 > ǫ̇0 → 0. Note that the slow strain
rate ǫ̇ → 0 corresponds to the static case, σ = Eǫ.

Remark 1.1 Other viscoelastic models

It should be noted here that there exist other, more involved models, linear as well as
nonlinear, for viscoelasticity. For these, however, more variables (in addition to the dis-
placements) need be solved for. The most common alternative is the Maxwell model, where
the elastic and viscous effects are coupled in series.

1.3 Semi-discretization

Following the steps in Eriksson (4), we shall now establish the semi-discrete equations
pertinent to a finite element discretization in space1 . As a point of departure, we consider
the weak form of the problem (1.6). First, we want to replace the expression for the
stress by using the constitutive relation in (1.13) and the kinematics defined in (1.9,1.10),

1
Semi-discrete defines the set of equations that are discretized in space, but still on the form of
differential equations in time.

6
1.3 Semi-discretization 7

whereby we obtain
Z  T  T Z Z Z
˜
∇v ˜ + ∇v
D∇u ˜ ˜ u̇ dV
G∇ = T
v t dS + T
v h(t) dS + v T b(t) dV
V Sg Sh V
for t ∈ (0, T ),
u = g(t) on Sg , t ∈ (0, T ), (1.18)
u = u0 in V, t = 0.

Following the general procedure, as presented in (7), we now introduce the approximation
in terms of a vector of unknown displacements and a matrix of shape functions,
u(x, y, z, t) ≈ N (x, y, z)a(t) (1.19)
where we note that the nodal displacements a vary only in time and the shape functions
N vary only in space. We recall the explicit expressions from (7) as
 
ux1
 uy1 
 
 
 uz1 
 
 ux2 
 
 
N1 0 0 N2 0 0 . . . Nn 0 0  
 uy2 
N =  0 N1 0 0 N2 0 . . . 0 Nn 0  , a =  . (1.20)
   
 uz2 
0 0 N1 0 0 N2 . . . 0 0 Nn  .. 
 
 . 
 
 uxn 
 
 uyn 
 

uzn
In order to introduce the approximation into the weak form, we shall now evaluate the
pertinent expressions for the strain and strain rate. First, the strain is approximated as
˜
ǫ = ∇u(x, ˜ (N (x, y, z)a(t)) = B(x, y, z)a(t),
y, z, t) ≈ ∇ (1.21)
˜ evaluates the
since a is independent of the spatial coordinates for which the operator ∇
derivatives. Here, we introduced the B-matrix identically as in (7) for the static case as
 
∂N1
0 0 . . . ∂N n
0 0
 ∂x ∂N1 ∂x
. . . 0 ∂N

 0 ∂y
0 ∂y
n
0 
 
˜
 0 0 ∂N1 ... 0 0 ∂N n 

B = ∇N =  ∂N1 ∂N1 ∂z ∂z 
(1.22)

∂Nn ∂Nn

 ∂y
 ∂x
0 . . . ∂y ∂x
0  
 ∂N1 0 ∂N1 . . . ∂N n
0 ∂Nn 
 ∂z ∂x ∂z ∂x 
∂N1 ∂N1 ∂Nn ∂Nn
0 ∂z ∂y
... 0 ∂z ∂y

7
8 1 Viscoelasticity - The Kelvin Model

For the strain rate, we obtain

∂ ∂ ∂
ǫ̇ = (ǫ) ≈ (B(x, y, z)a(t)) = B(x, y, z) (a(t)) = B ȧ, (1.23)
∂t ∂t ∂t
since B is independent of time.
The weak form (1.6) holds for any weight function v for the exact solution to the problem.
However, after introducing the approximation u ≈ N a we cannot expect the result to
hold for an arbitrary weight function. Following the Galerkin method, we thus introduce
the weight function as described by the same shape functions as our approximation, i.e.,

v(x, y, z) ≈ N (x, y, z)c, (1.24)

where c are arbitrary coefficients. Note here that, since the weak form in space should
hold for any instance t, we do not consider any time-variation of the weight function.
In order to introduce the restricted weight function in the weak form, we identify the
expression for the strain of the weight function,

˜
∇v = Bc. (1.25)

We are now in the position to introduce our approximation into the weak form and obtain
the equation
Z
cT B T DBa + cT B T GB ȧ dV =
Z Z V Z
T T
T
c N t dS + T
c N h(t) dS + cT N T b(t) dV (1.26)
Sg Sh V

