Sei sulla pagina 1di 10

COMPOSITES

SCIENCE AND
TECHNOLOGY
Composites Science and Technology 66 (2006) 27092718
www.elsevier.com/locate/compscitech

Mechanical properties of high density polyurethane foams: II Eect


of the ller size
Fabrice Saint-Michel, Laurent Chazeau *, Jean-Yves Cavaille
Groupe dEtude de Metallurgie Physique et Physique des Materiaux, UMR CNRS 5510, Bat. Blaise Pascal, 69621 Villeurbanne Cedex, France
Received 9 August 2005; received in revised form 4 March 2006; accepted 14 March 2006
Available online 5 May 2006

Abstract
This article presents the mechanical behaviour of rigid polyurethane reinforced by mineral llers, for a density range (qf/qs) >
0.27. The inuence of the particle size on the foam microstructure and its mechanical behaviour has been studied using calcium
carbonate with two dierent sizes (with average diameter of 1 and 30 lm, respectively) and crystallised silica particles with an
intermediate size. The microstructural characterisation has been performed by electron microscopy. It has shown closed spherical
cells whose size is decreased when the llers are added. The results of mechanical characterisation by large and small deformation
(mechanical spectrometry) tests have been compared to dierent modelling approaches. It appeared that a correct description of
the viscoelastic properties and of the yield stress needs to take into account the ller size compared to the wall size. Two modelling hypothesis which consider the composite foam as either a ller dispersion in a foam or a void dispersion in a lled polymer
have been made. The rst one is adapted to the description of PU lled with the big carbonate llers, when the second one is
better to describe the PU lled with the small carbonate llers. The use of the crystallized silica particles gives composite foams
with properties in between that of the two calcium carbonate composites, showing that the ller size in the range studied is a key
parameter to take into account in the reinforcement level.
 2006 Elsevier Ltd. All rights reserved.
Keywords: Polyurethane foam; A. Particle reinforced composites; B. Mechanical properties; B. Modelling

1. Introduction
Due to the presence of a skeleton made of cells (open or
closed) more or less regular [1,2] foams present very interesting mechanical properties such as a high capacity of
energy absorption, particularly useful for shock damping.
The mechanical response of these cellular materials depend
on the architecture and the intrinsic properties of the polymer constitutive of the cell walls [1]. The parameters of the
foam architecture are the wall width, the size distribution
and the shape of the cell.
For some years, their study have been extended to composite foams reinforced by the introduction of bres [3] or
*

Corresponding author.
E-mail address: laurent.chazeau@insa-lyon.fr (L. Chazeau).

0266-3538/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2006.03.008

metallic or mineral particles [47]. This modication of the


polymer matrix can lead to a better adsorption of the heat
released during the polymerisation reaction, an increase of
the matrix rigidity, with still a large capacity of energy
absorption, and an acceptable density. The granulometry,
the surface state, and the ller type are important parameters governing the mechanical properties. The foam architecture obviously depends on the ller dispersion, and the
wall width compared to the particle size. Works of Goods
et al. [5], have shown a more important increase of the elastic modulus when the foam density is higher. Moreover, the
mechanical properties of the foam being also governed by
the mechanical properties of the constitutive material, the
reinforcement level of the foam depends on the interaction
between the ller and the polymer matrix [4]. Eventually,
ller addition can lead to an embrittlement of the cell walls

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

and therefore favour their rupture instead of their


deformation.
Theoretical works on foam have mainly been devoted to
low density foams. The foam structure is then modelled as
a compact assembly of cells constituted of walls and struts.
The most famous modelling to predict the mechanical
behaviour of cellular materials is that of Gibson and Ashby
[1]. For higher density foams, these materials are made of
closed and isolated spherical cells, which lead to consider
them as porous materials. Then, the modelling can be performed using modelling developed for composite description. For instance, Siegmann et al. [7], have correctly
predicted elastic properties of polyurethane foam using
the Kerner equations [8] and the present authors [9] have
modelled polyurethane foams in the density range (0.3
0.85) using the Christensen and Lo [10] or Herve and Zaoui
equations [11].
The accounting for ller presence is a supplementary
diculty for the modelling of composite foam behaviour.
For this purpose, with lled foam in the density range
(0.30.6), Goods et al. [5] have combined the Gibson and
Ashby [1] modelling with the Kerner equation [8]. However, their determination of the PU modulus from the Gibson and Ashby modelling is questionable since the latter
involves a parameter which is not justied by the microstructural characterisation. Siegmann et al. [7] have also
studied foam in the same density range, lled with various
type of ller with size below 10 lm. They correctly modelled the linear behaviour of their materials using the Kerner equations [8] in two steps. However, they did not
provided microstructural characterisation of the foam
and no discussion was done about the possible inuence
of the ller size.
The aim of this article is to address this question by a
study of the mechanical behaviour of rigid polyurethane
foams reinforced by mineral particles with dierent sizes.
The llers mainly studied are calcium carbonate with
two dierent granulometries. Another ller type, crystallised silica, with an intermediate size is also used. The
morphology of the obtained foams are characterised by
scanning Electron Microscopy. Their mechanical characterisation associates large and small deformation tests.
The experimental data are then compared to dierent
modelling such as the Gibson and Ashby [1] and the
2 + 1 phases of Christensen and Lo [10] (or Herve and
Zaoui [11]) modelling which is modied to account for
the ller size eect. Like in a previous article [9], a
simulation of the non linear behaviour is also
proposed.

