You are on page 1of 8

Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15

Contents lists available at SciVerse ScienceDirect

Journal of Non-Newtonian Fluid Mechanics


journal homepage: http://www.elsevier.com/locate/jnnfm

A critical overview of elasto-viscoplastic thixotropic modeling


Paulo R. de Souza Mendes a,⇑, Roney L. Thompson b
a
Department of Mechanical Engineering, Pontifícia Universidade Católica-RJ, Rua Marquês de São Vicente 225, Rio de Janeiro, RJ 22453-900, Brazil
b
LCFT-LMTA-PGMEC, Department of Mechanical Engineering, Universidade Federal Fluminense, Rua Passo da Pátria 156, Niterói, RJ 24210-240, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The literature on thixotropy modeling is reviewed, with particular emphasis on models for yield stress
Received 11 February 2012 materials that possess elasticity. The various possible approaches that have been adopted to model the
Received in revised form 8 June 2012 different facets of the mechanical behavior of this kind of materials are compared and discussed in detail.
Accepted 24 August 2012
An appraisal is given of the advantages and disadvantages of algebraic versus differential stress equa-
Available online 2 September 2012
tions. The thixotropy phenomenon is described as a dynamical system whose equilibrium locus is the
flow curve, and the importance of using the flow curve as an input of the model is emphasized. Different
Keywords:
forms for the evolution equation for the structure parameter are analyzed, and appropriate choices are
Elasticity
Viscoplasticity
indicated to ensure a truthful description of the thixotropy phenomenon.
Thixotropy Ó 2012 Elsevier B.V. All rights reserved.
Modeling

1. General aspects of elasto-viscoplastic thixotropic models has been proposed that accommodates both true yield stress mate-
rials and apparent yield stress fluids [12].
The current status of modeling elasto-viscoplastic thixotropic The most usual way to model thixotropic behavior is to intro-
materials has achieved a matured stage, as a result of the develop- duce a structure parameter, typically called k, that represents the
ments of important research groups like the ones whose work is ci- level of organization of the material internal microstructure.
ted in this manuscript. Although there is still a long journey ahead The next step is generally to postulate a correlation between the
to reach the goal of predicting this kind of material behavior under structure parameter k and bulk properties such as viscosity and
complex motion, at least the viscometric response has achieved a elastic modulus (or yield stress). A common feature of thixotropy
stage, where a comprehensive critical overview is worth while. In models is to introduce an evolution equation for the structure
contrast to the general thixotropic reviews of Mewis [1], Barnes parameter to capture the time-dependent behavior of the material.
[2], Mujumdar et al.[3], and Mewis and Wagner [4], the present This is invariably formatted as a kinetic equation, where the com-
text is focused on thixotropic effects in elasto-viscoplastic materi- petition between a buildup and a breakdown term is represented
als. Hence we are particularly interested in identifying in the liter- [13]. An equilibrium or steady state condition is achieved when
ature and analyzing the approaches that were employed to model the rates of destruction and of construction of the microstructure
the thixotropic behavior combined with elasticity and plasticity. become equal.
The various models available in the literature can be roughly
grouped into two different types. In one of these, which we will call
Type I, the approach starts with a viscoplastic stress equation, to 2. Type I models
which elasticity and thixotropy are introduced, following Houska
[5] [e.g. 3,6]. In the other type, here denominated Type II, the ap- The Bingham model has a very strong influence on the visco-
proach starts with a viscoelastic stress equation, to which plasticity plasticity literature. Accordingly, a large number of thixotropy
and thixotropy are introduced [e.g. 7,8]. models available in the literature are of Type I, i.e. are based on
For modeling yield stress materials, in Type II models plasticity the Bingham model, to which other features are added.
is introduced by using a viscosity function that diverges at high The Bingham model clearly arises from the introduction of the
structuring levels [9,7]. For apparent yield stress fluids, a viscosity yield stress in the classic Newtonian fluid constitutive model.
function that possesses a high-viscosity Newtonian plateau in the The yield stress sy is an entity that traditionally appears in consti-
low strain-rate limit is employed [10,8,11]. A recent Type II model tutive equations of solid mechanics. When subjected to stress lev-
els below this yield stress, the material has a ‘‘solid-like behavior’’,
⇑ Corresponding author. Tel.: +55 2131141177; fax: +55 2131141165. and above the yield stress it possesses a ‘‘liquid-like behavior’’.
E-mail addresses: pmendes@puc-rio.br (P.R. de Souza Mendes), rthompson@ In the Bingham model, the solid-like and the liquid-like
mec.uff.br (R.L. Thompson). contributions are combined as additive parts of the total stress s.

