Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Modelling the Gas-loss Creep Mechanism in

EVA Foam from Running Shoes

N.J. Mills and M.A. Rodriguez-Perez*

School of Metallurgy and Materials, University of Birmingham, UK
*Department of Solid State Physics, University of Valladolid, Spain

Received: 5th February 2001; Accepted: 15th March 2001

SUMMARY
Compressive creep experiments were performed on an EVA foam from a
running shoe midsole of density 275 kg m-3. Analysis revealed a 50% air loss
from the cells on a time scale of 10 hours, and a significant polymer contribution
to the creep response. The recovery process after creep is slow and incomplete,
perhaps due permanent deformation or polymer recrystallisation. SEM shows
that surface cells are ‘permanently’ compressed. Modelling was performed of
gas diffusion perpendicular to the stress axis, to predict the creep curves, and
along the stress axis to predict the development of a compressed, gas-depleted
surface layer. The latter considers the interactions of cell strain, diffusivity, stress
sharing between the polymer and cell gas, and strain rate. The number of gas-
depleted cells is predicted to increase with the square root of the creep time.

INTRODUCTION
The majority of running shoes (trainers) have a midsole compression-
moulded from poly ethylene - co-vinyl acetate (EVA) foam. The reasons
for selecting EVA foam are probably related to the processing technology.
Steam-heated presses, for compression moulding rubber shoe soles, were
adapted to compression mould EVA foam, which crosslinks in the mould
at similar temperatures to rubbers. The foam density varies from 200 to
300 kg m-3, with higher densities being used in regions intended to give
extra support. The natural white colour of foam allows the incorporation
of pigments. Mineral additives may enhance the white colour. The mid-
sole does not need to be as wear resistant as the outsole, but it must bond
the shoe components together; the crosslinking of EVA, and its polar
nature, provides a strong adhesive bond to the other materials.
Shorten(1) investigated the dynamic compression of a midsole; his
pressure maps have peak pressures the order of 400 kPa. Hennig and
Milani(2) measured the pressure distribution in 20 types of shoes for
athletes running at a marathon pace of 3.3 m/s. For some shoes the peak
pressure reached 1 MPa, but more typically it was 500 kPa. The response

Cellular Polymers, Vol. 20, No. 2, 2001 79

 N.J. Mills and M.A. Rodriguez-Perez

of EVA foam deteriorates in a sequence of impacts. Misevich and

Cavanagh(3) repeated loaded an EVA foam at a frequency of 1 Hz with an
Instron, to a constant peak compressive stress of 1.15 MPa. The
curvature of the foam force-compression relationship increased with the
number of cycles, and the input energy decreased linearly with the
logarithm of the number of cycles. Kinoshita and Bates(4) measured the
temperature of midsoles of shoes worn by Japanese runners, and found
that it rose to a plateau value after 20 minutes of running. The foam was
heated by heat transfer from the hot tarmac road surface and the foot, and
hysteresis losses.
The crystalline phase in the EVA foam completes its melting at
Tm ≅ 70°C, with partial melting and recrystallisation at lower temperatures,
so its mechanical properties change as Tm is approached. Misevich and
Cavanagh(3) state that gas diffusion, from foam cells near the surface of
EVA foam, is observed several minutes after the foam is place under high
pressure. They attribute most of the reduction in performance to the
formation of preferred creases or hinges in the cell walls after a few
compressive cycles. Energy losses were also attributed to adiabatic
heating of the cell gas, and to lateral foam bulging under compression.
Bartlett(5) showed SEM micrographs of EVA foam midsoles after 3000
km of running and suggested that the flattened cells near the outsole had
suffered compression set (a phenomenon resulting from long term
constant compressive strain). He indicated that some cell walls fracture
and buckle.
The aims of this research were to characterise the EVA foam
microstructure, to identify the mechanism(s) for the changes in the EVA
foam midsole compressive response, and to model this mechanism.

MODELLING

The uniaxial compressive response

Tordon(6) measured the impact compression response of foam he
described as EVA65. Our inspection of his foam revealed that it was a
crosslinked EVA foam, produced by the Zotefoams process, with density
65 kg m-3. He computed both the isothermal and adiabatic response of
the air in the cells. The isothermal analysis was Gibson and Ashby’s(7)
slight modification of Rusch(8) analysis. They give the equation
p0 ε
σ = σ0 + (1)
1– ε – R
p0 is the absolute gas pressure in the undeformed foam cells (usually