2. Materials and experiments


2.1. Materials
2.1.1. Composite PU foam formulation
The composite foams are formulated from a polyol
matrix made of polyols, catalysts, a surfactant and mineral
llers. This matrix is mainly made of two polypropylene
triols (polyols) supplied by Shell Chemicals. The silicone
surfactant is supplied by Air Products and is a PDMS/
POE copolymer. The two catalysts used are salts supplied
by Air Products. The catalyst A is a dibutyltin dilaurate
(DBTDL). The catalyst B is a 2,2 0 -[(dimethylstannylene)bis(thio)]diacetate. Their role is to accelerate the reaction of chain polymerisation and regulate the reaction
kinetic during the foam expansion.
The calcium carbonate llers C1 and C2 are supplied by
Provencale SA. They are made of crushed angular particles. Their size have been characterised by laser granulometry with a Malvern apparatus. The size distribution of the
C1 calcium carbonate shows a large polydispersity. The
distribution spreads from 1 up to 200 lm. The size distribution of the calcium carbonate C2 goes from 0.7 up to 20 lm
and presents a peak at 4 lm. One can describe these two
distributions with the diameters d10, d50 and d90, which represent the diameters for which x% in volume of particles
have a diameter below dx. These values, as well as the densities, the specic surfaces and the average diameter are
reported in Table 1. The crystallised silica, supplied by
La Provencale SA, has also been used in the frame of this
study. Its size distribution is very large (from 0.1 up to

700

(mW/g)

600

f /s
0.63
0.27

500
0.1
400

Tan

2710

300
200
-10

10

30

50
70
90
Temperature (C)

110

0.01
130

Fig. 1. Tests by dierential scanning calorimetry (a) and by DMA (b) for
foams reinforced by calcium carbonate. (a) T = 10 at 130 C,
DT = 10 C/min. (b) f = 0.1 Hz, T = 10 at 130 C, DT = 1 C/min.

Table 1
Physico-chemical properties, elastic modulus and Poisson coecient of mineral llers used
Mineral llers

Density (g/cm3)

Specic
surfaces (m2/g)

d10 (lm)

d50 (lm)

d90 (lm)

Average
diameter (lm)

Modulus (GPa)

Poisson coecient

Calcium carbonate C1
Calcium carbonate C2
Crystallised silica

2.7
2.7
2.35

0.4
3.3
1

8.2
1.8
0.3

44
4.4
10.4

108.1
9.9
37.2

52
5.2
15

14
14
30.8

0.27
0.27
0.17

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

100 lm) and bimodal with two peaks at 0.3 and 30 lm. It
can be considered as an intermediate between the calcium
carbonate C1 and the calcium carbonate C2.
The foams are obtained by mixing the polyol matrix
with a polyisocyanate, diphenylmethanediisocyanate-4-4 0
(so-called MDI). Its reaction with the polyols products urethane bonds [12]. The water introduced with the polyol
reacts with the MDI to produce carbon dioxide at the origin of the foam expansion.
2.1.2. Processing
The initial water concentration being the main parameter regulating the nal foam density, its estimation and its
control are extremely important. The components of the
polyol matrix (polyols, surfactant, catalysts) are mixed at
600 rpm during 2 min with a mixer (Rayneri) with rotating
molecular sieve for
blades. Then this mix is dried on a 4 A
48 h to eliminate the residual water. The mineral llers are
kept in a drying oven during 24 h to eliminate water
adsorbed on the ller surface. For calcium carbonate the
temperature of drying is 200 C. This temperature is only
100 C for the silica to avoid any modication of its surface
chemistry by a degradation of the silanol groups. A volume
fraction /cc of calcium carbonate or of crystallised silica /s
is incorporated in the polyol matrix. This suspension is
mixed at 1500 rpm for 10 min. All these mixings are performed under a nitrogen gas ux to avoid the introduction
of water via the air humidity. The samples are then
degassed under vacuum during 24 h to eliminate air bubbles introduced in the resin during the mixing.
Distilled water and the polyisocyanate are then added to
the polyol matrix. The mix (suspension/polyisocyanate) is
then stirred at 1500 rpm for 20 s, and is poured into a cylindrical mould. The foam is obtained after free expansion in
this open mould. The expansion phase lasts about 3 min,
then the gelation phase begins (polymerisation and crosslinking). The foam is then unmoulded after 30 min.
The block obtained is then cut in three pieces, two with
45 20 30 mm dimensions and one with 80 30 40 mm
dimensions. The measurement of the apparent density (Dr)
is performed from the measurement of the volume and the
mass of this three blocks. Dr is dened as the ratio of the
foam density qf to that of the solid qs. This relative density
expresses the volume fraction of matter in the foam. It was
impossible to elaborate a material without bubbles and
therefore to measure qs. Indeed, the mixing of the suspension with polyisocyanate introduces bubbles which are
imprisoned during the polymerisation. qs has been estimated by calculation, considering the density of the dierent ingredients of the polyurethane foam, their quantity,
and a mixing law. Taking a volume fraction of calcium carbonate /cc equal to 21.6% (in the polymer) the calculated
density qs of the lled polyurethane is 1467 g l1. For the
polyurethane lled with 21.6 vol% of crystallised silica /s,
it is found qs = 1393 g l1. Then, the relative densities of
the dierent lled foams obtained can be calculated. They
vary from 0.27 up to 0.75.