Using the fact that the coefficient c, the nodal temperatures a and the nodal rates of
temperature ȧ do not depend on the spatial coordinates, we may extract them from the
integrals, whereby
Z Z
T
c T
B GB dV ȧ + B T DB dV a
V V
Z Z Z #
− N T t dS − N T h(t) dS − N T b(t) dV = 0. (1.27)
Sg Sh V

Finally, since we want the expression to hold for arbitrary coefficients c, we conclude that

C ȧ + Ka = f b + f l , (1.28)

8
1.3 Semi-discretization 9

where we introduced the damping matrix C, the stiffness matrix K, the boundary force
vector f b and the load vector f l as follows
Z
C = B T GB dV, (1.29)
Z V

K = B T DB dV, (1.30)
Z V Z
T
fb = N t dS + N T h(t) dS, (1.31)
Sg Sh
Z
fl = N T b(t) dV. (1.32)
V

We identify the stiffness matrix to be exactly the one used in the static case. Furthermore,
we see that the damping matrix for this problem has the same structure as a stiffness
matrix, with the only difference lying in the constitutive matrix G replacing the Hook
matrix D. Adding the two vectors on the right hand side into the force vector, f = f b +f l ,
we are in the position to state the system of differential equations in time as follows:

C ȧ + Ka = f . (1.33)

1.3.1 Essential boundary conditions in space

Just as for the static case described in (7), we introduce the essential boundary condi-
tions, u = g on Sg in terms of prescribing the pertinent components in the vector of
displacements a. In order to visualize this, we consider the schematic structure2 of the
nodal displacement vector a as follows:
" #
aC
a= , (1.34)
aF

where aC represents the the nodal displacements that are prescribed (Constrained) and
aF contains the remaining unknown (Free) nodal displacements. It should be noted that
the same sub-structuring of the right-hand-side gives
" #
fC
f= , (1.35)
fF
2
Note that, in practice, the nodes are not generally ordered such that the prescribed degrees of freedom
end up as the top part of the vector.

9
10 1 Viscoelasticity - The Kelvin Model

where the nodal force components f C (i.e. the bundary part f b,C ) contains the unknown
reaction forces, cf. (7).
Describing the values of g in terms of ag , we can formally introduce the essential boundary
condition as
aC = ag . (1.36)

Remark 1.2 Condensation

As an alternative, we note that, since only aF are sought for, we may perform a conden-
sation of our system, similarly to that presented in (7) for the static case, as
" #" # " #" # " #
C C,C C C,F ȧC K C,C K C,F aC fC
+ = , (1.37)
C F,C C F,F ȧF K F,C K F,F aF fF

which by using the second ”row” results in the reduced system

C F,F ȧF + K F,F aF = f F − C F,C ȧg − K F,C ag . (1.38)

Here, the right-hand-side is known and all components in aF are unknown functions in
time.

1.3.2 Initial conditions

In order to complete the semi-discrete form, we need to introduce the initial condition.
To this end, we simply describe the initial displacement field u0 in terms of the nodal
displacements a0 , whereby we state

a(0) = a0 . (1.39)

Collecting the system of ODE’s (1.33), the essential boundary conditions (1.36) and the
initial condition (1.39), we are in the position to state the complete semi-discrete form of
finding a(t) as

C ȧ + Ka = f (t) for t ∈ (0, T ),

aC = ag (t) for t ∈ (0, T ), (1.40)
a(0) = ao .

10
1.4 Discrete solution in time 11

1.4 Discrete solution in time

Finally, we shall adopt a numerical method to solve the system of differential equations
in time. Following (4), we directly subdivide the time domain into N + 1 time-nodes,
0 = t0 < t1 < . . . < tN −1 < tN = T . at each time node, we now wish to compute an
approximation an ≈ a(tn ). Since a(t0 ) = a(0) = a0 is known, we conclude that we can
solve for one time-node at the time. To this end, we consider the step of solving for an+1
when an is known. Introducing the time-step length, ∆t = t n+1 − tn , we may use the
method described in (4) and conclude that

[C + ∆tΘK] an+1 = [C − ∆t(1 − Θ)K] an + ∆tf̄ n+1 , (1.41)

where the choice of weight-function in time, w, defines

R ∆t τ
R ∆t
w(τ ) dτ w(τ )f (tn + τ ) dτ
Θ = 0R ∆t ∆t
, f̄ n+1 = 0 R ∆t , (1.42)
0
w(τ ) dτ 0
w(τ ) dτ
where τ = t − tn is a local time-coordinate defined inside the time-step.

