A constitutive model for drying of a partially saturated porous material

Ivan Vlahinic
a
, Hamlin M. Jennings
a,b,
*
, Jeffrey J. Thomas
b
a
Northwestern University, Department of Civil and Environmental Engineering, Evanston, IL 60208-3108, USA
b
Northwestern University, Department of Materials Science and Engineering, Evanston, IL 60208-3109, USA
a r t i c l e i n f o
Article history:
Received 6 February 2008
Received in revised form 21 October 2008
a b s t r a c t
A new constitutive model for drying of an elastic porous material is presented. The model is
derived by decomposing the boundary value problem of drying into parts for which an
explicit solution is readily available. The mathematical form differs from models developed
using either the concept of equivalent pore pressure or average pore pressure. Instead of
pressure averaging, the extent of weakening of the solid due to void inclusions, a material
parameter, is found to be critical for the proper predictions of the volumetric strains. The
model is applicable to drying under conditions where multi-layer adsorption of liquid per-
sists, i.e. where the effects of solid surface energy are negligible, and where capillary pres-
sure is the primary loading stimulus. A particular (and simple) form of the model is
compared with experimental data for partially dried cement paste and Vycor
TM
glass, and
is shown to provide a much better agreement than previous models based on average pore
pressure. Implications to the poromechanics and the effective stress of partially saturated
porous media are discussed.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
The constitutive behavior of porous materials has been
studied extensively. In their pioneering works, Paul Fillun-
ger in the 1910s and Karl von Terzaghi in the 1920s
explored the inuence of pore uid primarily on strength,
but also on one-dimensional deformation of soils (see
review by de Boer (1992)). In the late 1930s and 1940s,
Biot (1941) developed a linear theory of poroelasticity.
Since that time, both drained and undrained elastic behav-
iors (the two limit cases) of uid-lled porous materials
have been well grounded. The extension of Biots theory
to partially saturated porous materials has essentially re-
mained unchallenged in the elastic region since Bishop
(1959) introduced the concept of average pore pressure;
and yet satisfactory agreement with experimental results
has not been found even in relatively simple elastic mate-
rials such as Vycor
TM
glass (Bentz et al., 1998). And while
constitutive modelling of partially saturated materials
continues to be a rich subject of ongoing research
(e.g. Loret and Khalili, 2002; Gallipoli et al., 2003), with a
comprehensive review on the subject given by Gens et al.
(2007), the elastic formulation introduced by Bishop
(1959) is seldom re-examined. Notable exceptions can be
found in the contribution by Khalili et al. (2004) as well
as in the models that are based on the assumption of
multiple scales within the porous material (e.g. Chateau
et al., 2002; Coussy and Brisard, in press).
The authors modelling experience shows that experi-
mental results of elastic shrinkage during drying of hard-
ened cement paste are consistently and signicantly
greater than predictions by present formulations. Coussy
et al. (2004) recently addressed the lack of agreement by
modifying the capillary stress contribution to include an
additional surface energy term. However, above 50%
relative humidity (RH) capillary stresses should dominate
surface energy effects due to multi-layer adsorption on
the surfaces of emptied pores, and therefore it should be
possible to obtain an accurate model using only the
KelvinLaplace formulation for capillary stress.
0167-6636/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmat.2008.10.011
* Corresponding author. Address: Northwestern University, Depart-
ment of Civil and Environmental Engineering, Evanston, IL 60208-3108,
USA. Tel.: +1 8474915282; fax: +1 8474914011.
E-mail address: [email protected] (H.M. Jennings).
Mechanics of Materials 41 (2009) 319328
Contents lists available at ScienceDirect
Mechanics of Materials
j our nal homepage: www. el sevi er . com/ l ocat e/ mechmat
Here, a constitutive model is developed for partially sat-
urated media subjected to drying, where pore pressure is
known a priori. Vycor
TM
glass and hardened portland ce-
ment paste (OPC) dried to various relative humidities are
used as model materials. Cement paste in particular pro-
vides a unique model material because it has pores that
span the nanometer to micron size range, allowing sub-
stantial variation in pore pressure due to capillary forces
and fraction of pores lled. Thus, both the quantity of lled
pores and pore pressure can be varied by drying to specic
vapour partial pressures. Water saturation, the fractional
quantity of water-lled pores, is determined by weight loss
and the pressure is governed by the KelvinLaplace
equation.
2. General theoryeffective stress and constitutive law
2.1. Fully saturated porous media
The geomechanics community has developed the con-
cept of effective stress to model the behavior of soils. This
facilitates computations by incorporating the inuence of
pore pressure, p, as well as external loads, r
ij
, into a single
term, known as the effective stress, r
0
ij
. The problem of a
fully saturated porous media is then treated like any other
dry continuum problem, such that:
r
0
ij
r
ij
fpd
ij
1
Here, the positive values of the stress and pore pressure
indicate compression. Parameter f reects the fact that the
pore pressure does not equally inuence high and low
porosity materials. Disagreement over its value permeated
the literature well after Fillungers and Terzaghis introduc-
tion of effective stress and several different expressions for
isotropic materials were proposed (see Table 1).
Terzaghi (1923) originally argued that f should be equal
to the porosity, /

