Abstract
Background
Vibration in machinery leads to high cycle fatigue failure and decreases its
lifecycle. It costs the organization both time and money to maintain and replace parts over its lifespan. Damping can be defined as the process by which
a structure or structural component dissipates mechanical energy, usually
through friction at microscopic scale.
Woodhouse [1] classified damping broadly in following classes:
Energy dissipation distributed throughout the bulk material of structurematerial damping e.g. rubber. This is easier to model with help of
prony series/stress relaxation/creep data for the material and is widely
covered in literature.
Dissipation associated with interfaces and junction between two parts
of structure-Boundary damping e.g. fibre-matrix slip interface, interlaminar damping. These are highly non-linear effects and lack proper
understanding.
Dissipation associate with a viscous fluid in contact with structure.
The problem of vibration laminated beams and plates has been widely covered in the literature. The concept of Constrained layer damping was introduced in 1960s and is fundamental to study of laminated beams. Early
works in this field can be found in Ross,Ungar and Kerwin, Rao and Plunkett and Lee. Johnson and Kienholz[8] introduced the concept of the use
of modal strain energy(MSE) and FE formulations to be used to calculate
modal loss factors for a system with viscoelastic dampers. MSE concept is
widely used to quantify the damping capacity of structures.
However, in above mentioned analyses the base material was considered
perfectly elastic which is reasonable assumption for metals. Usually there is
a trade off between damping capacity and structural strength. Composite
materials have excellent structural properties and often an order of magnitude higher damping capacities than metals. The intrinsic damping in
composite is primarily due to viscoelastic nature of the matrix and interface
between fibre and matrix. This study will be focused understanding hysteric
damping which is dominant mechanism in undamaged composites at small
amplitudes.
Due to anisotropic nature of fibre reincofrced polymers, damping is negligible
in direction of fibres is much higher in tranverse direction. The proeprties
of composites are usually calculated using voight and reuss rules of mixtures. Hashin-shritkman dervided bounds based on variational principles
which provide a more accurate prediction of composite materials. Hashin
further developed analytical equations to calculate complex modulus of unidirectional FRP based on properties of matrix and fibres. Numerical works
by [Hwang and Gibson, Chandra have been presented based on strain energy method to predict damping in a composite ply. The strain energy
method states that of any system of linaer viscoelastic elements the loss
factor can be expressed as a rati of summation of the product of individual element loss factors and strain enrgy stored in each elemnt to the toal
strain energy. Damping and moduli of composites with short aligned fibres
can be predicted using Cox. Bert and chang used micromchanics and correspondence principle to characterize damping in a unidirectional composite.
Saravanos and chamis developed integrated micromechanics methodology
to completely characterize damping in comosites via six damping coeffients.
Kaliske and Rothert developed a modeol for material damping using a consistent micromechanical theory of a representative fibre matrix cell. Tsai
and chai explored the effects of packing array of fibres on overall damping
using method of cells. Brinson and fisher workers have compared the mechanical property predictions for a three phase viscoelastic coposites by use
of bensivite and mori tanaka method.
Given the damping properties of a lamina, there are several macromechanical theories to predict damping of a composite. Adams and bacon developed
macromechanical model to predict damping in compostes. They characterized overall damping as a summation of damping capasities due to longitudinal, tranverse and shear stresses. Adams and maheri used these results
to predict flexural damping and strength with respect to fibre orientation.
They showed that at lower angles(0-15), material behaved as hookean elastic material and damping can be predicted usinghysteric loops. However at
higher angles of (30-90) damping is governed by interaction between fibres
and matrix. CT SUN AND GIBSON used classical laminate theory (CLT)
and correspondence principle to predict damping in composite materials. A
state of plane stress can only be assumed if the lamina is thin. However for
prediction of damping in composites with integrated compliant viscoelastic
layers, first order shear deformation theory gives better results. A comparitve review of major macromechanical theories is given by Billups and
cavalli. Crane and Gillespie used CLT to calculate A-B-D matrix of composites integrating the frequency dependence of damping. In the above theories, an assumption fibres being elastic and non-dissipative has been made.
Saravanos and Chamis further developed their theory to predict damping
in composites using discrete layer damping mechanics for thick laminates.