0377-0257/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jnnfm.2012.08.006
P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15 9

Therefore, thinking in terms of a mechanical analog, we can say 2.1. Examining Type I models from the viewpoint of mechanical
that the two contributions are combined ‘‘in parallel.’’ An impor- analogs
tant consequence of this framework is that the viscosity goes to
infinity as the strain rate approaches zero. It is interesting to analyze Type I models from the mechanical-
The Newtonian fluid contribution may be replaced by a non- analog viewpoint. Note that the existence of some mechanical ana-
linear function of the strain rate. If this is done, then we get a mod- log for every model is a requirement of Newton’s second law.
ified Bingham model that keeps its essential features. For example, The two additive terms of Eq. (2) imply a mechanical analog
if a power-law dependence with the strain rate is assumed, then with two branches, one whose stress intensity is gc_ , and the other
the Herschel–Bulkley model is obtained. The Herschel–Bulkley whose stress intensity is Gce. Since c  ce, the equation for stress
model is perhaps the most representative constitutive relation of cannot be translated in to a mechanical analog in a straightforward
this kind, and also the most used one. manner. There is a need for a description of the mechanism that
The solid mechanicians usually assume that below the yield rules the deformation c  ce. A possible analog for Type I models
stress the material presents an elastic behavior, which is most of- is shown in Fig. 1. This analog possesses a mechanical component
ten ignored by the fluid mechanicians when modeling a viscoplas- in series with the elastic component Gce. Because, by definition, the
tic material. However, it is true that a number of fluid-mechanics elasticity of the model is given by the elastic component Gce alone,
oriented authors proposed ways to add elasticity in a Bingham-like it is reasonable to infer that this component—indicated by an inter-
model. rogation mark in Fig. 1—is a viscous mechanism of the form gh c_ h ,
The most intuitive way to introduce elasticity into a viscoplastic where c_ h ¼ c_  c_ e is its strain rate. The subscript h stands for ‘‘hid-
model was proposed by Oldroyd [14], by replacing the solid contri- den,’’ because this component does not appear explicitly in the
bution, i.e. the behavior below the yield stress, by an expression formulation.
that takes the form of a Hookean solid. Thus, the elastic modulus The presence of this hidden mechanical component is what
G—another traditional quantity in solid mechanics—comes into makes Eq. (2) different from the one describing a Kelvin–Voigt
play. In this case, the resulting constitutive equation is material. This hidden mechanical component governs the strain
 rate c_ h ¼ c_  c_ e . In other words, the evolution equation for ce is di-
s ¼ sy þ lp c_ if s P sy rectly related to an evolution equation for ch, which in turn is a
ð1Þ
s ¼ Gc if s < sy consequence of the physical nature of the hidden mechanical
component.
where lp is the plastic viscosity, c is the strain, and c_ is the strain Next we examine two thixotropic models representative of
rate. Interestingly, this additional elastic expression creates an Type I under the optics of the mechanical analog shown in Fig. 1.
equivalence between the yield stress and a yield strain, the latter In these models, like in the vast majority of the models available
being defined as the deformation that occurs at the yield point, gi- in the literature, the structure parameter is defined such that it
ven by cy = sy/G. varies within the range between 0 and 1; k = 1 indicates the high-
Thixotropic models of Type I follow Oldroyd’s general idea to est structuring level, while k = 0 corresponds to the lowest struc-
introduce elasticity into the stress equation. In these models, the turing level.
features of Eq. (1) are generally expressed in the following single
equation for the stress: 2.1.1. Finding the hidden mechanical component of the model of
Mujumdar et al. [3]
s ¼ Gs ce þ gs c_ ð2Þ
The stress equation of the thixotropy model proposed by
In this equation, Gs is a structural elastic modulus, gs is a struc- Mujumdar et al. [3] is summarized below:
tural viscosity, and ce is the elastic strain, i.e. the recoverable part
of the total strain. s ¼ kGo ce þ ð1  kÞK c_ n ð3Þ
(
Representative elasto-viscoplastic thixotropic models [e.g. 3,6] c_ e ¼ c_ if jce j < cy ðkÞ ðpre  yieldÞ
originate from Oldroyd’s approach described above. Four ingredi- ð4Þ
jce j ¼ cy ðkÞ ðpost  yieldÞ
ents compose the models of this type: (i) the equation for stress,
(ii) the relation between the material properties and the structure where cy ðkÞ ¼ cy0 km ð5Þ
parameter, (iii) the evolution equation for k, and (iv) an equation
In the equations above, Go is the structural elastic modulus of
for the evolution of elastic strain ce. The three first ingredients also
the fully structured material, K is the consistency index, n is the
appear in Type II models (to be discussed later), while the last
power-law index, and cy0 is the elastic strain of the fully structured
ingredient deserves special attention, due to its crucial role in the
material.
mechanical behavior predicted by Type I models.
It is clear that the stress acting at the hidden mechanical com-
The physical reasoning that leads to a specific type of evolution
ponent is the same as the stress acting at the elastic component. As
equation for ce is typically not discussed in enough detail in the lit-
discussed earlier in a general context, since there is no elasticity in
erature, in contrast to the prolific discussions often found to de-
scribe the underlying physics of the other model ingredients. For
example, it is common that the evolution equation for the struc-
ture parameter be described as a competition between a break-
down term and a buildup term. It follows that some general
features arise from this viewpoint which provide guidance to the
construction of a new model. The process of building a specific evo-
lution equation for k is thus reduced to deciding what characteris-
tic thixotropic time will be used, which parameters should be
chosen to appear in the buildup term, what physical quantity is
responsible for the destruction of the material microstructure,
and so on, as it will be discussed later on in this text. There are
no counterparts, clearly stressed in the literature, for the evolution
equation of ce. Fig. 1. A possible mechanical analog for Type I models.
10 P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15