80 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

atmospheric pressure), ε is the compressive strain, and R the foam relative

density. The assumptions were zero lateral expansion of the foam,
isothermal gas compression, and an incompressible polymer. The polymer
contribution σ0 to the yield stress is assumed to be constant, irrespective
of the deformation mechanisms (which may change with the strain). The
variable ε/(1- ε - R) will be referred to as the gas volumetric strain. When
Mills and Gilchrist(9) repeated the analysis for foam with a non-zero, but
constant, Poisson’s ratio ν, the slope of a graph of applied stress against
the gas volumetric strain was shown to be p0 (1-2ν). To cover both cases,
the slope of the graph will be referred to as the effective initial cell gas
pressure. σ0 can be evaluated by fitting a straight line to the post-collapse
part of the graph, and extrapolating to zero strain. To apply this analysis
to creep data, the creep stress must be plotted against the gas volumetric
strain at a given creep time.
When Tordon(6) subtracted the isothermal gas response from the EVA
foam response, there appeared to be a constant polymer contribution of
140 kPa when the strain was increasing, and –30 kPa when the strain was
decreasing. He concluded that the gas compression process is close to
isothermal, a view supported by the analysis of Mills and Gilchrist(9).
Lemmon et al(10) and Thompson et al(11) used the hyperelastic foam
model in the ABAQUS finite element analysis (FEA) program to compute
the stress distribution in the foam in respectively the forefoot and heel of
a running shoe. However, the properties of the material model remain
constant with time, and elastic models cannot explain loss processes. The
validation of FEA hyperelastic models for elastomeric foams is in its
infancy(12) and it is a drastic approximation to treat closed cell EVA foams
as hyperelastic materials. In this paper, because we are interested in time-
dependent behaviour, uniaxial compressive loading is considered.
Figure 1 shows a typical compressive stress-strain curve, at impact
strain rates, for EVA foam from a trainer midsole of density 250 kg m-3.
An attempt was made to analyse the shape of the compressive impact
curve, using equation (1). However, the loading part of the response has
strong negative curvature. In this, it differs from the lower density foams.
The unloading part of the curve is nearly linear with fitted line having the
constants σ0 = -200 kPa and p0 (1-2ν) = 131 kPa. As the initial cell gas
pressure is 100 kPa, the higher apparent value suggests that the polymer
contribution to the stress is not constant with strain. The imposed stress
history ramps from zero to maximum load and back over a 100 ms period.
In running, there is a 200 ms zero-load interval before the cycle repeats.
This stress history is complex, and beyond the current capacity of our
modelling. Consequently, we chose to examine the EVA foam in constant
stress creep, to establish the deformation mechanisms.

Cellular Polymers, Vol. 20, No. 2, 2001 81

 N.J. Mills and M.A. Rodriguez-Perez

Figure 1 Compressive stress-strain curve for loading and unloading EVA foam
of density 275 kg m-3; impact from 0.4 m. The straight line is the fit of equation
(1)
2

1.6

1.2
Stress MPa

0.8

0.4

0
0 1 2 3 4 5
Gas pressure strain ratio

Compressive creep - gas diffusion perpendicular to the

applied stress axis
Mills and Gilchrist(13) analysed the creep mechanism of air loss from
polyolefin foams, and studied a 32 kg m-3 density EVA foam of 18% vinyl
acetate (VA) content. The direction in which the gas diffused was assumed
to be perpendicular to the axis of the compressive stress. The diffusion
problem treated was one-dimensional, i.e. the foam block was much
longer than it is deep. However, the results can be applied qualitatively
when there is gas diffusion along both axes perpendicular to the stress
axis. The foam is split into a number >40 of vertical (parallel to the stress
axis) layers. Each layer suffers the same applied strain, and is a parallel
load transfer path, i.e. it is a parallel model. The cell gas content can differ
from layer to layer, so the compressive stress can vary. In this finite-
difference model, the cell gas pressure is assumed uniform inside each
‘layer’. The assumptions were:

a) The diffusion coefficient for the gas in the polymer is so high, and
the cell face so thin that the gas flow through the face reaches steady
state conditions within a negligible time.
b) The gas flow occurs in one direction, perpendicular to the axis of the
compressive stress.

82 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

c) The compressive yield (or collapse) stress σY of the polymeric

structure of the foam is given by
ε −n
σ Y (t) = Y t (2)
J(1)

where εy is the constant yield strain of the foam, taken to be 0.1, t is time,
n is a constant, and J(1) is the 1 sec foam creep compliance. The values
of n and σY are fitted to the creep data.

d) The compressive stress applied to the foam is shared between the

time dependent polymeric stress of equation (2) and the cell gas
pressure.
e) There is no modelling of cell face touching at high compressive
strains, so the simulation is stopped when the foam density reaches
the polymer density.