Gibson and Guan used CLT to predict damping in woven fabric-reinforce
polymer matrix composites using saravanos and chamiss equations. There
are several studies in lieterature which further extend the findings are macroscopic scale into structural level. However, they are not considered at this
stage of in this project.
3
3.1
Introduction
Damping
Damping is defined as the ability of a material or structure to dissipate energy under oscillatory motion. It is broadly classified into Active damping-in
which feeback forces in a control loop are used to limit deflections. Piezoelectic actuators are being widely researched and used for this purpose. Passive
damping-which will be the focus of this research, is related to the iherit ability of the material or system to dissipate energy under oscillatory loading.
Passive damping is further divided in two branches Material or Hysteric
damping and Structural damping.
Material damping is related to the energy dissipation inherent to the ma3
3.2
Measurement of damping
W
W
(1)
Where,
W:Energy dissipated in one cycle
W:Total Energy Stored in one cycle
There are several choices of W which have been used in literature.
[Lazan, B.L., Damping of Materials and Members in Structural Mechanics, Permagon Press Ltd., New York, 1968.] defined W to be total
strain energy for the entire specimen at maximum deformation. Other
definitions include, peak potential energy solely of the elastic component of the model. However, [Ungars] definition of stored energy equal
to the totalstrain energy at maximum displacement is widely used. As
for any structure, the total strain energy and hence the damping will
be related to the global deformation. The concept of Modal damping Capacity is widely used to simply characterization of damping in
structures undergoing small deflections. Modal SDC with n mode is
according to Chamis/saravanos 1991:
Psi n
4
Figure 1
Loss Factor Eta Loss factor is defined as the tangent of the loss angle
Phi. It is also related to the SDC as follows. Loss factors are also
related to the complex modulus which will be discussed in more detail
in subsequent sectons
Logarithmic Decrement This method is useful to characterize damping
for free vibrating structures in time domain. Logarithmic decrement is
defined as the natural log of the ratio of amplitudes of any successive
peaks.
=
4
4.1
1
x1
ln
n xn+1
(2)
In order to fully characterize damping in composite its essential to understand the theory of linear viscoelasticity and the constituitive models that
allow mathematical characterization of damping. The constituitive relationship between stress and strain for elastic materials in one dimensiom is given
by hookes law as follows.
= E
(3)
However, it is well known that real materials exhibit both elastic as well
as some viscous properties. These Viscoelastic materials exhibit properties such as relaxation, creep, frequency dependent stiffness and dissipative
characteristics as well as strain rate dependent hysteric behavious. Several
texts such as Ferry, Christensen and aklonis and macknight describe the
fundamental theory of viscoelasticity. Readers are also recommended text
by hal and catherine brinson for an introduction to viscoelasticity and more
information about microscopic properties. The dissipative property of viscoelastic polymer matrix arises from its long chained polymer structure. The
movement of polymer are highly dependent on temperature and frequency.
In order to develop constitutive relations for viscoelastic materials, it is
important to understand relaxation and creep phenomena.
4.2
Stress Relaxation
(t)
= 0 E(t)
0
(4)
Where,
E(t) = RelaxationM odulus
The latter equation is analogous to Hookes law for material that is time
dependent but only at constant strain. Similar relaxation functions can be
obtained for Shear G(t) as well as Bulk Moduli(t). It is also helpful to define
the Instantaneous modulus, Eo at t=0 and Equilibirium modulus when t=.
4.3
Creep
(t)
or (t) = 0 D(t)
0
Where,
D(t) = CreepCompliance
(5)
Figure 2
Figure 3
4.4
Linearity
In order to determine whether under specific loading conditions, the viscoelastic response of the material is linear or non-linear. This can be done
by performing creep or relaxation tests at three different stress(or strain)
levels and plotting the isochronous Stress-Strain diagrams as shown below.
If the isochronous graph is linear for any given time, the material response
is also linear.
4.5
So far, the response of viscoelastic materials to step stress or strain have been
defined. To develop constituitive relationship at any given stress/strain we
use the Boltzman superposition principle. Boltzman superposition principle
states that the effect of a compound cause is the sum of the effects of the
individual causes.