the hidden component, then it is treated as a viscous component. In the case that the structural elastic modulus does not depend
Hence, on the structuring level, then Eq. (12) takes the same form of Eq.
(11), even if the structural viscosity gs is strongly dependent on k.
kGo ce
kGo ce ¼ gh c_ h ) gh ¼ : ð6Þ This fact was explored by several authors [e.g. 19,10,20] to construct
c_ h models in which the polymeric contribution is given by an equation
From Eqs. (4) and (5) we have that of the form of Eq. (11), with the Maxwell viscosity l being replaced
8 with a structural viscosity gs(k). Some of these authors have also
< gh ! 1 ðpre  yieldÞ
added a matrix contribution, and ended up with a Jeffreys-like
kGo cy0 km ð7Þ
: gh ¼ _ m1 dk ðpost  yieldÞ model. Quemada [10] goes a little further and also considers a
cc y0 mk dt
Burger’s-like model. Clearly, all these models are restricted to mate-
Eq. (7) illustrates that the physical nature of the hidden rials whose structural elastic modulus is independent (or at most a
mechanical component of this model is not clear. very weak function) of the structuring level (dGs/dk ’ 0).
One of the few models for elasto-viscoplastic thixotropic mate-
2.1.2. Finding the hidden mechanical component of the model of rials based on Eq. (12) was proposed by Yziquel et al. [7], who
Dullaert and Mewis [6] adopted a full 3D approach by replacing the time derivative with
The stress equation of the thixotropy model proposed by an upper convected one. They also introduced a solvent viscosity
Dullaert and Mewis [6] is summarized by the following equations: to come up with an Oldroyd-B-like equation. An important contri-
bution of this work is a comparison between different origins of
s ¼ kGo ce þ ðkgs0 þ g1 Þc_ ð8Þ structure destruction. This aspect is discussed in more detail
 b below.
k
c_ e ¼ 4 ½sðk; c_ Þcy0  seq ðc_ Þce  ð9Þ A recent model of Type II was proposed by de Souza Mendes
t
[8,11]. In these works, a novel assumption has lead to a
where gs0 is the structural viscosity of the fully structured material, Maxwell-like [8] or Oldroyd-B-like [11] stress equation, with no
g1 is the viscosity of the material at the lowest possible structuring restrictions regarding the dependence of the structural elastic
level, t is the time, seq ðc_ Þ is the equilibrium (steady-state) stress cor- modulus on the structuring level.
responding to c_ , and k4 and b are constant parameters. The two parts of this assumption can be summarized as follows.
Following the same procedure that we used above for the model Firstly, a neutral configuration is introduced, represented by the
of Mujumdar et al. [3], we can determine the hidden mechanical neutral elastic strain cen. All configurations are defined with re-
component for the model of Dullaert and Mewis [6]. The following spect to any arbitrary but fixed reference configuration. This neu-
expression for the hidden viscosity gh is obtained: tral configuration corresponds to a state of zero elastic stress,
kGo ce and hence the elastic stress is computed as
gh ¼  b
k4
c_  t
½Go ce cy0 ðk  keq Þ þ gs0 c_ ðkcy0  keq ce Þ þ g1 c_ ðcy0  ce Þ se ¼ Gðce  cen Þ: ð13Þ
ð10Þ The second part of the assumption consists of postulating that