Remark 1.3 Essential boundary conditions in space

Note that the system of equations in (1.41) is similar to that of a static FEM-problem in
linear elasticity. It is important to note that, in contrast to the situation for e.g. transient
heat conduction, C has the same structure as K and, thus detC = 0. This means that we
have to introduce essential boundary conditions in order for the system to be solvable. In
other word, rigid-body motion must be prevented in all quasi-static equilibrium problems.

We are now in the position to formulate the entire FEM-problem in space-time as follows:

[C + ∆tΘK] an+1 = [C − ∆t(1 − Θ)K] an + ∆tf̄ n+1 for n = 0, 1, . . . , N − 1,

an+1,C = ag (tn+1 ), (1.43)

where a0 is known in terms of an initial condition.

Remark 1.4 Just as for the semi-discrete case, we may study the explicit solution of
(1.43) by condensation of the system as

[C F,F + ∆tΘK F,F ] an+1,F = [C F,• − ∆t(1 − Θ)K F,• ] an + ∆tf̄ n+1 (1.44)
− [C F,C + ∆tΘK F,C ] ag (tn+1 ) for n = 0, 1, . . . , N − 1,

11
12 1 Viscoelasticity - The Kelvin Model

where we introduced the notation

h i h i
C F,• = C F,C C F,F , K F,• = K F,C K F,F , (1.45)

in order not to split up an .

12
Chapter 2

Dynamics

One important problem in engineering practice is that of structural dynamics. The anal-
ysis of dynamical behavior of structures requires to take into account the inertia effects
of the mass in the system. Applications where dynamics are studied are numerous. Some
examples are those of short time-intervals under rapid loading, such as collision analysis
of vehicles, drop tests of cellular phones etc., or vibration analysis of systems under longer
times, such as vibration of a structure due to wind loads or the response of a structure
due to earthquake excitation. In this chapter, we shall construct the semi-discrete form
of a dynamic problem by introducing a finite element approximation in space. For a more
in-depth analysis of the semi-discrete system, we refer to, e.g., Abrahamsson (1), Craig
(3) or Géradin and Rixen (5).

2.1 Strong form of the equation of motion

We shall start by briefly deriving the strong form of the momentum equations in a contin-
uum1 . To this end, we shall consider an arbitrary volume Ṽ , sustained to loading in terms
of a body force b [N/m3 ] in Ṽ and traction t [N/m2 ] applied on its surface S̃. Following
the procedure of using Gauss’ theorem, cf. (7), we can describe the force acting on this
body as
Z Z Z
F = b dV + t dS = ˜ dV,
b + ∇σ (2.1)
Ṽ S̃ V˜
1
A continuum is the notion of a continuously ”smeared” material in space, as opposed to a discrete
system of point masses.

13
14 2 Dynamics

where we introduce the stress vector σ and the differential operator matric ∇˜ as defined
in (1.3). In order to obtain the equality in (2.5) we used the relation between the stress
tensor (matrix) S, that contains the components of σ, and the outwards pointing normal
n. Next, we shall assume a density ρ [kg/m3 ] whereby we can formulate the mass of the
volume as
Z
m= ρ dV. (2.2)
Ṽ

Now, we use Newton’s second law for determining the average acceleration, ā, of the body

F = mā. (2.3)

˜ T σ and
Now, we let Ṽ = dV be a small volume around a point (x, y, z). Assuming b, ∇
ρ to be approximately constant in the infinitesimal volume, namely that of the values in
the point (x, y, z), we can write
 
˜
dF = b + ∇σ dV, (2.4)
dm = ρ dV. (2.5)

Again, using Newton’s second law, we can formulate an expression for the acceleration -
∂2
the second order derivative in time of the displacement, ü = ∂t2 u - as

dF = dmü. (2.6)

Finally, dividing by the volume yields

dF dmü ˜ T σ = ρü,
= → b+∇ (2.7)
dV dV

which is a differential equation in space and time - the equation of motion.