, so that in the non-porous material the

effect of pore pressure would vanish (Garg and Nur,
1973); and yet experimentally, he found f 1 for several
porous materials (Nur and Byerlee, 1971). Geertsma
(1957), and later Skempton (1961), independently dened
f in terms of K
b
, the (drained) bulk modulus of a porous
material, and K
s
, the bulk modulus of the solid or the grain.
In this form, f is often referred to as the Biot coefcient.
Suklje (1969) proposed another expression based on
porosity, although rigorous proof was not provided. Nur
and Byerlee (1971) later conrmed the result of Geertsma
using a uniqueness theorem for the stress boundary value
problem. For a general case of non-isotropic materials, the
value of f depends on the constitutive material matrix
(Zienkiewicz et al., 1999).
It is now instructive to write the constitutive equations
for an isotropic uid-saturated porous solid (Biot, 1962),
where l
b
and k
b
are Lams constants of a dry material
and e
v
is a volumetric strain dened as e
v
e
xx
e
yy
e
zz
:
r
xx
1 K
b
=K
s
p 2l
b
e
xx
k
b
e
v
r
yy
1 K
b
=K
s
p 2l
b
e
yy
k
b
e
v
r
zz
1 K
b
=K
s
p 2l
b
e
zz
k
b
e
v
r
xy
l
b
e
xy
r
yz
l
b
e
yz
r
zx
l
b
e
zx
2
From the constitutive equations, it is evident that pore
pressure inuences only normal strains (not shear), and
hence only volumetric deformation (and not deviatoric).
Physically, this conclusion also naturally stems from the
fact that a uid by denition cannot resist shear.
1
Thus,
absent of externally applied deviatoric stresses, the constitu-
tive Eq. (2) can be reduced to Eq. (3), where a volumetric
stress,

r, is dened as

r 1=3r
xx
r
yy
r
zz
and where
the bulk modulus can be expressed in terms of the Lams
constants of the dry material, i.e. K
b
1=32l
b
3k
b
:

r 1 K
b
=K
s
p K
b
e
v
3
It is often convenient to reorganize Eq. (3) and express
the volumetric strain in terms of the applied stresses, as
shown in Eq. (4). The latter constitutive equation appears
to have been rst published by Mackenzie (1950), although
not in the context of the effective stress formulation.
e
v

r
K
b
p
1
K
b

1
K
s

4
Thus far, only the drained and undrained responses of
saturated materials have been examined. The saturated
drained response corresponds to a state of zero excess pore
pressure throughout the pore space. In this case pore uid
can ow in and out of the pore space without restriction
and the material therefore deforms as if the pore uid
is not present. In the case of the saturated undrained
response, uid drainage is impeded and the excess pore
pressure is nite.
2.2. Partially saturated porous media
In partially saturated porous materials, more than one
uid occupies the pore space and at least one uid is under
a nite pressure. The discussion here is limited to cases
where air and water occupy all the porosity. For the major-
ity of proposed formulations, the constitutive laws for the
partially saturated system can be found directly from the
fully saturated system. This is achieved by replacing the
pore pressure term p that appears in Eq. (4) with an effec-
tive pore pressure obtained by averaging the pressure of all
the uid constituents.
The rst expression was developed by Bishop (1959),
who averaged the pore pressures according to the contact
area, v, between the uid and the solid. Because of the
inherent difculty in quantifying the effective uid contact
area, v is usually expressed as some function of the degree
Table 1
Proposed f coefcients in effective stress formulation Eq. (1).
f / Terzaghi (1923)
f 1 K
b
=K
s
Geerstma (1957)/Biot (1962)
f 1 1 /K
b
=K
s
Suklje (1969)
1
This is true under quasi-static loading, i.e. where the shear stress
components that result from viscous effects in a uid are negligibly small.
320 I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328
of uid saturation, S, a quantity that can be conveniently
measured, such that v vS for each uid phase. The
effective areas of water and air are related, i.e.
v
w
v
a
1, and as a result the effective pore pressure
can be described in terms of the effective water contact
area only, v
w
, commonly referred to as the Bishop
parameter:
p v
w
p
w
1 v
w
p
a
5
Instead of averaging the uid contact areas, the formu-
lation can be simplied by direct volume averaging. In his
Ph.D. Thesis, this approach lead Schreer to the same
expression, with v
w
S
w
S
w
where S
w
is the degree of
water saturation, i.e. the volume fraction of pore space
occupied by water; this work was later published by
Lewis and Schreer (1987). Hutter et al. (1999) came to a
similar conclusion based on the thermodynamic theory of
mixtures. This form has since been adopted by several
investigators (e.g. Gallipoli et al., 2003; Coussy, 2004),
and will be employed in this work for the purposes of mod-
el comparison. This is not to say that other expressions
cannot be found in the literature, although they are typi-
cally derived on an empirical basis.
Absent of externally applied deviatoric stress, the con-
stitutive law for a partially saturated system based on
the effective pressure concept is shown below.
e
v

r
K
b
S
w
p
w
1 S
w
p
a

1
K
b

1
K
s

6
2.3. Generating pore pressures in a partially saturated system
In a porous material, partial saturation can be intro-
duced in several ways. Starting with a saturated specimen,
load cells can be coupled with pumps to induce water
egress and/or air entry and thus provide a great deal of
control over the uid phases (e.g. Sun et al., 2004). Instead
of using pumps to generate suction pressures, water egress
can also be induced via reverse osmosis. Furthermore, elec-
trolysis (the ionization of hydrogen and oxygen when elec-
tric current is passed through the water) has been shown
to allow for partial dewatering of the soils by introducing
gas into the pore spaces (Yegian et al., 2007).
In contrast to the above mentioned experimental meth-
ods, drying provides perhaps the simplest way of generat-
ing partial saturation in a porous material. As pore water is
removed by lowering the external relative humidity, capil-
laries form at the uiduid interface and control the value
of the pore pressure. This means that the pore pressure
need not be experimentally measured as, at equilibrium,
it can be solved a priori via the KelvinLaplace equation
shown in Eq. (7), with radius of meniscus, r
m
, dened in
Eq. (8) and variables described in Table 2.
p
c
RT=Mlnh 7
r
m
2c=p
c
8
In the case of the materials examined in this work, the
two uids are water and air, while the corresponding cap-
illary pressure, p
c
, is dened as the difference between air
pressure and water pressure, i.e. p
c
p
a
p
w
. Furthermore,
jp
a
j jp
w
j as the absolute value of the pressure in the
water quickly exceeds atmospheric pressure during drying,
climbing above 10 MPa (100 atm) at only 92% RH. Consid-
ering now the case of no external load, the constitutive law
Eq. (6) reduces to:
e
v
S
w
p
c
1
K
b