the neutral configuration changes if and only if the structuring le-
where keq is the equilibrium (steady-state) value of the structure
vel changes, which leads to
parameter corresponding to c_ . Thus, the physical nature of the hid-
den mechanical component of this model is even more complex.
1 dGs dk
c_ en ¼ ðc  cen Þ ð14Þ
Gs dk dt e
3. Type II models
Eq. (14) says that the neutral configuration remains unchanged
Another approach for developing the stress equation for an in three (non-exclusive) cases: (i) when the elastic modulus does
elasto-viscoplastic thixotropic model is to start from a viscoelastic not depend on the structuring level; (ii) when the structuring level
model for the stress equation. In this case, the Maxwell model— is not changing, or (iii) when the present configuration coincides
the most fundamental of the viscoelastic constitutive with the neutral configuration, even if the structuring level is
equations—plays the counterpart of the Bingham model when changing.
the goal is to combine solid-like and liquid-like behaviors. Instead Note that, in all the other models mentioned above, the exis-
of combining plasticity with viscous effects, the Maxwell model tence of a neutral configuration that depends on the structuring le-
combines elasticity with viscous effects. In this case, the mechan- vel has not been given the deserved attention. It is well known that
ical analog that represents the combined effects consists of a this issue plays a major role in the description of the mechanical
spring and a dashpot in series, and the resulting equation is well behavior of the type of materials of interest in this text [21–
known: 23,8,11], and neglecting this fact is expected to lead to unphysical
l predictions.
s þ s_ ¼ lc_ : ð11Þ To illustrate this, let us imagine an elasto-viscoplastic thixotro-
G
pic material that is initially fully structured, stress-free and at rest.
where l is the Maxwell viscosity. The way of introducing plasticity At time t = 0, it is subjected to a very high stress which is main-
and thixotropic effects is inspired in the series of papers by tained for a long time, such as to bring its structuring to the lowest
Marrucci et al. [15], and Acierno et al. [16–18], who deal with level, corresponding to a state of no elasticity and low viscosity.
viscoelastic models for non-dilute regimes, where entanglement be- Then, the material is brought to rest. A gradual buildup of the
tween the macromolecules has a significant role on the mechanical microstructure is then observed, and, during this process the mate-
response. If the parameters of the Maxwell model are assumed to rial is expected to remain at rest and stress-free. After a long en-
depend on the structure parameter k, the following constitutive ough period of time, the material initial condition of fully
equation is obtained: structured, stress-free and at rest is expected to be achieved. If
" # the neutral configuration issue is not treated appropriately, how-
gs ðkÞ dGs dk gs ðkÞ _
s 1 þ s ¼ gs ðkÞc_ : ð12Þ ever, a stress buildup may be predicted during the just described
Gs ðkÞ2 dk dt Gs ðkÞ
microstructure buildup process [e.g. 3].
P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15 11