2711

2.2. Experiments
2.2.1. Dierential scanning calorimetry (DSC)
The apparatus to characterise the glass transition temperature of the samples is a Pyris Diamond from Perkin
Elmer. To erase the thermal history of the samples, a rst
temperature ramp is applied with a heating rate of 10 C/
min from 10 up to 130 C. After cooling, a second ramp
is performed in the same conditions.
2.2.2. Electron microscopy
The characterisation of the cellular microstructure and
of the cell size of the polyurethane foam is performed with
a scanning electron microscope (SEM) JEOL 840 ALGS,
with a lament tension of 10 kV. First, the samples are broken in liquid nitrogen. Then the fracture surfaces are gold
coated to make them conducting. The micrographs
obtained are treated by image analysis with the software
Scion Image to quantify the size distribution of the voids
in the foam. The apparent pore diameters are then determined with the assumption of their sphericity. The Saltikov
method [13] enables the determination of the distribution
of the real diameter of the voids.
2.2.3. Mechanical tests
Preliminary tests on unlled foams have shown an eect
of physical ageing on their mechanical properties [9]. That
is why a thermal treatment has been performed to be sure
that the thermal history of each sample is the same. Before
any mechanical test, the samples are kept 10 min at 80 C
(rejuvenation above Tg) then 24 h at 25 C.
The elastic (G 0 ), and viscous (G00 ) moduli, and tan /
have been measured by dynamic mechanical analysis on
an apparatus Mecanalyser (Metravib S.A.) developed in
the laboratory [14]. It is an inverted torsion pendulum
working in forced harmonic regime, at low frequency,
from 105 up to 5 Hz, in a temperature range from 170
up to 400 C. The samples tested are parallelepipeds of
15 mm length, 5 mm width and 1.5 mm thickness. They
are submitted to sinusoidal torsion, with controlled deformation at a frequency f = 0.1 Hz. A temperature ramp is
performed during the test from 6 up to 130 C with a
heating rate of 1 C/min. Preliminary tests have shown
that humidity does not aect the mechanical properties
of the foams.
Uniaxial compression tests have been performed on a
tensile test machine Instron 8510. The compression plateaus are maintained parallel by a guidance device with ball
bearings. This machine possesses a thermo-regulated chamber to perform tests at temperature in between 90 and
230 C. The tested samples are cylinders with 10 mm diameter and 20 mm height (a compromise to avoid buckling
and barrel deformation) [15]. The samples are carefully
prepared so that their contact surfaces be strictly parallel.
To decrease friction, they are polished and a lubricant is
used (molybdenum bisulphur). Tests have been performed
at 30 C with a constant crosshead speed corresponding to

2712

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

an initial strain rate of 103 s1. The results are presented


in nominal stress and strain.
3. Results and discussions
3.1. Physical characterisation
The foams have been analysed by DSC. For the unlled
foams, we had measured a glass transition temperature (Tg)
equal to 74 2 C [9]. As illustrated by the Fig. 1 for foams
containing 21.6 vol% calcium carbonate C1, Tg is still equal
to 74 2 C and is therefore not modied by the ller
addition.
Moreover, measurements performed by Dynamic
Mechanical Analysis (Fig. 1) show a similar temperature
of the main relaxation, measured at the maximum of the
tan /, equal to 83 C whatever the density. In the case of
unlled foam with low density, (qf/qs = 0.33) a supplementary very wide relaxation had been detected at around
10 C [9]. The occurrence of this secondary relaxation has
been attributed to the formation of domains made of reaction sub-products. Indeed, urea, formed during the reac-

tion between the water and the isocyanate can then react
with the isocyanate to form ureas. However, though the
existence of this supplementary relaxation, the very good
modelling of the viscoelastic behaviour of the unlled foam
with the assumption of no modication of the matrix indicated that this slight modication is negligible. For lled
foams, this secondary relaxation is not observed. The introduction of llers enables a better absorption of the heat
released during the polymerisation reaction. This could
induce a decrease of the kinetic of formation of these byproducts which normally occur at temperature above
100 C. From these results we can conclude that the dierent foams processed have a similar polymer matrix.
3.2. Morphological characterisation
Fig. 2 presents an example of SEM micrographs
obtained for foams reinforced by both calcium carbonates
C1 and C2 and by the crystallised silica respectively. In the
density range studied (0.27 < q*/qs < 0.75), the cells basically appear spherical and closed. Whatever the ller type
and the foam densities, particles are correctly dispersed.