1
K
s

9
Bentz et al. (1998) applied Eq. (9) to Vycor
TM
glass and
found that it signicantly underestimates the measured
deformations below 80% relative humidity.
3. New constitutive formulationdrying of a partially
saturated system
3.1. Constitutive law for a fully saturated porous system
Following the method of Nur and Byerlee (1971), we
rst derive the constitutive law for a fully saturated porous
system. The porous system is composed of a linear elastic
solid with connected pores of arbitrary shape. The pores
are assumed homogeneously distributed.
For simplicity, we consider the porous system subject to
a uniform pore pressure p p
c
and free of any external
loads. This boundary value problem is largely intractable
on its own, but becomes tractable when decomposed into
individual parts for which an explicit solution is readily
available. Hence we consider a convenient superposition
of stresses, as shown in Fig. 1a. We rst apply an external
hydrostatic compression

r p
c
, which results in a strain of
p
c
=K
b
(note that compression is dened positive). Next, we
apply an external hydrostatic pressure

r p
c
and an
equal pore pressure p p
c
. This results a uniform stress
in the solid (e.g. Love, 1944, p.86) and a strain equal to
p
c
=K
s
.
Since we assume the porous system exhibits linear elas-
tic behavior, the strain elds resulting from the two steps
can be added and the constitutive law reduced to:
e
v
p
c
1
K
b

1
K
s

10
Invoking a uniqueness theorem for the stress boundary
value problem, Nur and Byerlee (1971) used the preceding
superposition of stresses to deduce a theoretically exact
expression for the effective stress. Note that Eq. (10) is
identical to Eq. (4), absent of externally applied loads.
3.2. Constitutive law for a partially saturated porous system
We now apply a similar strategy to a partially saturated
porous system, where only select pores are subject to a
Table 2
Description and constants used in KelvinLaplace Eq. (7).
Constant Symbol Value unit
Surface tension c 0.0728 N/m
Molar volume M 1.802 10
5
m
3
/mol
Univ. gas constant R 8.314 Nm/Kmol
Abs. temperature T 293 K (OPC)
284 K (glass)
Relative humidity h 0100%
I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328 321
uniform pore pressure p p
c
. Conceptually, we dene a
boundary of a new effective solid delineated by the sur-
faces of pores that apply pressure, i.e. the remaining full
pores. In this manner, we can reduce a partially saturated
system to a fully saturated system, provided we update
the material constants of the new solid. Physically, the load
imposed by the capillaries is still carried by the solid,
although the solid is now weakened and behaves as if it
were porous itself.
To arrive at the solution, we similarly decompose the
boundary value problem described above into two loading
steps as shown in Fig. 1b. In the rst step, we apply posi-
tive (normal) tractions

r p
c
on the external boundary
of the porous body. It is apparent that under this loading,
the volumetric strain is equivalent to that caused by exter-
nal load p
c
in a fully drained porous system and is equal to
p
c
=K
b
regardless of the degree of drying. In the second step,
we impose equal tractions on internal and external bound-
aries. This results a uniform state of stress in the effective
(weakened) solid and strain of p
c
=K. The effective bulk
modulus, K describes the bulk modulus of the effective so-
lid and varies during drying. It can be conveniently ex-
pressed in terms of the degree of water saturation, S
w
, as
the volume of water-lled and empty pores are closely re-
lated. Finally, combining the strain elds, the constitutive
law of a partially saturated system reduces to:
e
v
p
c
1
K
b

1
KS
w

!
11
Before proceeding, a brief check of the behavior at ex-
tremes should be performed. In the case of zero saturation,
water occupies negligible porosity. Hence, K !K
b
as evi-
dent from Fig. 1b, and the volumetric strain given by Eq.
(11) reduces to zero as expected. In the case of full satura-
tion, water occupies all the pores, K !K
s
, and Eq. (11) cor-
rectly reduces to Eq. (10).
Similar linear superposition of stresses can be estab-
lished for systems that include external loads as well as
two or more uid phases where each uid is under nite
pressure. Presently, for simplicity, we include only a 2-
phase uid system free of external boundary stresses and
assume the inuence of the gas phase (air) to be
inconsequential.
From Eq. (11), it is clear that the constitutive law for
a partially saturated system does not necessitate the use
of pressure averaging over all uid constituents, in direct
contrast to Eq. (9). Rather, the extent of weakening of
the effective solid skeleton, a material parameter, must
be deduced to determine the volumetric deformation.
This is consistent with the experimental observation
that the Bishop parameter v is often found to be
strongly inuenced by the material fabric (Coleman,
1962).
It should also be noted that in the geomechanics com-
munity, it is customary to assume that the solid or grain
is incompressible. In a fully saturated system, this assump-
tion has a clear meaning, namely that K
b
K
s
when
S
w
1. As such, the assumption also allows for a natural
simplication of Eq. (10). In a partially saturated system,
however, the assumption of solid incompressibility does
not easily simplify Eq. (11) because, in general, the effec-
tive bulk modulus K cannot be assumed incompressible
even if K
b
K
s
. This is because even a few pores can give
Fig. 1. Stress superposition schematic of (a) fully saturated and (b) partially saturated system. Capillary pressure, p
c
, is shown to act on the surface of the
internal solid boundary. Inuence of atmospheric air pressure is assumed inconsequential.
322 I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328
rise to an appreciable compressibility in an otherwise
incompressible medium.
3.3. Quantifying the loss of stiffness of the effective solid
Before analyzing the inuence of void inclusions on bulk
properties of the effective solid, it is useful to relate howthe
porosity of the effective solid varies as a function of the de-
gree of saturation. To this end, we dene a volume fraction,
n
a
, of the solid, water, and air phases, with a s; w; a corre-
spondingly. We further note that the degree of (water) satu-
ration can be expressed as S
w
n
w
=n
w
n
a
and that the
fractionof thematerial occupiedbytheair andwater phases,
the total porosity, is given by /

n
w
n
a
. At any stage of
drying, the porosity of the effective solid, u, then amounts
to u n
a
=n
s
n
a
(see Fig. 1b) and in terms of the degree
of saturation simplies to:
uS
w