The analysis(13) showed that the diffusivity of the undeformed foam Df0 is
given by
6P p0
Df = (3)
φR

where P is the polymer permeability, p0 the initial gas pressure in the cells,
φ the fraction of polymer in the foam cell faces, and R the foam relative
density.
An error in the model was corrected. It had been assumed that the foam
diffusivity was proportional to (1- ε) when a compressive strain ε is applied
(equation (3) links the undeformed foam relative density and diffusivity).
However, the gas flow is only impeded by permeation through the cell
faces, which do not change in thickness during compression, or in area
as the faces wrinkle. As the cell gas volume is proportional to (1- ε - R),
the foam diffusivity should be proportional to 1/(1- ε - R). In the limit when
the foam becomes a solid polymer, its diffusivity becomes equal to that of
the polymer (this stage will not be reached for typical creep tests).

Compressive creep - gas diffusion parallel to the applied

stress axis
SEM evidence, that foam cells close to the lower surface of the foam
flatten after creep, suggests that some cell gas diffuses in the direction of
stress (the y axis in Figure 2a). In this series load transfer model, the
applied stress is constant throughout the foam, while the strain varies from
cell to cell. The applied pressure is shared between the polymer structure

Cellular Polymers, Vol. 20, No. 2, 2001 83

 N.J. Mills and M.A. Rodriguez-Perez

Figure 2 (a) Schematic of gas diffusion, parallel to the stress axis, through the
faces of cubic cells, (b) the assumed contributions to the creep stress, from the
cell gas and the polymeric structure, as a function of compressive strain

Stress

(a)
Centre line
Cell 1

Cell 2 L

Cell n-2 Gas

diffusion
through
faces
Cell n-1

Cell n
Stress

(b)

Stress

Polymer
σ0 contribution

Gas
contribution

Compressive strain 0.7 1–R

84 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

and the pressurised cell gas. When a constant creep stress is applied, and
the compressive strain is in the range 0.1 (above yield) to 0.7 (before faces
touch), there is assumed to be a constant polymeric contribution (Figure
2b) at a particular time. This means there must be a constant pressure p
in the cells, irrespective of their y co-ordinate. Hence, the pressure in
closed cells, at the surface in contact with the loading plates, must remain
at p until most of the cell gas has escaped.
There are interactions between the cell strain, the gas diffusivity, cell
deformation (in particular cell face touching at high strains), and the strain
rate in the polymer. Hence, the modelling, which considers all these
factors, is more complex than that previously considered. If a small time
interval is used, the assumption is made that the factors can be considered
in sequence, with the interaction of the factors developing with the
increase of time. Initially all cells have a gas content pcell(n) = 100 kPa.
This is the cell pressure if the cell is returned to zero strain, and is
equivalent to the percentage of the initial cell gas content. The initial
diameter L of the cubic cells was chosen to be equal to the mean cell
diameter in the EVA foam.
The sequence of 6 steps is as follows (the order is arbitrary):
a) Diffusion, using a time step dt small enough for the finite difference
recurrence relationship for the pressure in cell x, of initial diameter
L, after an interval dt, to be stable for all cells.
pnew (n) = (1 – 2D)pold(n) + D(pold(n – 1) + pold(n + 1)) (4)
where
2
L
dt = (5)
2Df max
and
Df (n)
D= (6)
2Df max

Dfmax is the maximum of the cell diffusivities Df(n).

b) The gas content of each cell pcell(n) from the previous value,
multiplied by pnew(n)/pold(n).
c) The absolute pressure pabs applied to the cell gas in the foam. It is
related to the creep time t by
pabs = σ + pa – σ Y ( t ) (7)
where σ is the creep stress, and σY (t) is given by equation (2).

Cellular Polymers, Vol. 20, No. 2, 2001 85

 N.J. Mills and M.A. Rodriguez-Perez

d) The strain ε(n) in each cell, using

 pcell(n) 
ε(n) = (1 – R) 1 –  (8)
 pabs 
There is a correction for the strain rate of the cell. If it is high, then the
cell strain is not allowed to increase as much as if the strain rate is low.

e) The effect of cell faces touching. The progressive contact of cell

faces, once the foam strain exceeds a limit (taken as 0.7), must cause
the percentage of stress carried by the polymer to increase. It reaches
100% when the foam strain is 1- R, i.e. when the foam is solid. A
face-touching factor F was used to multiply the absolute gas pressure
of equation (7). F was assumed to have a value 1 at 0.7 strain and
0 at strain 1- R, with a linear decrease in the interval (Figure 2b). The
absolute pressure of each cell pold(n) was computed using pold(n) =
F pabs.