Just as for the case of the quasi-static equilibrium, we shall now choose to study a body
occupying the volume V in space under the time interval 0 < t < T . Introducing the
boundary conditions of conventional type, we may prescribe the displacements u on one
part of the boundary surface, Sg , while prescribing the traction t on the remaining part,
Sh . Since (2.7) contains second order derivatives in time, we have to introduce two initial
conditions, on the displacement u and on the velocity u̇. Hence, we formulate the strong

14
2.2 Weak form in space 15

form of the equation of motion as

˜ T σ = b(t) in V,
ρü − ∇ t ∈ (0, T ),
t = h(t) on Sh , t ∈ (0, T ),
u = g(t) on Sg , t ∈ (0, T ), (2.8)
u = u0 in V, t = 0,
u̇ = u̇0 in V, t = 0,

where u0 and u̇0 are known functions in space defining the initial values of the displace-
ment and velocity, respectively.

2.2 Weak form in space

We shall now construct the weak form in space of (2.8). To this end, we multiply by a
weight function2 v = v(x, y, z) and integrate over the volume V , whereby

Z  
T
v T ˜
ρü − ∇ σ − b(t) dV = 0, (2.9)
V

which should hold for any time-instance t. The next step in deriving the weak form is
constituted by balancing the derivatives in space on v and u. Although we have not
yet introduced the constitutive relation for the stress, we shall assume that it contains
first-order derivatives in space on the displacement, which is the case both for ordinary
elasticity (7) and for visco-elasticity as presented in the previous chapter. Hence, the only
term we have to target is the one containing the stress. Since this term occurs in static
elasticity, we refer to (7) and conclude that

Z   Z Z Z  T
T ˜
v ∇σ dV = T
v h dS + T
v t dS − ˜
∇v σ dV (2.10)
V Sh Sg V

2
Note that, since we wish to construct the weak form in space, the weight function does not depend
on time. This procedure should be viewed as performed for any time-instance t.

15
16 2 Dynamics

where we inserted the natural boundary conditions on the part Sh where the traction is
prescribed. We thus obtain the following weak form of the equation of motion:
Z Z  T Z Z Z
T
v ρü dV + ˜
∇v σ dV = T
v h dS + T
v t dS + v T b dV
V V Sh Sg V
for t ∈ (0, T ),
u = g(t) on Sg , t ∈ (0, T ), (2.11)
u = u0 in V, t = 0,
u̇ = u̇0 in V, t = 0,

where it only remains to introduce the constitutive relation for the stress.

2.3 Semi-discretization

For the sake of simplicity, we shall now consider trhe (rate-independent) linear Hooke
elasticity model,
σ = Dǫ. (2.12)
The combination of the two chapters in this compendium - dynamics with Kelvin visco-
elasticity - is straightforward and left to the reader as an exercise.
Our aim is now to introduce a finite element approximation in space, and the pertinent
weight functions, to obtain the semi-discrete form of the elastodynamics problem. In
other words, we wish to establish the discrete set of ordinary differential equations in
time that follow a finite element approximation in space. To this end, we introduce our
approximation as in the previous chapter to be that of

u(x, y, z, t) ≈ N (x, y, z)a(t), (2.13)

where we note that the nodal displacements a vary only in time and the shape functions
N vary only in space. Firstly, we identify the expression for the strain identically to that
in (7),
ǫ ≈ B(x, y, z)a(t). (2.14)
The matrices N and B where defined in the previous chapter. Secondly, the acceleration
term, i.e. the second order derivative in time of the displacement, can be evaluated as
∂2 ∂2
ü ≈ (N (x, y, z)a(t)) = N (x, y, z) (a(t)) = N ä, (2.15)
∂t2 ∂t2
16
2.3 Semi-discretization 17

in terms of the second order time-derivative of the nodal displacements. Finally, we use
the Galerkin method, following (7), whereby we restrict the weight function to be a linear
combination of the shape functions used for displacement approximation,

v(x, y, z) = N (x, y, z)c, (2.16)

where the arbitrary coefficients in c determine the restriction of the weak form. In order
to introduce the restricted weight function in the weak form, we need also to identify the
expression for the strain of the weight function,
˜
∇v = Bc. (2.17)

We are now in the position to introduce our approximation into the weak form and obtain
the equation
Z
cT N T ρN ä + cT B T DBa dV =
Z Z V Z
T T
T
c N t dS + T
c N h(t) dS + cT N T b(t) dV (2.18)
Sg Sh V

Using the fact that the coefficient c, the nodal temperatures a and the nodal rates of
temperature ȧ do not depend on the spatial coordinates, we may extract them from the
integrals, whereby
Z Z
T
c T
N ρN dV ä + B T DB dV a
V V
Z Z Z #
− N T t dS − N T h(t) dS − N T b(t) dV = 0. (2.19)
Sg Sh V