1 S
w
/

1 S
w
/

where 0 6 uS
w
6 /

12
Now only the inuence of porosity on the solid com-
pressibility remains to be quantied. Before proceeding,
it should be mentioned that what is referred to as porosity
in the effective solid is in fact the space that is occupied by
the gas phase (air). While it may be argued that air may
offer a role of structure and thus limit the compressibility
of the effective solid, this role is largely negligible; the
compressibility of a solid is typically orders of magnitude
less than that of gas. Therefore, for the purpose of estimat-
ing the effective solid properties, the gas (air) phase may
be treated as a void or a cavity rather than a material phase
with nite compressibility.
The experimentally derived relationships can certainly
provide the solid properties for varying values of porosity.
Oftentimes, however, such relationships are not readily
available and theoretical approaches may need to be em-
ployed. The simple rule of mixtures can establish upper
and lower bounds for parallel (Voight) and series (Reuss)
congurations. Unfortunately, as void inclusions are in-
nitely compressible, the lower bound estimate invariably
yields K 0 for any value of porosity, a conclusion that
also applies to the classic HashinStrickman lower bound
(Hashin, 1983).
Using a concept of averaged stress, many successful
micromechanical schemes have been built on the founda-
tions of the Eshelby (1957) tensor. Among these, perhaps
the most prominent are the MoriTanaka (e.g. Mori and
Tanaka, 1973; Weng, 1984) and self-consistent (e.g. Her-
shey, 1954) approximations. The former have proven use-
ful in a variety of composite materials while the latter
have found success with the porous materials whose skel-
eton is polycrystalline or granular; for example, Constan-
tinides and Ulm (2007) recently found the self-consistent
scheme useful in modeling the elastic behavior of ce-
ment-based materials at the lower material scales. For a
more detailed treatment of the aforementioned schemes
(and others), Christensen (1990) provides a good overview
and comparison.
Predictions can be greatly improved when values of
K
b
andK
s
are established experimentally and used as
bounds thereafter. From Fig. 2, it is evident that a straight
line between the two experimentally measured moduli
yields a rst-order estimate of the effective properties for
the porous material in question. In this regard, low poros-
ity materials enjoy an inherent advantage: as porosity de-
creases, the rst-order estimate approaches the true
properties of the effective medium. The experimentally
aided estimate of K is given by:
K K
s