f) The diffusivity of the compressed cell Df(n), from the cell strain. As
for the parallel model,
Df0
Df (n) = (9)
1 – ε(n) – R
The increase in the foam diffusivity on compression was limited to a
factor of 10.
The model predicts that the gas pressure changes over a short distance,
from fully pressurised to completely empty cells. Such a pressure
differential, across a cell face, will cause the face to develop biaxial tensile
stresses and bow. However, the mechanics of the face bowing was not
modelled, since it depends on cell shape. Therefore, the pressure gradient
in the real foam may differ somewhat from that predicted by the cubic
foam model.
The air diffusivity of solid EVA, should be determined. Although the
permeabilities of EVA films are known, it was not possible to find direct
measurements of the air diffusivity. Extrapolating from the diffusion
constant of 46 p m2s-1 for oxygen in LDPE(14), the value for EVA
containing 18% VA for air is estimated to be the order of 100 p m2s-1.
This value is similar to that estimated for the undeformed foam of density
275 kg m-3.

86 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

EXPERIMENTAL
Material
Samples were cut from one running shoe (midsole) which had been
supplied for demonstration purposes as a separate component with a
moulded marking C-CAP.

Thermogravimetric analysis (TGA)

This technique was used to estimate the percentage of inorganic material
in the foam. The material was heated from 100°C to 900°C at 20°C/min.
Figure 3 shows there is an inorganic content of approximately 18%.

Figure 3 TGA trace showing 18% inorganic content in the foam after heating
to 800°C

120

100
Weight (wt.%)

80

60

40

20

0
0 100 200 300 400 500 600 700 800 900
Temperature (°C)

Scanning Electron Microscopy (SEM) and X-ray microanalysis

Micrographs were taken of the original untested EVA foam and of a
sample 6 (with 29% permanent compressive strain, 12 months year after
creep testing, in which the maximum creep strain was 70 %). The foams
were cooled in liquid nitrogen then cut with a razor blade. The foam was
sectioned both in a plane that included the thickness direction (from the
upper to the lower surface of the midsole), and in the plane perpendicular
to the thickness direction. Some sections were subjected to permanganate
etching, in an attempt to reveal the spherulitic microstructure of the
polymer.

Cellular Polymers, Vol. 20, No. 2, 2001 87

 N.J. Mills and M.A. Rodriguez-Perez

In the undeformed foam, a non-homogeneous cellular structure was

observed (Figure 4a) in the plane that included the thickness direction. Far
from the moulded surface, the cells appear equiaxed, but next to it, the
cells are elongated in a direction parallel to the surface. In the section
perpendicular to the thickness direction, taken at mid-depth, the cells
appear equiaxed. Thus at the mid-depth, the cells must be equiaxed. The
mean cell diameter was 105 µm, measured at mid-depth in the foam. The
compression moulding process for the midsole has distorted the shape of
cells near the surface. Comparing the relative thickness of the cell faces
and edges in this foam shows that the fraction of polymer in the cell faces
φ > 0.7. The cell faces are relatively thick, and there is little sign of
connected vertexes, forming stiff structures.
After permanganate etching, many platelet shaped particles were seen
(Figure 5), the larger ones having diameters of approximately 10 µm.
X-ray microanalysis shows that they contain Si and Mg. Their habit and
chemical content suggests that some particles are talc. The large number
of particles observed is consistent with a filler content greater than 5% in
the EVA foam. Preliminary tests, on a range of EVA foams from many
brands of trainer, suggest that there is often a mineral content, but seldom
such a high content by weight.
In the deformed foam after recovery. A series of photographs were
taken at different distances from the lower surface of the midsole for the
sample having 29% permanent compressive strain. Figure 4b shows the
cells are almost isotropic, near the centre of the midsole, but elongated
close to the moulded surface. The aspect ratio of these elongated cells is
higher than in the initial moulding. In the section perpendicular to the
thickness direction, at the mid-thickness, the cells still appear almost
equiaxed.