Fig. 2. SEM micrographs of foams reinforced by (a) calcium carbonate C1 (qf/qs = 0.31, 70), (b) calcium carbonate C1 (qf/qs = 0.51, 300), (c) calcium
carbonate C2 (qf/qs = 0.51, 1000), (d) crystallized silica. (qf/qs = 0.65, 300).

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

3.3. Mechanical properties


The evolution of the G 0 modulus as a function of temperature has been measured by DMA on the dierent samples (Fig. 4). For a better clarity of the gures, only two
Average real diameter (m)

140
with C2
with C1
with silica
unfilled

120
100
80
60
40
20
0
0.3

G''

0 .4

0.5

0 .6
f / s

0.7

0 .8

0.9

Fig. 3. Average real diameter of the cells foams as a function of relative


density.

106
105

G'

0.62
0.27
10

104

Assumption 1
30

50
70
90
Temperature (C)

103
130

110

(b) Foams with C2

108

1010

G' '
10

10

107
106

/
f

107
106
-10

0.75
0.35
10

105

G'' (MPa)

106
-10

G'' (MPa)

107

108
107

where D is the foam relative density and A and B are two


adjustable parameters expressed in microns. By this linear
least square regression method, A is found equal to 76,
100, and 52 and B is found equal to 83, 88 and 59 for
the C2 lled PU, the C1 lled PU and the silica lled
PU, respectively. Barma et al. [4] have also observed such
a decrease in the case of silica llers. Conversely, Goods
et al. [5] and Siegmann et al. [7] did not observe it. Goods
et al. [5] have studied alumina lled foams. Three types of
ller have been studied by Siegmann et al. [7]: graphite particles, glass bres and glass beads. Note that in all these
works, the ller concentration, expressed in mass fraction,
is below 30% , that means volume fraction below that of
our study. Moreover, their processing conditions are dierent, such as the mixing time of the lled suspension with
the isocyanate, the mould size, the ingredients, on which
the cellular structure is very dependent. In our case, the decrease of the cell size is probably due to a nucleating eect
of the llers. Moreover, this eect can be enhanced by the
residual water at the particle surface which induces a supplementary CO2 release. However, conversely to our expectation, the particle size does not inuence the cell size. This
suggests that the nucleating eect, if it exists, is astonishingly not proportional to the particle surface. The authors
have presently no explanation for this.

108

109

G' (MPa)

L A  D B;

(a) Foams with C1

1010

G' (MPa)

The apparent cell size distribution as a function of the foam


density was evaluated from image analysis. A calculation
by the Saltikov method [13] allowed to deduce the real size
distributions (Fig. 3). These results have been compared
with those obtained with unlled foams [9]. The ller addition does not induce any modication of the cell morphology but a decrease of its size. In the density range studied,
the cell size L, expressed in microns can be roughly tted by
the linear equation:

2713

G'
104

Assumption 2
30

50
70
90
Temperature (C)

110

103
130

Fig. 4. Evolution of elastic modulus (G 0 ) and viscous modulus (G 0 ) as a


function of temperature obtained by DMA for foams reinforced by:
(a) calcium carbonate and (b) calcium carbonate C2. The values on the
curves represents the relative density. The lines represents the predicted
values with the 2 + 1 phases modelling [10]. f = 0.1 Hz, T = 10 at
130 C, DT = 1 C/min.

densities are presented on each one. The curves with dierent densities seem parallel and only shifted by a factor (in
logarithmic scale). The G 0 modulus measured at 30 C as
a function of the relative density is reported on the
Fig. 5. To evaluate the inuence of the ller addition,
experimental data obtained with the unlled foams [9] are
also reported. Whatever the ller type, one can observe a
decrease of the G 0 modulus with the decrease of the density.
For C1 lled foams with relative density below 0.5, the
addition of llers does not inuence the modulus value.
The increase of the modulus with the ller addition is only
visible for higher density, but this increase is relatively low.
Conversely, the C2 lled foams present G 0 values above
those of the unlled foams and the C1 lled foams, whatever the density. The silica lled foams with relative density
below 0.5 presents G 0 values in between those of the C1
lled foams and those of the C2 lled foams. For higher
density, their G 0 values tends toward those of the C2 lled
foams. These experimental results clearly show an eect of
the ller size on the elastic behaviour of the lled foams:
the modulus values increase when the average size of the
llers decreases.
Figs. 6a,b and 7 present the compression curves of PU
foams of dierent density lled with calcium carbonate

2714

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

(a) 1000

(a) 35

with C2
with C1
unfilled
with silica

30

with
= 1.03

25

with = 1.18

600
400

Stress (MPa)

G' (MPa)

800

with = 0.89

200

0.31

20

0.62

15

0.5

10

0.42

0
0.3

0.4

0.5

0.6

0.7

0.8

0
0

0.9

0.1

0.2

f / s

with C2
with C1
unfilled
with silica

0.7

0.52 0.42

0.35

25

Assumption
(2)

400

Assumption
(1)

200
0
0.3

0.6

30

Stress (MPa)

G' (MPa)

800

0.3 0.4
0.5
Nominal strain
0.75

(b) 35

(b) 1000

600

0.27

20
15
10
5

0.4

0.5

0.6

0.7

0.8

0.9

f / s
Fig. 5. Evolution of elastic modulus (G 0 ) as a function of relative density
obtained by DMA. Unlled and reinforced foams. (a) The lines represent
the predicted values with the Gibson and Ashby modelling [1]. (b) The
lines represents the predicted values with the 2 + 1 phase modelling [10].
f = 0.1 Hz, T = 30 C.