K
s
K
b
/

uS
w
13
It should be noted that Eq. (13) may not provide a good
estimate for solids whose effective properties are affected
greatly when voids are introduced. One such limitation
may be especially prominent in solids often referred to as
incompressible for which K
b
K
s
as noted earlier. For
these materials, voids clearly have a detrimental effect on
the overall volumetric stiffness, as evidenced by the very
fact that K
b
is orders of magnitude smaller than K
s
. A large
loss of stiffness can be captured by micromechanical esti-
mates if highly irregular pore shapes, e.g. crack-like cavi-
ties, or a high concentration of voids are present. In such
cases, a better estimate than the one shown in Eq. (13)
should be used for more accurate predictions of volumetric
deformations via Eq. (11); however, this is beyond the
scope of this work.
4. Experimental comparison
4.1. Drying strains of hardened cement paste on second
drying
Our formulation can be perhaps most easily compared
against theexperimental results of apartiallydriedmaterial,
whichprovides favorable conditions for several reasons. The
induced capillary stresses are purely hydrostatic. For a
Fig. 2. Effect of void inclusions on elastic properties of a solid. In absence
of an actual relationship, a rst-order estimate of the properties of the
effective solid can be established using the experimentally measured
values of the drained bulk modulus K
b
and the solid bulk modulus K
s
.
I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328 323
homogeneous and isotropic materials the deformations
then translate to uniform strains in all directions, avoiding
a necessity of volumetric measurements. Linear deforma-
tion, i.e. drying shrinkage, in any one direction equates to
1=3 of the total volumetric deformation. Differential shrink-
age, which gives rise to cracking, can be avoided by slow
step-wise drying, i.e. by gradually lowering the relative
humidity to the desired nal value, and by the use of speci-
mens with small cross section.
Because a new equilibrium can be obtained for each
partial saturation, only these values of length change at
each partial vapour pressure need to be recorded exper-
imentally. At equilibrium, exchange of pore liquid with
the environment also stops. By recording weight differ-
ence between successive relative humidity points,
preferably for the entire range of saturation states, the
degree of saturation S
w
in the porous system may be
interpolated.
It is very important to note that hardened cement paste
does not deform elastically the rst time it is dried as the
rst drying cycle certainly induces permanent dimensional
changes. However, each successive cycle is nearly fully
reversible in terms of water sorption and shrinkage, as
shown experimentally by e.g. Helmuth and Turk (1967),
Verbeck et al. (1968), and others. It is on this basis that
elastic shrinkage formulations may be compared against
experimental data using the 2nd drying cycle of pastes.
Published data on 2nd drying of cement paste is scarce.
Recent experiments at Northwestern are starting to ll this
void. For example, data on pastes with water to cement
ratio by mass of 0.5 has been acquired by Gevrenov
(2005). The water retention curve
2
is displayed in Fig. 3;
each data point is an aggregate of measurements from
several specimens.
3
4.1.1. Material properties of hardened cement paste
Before comparing shrinkage models with experiments,
a brief review of available data on elastic constants is
necessary, as their values are not readily available. Direct
measurements of the bulk modulus of the solid constitu-
ent K
s
of cement paste with varying w/c ratios and ages
were performed by Grasley et al. (2006). A small scatter
of data was observed, with values of K
s
in the range of
4050 GPa. Fortunately, such variation does not result
in signicant changes in the model predictions of shrink-
age strain (up to 5%). Therefore, K
s
45 GPa is used
here.
Values of K
b
, on the other hand, signicantly inuence
the nal shrinkage values and, therefore must be used
judiciously. Grasley et al. (2006) calculate K
b
within the
range of 8.39.0 GPa, using an approach presented in
its companion publication (Scherer, 2006). Estimates of
K
b
can also be made based on available tests of Youngs
modulus and Poissons ratio. Measurements by Ghosh
and Timusk (1974), which are consistent with those of
Feldman (1968), show E
b
12 GPa for Youngs modulus
of hardened paste with w=c 0:5. Poissons ratio, as re-
ported by Parrott (1974) for w=c 0:5, was shown to
vary between m
b
0:2200:275, depending on the de-
gree of saturation. These numbers correspond to
K
b
6:98:6 GPa, slightly lower than those estimated
by Grasley et al. (2006). Because of the comparatively
large effect of bulk modulus on nal shrinkage predic-
tions, the two extremes of K
b
7andK
b
9 GPa are in-
cluded in our predictions.
How bulk modulus varies with porosity remains uncer-
tain. At present, the only known studies of the inuence of
porosity on paste modulus were performed by Feldman
(1968) who found that hardened cement paste exhibits a
semi-logarithmic relationship between elastic modulus
and porosity. Unfortunately, the published data set con-
tains two signicant shortcomings:
(1) The author fails to indicate the method with which
porosity of paste was measured. This issue is of crit-
ical importance due to sensitivity of results to exper-
imental methods and controversies over methods of
measuring capillary porosity (Jennings, 2008). Sensi-
tivity lies in the dominant constituent of hardened
cement paste, calcium silicate hydrate (CSH)
which is highly hydrophylic and contains structural
water.
(2) Modulus data in the porosity range below 20% was
not measured directly, a range most useful at pres-
ent. Instead, it was interpolated from higher porosity
cement paste samples.
For these reasons, applicability of Feldmans conclu-
sions to this work remains uncertain. Instead, an interpola-
Fig. 3. Water retention data for Vycor
TM
glass (Corning 7930) and
hardened cement paste (2nd drying, 56 days old, w=c 0:5) based on
experiments by Amberg and McIntosh (1952), Gevrenov (2005) respec-
tively. In Vycor
TM
glass water retention is fully reversible for all drying
cycles while in hardened cement paste it is nearly fully reversible after
the rst drying cycle. Percent saturation was interpolated from the
weight the samples lost on drying to specic relative humidities. See text
for more details.
2
The water retention curve is shown as a function of relative humidity.
From the displayed data, the values of the capillary pressure may be
deduced via Eq. (7).
3
Data on the 1st drying cycle is available for comparison in literature. In
particular, careful measurements by Roper (1966) compare favorably to
those of Gevrenov (2005), where both 1st and 2nd drying cycles are
obtained using the same experimental methods. This provides additional
condence in the referenced data set.
324 I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328
tion of effective bulk modulus based on the experimental
values of K
b
and K
s
will be used. For detailed discussion
of the method, refer to Section 3.3.
4.1.2. Types of water in cement paste and applicability to
capillary tension
Water plays a critical role in hardened cement paste,
and its presence (or lack thereof) denes essentially all of
its material properties. For this reason, it is important to
address the inuence of different water types.
Due to the chemical properties and hydrophilic nature
of CSH, not all water remaining in hardened cement
paste during drying is under hydrostatic stress. Chemically
combined water in CSH is, in fact, part of the solid struc-
ture.
4
It is non-evaporable, and certainly not subject to
capillary pressures. Water that is adsorbed onto highly
hydrophilic layers and surfaces of CSH also does not
maintain all the properties of bulk water due to solid sur-
face forces.
Beyond the reach of surface forces, water in pores be-
haves as bulk and thus maintains the full capillary pressure
p
c
imposed by the meniscus. It is established experimen-
tally that on average, one molecular layer of water coats
the surfaces of hardened cement paste at 11% RH, and
two layers at 35% RH ( Badmann et al., 1981; Jennings,
2008). Adsorption of additional layers results in negligible
change in solid surface energy (e.g. Mindess and Young,
1981, p.490). This suggests that, during drying, an estimate
of water subjected to capillary stress can be obtained by
measuring the water that evaporates from the sample be-
tween full saturation and 35% relative humidity. The de-
gree of water saturation can then be estimated by
dividing the current weight change of the sample from sat-
uration (100% RH) by the weight change on drying to 35%
RH.
In this work, we dene 33% RH as a zero saturation
point. This value is used simply for convenience because
it is part of the referenced data set. In addition, we nd
the total (capillary) porosity to be /

0:28 (Tennis and

Jennings, 2000).
4.2. Drying strains of Vycor
TM
glass
In contrast to hardened cement paste, Vycor
TM
glass
(Corning 7930) has an extremely narrow distribution of
pore sizes, with an overwhelming majority of the pores
on the order of nanometers and with the pores of micron
size nearly entirely absent ( Nordberg, 1944; Scherer,
1986). This provides a distinct model material in which
the separation of scale between the different porous re-
gions is largely absent.
Due to its morphology, the porous glass tends to retain
the saturating liquid during the initial stages of drying
while precipitously losing much of the liquid in a narrow
range of relative humidity values, as evident by the water
retention data displayed in Fig. 3. Desorption and adsorp-
tion isotherms also start to merge near the lower end of
this range, indicating a loss of capillaries. The detailed dry-
ing experiments by Amberg and McIntosh (1952) place the
loss of capillary stability near 60% relative humidity.
The desorption isotherms of Vycor
TM
glass are also
fully reversible for all drying cycles, as shown by Nord-
berg (1944), Amberg and McIntosh (1952), and others,
suggesting that all measured strains are fully elastic.
The dynamic Youngs and shear moduli were measured
by Scherer (1986) who found that E
b
14:7 GPa and
G
b
6:34 GPa respectively; this corresponds to
K
b
7:2 GPa. In the literature, no indication exists that
anyone has ever directly measured the properties of
the glass skeleton. The work by Bentz et al. (1998) sug-
gest values of E
s
45 GPa and m
s
0:1 for the Youngs
modulus and Poissons ratio of the solid respectively
based on micromechanical analysis, while Amberg and
McIntosh (1952) suggest values of E
s
40 GPa and
m
s
0:25. Presently, we use m
s
0:20, a value that is typ-
ical of most glass types, along with an average of the re-
ported values of Youngs moduli; this yields
K
s
23:6 GPa. The manufacturer reports a total porosity
value of /