Figure 5 SEM photographs of cell faces, after permanganate etching,

revealing particles that may be talc

88 Cellular Polymers, Vol. 20, No. 2, 2001

 Figure 4 SEM of (a) new foam- near middle, near lower surface etc, compared with (b) foam after creep at 250 kPa and recovery
for 1 year. Sections contain the through-thickness direction

Thickness direction
> >

d = distance from the lower surface

(a) Original foam

Cellular Polymers, Vol. 20, No. 2, 2001

Top surface d=5 mm d=10 mm d=5 mm lower surface

(b) Foam after creep and recovery

Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Top surface d=5 mm d=10 mm d=5 mm lower surface

89
 N.J. Mills and M.A. Rodriguez-Perez

Differential Scanning Calorimetry (DSC) and density

measurements
The crystallinity of the foam was approximately 20% and the final melting
point 82°C, which imply a VA content of approximately 18% in the base
polymer. A Dupont brochure(15) shows how the melting temperature Tm
and density of EVA copolymers change with the VA content. The
response for the initial heating run of the DSC (Figure 6) reveals a shoulder
in the melting endotherm at 40°C. This suggests that secondary
crystallisation has occurred while the foam was stored at room temperature.
The shoulder disappears when the sample is reheated after the initial
heating/cooling cycle, and reappears after storage for some hours.

Figure 6 DSC trace for the initial heating of the foam. The small arrow
indicates melting of recrystallised polymer, while the large arrow indicates the
main melting peak

1 Mass 5.48 mg
Crystallinity 20.45%
0.5 Integral -58.88 J/g
Onset 24.05°C
0 Peak 81.60°C
Left limit -0.42°C
Heat flow (mW)

-0.5 Right limit 130.02°C

-1

-1.5

-2

-2.5
-75 -25 25 75 125 175 225
Temperature (°C)

The density of the foams was 275.4 kg/m3 (mean value of twelve
measurements). Two solid samples were produced by compression
moulding of the foam at 125°C, for 12 minutes, at pressures of 250 and
500 kPa. The density of the first moulding was 1070 kg m-3 and the
second 1030 kg m-3. The crystallinity and melting point of these two
materials were similar to those of the foam. The moulding densities
confirm a filler addition, since EVA of 18% VA content has a density of
948 kg m-3 (15). The 1070 kg m-3 density suggests a talc content of 7
weight %, but, if there is porosity in the moulding, the talc content will be
underestimated.

90 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Creep tests
A compressive creep apparatus(13) was used to measure the creep and
recovery of EVA foam blocks 16 x 16 mm x 20 mm high, for creep
stresses between 50 and 500 kPa. The samples were oriented in the same
way as in a shoe, with the compressive stress aligned with the midsole
thickness direction. The experiments were performed at 35°C, because
the skin temperature is about 36°C, and hysteresis losses in running can
cause heating of the midsole. Figure 7 shows creep curves for a range of
creep stresses.

Figure 7 Creep graphs for a range of stresses (kPa) at 35°C

70

60

50

40
Strain %

30

20

10

0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Log (time sec)

Isochronous stress-strain cross-plots (Figure 8) were obtained from the

creep tests at a series of times. The gas volumetric strain was used as the
strain measure, so the slope of the linear region gives the effective cell gas
pressure p0(1-2ν). The graphs have negative curvature, so the slope was
taken at moderate values of the gas volumetric strain. Figure 9 shows the
effective cell gas pressure decreases nearly linearly with log t. The values
for creep times of 1000 s or less exceed atmospheric pressure, implying
that part of the foam hardening is due to deformation of the polymeric cell
structure. The intercept on the stress axis in Figure 8, the polymer initial
collapse stress σ0 was also plotted against the creep time in Figure 9. The
value at 1 second was approximately 120 kPa.

Cellular Polymers, Vol. 20, No. 2, 2001 91

 N.J. Mills and M.A. Rodriguez-Perez

Figure 8 Isochronous stress vs gas pressure strain graphs at times 5, 10, 100
s etc, with straight lines fitted
400

300
Creep stress kPa

5s
200
10 s
100 s
1000 s
100 10000 s
63000 s

0
0 1 2 3 4 5
Gas pressure strain e/(1-e-R)

Figure 9 The effective gas pressure, and initial collapse stress vs log creep
time, derived from Figure 8
180

160 p0 effective
Stress or pressure kPa

140

120 σ

100

80

60
0 1 2 3 4 5
log10 times

Creep recovery measurements were made after creep tests at stresses

of 400 and 500 kPa lasting 106 s. The recovery process is slow to start,
with the main recovery occurring between 104 and 106 s. This suggests
that gas diffusion back into the foam is the main determining factor. Even
after 106.5 s the foam recovery was incomplete (Figure 10).