C1, calcium carbonate C2 and crystallised silica respectively. They are typical of that of a PU foam, with an elastic domain, followed by a densication plateau, a
hardening domain and a nal rupture. It is noticeable that
the plateau occurs at constant strain level around 0.4. The
C2 lled foam appeared to be more brittle, as shown by
instabilities on the curve (Fig. 6b). All the curves show,
at a given strain level, a decrease of the stress value with
the decrease of density. Compression elastic modulus
of the dierent samples can be deduced from the slope at
the origin of their compression curve. However, due to
the foam nature of the material, the error made to estimate
it is large. In fact this leads to an underestimation of the
real value. The important result is that they show the same
trend as the results with G 0 and lead to the same comment
on the inuence of the ller size. The yield stress (rf) of the
dierent lled foams can be more precisely estimated. Their
values are plotted as a function of the density for the dierent llers on Fig. 8. They increase with the density. For
density below 0.6, the addition of C1 ller does not induce
any signicant change of the yield stress value. However,

0
0

0.1

0.2

0.3 0.4
0.5
Nominal strain

0.6

0.7

Fig. 6. Compression tests for dierent relative density lled foams:


(a) foams reinforced by calcium carbonate C1 (b) foams reinforced by
calcium carbonate C2. The values on the curves represents the relative
density. The dash lines represents the predicted values with the 2 + 1
phases modelling [10]. Initial strain rate = 103 s1, T = 30 C.

one can observe that the addition of C2 calcium carbonate


or of crystallised silica induces an increase of the rf value
for foam density above 0.5.
3.4. Modelling of the mechanical behaviour
The most common modelling to predict the mechanical
behaviour of cellular materials is the Gibson and Ashby
approach [1]. In this modelling, the foam is described as
an arrangement of cubic cells made of struts and walls. It
leads to a simple expression of the elastic modulus Gf of
the foam as a function of its density (qf/qs), the solid fraction in the struts / and the mechanical properties of the
constitutive material (Gs), (in the present case the modulus
of the lled polyurethane):
 2
Gf
q
2 qf
/
1  / f .
2
Gs
qs
qs
To obtain this equation, Gibson and Ashby [1] made the
assumption that the Poisson coecient is constant whatever the density (mf = 1/3). For closed cell foam, / is in

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718


0.65

(a) 35

0.55

0.43 0.36

30

Stress (MPa)

25
20
15
10
5
0
0

(b)

0.1

0.2

0.3 0.4
0.5
Nominal strain
0.65

35

0.55

0.6

0.7

0.43 0.36

30

Stress (MPa)

25
20
15
10
5
0
0

0.1

0.2

0.3 0.4
0.5
Nominal strain

0.6

0.7

Fig. 7. Compression tests for dierent relative density lled foams


reinforced by crystallized silica. The values on the curves represents the
relative density. The dash lines represents the predicted values with the
2 + 1 phases modelling [10]: (a) assumption 1; (b) assumption 2. Initial
strain rate = 103 s1, T = 30 C.

between 0.6 and 0.9. For open cell foam, / is equal to 1.


This modelling is normally adapted to foam with relative
density below 0.4.

An alternative approach considers the polyurethane


foam as a composite made of void inclusions in a polymer
matrix. Using the 2 + 1 phases from Herve and Zaoui [11],
it is possible to predict the elastic modulus and the yield
stress of unlled foam as a function of the density [9]. This
approach models the material from a representative elementary volume made of a void inclusion embedded in a
shell of the polymer matrix, itself embedded in the
unknown eective medium. In the case of calcium carbonate lled foams, the modelling must account for the ller
presence. Moreover, the eect of the experimentally
observed ller size must also be considered. Two dierent
assumptions, depicted on Fig. 9, have been tested to
account for this eect. The rst assumption (1) depicts
the lled foam as a ller dispersion in a foam then considered as an homogenised medium. The ller volume fraction
is related to the whole material, and has to be recalculated
as a function of the foam density. This leads to ller fraction values which are lower and lower when the density
decreases. In this modelling, the rst step is the calculation
of the void fraction (the inclusions) in the material made of
the unlled foam (the matrix), and of its modulus. The second step is then the calculation of the modulus of this foam
(the matrix) reinforced by the llers (the inclusions). The
second assumption (2) considers the composite foam as a
void dispersion in a lled polymer. In this case, the ller
fraction is that in the polymer. It is constant and equal
to that initially calculated, 21.4%. The rst step calculates
the lled PU modulus (the llers are the inclusions and
the matrix is the PU), then the second step consists in calculating the modulus of this material (the matrix) lled wit
voids (the inclusions).
The application of the equivalence principle of Dickie
[16], used later by Hashin for the self-consistent scheme
[17] , enables the extension of this modelling to the description of the viscoelastic behaviour. It can be simply done by
replacing the elastic terms in the equations by their complex counterpart.
Assumption
(1)