0:28.
The composition of Vycor
TM
glass is fundamentally dif-
ferent than that of cement paste, in that water is not an
integral part of the solid phase. Here we dene a zero sat-
uration point at 5% relative humidity, where on average
slightly more than one molecular layer coats the surface
of the glass (Emmett, 1942); this RH point is also a conve-
nient part of the referenced data set. As very little water is
lost from Vycor
TM
glass at such low humidity values, the
inuence of the zero saturation value on the predicted val-
ues of strain is small.
4.3. Predictions of the proposed drying model
Combining the proposed constitutive law given in Eq.
(11) with the expressions for p
c
and K dened in Eqs. (7),
(12) and (13), the drying shrinkage in any one direction re-
duces to Eq. (14). Similarly, the drying shrinkage based on
the constitutive law using an averaged pore pressure Eq.
(9) together with Eq. (7) reduces to the classic Biot-Bishop
constitutive law Eq. (15):
e
sh

1
3
RT
M
lnh
1
K
b

1
K

14
where K K
s

K
s
K
b
1 S
w

1 S
w
/

e
sh

1
3
RT
M
lnhS
w
1
K
b

1
K
s

15
In Figs. 4 and 5, we compare the predictions
5
of our new
formulation Eq. (14), with that of the classic averaged pore
pressure model Eq. (15). The proposed model exhibits much
better agreement with experimental data for two very dis-
tinct materials. That the pore structures of the materials
greatly differ from each other is evident by the fraction of
water each material retains at a specic relative humidity
point, as shown by the water retention data in Fig. 3. The
4
The water content of solid CSH was recently determined by Allen
et al. (2007).
5
The predictions are based on the rst-order estimate of the effective
solid properties, aided by the experimentally measured bounds (see Section
3.3). This estimate may need to be improved if porosity is found to
signicantly inuence the overall volumetric stiffness of the effective solid.
I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328 325
predicted strain values for K
b
7 and K
b
9 GPa fall on
either side of the data in cement paste and the predicted
strain values for Vycor
TM
glass are able to better capture
the trend of the measured data.
Figs. 4 and 5 also show the extent to which the classic
model Eq. (15) underpredicts experimental measurements.
The disagreement is especially prominent when the mate-
rials are truly partially saturated. This approximately cor-
responds to relative humidity range below 75% in both
materials, precisely the range in which capillary mecha-
nisms should be the most dominant. While the proposed
model still appears to underpredict the large drying strains
captured by the experiments, albeit somewhat less so in
the case of hardened cement paste, the improvement over
the classic model is readily apparent.
Recently, a new formulation was proposed by Coussy
et al. (2004) that accounts for the effects of surface energy
in cementitious materials. It was suggested that surface
energy serves to augment capillary effects, resulting in
deformations up to 30% greater than those predicted by
the classic Biot-Bishop model at 50% relative humidity.
However in light of the discussion in Section 4.1.2, the con-
tribution of the solid surface energy to deformation is
likely negligible above 50% relative humidity due to the
presence of adsorbed water.
It should further be added that it has long been recog-
nized that the complete mechanical behavior of partially
saturated materials cannot be described adequately via
use of the effective stress alone (a tensorial variable) and
that an additional variable (a scalar) is often required. In
the case of the elastic material behavior, the need for an
additional scalar variable largely stems from the observa-
tions that capillary pressure appears to augment the force
of contact between the adjacent particles or grains (e.g.
Bishop and Eldin, 1950). As the (normal) contact force be-
tween grains increases due to capillary pressure, so does
the ability of the grains in contact to withstand shear
forces (and sometimes higher tensile stresses). It is noted
that while for some materials this effect is extremely
important (e.g. one can build castles out of partly wet sand
whereas a highly saturated sand and even dry sand crum-
bles under its own weight) for others it is less so (e.g. rel-
ative humidity appears not to affect the shear strength of
cement paste, although it does affect fracture properties
(Beaudoin, 1982).
In this work, we present a new constitutive model for
the volumetric response of a porous material during dry-
ing. From the macroscopic arguments, we suggest a mod-
ication to the classic form of the constitutive law, and by
extension to the classic form of the effective stress, gener-
ally a tensorial variable. The increase in shear resistance
due to capillaries is inherently a micromechanical phe-
nomenon and is not captured by the present formulation
just as it is not captured by the classic Biot-Bishop expres-
sion herein described. In the simplest way, the increased
shear resistance can often be accounted for via an addi-
tional (scalar) variable which depends on the capillary
pressure (suction) as introduced by e.g. Gallipoli et al.
(2003). In the case of inelastic material behavior, addi-
tional considerations are likely needed. The basic argu-
ments present in this work intend to illuminate the
subject in the context of linear elasticity, which forms a
theoretical basis for the more advanced constitutive
models.
5. Conclusion
By decomposing a boundary value drying problem, a
new constitutive model is developed to predict the elastic