Finally, since we want the expression to hold for arbitrary coefficients c, we conclude that

M ä + Ka = f b + f l , (2.20)

where we introduced the mass matrix M , the stiffness matrix K, the boundary force
vector f b and the load vector f l as follows
Z
M = N T ρN dV, (2.21)
ZV

K = B T DB dV, (2.22)
V
Z Z
T
fb = N t dS + N T h(t) dS, (2.23)
Sg Sh
Z
fl = N T b(t) dV. (2.24)
V

17
18 2 Dynamics

We identify the stiffness matrix to be exactly the one used in the static case. Furthermore,
we see that the mass matrix has the same structure as the capacity matrix in the transient
heat flow problem in (4), with the only difference lying in not introducing the heat capacity
c. Adding the two vectors on the right hand side into the force vector, f = f b + f l , we
are in the position to state the system of differential equations in time as follows:

M ä + Ka = f . (2.25)

Remark 2.1 Essential boundary conditions in space

The mass matrix has the inherent property that detM 6= 0. This has the effect that
we need not prescribe any essential boundary conditions in space for the dynamic case.
Rigid body motion will not give rise to an unsolvable system, but may be part of a time-
integrated solution, cf. the case for transient heat conduction in (4) where no essential
boundary conditions on the temperature are needed.

Adding the essential boundary conditions on the displacements, cf. (1.36), and the initial
condition in terms of known initial displacement a0 and velocity ȧ0 , we are in the position
to state the complete semi-discrete form of finding a(t) as

M ä + Ka = f (t) for t ∈ (0, T ),

aC = ag (t) for t ∈ (0, T ), (2.26)
a(0) = ao ,
ȧ(0) = ȧo .

Note that f (t), ag (t), a0 and ȧ0 are independent known functions/vectors.

2.4 System of first order ODE’s

In order to adopt a numerical scheme for the approximate solution in time for the semi-
discrete problem (2.26), it can be useful to rewrite the problem into a first order system.
In this manner, our techniques developed for, e.g., the transient heat flow problem can
be re-used for the dynamics problem. The procedure below follows the brief description
in Eriksson (4).
As a first step, we introduce the velocities of the nodes as a new vector of unknowns,

p = ȧ. (2.27)

18
2.4 System of first order ODE’s 19

Introducing this assumption together with (2.26), we obtain the enlarged system of equa-
tions
" #" # " #" # " #
0 M ȧ K 0 a f (t)
+ = for t ∈ (0, T ),
I 0 ṗ 0 −I p 0
aC = ag (t) for t ∈ (0, T ), (2.28)
pC = ȧg (t) for t ∈ (0, T ),
a(0) = ao ,
p(0) = ȧo ,

which is obviously a 1-st order system that can be expressed as

Ĉ ż + K̂z = f̂ for t ∈ (0, T ),

" # " #
aC ag
zC = = for t ∈ (0, T ), (2.29)
pC ȧg
z(0) = z 0 .

Here, we introduced the vector of unknown displacements and velocities,

" #
a
z= . (2.30)
p

Furthermore, for brevity, we established the effective matrices

" #
0 M
Ĉ = ,
I 0
" #
K 0
K̂ = ,
0 −I
" #
f (t)
f̂ (t) = .
0

Note that, since any possible essential boundary conditions in space ag (t) must be known
as functions of time, whereby
∂ ∂
pC = aC = ag (2.31)
∂t ∂t
is trivially known.

19
20 2 Dynamics

Remark 2.2 Alternative formulation

−1
In (4), the 1st order system is stated by pre-multiplying the equations by Ĉ , which is
invertible due to the property of M as
" #
−1 0 I
Ĉ = . (2.32)
M −1 0

The multiplication of the right- and left-hand sides of (2.28) results in the equivalent
equation
" # " #" # " #
ȧ 0 −I a 0
+ = for t ∈ (0, T ).
ṗ −1
M K 0 p M f (t)
−1

This presented formulation of collecting displacements and velocities in a joint vector of

unknowns is sometimes denoted a state-space formulation. Based on this formulation, we
can adopt a time-integration scheme for 1st order systems following, e.g., (4). For a more
detailed description on the analysis of the semi-discrete system, cf. (1, 3, 5).