35
with C2

30

with C1

25

with silica

PU matrix

Step 1

f (MPa)

unfilled

20
15

2715

Assumption
(2)

Void

Homogeneous medium

Assumption
(2)
PU matrix
Filler

Homogeneous medium

10
Assumption
(1)

PU foam

Step 2
0
0.3

0.4

0.5

0.6
0.7
f / s

0.8

0.9

Fig. 8. Evolution of the stress value at the beginning of the plateau (rf) as
a function of relative density obtained by compression tests. Unlled and
reinforced foams. The line represents the predicted values with the 2 + 1
phases modelling [10]. Initial strain rate = 103 s1, T = 30 C.

Filler

Homogeneous medium

Filler dispersion in a
foam

Filled PU
Void

Homogeneous medium

Void dispersion in
a filled polymer

Fig. 9. Schematic representation of homogenisation method used for the


2 + 1 phases modelling [10]. Assumptions and steps used for the modeling
of the lled foams.

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

3.4.1. Linear domain


It was impossible to process polyurethane without bubbles and therefore to measure the mechanical property of
the constitutive polymer. The shear modulus of the polyurethane polymer has been estimated from the dilute
sphere modelling, which is equivalent to the 2 + 1 phase
modelling at low inclusion fraction and easier to invert
[9]. The deduced value from the modulus of the 0.85 density foam was found equal to Gs = 920 MPa. The value of
the elastic shear modulus of the calcium carbonate lled
polymer has then been calculated from the 2 + 1 phases
modelling, taking for the ller shear modulus and Poisson
coecient, 14 Gpa and 0.27, respectively. It was found
Gs = 1400 MPa and ms = 0.3. Taking 30.8 GPa and 0.17
for the crystallised silica shear modulus and Poisson coefcient, the silica lled polymer modulus has been found
equal to 1455 Mpa, and its Poisson coecient equal to
0.3.
These values have been introduced in the Gibson and
Ashby equation [1] to calculate the composite foam modulus. The elastic shear modulus of the C1 and C2 lled
foams as a function of the relative density is correctly predicted by the Gibson and Ashby [1] equation when / is
equal to 1.03 and 1.18, respectively (Fig. 5a). For a closed
cell foam, the / value is generally between 0.6 and 0.9 [1].
When / = 1, the foam is made of open cells. However,
electron microscope observations show closed cell foam
for all the relative density range studied (0.27 < qf/
qs < 0.75). Thus, as for the study of unlled foam [9]),
the / parameter has no clear signicance and must be considered as an adjustable parameter. For instance, it is
found / = 1.07 for the foam containing the calcium carbonate C2 when the density range considered is reduced
between 0.35 and 0.42. Therefore, it can be concluded that
this modelling is not adapted in the case of our material.
The modulus values of the composite foam calculated
from the 2 + 1 phase modelling applied in two steps are
reported on Fig. 5b. For the C1 lled foams, the moduli
are more correctly evaluated when the assumption (1) is
chosen. Conversely, the moduli of the C2 lled foams are
more properly evaluated when the assumption (2) is made.
However, in the latter case, one can see that the experimental data tends to diverge from the theoretical t and to
become closer to the theoretical calculation with the
assumption (1). The same gure presents theoretical data
with assumption (1) and assumption (2), and experimental
modulus of the silica lled foam (the dash lines). In the
density range studied, the experimental moduli are in
the transition zone between the assumption (1) t and the
assumption (2) t.
It is possible to estimate the ller fraction value at which
the particle size is equal to the width of the wall. With the
approximation of centred cubic arrangement of the voids
into the foam, using the Eq. (1) giving the particle size as
function of the relative density, the width of the cell wall
e (taken as the smallest distance between particles and
expressed in microns) can be written as:

"
e

p
31  D

#
13 p
3
 1 AD B.
2

The calculated cell wall as a function of the density is presented in Fig. 10 for the dierent ller types. This can be
compared to the experimental average particle size of the
dierent llers. As expected, in the density range studied,
the C1 llers are in average much bigger then the cell wall
thickness of the C1 lled foams. Conversely, the cell wall
thickness of the C2 lled foams are in the same range than
the calculated cell wall thickness. This explains why one
can observe for the modulus value of these foams the
beginning of the transition from a value predicted with
assumption 2 to one predicted with assumption 1. Concerning the crystallised silica, the calculation is less convincing since the average particle size is above that of the
calculated wall thickness. However, one must notice that
the silica size distribution present two peaks, whose one
is below 1 lm, i.e. well below the calculated wall thickness.
This probably explains the modulus evolution of the silica
lled foam as function of the density, which is always in between the predictions with the two dierent assumptions.
The modelling has been extended to the viscoelastic
behaviour. Fig. 4 compares experimental data and theoretical curves and shows that the previous modelling, suitable
in the glassy domain does not correctly predict the G 0
(Fig. 4a) and G00 (Fig. 4b) values at temperature above
Tg. It largely underestimates the moduli in the rubbery
domain. However, it was previously shown [9] that this
modelling correctly predicts the G 0 modulus of the unlled
foams in the whole density range and in the whole temperature range studied (from 10 up to 130 C). One assumption to explain the underestimation observed in the case of
lled foams is that the introduction of ller leads to an
increase of the resin crosslink density: an inuence of the
ller in the foam processing is already evidenced by the
change of size distribution of the bubbles, as discussed previously. Moreover, CaCO3 ller surface is highly hydrophilic and probably absorbs a large part of the water
added in the formulation. This might modify the complex