Consider any arbitary strain history as given in FIGURE. The idea here is
to decompose entire strain history into small pulses. The stress at time t
is the summation of the stress effects of each of these pulses. The principle
of causality which states that the cause should precede the effect allows us
to consider oly the pulses upto present time. The Boltzman superposition
integral can be shown to be, where t is time and T is time variable of
integration.
Z t
d(T )
E(t T )
(t) =
dT
(6)
dT
0
4.6
Correspondence Principle
In order to mathematically model viscoelasticity and gain a better understanding of the physical relation between stress and strain, combination of
spring and dash-pots are used. Maxwell model consists of a purely elastic
8
Figure 4
spring and purely viscous damper in series. Kelvin model consist of a spring
and dashpot connected in parallel.
5.1
(7)
(8)
s = Es s
(9)
d = d d
(10)
For spring,
For dashpot,
Hence, the differential equation is given by,
= Es s + d d
(11)
Considering Creep loading i.e. step stress input, The Strain is given as,
t
0
(t) =
1 e
(12)
E
and the corresponding creep compliance is,
t
1
D (t) =
1 e
E
9
(13)
5.2
Standard linear Solid (SLS) model in the simplest model which can predict
both relaxation as well as damping phenomena effectively. The equilibirium
condition and kinematic conditions are,
= maxwell + 1
(14)
= maxwell = 1
Considering the Maxwell arm, the equilibrium and kinematic conditions are,
maxwell = s + d
(15a)
maxwell = s = d
(15b)
Differentiating 15a and substituting equations for spring 9 and dashpot 10,
the constituitive relationship for the maxwell element can be obtained,
.
.
+ =
E2
(16)
Using 16 and 14 the constitutive equation for the SLS is as given below,
1
E
.
.
+ = (E1 + E2 ) + 1
(17)
Where,
Relaxation time =
E2
a0 + a1 = b0 + b1
10
(18)
5.2.1
(19)
E1 it
= iw0 (E1 + E2 ) eit +
0 e
(20)
Solving the above differential equation and ignoring the transient terms, For
steady state oscillations stress as a function of loading frequency is given by,
2 2
iE2
it
(t) = 0 e
E1 + E2
+
(21)
1 + 2 2
1 + w2 2
Where;
2 2
t
Storage Modulus=E 0 : E1 + E2 ( 1+
2 2 )
Loss Modulus:E 00 =
iE2
1+w2 2
Complex Modulus: E = E 0 + iE 00
The storage modulus(Youngs modulus) is defined as the component in phase
of the applied strain while other component, Loss modulus is related to the
.
energy lost by the material and hence is out of phase by 90 deg (t) =
iw0 eit .From 21 it is clear the the stress lags behind strain by certain angle,
called loss angle . The storage modulus represents purely elastic response
of the material, while loss modulus represents purely viscous response which
is out of phase by 90 deg. Similar to Storage and Loss Modulus, the Storage
and Loss compliance can be shown to be
StorageCompliance, J 0 =
LossCompliance, J 00 =
11
E1 + 2 t2 (E1 + E2 )
EI2 + 2 2 (E1 + E2 )2
E2
E12 + 2 t2 (E1 + E2 )2
(22)
(23)
Storage and Loss modulus are directly related to the energy dissipation
of the material at certain loading frequency via previously defined quantity,
loss factor as follows,
LossF actor : () = tan () =
E 00
E0
(24)
Where,
:Loss angle or phase angle between stress and strain
The Steady state reponse of viscoelastic material to oscillatory strain is
given via 25
(t) = 0 eit E 0 + iE 00
(25)
(t) = e0 E eit
(26)
The above equation is for any general sinusoidal loading. For sin strain
input, the stress function can be extracted as follows,
= 0 sin (t)
(t) = Im 0 E eiwt
using Eulers Identity we get,
(t) = 0 Im E 0 + iE 00 [cos (wt) + isin (t)]
(t) = 0 E 0 sin (wt) + E 00 cos (t)
5.2.2
(27)
(28)
(29)
(30)
Using 21, storage modulus, loss modulus and loss factor can be plotted for
certain SLS solid at various frequency.