20
Chapter 3

Exercises

Exercise 1 A uniaxial creep test is simulated by prescribing the stress, σ, and evaluating
the response in terms of the strain, ǫ. The material is modelled as a linear viscoelastic
solid of Maxwell type, i.e.,
E
σ̇ + σ = E ǫ̇.
µ
(a) Construct c, k and f corresponding to Equation (13) in Eriksson (2005).
(b) Assume an initially undeformed specimen. What is the initial condition at t = 0?
(c) For a constant stress, σ = σ̄, determine the response using two different methods; The
Backward Euler rule and the Forward Euler rule. Solve the problem analytically. Compare
and comment the results.

Exercise 2 A uniaxial relaxation test is simulated by prescribing the strain, ǫ, and eval-
uating the response in terms of the stress, σ. The material is modelled as a linear vis-
coelastic solid of Maxwell type, i.e.,
E
σ̇ + σ = E ǫ̇.
µ

(a) Construct c, k and f corresponding to Equation (13) in Eriksson (2005).

(b) Assume an initially undeformed specimen. What is the initial condition at t = 0?
(c) For a constant strain, ǫ = ǭ, determine the response using two different methods; The
Backward Euler rule and the Forward Euler rule. Solve the problem analytically. Compare
and comment the results.

21
22 3 Exercises

Exercise 3 Establish the Quasi-static FE-formulation in 3D for a Kelvin type viscoelastic

solid using the Backward Euler rule.
The quasi-static equilibrium equation is given in Ottosen & Petersson (1992) as

T
˜ σ = b,
−∇

and a Kelvin type material is modeled by the constitutive relation

σ = Dǫ + Gǫ̇,

where D and G are two constitutive matrices of size (6 × 6).

Exercise 4 Establish the semi-discrete FE-formulation for elasto-dynamics in 3D. The

equations of motion is the expansion of the quasi-static equilibrium (cf. Ottosen &
Petersson (1992)) given by

˜ T σ = b,
ρü − ∇

where ρ is the density and ü is the acceleration.

Exercise 5 Construct the strong, weak and FEM formulations of transient heat conduc-
tion in 1D, cf. Ottosen & Petersson (1992)).

Exercise 6 Consider transient heat conduction in 1D. Construct the element matrices
C e and K e for one element using the lowest possible order of approximation for the
temperature. Evaluate det(C e ) and det(K e ). Comment the results.

Exercise 7 Compute the weight parameter Θ for the weight function w = 2 in Equation
(16) in Eriksson (2005). The result coincide with another choice of w in Table 1. Why?
Will the corresponding f̄ also coincide for the two choices?

Exercise 8 The dynamic version of the equilibrium equation for a bar, i.e. the equation
of motion for a 1D bar, reads
d
ρAü − (Aσ) = b,
dx
22
 23

cf. Ottosen & Petersson (1992), Section 4.2. Consider the case when the constitutive
model is that of Kelvin visco-elasticity, i.e.,

σ = Eǫ + µǫ̇.

Construct the matrices K, C and M such that the semi-discrete FE formulation reads

M ä + C ȧ + Ka = f .

23
24 3 Exercises

24
Bibliography

[1] Abrahamsson T., 2000, Linear System Theory in Vibration Engineering, Chalmers
University of Technology, Department of Applied Mechanics.

[2] Bathe K.-J., 1982, Finite Element Procedures in Engineering Analysis, Prentice Hall,
Englewood Cliffs, NJ.

[3] Craig R.R.Jr. and Kurdila, A.J., 2006, Fundamentals of Structural Dynamics, 2nd
Edition, John Wiley & Sons.

[4] Eriksson J., 2005, Notes on transient heat conduction and similar problems, Chalmers
University of Technology, Department of Applied Mechanics.

[5] Géradin M. and Rixen D., 1998, Mechanical Vibration, John Wiley & Sons, Chich-
ester.

[6] Hansbo P., 2004, Beyond the elements of finite elements: General principles for solid
and fluid mechanics applications, Lecture Notes.

[7] Ottosen N.S. and Petersson H., 1992, Introduction to the finite element method,
Prentice Hall, Wiltshire.

[8] Runesson K., 2006, Constitutive Modeling of Engineering Materials - Theory and
Computation. The Primer, Chalmers University of Technology, Department of Ap-
plied Mechanics.

[9] Zienkiewicz O.C. and Taylor R.L., The Finite element method, Volume 2 Solid Me-
chanics, Butterworth Heinemann.

25