Type II models also require additional equations to relate the of the terms ‘‘increasing elasticity’’ and ‘‘increasing elastic modu-
structural viscosity and the structural elastic modulus to the struc- lus’’, which is not always correct. The increase of the elastic mod-
ture parameter, and also an evolution equation for the structure ulus represents an increase on the stiffness only, which by itself is
parameter itself. However, Type II models do not need an addi- not determinant of the level of elasticity, as discussed in the previ-
tional evolution equation or rule for ce, in contrast to Type I mod- ous paragraph. This misunderstanding is certainly responsible for
els, because this information is already embedded in their stress the fact that, with rare exceptions [8,11], most Type II models em-
equation. This additional information is needed by Type I models ploy expressions for the structuring-level-dependent elastic modu-
because their stress equation does not provide it; it is exactly this lus that, in the limit of zero elasticity, go to zero instead of infinity.
information that is needed to fully construct their mechanical ana- A striking consequence of the opposite trends of the analog
log (see Eqs. (7) and (10)). arrangements discussed above is that the Type I models do not re-
In conclusion, models of Type II are capable of predicting elastic duce to their inelastic counterpart in the limit of zero elasticity, as
and thixotropic behavior, and plasticity is readily introduced by it would be expected. In other words, in the framework of Type I
expressing the bulk parameters as functions of the structuring le- models, in applications, where elasticity is not considered, the
vel such that the structural viscosity is unbounded when the mate- yielding term is just a structuring-level-dependent yield stress,
rial is fully structured [19,7,12]. while when elasticity is considered the yield stress is written as
Gsce. Note that, in the case of Type I models, the limit of zero elas-
4. Comparing the stress equations of Type I and Type II models ticity implies a null elastic modulus (Gs ? 0). Thus the resulting
model in this limit is at most a thixotropic model with no yield
It is not difficult to observe that the stress equations of Type II stress, because both the yield stress and elasticity appear in the
models can be represented by the mechanical analog shown in same term. Fig. 1 may be useful to further illustrate this point.
Fig. 2. In contrast to what was observed in Type I models, the infor- Therefore, a clear inconsistency arises at low elasticity levels. This
mation contained in the analog is fully contained in the stress is a major limitation that indicates that the main hypotheses
equation. adopted by Type I models are not representative of the underlying
If g1 is set to zero, then the analog reduces to Maxwell-like physics of the phenomena involved.
form. Note that g1 is the viscosity of the material corresponding On the other hand, Type II models reduce nicely to inelastic
to its lowest structuring level, and hence setting it to zero repre- viscoplastic thixotropic models in the limit of zero elasticity
sents a rather unphysical assumption that is hardly defensible in (G ? 1), as readily inferred from Fig. 2. This is so because the yield
a constitutive model that intends to possess a broad range of stress effects appear in the structural viscosity (which diverges at
applicability. high structuring levels), being thus decoupled from the elastic
From the mechanical analog viewpoint, the main difference be- terms. Thus it can be concluded that the in-series analog consti-
tween the two types of models is the way that the structure- tutes a more truthful representation of the microstructure
dependent mechanical elements are arranged in the analog. In mechanical behavior of these systems.
Type I models, the mechanical element related to the structural Mewis and Wagner [4] analyzed some aspects of models of
viscosity and the one related to the structural elastic modulus ap- Type II when compared to models of Type I. Regarding the model-
pear in parallel, while in Type II models they are arranged in series. ing of elasto-viscoplastic thixotropic behavior, in their general
Hence, in Type II models the structure-dependent mechanical ele- discussion they clearly recommend models of Type I. The argu-
ments are subjected to the same stress level, while in Type I mod- ments invoked to support such preference are expressed in
els they are subjected to neither the same stress nor the same sentences like ‘‘close link between thixotropy and yield stress,’’
strain rate. and ‘‘nearly all thixotropic materials have a yield stress.’’ These
It is important to notice that these differences lead to opposite statements intend to justify the fact that, in the formulation of
responses to changes of the structural elastic modulus and struc- all Type I models, most of the dependence on the structuring level
tural viscosity. When the elastic mechanical element is in parallel enters the formulation in the term of the stress equation that
with the viscous one, the elastic modulus and the viscosity act as resembles a yield stress.
enhancers of elasticity and dissipative behavior, respectively. On They also state that ‘‘The lack of a yield stress is a serious draw-
the other hand, when they are combined in series, the elastic mod- back of the models derived from the Maxwell model,’’ but in the
ulus and viscosity act as inhibitors of elasticity and viscous effects, sequence they mention that it is possible to eliminate this limita-
respectively. In other words, as the characteristic time g/G ? 1, a tion by using a structural viscosity that diverges at high structuring
purely viscous behavior is obtained when the elastic and viscous levels.
elements are in parallel, and a purely elastic behavior is obtained Their ultimate argument in favor of Type I models and against
when they are arranged in series. When g/G ? 0, the trends are Type II models is now quoted: ‘‘It should be pointed out that a con-
of course reversed. stant yield stress precludes any flow below that value. As discussed
The importance of this fact cannot be overemphasized, since in above this is not consistent with the available experimental evi-
the literature it is rather common to find the interchangeable usage dence that suggests that yielding should depend on the shear his-
tory.’’ It is important to observe that this argument is flawed
because it is a circular argument.
The reason is that the mentioned experimental evidence is ana-
lyzed under the optics of the Type I model paradigm. In other
words, to conclude that yielding should depend on the shear his-
tory implies the prior assumption that the correct stress equation
is the one adopted in Type I models, where most of the burden
of describing thixotropy falls on the yield-stress-like term. Under
the optics of Type II models, in which formulation most of the
dependence on the structuring level is introduced in the structural
viscosity, the above quoted argument would not make sense, be-
cause yielding in the classical sense does not depend on shear his-
Fig. 2. The mechanical analog for Type II models. tory, as discussed in the next section.
12 P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15