92 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Figure 10 Recovery after creep for 106 sec as a function of time, at 35°C. Prior
creep stresses in kPa indicated
70

60

50

40
Strain %

30

20

10

0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
Log (Time sec)

PREDICTED CREEP CURVES

Model for gas diffusion perpendicular to the stress axis

The parameters, used to fit the experimental data, were:

Block half width = 8 mm, p0 cell pressure = 100 kPa

Yield stress = 120 kPa, Yield strain = 0.1
1 s compliance = 600 x 10-9 Pa-1 , n (compliance slope) = 0.06
Relative density = 0.26, Fraction of polymer in cell faces = 1.0, Poisson’s
ratio = 0.07

The foam diffusivity was varied, and a reasonable fit between the
experimental results and the prediction obtained for a foam diffusivity of
100 p m2s-1 (Figure 11). The slope of the creep curves is close to
experiment, and the increase in slope after 105 s is matched. However,
the separation of the predicted creep curves, for a 50 kPa increment of
creep stress, is too large for creep stresses in the range 100 to 200 kPa,
when the creep strains are in the range 10% to 40%. It suggests that the
polymer structure contribution to the creep stress is not constant for
0.7 > ε > 0.1 as in Figure 2b, but increases with the compressive strain.
Further, the predicted creep rate does not slow when the strain exceeds

Cellular Polymers, Vol. 20, No. 2, 2001 93

 N.J. Mills and M.A. Rodriguez-Perez

Figure 11 Predicted creep curves, modelling gas diffusion perpendicular to

the stress axis, with foam diffusivity 100 p m2s-1. Stress levels in kPa

0.7

0.6
Compressive creep strain

0.5

0.4

0.3

0.2

0.1

2 3 4 5 6
log10 (creep times)

60%, whereas experimentally it does. This suggests that cell face

touching at high strains, ignored in the model, affects the high strain
creep.
Figure 12 shows the predicted average of cell gas content pcell(n)
versus the creep time. Three values of the foam initial diffusivity are
considered. Comparing the predictions with the experimental data in
Figure 8, in both the gas content drops most rapidly in time interval from
105 to 106 s. However, the effective initial gas pressures in Figure 8 are
greater, as a result of the polymer contribution to the total response. The
EVA foam diffusivity is deduced to be of the order of 100 p m2 s-1, but
the margin of error is quite large.

Model for gas diffusion parallel to the stress direction

The data was as for the transverse gas diffusion model, except that the
foam Poisson’s ratio = 0.0, and the cell diameter = 0.1 or 0.2 mm. The
dashed curves in Figure 13 show that a sudden strain gradient develops
in the foam cells close to the loading anvils, for log10 time > 3 (times >
1000 s), due to positive feedback between the diffusivity increase and the
strain increase. At a time of 10,000 s, the majority of the foam is at a

94 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Figure 12 Predicted average of the cell gas content vs. log time, modelling gas
diffusion perpendicular to the stress axis, for the foam diffusivity values in p
m2s-1 indicated

100

80
Mean cell gas pressure kPa

60

40

20

0
2 3 4 5 6 7
log10 creep times

Figure 13 Predicted cell gas content and layer strains versus y co-ordinate for
the series model, for log10 (creep times) indicated. Cell diameter 0.1 mm and
creep stress 200 kPa

100
100

80 80
Cell gas content kPa

60 60
Strain %

40 40

20 20

0 0
0 1 2 3 4 5
Distance from midplane mm

Cellular Polymers, Vol. 20, No. 2, 2001 95

 N.J. Mills and M.A. Rodriguez-Perez

constant strain of about 40%, but 4 cells, furthest from the midplane
(closest to the surface), are at a strain of about 70%, close to the limiting
value of 1- R. The solid curves in Figure 13 show that the cell gas content
(the absolute pressure that would exist if the cell were returned to zero
strain) is 100 kPa in the central part of the foam, but 15 kPa or less in
the surface cells. There is a small pressure gradient across the depleted
cells. When strain rate effects were included in the model, the development
of the depleted-cell layer was delayed slightly, but there was no effect on
the predictions for the longer times.
In Figure 13 shows there are 4 gas depleted cells at a loading time of
104 s, with more at longer times. The kinetics of the thickening of the fully
compressed, gas-depleted layer is shown in Figure 14, where the number
N of cells with less than 50 kPa gas pressure is plotted against time t on
logarithmic scales. The best-fit equation is
0.525
N = 0.0457 t (10)

which suggests a depletion rate proportional to the square root of time,

a result expected from Fickian diffusion processes.