14

Calculated
average cell width (m)

2716

12
10
8
6
4
2
0
0.3

0.4

0.5

0.6
0.7
f / s

0. 8

0.9

Fig. 10. Calculated width of the cell walls as a function of the density, for
the C1 lled foams (- - -), the C2 lled foams ( - -), and the silica lled
foams ().

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

kinetic balance between the chemical reactions involved in


the CO2 release (and therefore the bubble formation) and
those involved in the resin polymerisation. Of course, this
crosslink density modication is only detectable above
the glass transition temperature (when the crosslink density
is the key parameter of the modulus level).
3.4.2. Non linear domain
Gibson et Ashby [1] have linked the foam yield stress
(rf) to the yield stress of the constitutive material (rs) with
the assumption that the deformation is the result of the
buckling of the struts:

3
rf
q 2
q
 0:3 / f 0:41  / f .
4
rs
qs
qs
This assumption of buckling is justied in the present study
because of experimental results of tests of deformation
recovery. After heating treatment above the glass transition
temperature of the foam, a total recovery of compressed C1
lled polyurethane foam has been observed for deformation level below 0.5. However, like in the linear domain,
the use of such modelling with our material has been disappointing. In particular, the / parameter has to be adjusted
without physical justication, and the yield stress rs deduced from the same t is much higher than the experimental one.
In the case of composite materials, the mechanical
modelling in the nonlinear domain, are often the extension
of the modelling used in the linear domain [18,19]. Besides
the choice of the most appropriate modelling to describe
the linear domain, the choice of a linearization method is
needed. This passage between linearity and non linearity
can be done using the tangent modulus, the secant modulus, or the ane method along the compression curves
[20]. In this manner, the problem to solve is replaced by
a symbolic problem of heterogeneous elasticity with initial
deformation. In Ref. [10], the authors tested the secant
method, which did not provide a correct modelling of the
non-linear behaviour. However, in the same article, it
was found that a phenomenological approach based on
the 2 + 1 phase modelling correctly simulates the behaviour of the unlled foams. The more general question of
the modelling of PU foam in the non-linear domain is
not in the frame of the present article and will be addressed
in future publication. However, one can use this phenomenological approach as a discussion basis of the experimental results.
In this simulation, the assumption is made that the
secant modulus of the foam at a given strain is linked to
the secant modulus of the polyurethane matrix at the same
strain via the 2 + 1 phase modelling. One writes:


rf
rm qf
F
;
5
ef em
em qs
with F the 2 + 1 phase modelling applied to the matrix
modulus at the strain level em for a foam with a relative
density qf/qs. The compression curve of the unlled poly-

2717

urethane is that calculated from the study of the unlled


polyurethane foam in Ref. [10].
The rst assumption (1) considering the lled foams as a
ller dispersion in a foam, the rst step consists in calculating for each composite materials the compression curve of
the foamed polymer matrix, with its void fraction related to
the material without the llers. This uses Eq. (5). The second step calculates the compression curve of the lled foam
from this calculated curve, with the ller fraction related to
the whole material using the following equation:


rfc
rf
F0
;c
6
efc ef 1  c
ef
with F 0 the 2 + 1 phase modelling applied to the foam modulus at the strain level ef for a composite foam with a ller
volume fraction c, and a ller modulus and Poisson coecient taken from Table 1. Note that the non deformability
of the ller is assumed by writing that the composite foam
deformation is (1  c) times the deformation of the foam.
For easier calculation, the Poisson coecient has been taken constant whatever the deformation since it has been
checked that its value m, when between 0.35 and 0.4999,
has no signicant inuence on the results.
The second assumption (2) considers the composite
foam as a bubble dispersion in a lled polyurethane. In
the rst step, the compression curve of the lled matrix is
calculated from the unlled polyurethane compression
curve and the 2 + 1 phases model. The inclusion is the ller
and the assumption is again used that the llers do not participate to the composite deformation. The inclusion fraction is the ller fraction in the polymer and is equal to
21.6%. From the compression curve of this lled polyurethane and the 2 + 1 phase model, the stress-strain curves
of the lled foam are then deduced with the voids considered as the inclusion (Eq. (5)).
The results of these calculations with the assumption
(1) for all the densities studied are compared to the C1
lled foam stressstrain curve. As shown on Fig. 6a,
for strain level above 0.4, the modelling overestimates
the stress level. Such an overestimation was also
observed when modelling the unlled foam stressstrain
curves [9]. However, there is a good agreement between
experimental yield stress and calculated ones (Fig. 8).
For C2 lled foams, conversely to the results in the linear domain, the modelling overestimates the stress level
whatever the density range (Figs. 6b and 8). The yield
stress is in fact more properly described with the assumption (1). These results might be explained by the important brittleness of the C2 lled foam that is not taken
into account in the simulation approach. Indeed, this
brittleness has been observed during the compression
tests. In the C1 materials, some struts in the foam architecture do not contain ller (as shown on Fig. 2). Therefore, some plastic strain localisations are more likely to
occur. This explains their better toughness compared to
the C2 lled foams, in which all the struts can be embrittled by the llers they contain.