5.3
Wiechert Model
As seen from the spectrum of storage and loss modulus, a single Maxwell
element does not capture behaviour of real materials as the storage modulus
decays too rapdly. This can be overcome by using multiple Maxwell models
12
Figure 5
in the SLS, also called as Wiechert model. The solution for stress relaxation
and the relaxation modulus is given by,
!
n
X
t
(t) = 0 E +
(31)
Ei e i
i=1
n
X
Ei e i
(32)
i=1
The above series is called Prony series via which, experimental curve can
be fitted with many such maxwell models. An alternate form which relates
elastic modulus to equilibrium modulus is also useful,
E(0) = E +
n
X
Ei
(33)
t
Ei 1 e i
(34)
i=1
Therefore,
E(t) = E0
n
X
i=1
5.4
Viscoelasticity in ABAQUS
Since Abaqus FE is the primary FE tool used for this study, it is important
to understand how it handles viscoelasticity. Abaqus assumes time-domain
viscoelasticity to be defined by Prony series expansion. Prony series terms
can be indiviudally defined, or can be caluculated by Abaqus based on given
Creep test data, Relaxation test data or Frequency dependent data. It also
allows the definition of instantaneous or equilibrium youngs modulus. The
inpur parameters of prony series are given by normalized tems Abaqus uses
the time domain viscoelastic data only in the Dynamic Implicit analysis
step and frequency domain data in Steady State Direct dynamic analysis, subspace-based steady state, nautral and complex frequency extraction
steps. Dynamic implicit analysis step was chosen for this study to make the
anaylsis very general, allowing inclusion of study of heat generation due to
vibration in composite structures at a later stage. The prony series input for
abaqus is required in normalized form as follows from 34 for Shear modulus,
!
N
X
t
G(t) = G0 g +
gi e
(35)
i=1
Micromechnical Theory
Micromechanical analysis is essential in developing macroscopic constitutive equations for heterogeneous materials. Micromechanical analysis helps
to understand effect of fibre size, shape arrangement, volume fraction as
well as give information about the average and localized deformation. The
fundamental problem in micromechanical of heteogenous materials is the
determination of stress and strain concentration tensors averaged over the
the volume.
The average or homogenized stress and strain fields are extracted as
below,
Z
1
ij =
ij (x, y, z) dV
(36)
V V
14
Figure 6: Geometric representation of a)Statistically heterogenous microstructure charactirized by RVE b)Periodic Microstructure Characterized
by RUC
and,
1
ij =
V
Z
ij (x, y, z) dV
(37)
Figure 7
This enforces certain displament on every point on a boundary.
Static uniform boundary condition (Traction boundary condition):
The traction vector is prescribed at the boundary
tij = ij nj
x V
(39)
(40)
7
7.1
Aim of this study was to understand variation of damping and elastic modulus in a two phase composite with respect to volume fraction and geometrical
16
Figure 8
Material
Fibre
Matrix
E GPa
50
2.76
v
0.1
0.38
Density (kg/m3)
2580
1200
arrangement of fibres subject to loading at particular frequency and periodic boundary conditions. Only Axial and transverse damping and elastic
moduli were studied with an aim to extend to shear loading in future.
7.1.1
A FE model was created in Abaqus CAE with shown in ??. Modelling was
done with help of python scripts, via which the effect of volume fraction
could be modelled easily. The volume fractions of 0.2,0.4,0.6 and 0.7 were
studied for both square and hexagonal array. Glass fibre of diameter 10
was modelled as being elastic. A hypothetical material was considered as
matrix with a single term prony series.
As previously discussed, Abaqus requires normalized prony terms as input. Viscoelastic Property of matrix was input in time domain as follows
gi
0.6
ki
0.6
20
17
Volume Fraction
0.2
0.4
0.6
0.7
Square
1860
1680
1680
1680
Hexagonal
18700
7500
4200
3421
Figure 9
7.1.2
Mesh
8-node linear brick 3D stress elements were used for meshing. A mesh convergence study was performed and the required elements are given in the
table ??
7.1.3
A harmonic sin strain at a circular frequency of 0.02 rad/s with an amplitude of 1%, applied on Z+ and Y+ face (??) were considered in two separate
analysis. In each analysis, the Y- face was constrained in Y direction. Periodic Boundary conditions were employed in the X+ and X- face for each pair
of node opposite to each other. This was done via python script, given in
Appendix. Nodal reaction forces at the Z+(or Y+) faces were summed and
divided by the area to obtain the average stress response. Given the strain
response, the storage and loss modulus were calculated using the hysteresis
loop approach 10. Damping was obtained in terms of loss factor 24.