Figure 14 Predicted number of gas-depleted cells versus creep time,

modelling gas diffusion along the stress axis, with foam diffusivity 10 p m2s-
1

100
Number of depleted layers

10

1
1000 104 105 106
Creep times

96 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

Figure 15 shows the predicted creep graphs, for an initial foam

diffusivity of 100 p m2 s-1. The slowing of the creep rate when the creep
strain reaches 0.74 is reproduced reasonably, but the separation of the
creep graphs is still too great in the creep stress range from 100 to 200
kPa. In many ways, the predictions of the series load transfer model are
similar to those of the parallel load transfer model, given that the foam
block shape was initially nearly a cube.
The predictions are the same if the cell size is increased from 0.1 to
0.2 mm, while the foam dimensions are kept constant. If the creep stress
is 400 kPa, the diffusion losses are predicted to be slightly faster, with total
gas loss from the 16 mm thick foam block shortly before 106 s.
The effect of cell face touching on the cell pressure (step (e) in the
model) is critical to the results. If it is ignored, the gas content of the outer
cells falls to zero, preventing the escape of further gas. As the mechanics
of face touching has not been studied in detail, the predictions are subject
to change. Initial experimental measurements with a low density EVA
foam (R. Verdejo, unpublished) confirm the development of densified
surface layers, in contact with the surfaces that load the compressed
foam.

Figure 15 Predicted creep curves, modelling gas diffusion along the stress
axis, with cell size 0.1 mm, block thickness 16 mm, initial foam diffusivity 100
p m2s-1. Stress levels in kPa

0.7

0.6

0.5
Compressive strain

0.4

0.3

0.2

0.1

0.0
2 3 4 5 6
log10 creep times

Cellular Polymers, Vol. 20, No. 2, 2001 97

 N.J. Mills and M.A. Rodriguez-Perez

DISCUSSION
The cellular microstructure of the original foam moulding is inhomogeneous,
with elongated cells close to the lower surface of the midsole. This makes
the interpretation of ‘damage’ in shoes after use(5) difficult, since reference
samples were not available for comparison. The anisotropic cell shape
probably affects the mechanical properties of this surface layer. The EVA
foam studied contained 18% by weight of mineral filler particles. Possible
reasons for adding the filler are: to reduce costs and the permeability of
the cells to air; to improve the foam stiffness and thermal resistance; to
protect the polymer from degradation from UV light exposure.
The analysis of the isochronous stress-strain curves reveals that the cell
gas provides most of the foam hardening in compression, for creep times
> 1000 s. At shorter times, the polymer contribution to the hardening is
significant. Air loss from the cells, after a period of creep, causes the foam
hardening to decrease. These results should be applicable to EVA foam
subjected to repeated impacts, suggesting a mechanism for trainer
midsole fatigue.
Creep modelling has been extended, from the previously proposed
parallel load-path model, to a series model for diffusion along the stress
axis. For the latter the predicted gas content profile is non-Fickian, with
a sudden change of gas content at a well-defined front. However, the
advance of this front is predicted to be proportional to the square root of
time, as expected with Fickian diffusion. A gas-depleted surface layer
appears in trainer midsoles after prolonged use, in spite of their loading
history differing from creep loading. The relative proportion of gas
diffusion along and perpendicular to the stress axis, must depend on the
geometry of the loaded foam block. Gas diffusion parallel to the stress is
the major effect for foam products with thickness much less than either
lateral dimension (e.g. a shoe midsole). A complete gas loss model should
consider the three-dimensional diffusion of gas.
It is usual in trainer midsoles to use EVA foam of density 200 to 250
kg m-3, but the reasons for the foam density selection have not been
explained. The shoe industry uses the foam compression set as one
selection criterion. The European test standard(16) requires a constant
compressive strain of 50% to be maintained for 22 hours, at 23 or 70°C,
then the residual strain measured 30 minutes after releasing the sample.
The test pieces are 50 mm wide and long, and 25 mm thick. The values
for EVA foam will be affected both by the cell air loss in 22 hours and the
incomplete recovery process after 30 minutes. The air diffusivity of the
275 kg m-3 density EVA foam, estimated as approximately 100 x 10-12
m2 s-1, is an order of magnitude less than the 1000 x 10-12 m2 s-1 value