2718

F. Saint-Michel et al. / Composites Science and Technology 66 (2006) 27092718

As expected, the modelling of the silica lled foams leads


to curves, which, either overestimate or underestimate the
experimental stress-strain curves, depending on the model
assumptions.

ratory, in the frame of a GDR solid foam [21] and will be


the topic of future publications [22].

4. Conclusion

The authors acknowledge S. Youssef for his contribution to the experimental work and G. Vigier for fruitfull
discussions.

In the density range studied, the viscoelastic properties


of polyurethane foam reinforced by a 21.6% volume fraction of mineral llers such as calcium carbonate and crystallised silica can be correctly simulated as far as the
modelling takes into account the ller granulometry.
Indeed, depending on the particle dimension compared to
the average wall thickness, these composite foams can be
modelled with the following assumptions:
(i) when the llers have a large size, these foams are considered as a ller dispersion in an homogeneised medium with the foam characteristics.
(ii) When the llers have small sizes compared to the wall
thickness, the composite foams are considered as a
bubble dispersion embedded in a lled polyurethane.
From a practical point of view, the results shows that
the reinforcement of foam (in the non linear domain) is
not ecient if the ller added are bigger than the bubble
size. Moreover the comparison with the model above Tg
suggests that the llers lead to an increase of the crosslink
density of the matrix.
In the non-linear domain, in spite of its roughness, the
method proposed leads to a correct simulation of the curve
of the C1 lled foams for deformation below 0.5. Moreover, it enables to dierentiate the silica and the C2 lled
foams from the C1 lled materials. However a deeper
experimental and theoretical work is needed to correctly
simulate the large deformation behaviour. For instance,
mechanisms of deformation localisation, as well as the
development of the microstructural anisotropy during the
deformation and the evolution of the void fraction have
to be addressed and taken into account in the mechanical
description. These points are currently studied in the labo-

Acknowledgements

References
[1] Gibson LJ, Ashby MF. Cellular solids: structure and properties.
Oxford: Pergamon Press; 1988.
[2] Schwartz D, Shi D, Evans A, Wadley H. MRS symposium San
Francisco 1998:521.
[3] Cotgreave TC, Shortall JB. J Mater Sci 1977;12:708.
[4] Barma P, Rhodes MB, Salovery R. J Appl Phys 1978;49:4985.
[5] Goods SH, Neuschwanger CL, Whinnery LL, Nix WD. J Appl
Polym Sci 1999;74:2724.
[6] Vaidya NY, Khakhar DV. J Cell Plast 1997;33:587.
[7] Siegmann A, Kenig S, Alperstein D, Narkis M. Polym Compos
1983;4:113.
[8] Kerner HE. Proc Phys Soc 1956;B69:808.
[9] Saint-Michel F, Chazeau L, Cavaille JY, Chabert E. Comp Sci
Technol [in press].
[10] Christensen RM, Lo KH. J Mech Phys Solids 1979;27:315;
Erratum Christensen RM, Lo KH. J Mech Phys Solids 1986;34:639.
[11] Herve E, Zaoui A. Int J Eng Sci 1973;31:1 (1993).
[12] Dieterich D, Uhlig K. Ullmanns Encyclopedia Indus Chem
1992;A21:665.
[13] Saltikov SA. Stereometric metallography. Moscou: Mettallurgizdat;
1958.
[14] Cavaille JY, Salvia M, Merzeau P. Spectra 2000 1988;16:37.
[15] GSell C, Jonas JJ. J Mater Sci 1981;16:1956.
[16] Dickie RA. J Appl Polym Sci 1973;17:65.
[17] Hashin Z. J Appl Polym Sci 1983;50:481.
[18] Shaterzadeh M. Etude et modelisation physique et mecanique du
comportement viscoelastique et plastique de composites particulaires
a` matrice polyme`re. The`se presente a` lINSA de Lyon; 1997.
[19] Chazeau L, Cavaille JY, Perez J. J polym Sci PartB Polym Phys
2000;38:383.
[20] Zaoui A., Plasticite: approches en champ moyen. Homogeneisation
en mecanique des materiaux 2: comportement non lineaires et
proble`me ouverts, Hermes Science publication; 2001.
[21] GDR, Research Group on Solid Foam, SPI_SC CNRS, contact
Gerard.Vigier@insa-lyon.fr.
[22] Youssef S, Maire E, Gaertner R. Acta Mater 2005;53(3):719.

Potrebbero piacerti anche