shrinkage due to capillary stresses during drying of a por-
ous material. The model is valid under conditions where
solid surface energy does not play an important role in
deformations and where capillary stresses due to air
water menisci are present. For materials discussed in this
work, this translates to relative humidity values above
about 50% and 60% for hardened cement paste and Vycor
TM
glass, respectively.
Fig. 5. Experimental data for Vycor
TM
glass (Amberg and McIntosh, 1952)
with predicted drying shrinkage curves. Each model is shown for
K = 7.2 GPa and K
s
= 19 GPa, modulus of the bulk and the solid skeleton,
respectively. Zero-point bulk water saturation is assumed at 5%; an exact
value is of little inuence on predicted strain values in glass.
Fig. 4. Experimental data for 56-day-old paste Gevrenov, 2005 with
predicted drying shrinkage curves. Each model is shown for K = 7 GPa and
K = 9 GPa, i.e. limits of reported bulk modulus values; K
s
= 45 GPa. Zero-
point bulk water saturation is assumed at 33%.
326 I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328
The predictions of the newmodel were evaluatedagainst
experimental measurements of linear drying strains. Hard-
ened cement paste and Vycor
TM
glass were chosen as model
materials for their distinct pore size distributions and rela-
tively large shrinkage strains that ease collection of experi-
mental data. Additionally, cement paste exhibits nearly
elastic behavior during the 2nd drying cycle and Vycor
TM
glass during all drying cycles. Good agreement was found
above50%relativehumidityfor cement paste, whilethepre-
dictions were much improved above 60% for porous glass,
without use of any other energy terms, whereas the classic
model based on average pore pressure signicantly under-
predicts the measured drying strains for both materials.
The form of effective stress in partially saturated porous
media is also discussed in light of the present formulation.
It is found that a constitutive relation for a porous system
inltrated by more than one uid does not necessitate pore
pressure averaging, as is the case for all previous formula-
tions. Instead, a material parameter needs to be evaluated,
the extent of weakening of the effective solid that results
from partial emptying of pores (this is in agreement with
experimental observations that the Bishop parameter v is
often found to be strongly inuenced by the material fab-
ric). In turn, it is required that the inuence of voids on
the solid properties are known. Presently, we estimate
these properties with the help of experimentally measured
limit values. Improved estimates of the effective solid
properties may be needed for materials containing a solid
phase whose compressibility is signicantly affected by
the presence of the voids.
Acknowledgements
This work was supported by the National Science Foun-
dation under Grant CMS 0409571. The authors would also
like to thank Zdenek Bazant, Franz-Josef Ulm, Jose Andrade,
and Dale Bentz for their critical thoughts during prepara-
tion of the manuscript.
References
Allen, A., Thomas, J., Jennings, H., 2007. Composition and density of
nanoscale calciumsilicatehydrate in cement. Nature Materials 6 (4),
311316.
Amberg, C., McIntosh, R., 1952. A study of adsorption hysterisis by means
of length changes of a rod of porous glass. Canadian Journal of
Chemistry 30 (12), 10121032.
Badmann, R., Stockhausen, N., Setzera, M.J., 1981. The statistical thickness
and the chemical potential of adsorbed water lms. Journal of Colloid
and Interface Science 82 (2), 534542.
Beaudoin, J.J., 1982. Effect of humidity and porosity on fracture of
hardened portland cement. Cement and Concrete Research 12 (6),
705716.
Bentz, D.P., Garboczi, E.J., Quenard, D., 1998. Modelling drying shrinkage
in reconstructed porous materials: Application to porous vycor glass.
Modelling and Simulation in Material Science and Engineering 6,
211236.
Biot, M.A., 1941. General theory of three-dimensional consolidation.
Journal of Applied Physics 12, 155164.
Biot, M.A., 1962. Mechanics of deformation and acoustic propagation in
porous media. Journal of Applied Physics 33 (4), 14821498.
Bishop, A., 1959. The principle of effective stress. Teknisk Ukeblad 39,
859863.
Bishop, A., Eldin, A., 1950. Undrained triaxial tests on saturated sands and
their signicance in the general theory of shear strength.
Geotechnique 2 (1), 1332.
Chateau, X., Moucheront, P., Pitois, O., 2002. Micromechanics of
unsaturated granular media. Journal of Engineering Mechanics of
ASCE 128 (8), 856863.
Christensen, R., 1990. A critical evaluation of a class of micro-mechanics
models. Journal of the Mechanics and Physics of Solids 38 (3), 379
404.
Coleman, J.D., 1962. Stressstrain relations for partially saturated soils.
Geotechnique 12 (4), 348350.
Constantinides, G., Ulm, F.-J., 2007. The nanogranular nature of CSH.
Journal of Mechanics and Physics of Solids 55, 6490.
Coussy, O., 2004. Poromechanics, second ed. Wiley-VCH, Berlin.
Coussy, O., Brisard, S., in press. Prediction of drying shrinkage beyond the
pore iso-deformation assumption. Journal of Mechanics of Materials
and Structures.
Coussy, O., Dangla, P., Lassabatre, T., Baroghel-Bouny, V., 2004. The
equivalent pore pressure and the swelling and shrinkage of cement-
based materials. Materials and Structures 37 (1), 1520.
de Boer, R., 1992. Development of porous media theoriesa brief
historical review. Transport in Porous Media 9, 155164.
Emmett, P., 1942. Measurement of Surface Areas of Finely Divided or
Porous Solids by Low-Temperature Adsorption Isotherms in Advances
in Colloid Science, vol. 1. Interscience Publishers Inc., New York.