98 Cellular Polymers, Vol. 20, No. 2, 2001

 Modelling the Gas-Loss Creep Mechanism in EVA Foam from Running Shoes

found for 32 kg m-3 density EVA foam(13). The gas diffusion rate from
EVA foam appears to be inversely proportional to the foam density, as
the proportion of polymer in the cell faces appears to be independent of
foam density. The mineral filler content in the foam cell faces will reduce
the air diffusion rate, but the magnitude of this effect is unknown. It is
possible that shoe manufacturers select the foam density to meet a target
value of compression set being less than X%.
It could not be established if there was ‘permanent’ plastic deformation
of the thicker parts of the cell microstructure. Recovery after creep was
very slow, and the process was incomplete after several weeks at 35°C,
or after a year at roughly 20°C. The maximum driving pressure for
recovery would be a vacuum in the cells (–100 kPa relative pressure), but
it is unlikely that the driving pressure is that great; this compares with a
driving pressure for gas loss during creep up to 500 kPa. There was no
sign of fracture of cell faces. When the foam is stored at 20°C it
recrystallises, and this recrystallisation may interfere with the recovery
process. Another possibility is that the strains in the bent cell faces may
exceed the limit for yielding. At present there is no complete model of the
micromechanisms in compressed closed-cell foams. Mills and Zhu’s(17)
model, in which some cell faces buckle immediately on compressive
loading, provides good predictions of the compressive stress strain curve
of low-density polyethylene foams, but does not estimate the maximum
bending strains in the faces. The filler in the EVA foam faces may make
the faces more prone to plastic deformation than unfilled polymer.
To date there has been no modelling of foam recovery after creep. If
the cell walls do not fully recover from the bending and or in-plane
stretching that occurs in the creep phase, they cannot drive the recovery
to completion. The cell gas content, if constant, could cause complete
recovery if the cell walls do not oppose the process. However, if some cell
gas escapes during creep, it is possible on recovery for the cell gas
pressure to return to atmospheric, while the foam has a residual strain;
the permanently deformed cell faces exert no recovery forces. If this
occurs at a time shorter than the gas equilibration time for the particular
size foam block (approximately 4 hours for the 16 mm specimen width)
then, on a longer time scale, there will be no further change. The
‘damaged’ region in the recovered foam may be the layer of elongated and
gas-depleted cells adjoining the moulding surface, i.e. the damage is not
uniform throughout the foam.
We intend in future to model the repeated impact of EVA foams, as in
running. The diffusion geometry differs from that considered here; gas
diffusion towards the road surface may be impeded by the solid rubber

Cellular Polymers, Vol. 20, No. 2, 2001 99

 N.J. Mills and M.A. Rodriguez-Perez

outsole, while the midsole is wider than it is thick. Nevertheless, it is likely

that cell gas loss causes the change in the midsole response with repeated
impacts.

REFERENCES
1. Shorten M.R., The energetics of running and running shoes, J.
Biomechanics, 26, suppl. 1, (1993) 41-51
2. Hennig E.M. and Milani T.L., In shoe pressure distribution for
running in various types of footwear, J. Appl. Biomech., 11, (1995)
299-310
3. Misevich K.W. and Cavanagh P.R., Material aspects of modelling
shoe/foot interaction, Ch 3, Sports shoes and playing surfaces,
Frederick E.C. Ed. (1984) Human Kinetics Publ. Inc.
4. Kinoshita H. and Bates B.T., The effect of environmental temperature
on the properties of running shoes, J. Appl. Biomech., 12, (1996)
258-268
5. Bartlett R., Sports Biomechanics, (1995) Spon, London
6. Torden M.J., Evaluation of the cushioning of closed-cell plastic
foams, Dynamic loading in manufacturing and service, (1993) Inst.
Eng. Aust. Conf., Melbourne, 11pp
7. Gibson L.J. and Ashby M.F., Cellular Solids, (1988) Pergamon,
Oxford
8. Rusch K.C., J. Appl. Polym. Sci., 14, (1970) 1263
9. Mills N.J. and Gilchrist A., The effect of heat transfer and Poisson’s
ratio on the compressive response of closed cell polymer foams,
Cell. Polym., 16 (1997), 87-119
10. Lemmon T. et al., the effect of insoles in therapeutic footwear - a
finite element approach, J. Biomechanics, 30, (1997), 615-620
11. Thompson R.D. et al., Modelling and performance of training shoe
cushioning systems. Sport Engng. 2, (1999) 109-120
12. Mills N.J. and Gilchrist A., Creep and recovery of polyolefin foams
- deformation mechanisms, J. Cell. Plast., 33 (1997a) 264-292
13. Mills N.J. and Gilchrist A., Modelling the indentation of low density
polymer foams, Cell. Polymer, 19, (2000) 389-412
14. Mills N.J., (1993) Plastics, 2nd edition, Edward Arnold, London, 257
15. www.dupont.com/industrial-polymers/elvax, Thermal properties
of Elvax measured by Differential Scanning Calorimetry
16. EN ISO 1856 (1996) Polymeric materials, cellular flexible –
Determination of compression set, CEN, Brussels

100 Cellular Polymers, Vol. 20, No. 2, 2001