18
7.2
This resulted in average of 54, 92 and 130 fibres for 0.3,0.5 and 0.7 volume
fractions respectively. Four noded Plain strain elements were used for the
2D simulations. Based on mesh convergence study results, the global mesh
size was set at 1.2 and 39 nodes were created along the circumference of
each fibre. Also a minimum size control fraction was set at 0.01. This
resulted in 29869, 31250, 32295 elements for volume fractions of 0.3,0.5 and
0.7 respectively.
Two different realization of model at volume fractions of 0.3,0.5 and 0.7
are as given below.
7.3
Checkerboard Model
Explicitly modelling fibres is more accurate but computationally very expensive. A binary voxel model is proposed in which the RVE is discretized
into square grid, where each element represents either matrix or fibre. This
model is not expected to predict correct localizations due to square inclusions. However, the homogenized properties are of interest and would result
in substantial decrease in computational requirements. The matrix and fibres are considered perfectly bonded. RVE of 0.125mm was discretized into
square grid. Fibres were disperesed randomly using in the RVE based on
volume fraction. The model is as shown below,
At higher volume fraction, substantial clustering of fibres is evident
which doesnt reflect true geometry of the composite material.
8.1
Analytical results
Rules of Mixtures
20
(a) Vf=0.3
(b) Vf=0.3
(c) Vf=0.5
(d) Vf=0.5
(e) Vf=0.7
(f) Vf=0.7
Figure 11: 3 x 2
21
(a) Vf=0.3
(b) Vf=0.5
(c) Vf=0.7
= vf Ef + vm Em
Eaxial
(41)
(42)
Where,
0
00
Eaxial
= Eaxial
+ iEaxial
0
Eaxial
00
Eaxial
Etranverse
vf
vm
+
Ef
Em
(43)
22
8.1.2
KH
= Kf +
GL = Gm +
GH = Gf +
vm
1
Km Kf
3vf
3Kf +4Gf
vf
1
Gf Gm
vm
1
Gm Gf
(46)
(47)
(48)
Where subscripts L and H correspond to lower and higher bounds and m and
f subscripts refere to matrix and fibres respectively. Corresponding bounds
on complex Youngs modulus were obtain by using correspondence principle
9KG
and 3K+G
FE Results
The FE models were used to study the effect of volume fraction and fibre
arrangement on axial and transverse elastic modulus and damping. In this
section, the stress plots for yy and xy for square, hexagonal and random
array for different volume fractions are as given below are given below. The
FE results are then compared to Rules of mixtures and Hashin prediction.
, there is no significant influence of fibre arrangement
From figure
on the axial storage modulus. The FE results are in excellent agreement
with ROM and Hashin prediction. In the direction of fibres, the property of
composite is highly dependent on fibres even at low volume fractions. Since
the fibres are modelled as elastic, axial loss factor falls quickly to about 10 %
that of the matrix at volume fraction of 0.2. Negligible damping is expected
23
24
(a) Vf=0.4
(b) Vf=0.4
(c) Vf=0.6
(d) Vf=0.6
25
(a) Vf=0.4
(b) Vf=0.4
(c) Vf=0.6
(d) Vf=0.6
26
(a) Vf=0.3
(b) Vf=0.3
(c) Vf=0.5
(d) Vf=0.5
(e) Vf=0.7
(f) Vf=0.7
27
(a) Vf=0.3
(b) Vf=0.3
(c) Vf=0.5
(d) Vf=0.5
(e) Vf=0.7
(f) Vf=0.7
28
0
E11
Ef0
ROM
Hashin
Square
Hex
1
0.8
0.6
0.4
0.2
0
22
m
0.2
0.4
0.6
0.8
Volume Fraction
ROM
Hashin
Square
Hex
1
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Volume Fraction
29
0
E22
0
Em
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Volume Fraction
22
m
ROM
Hashin
Square
Hex
Random
Checkerboard
ROM
Hashin
Square
Hex
Checkerboard
1
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Volume Fraction
30