Eshelby, J.D., 1957. The determination of the elastic eld of an ellipsoidal
inclusion and related problems. Proceedings of the Royal Society of
London Series A 241 (6), 376396.
Feldman, R., 1968. Structure and physical properties of cement pastes. In:
Proceeding of the 5th International Symposium on the Chemistry of
Cement, vol. 3. Tokyo, Japan, p. 53.
Gallipoli, D., Gens, A., Sharma, R., Vaunat, J., 2003. An elastoplastic
model for unsaturated soil incorporating the effect of suction and
degree of saturation on mechanical behaviour. Geotechnique 53 (1),
123135.
Garg, S., Nur, A., 1973. Effective stress laws for uid-saturated porous
rocks. Journal of Geophysical Research 78, 59115921.
Geertsma, J., 1957. The effect of uid pressure decline on volumetric
changes of porous rocks. Transactions of AIME 210, 331.
Gens, A., Snchez, M., Sheng, D., 2007. On constitutive modelling of
unsaturated soils. Acta Geotechnica 1 (3), 137147.
Gevrenov, J. S., 2005. Microstructure and drying shrinkage of ordinary
portland cement and alkali-activated slag pastes as a function of
relative humidity Masters Thesis, Northwestern University.
Ghosh, R.S., Timusk, J., 1974. Effect of sustained loading at early ages on
the modulus of elasticity of cement paste. Materials and Structures 7
(5), 335340.
Grasley, Z.C., Scherer, G.W., Lange, D.A., Valenza, J.J., 2006. Dynamic
pressurization method for measuring permeability and modulus: II
Cementitious materials. Materials and Structures 40 (7), 711721.
Hashin, Z., 1983. Analysis of composite materialsa survey. Journal of
Applied Mechanics 50.
Helmuth, R., Turk, D., 1967. The reversible and irreversible drying
shrinkage of hardened portland cement and tricalcium silicate
pastes. Journal of the Portland Cement Association, Research and
Development Lab 9 (2), 821.
Hershey, A., 1954. The elasticity of an isotropic aggregate of anisotropic
cubic crystals. Journal of Applied Mechanics 21, 236240.
Hutter, K., Laloui, L., Vulliet, L., 1999. Thermodynamically based mixture
models of saturated and unsaturated soils. Mechanics of Cohesive-
Frictional Materials 4, 295338.
Jennings, H.M., 2008. Renements to colloid model of CSH in cement:
CMII. Cement and Concrete Research 38, 275289.
Khalili, N., Geiser, F., Blight, G.E., 2004. Effective stress in unsaturated
soils: review with new evidence. International Journal of
Geomechanics 4 (2), 115126.
Lewis, R., Schreer, B., 1987. The Finite Element Method in The
Deformation and Consolidation of Porous Media. John Wiley and
Sons, Chichester.
Loret, B., Khalili, N., 2002. An effective stress elasticplastic model for
unsaturated porous media. Mechanics of Materials 34, 97116.
Love, A., 1944. A Treatise on The Mathematical Theory of Elasticity, 4th
ed. Dover, New York.
Mackenzie, J.K., 1950. Elastic constants of a solid containing spherical
holes. Proceedings of the Physical Society Section B 63 (1), 211.
Mindess, S., Young, F.J., 1981. Concrete. Prentice-Hall, Englewood Cliffs,
NJ.
Mori, T., Tanaka, K., 1973. Average stress in matrix and average elastic
energy of materials with mistting inclusions. Acta Metallurgica 21,
571574.
Nordberg, M., 1944. Properties of some vycor-brand glasses. Journal of the
American Ceramics Society 27, 299305.
I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328 327
Nur, A., Byerlee, J., 1971. An exact effective stress law for elastic
deformation of rock with uids. Journal of Geophysical Research 76,
64146419.
Parrott, L.J., 1974. Lateral strains in hardened cement paste under short- and
long-term loading. Magazine of Concrete Research 26 (89), 198202.
Roper, H., 1966. Dimensional change and water sorption studies of
cement paste. Symposium on Structure of Portland Cement Paste and
Concrete. Washington, D.C., pp. 7483.
Scherer, G.W., 1986. Dilatation of porous glass. Journal of the American
Ceramics Society 69 (6), 473480.
Scherer, G.W., 2006. Dynamic pressurization method for measuring
permeability and modulus: I theory. Materials and Structures 39
(10), 10411057, 1059.
Skempton, A., 1961. Effective stress in soils, concrete, and rock. In:
Proceedings of Conference on Pore Pressure and Suction in Soils.
Butterworth, London, pp. 416.
Suklje, L., 1969. Rheological Aspects of Soil Mechanics. Wiley, New York.
p. 123.
Sun, D.A., Matsuoka, H., Xu, Y.F., 2004. Collapse behavior of compacted
clays by suction controlled triaxial tests. ASTM Geotechnical Testing
Journal 27 (4), 362370.
Tennis, P.D., Jennings, H.M., 2000. A model for two types of calcium
silicate hydrate in the microstructure of portland cement pastes.
Cement and Concrete Research 30 (6), 855863.
Terzaghi, K.v., 1923. Die berechnung der durchlassigkeitsziffer des
tones aus dem verlauf der hydrodynamischen spannungsersch-
einungen. Sitzungsberichte der Akademie der Wissenschaften Wien
132, 125.
Verbeck, G. J., Helmuth, R. H., 1968. Structure and physical properties
of cement pastes. In: Proceeding of the 5th Inetrnational
Symposium on the Chemistry of Cement, vol. 3. Tokyo, Japan, pp.
132.
Weng, G.J., 1984. Some elastic properties of reinforced solids, with
special reference to isotropic ones containing spherical
inclusions. International Journal of Engineering Science 22 (7),
845856.
Yegian, M.K., Eseller-Bayat, E., Alshawabkeh, A., Ali, S., 2007. Induced-
partial saturation for liquefaction mitigation: Experimental
investigation. Journal of Geotechnical and Geoenvironmental
Engineering 133 (4), 372380.
Zienkiewicz, O., Chan, A., Pastor, M., Schreer, B., Shiomi, T., 1999.
Computational Geomechanics. John Wiley and Sons, Chichester.
328 I. Vlahinic et al. / Mechanics of Materials 41 (2009) 319328