5. Comments on yielding As a final comment on equations of the type of Eq. (15)—which,


as mentioned above, originate from the classic Herschel–Bulkley
The yield stress sy was firstly introduced as a steady-state flow flow curve—it is interesting to observe that the flow curve corre-
concept, being defined as the stress below which no unrecoverable sponding to these equations is not the Herschel–Bulkley one, inas-
deformation is observed. It is thus a single fixed parameter of the much as sy, K, and n will be functions of the shear rate:
flow curve, and below the yield stress there is no steady state flow
(c_ ¼ 0 in the flow curve). It is simply given by
s ¼ sy ðkeq ðc_ ÞÞ þ Kðkeq ðc_ ÞÞc_ nðkeq ðc_ ÞÞ ð17Þ
sy ¼ limt!1 ½limc_ !0 sðc_ ; tÞ. The yield stress is also the minimum where keq ðc_ Þ is the structural parameter under equilibrium (steady-
stress required to trigger a major microstructure breakdown in state) conditions (see Eq. (19) below for an example of keq ðc_ Þ). Con-
an initially fully-structured material. It is thus a quantity related sequently, the thixotropy models that employ stress equations like
to the fully structured state only. Eq. (15) do not reduce to the Herschel–Bulkley equation as thixot-
Hence, to use the term yield stress to denote a quantity that is a ropy becomes negligible. This important fact is not discussed in
function of the structuring level is clearly not consistent with the the literature, and, actually, flow curve equations are seldom given
concept of yield stress. This inconsistency is commented by Mewis explicitly.
and Wagner [4]: ‘‘It could be argued on physical grounds that
attributing a Bingham yield stress to a structure that exists during
6. The role of the flow curve in thixotropy modeling
flow is not suitable. In practice it is often found to be adequate
parameter to fit constant k curves.’’
6.1. General aspects
As discussed earlier, the strategy usually employed by the most
representative Type I models in order to incorporate thixotropy ef-
Some theories for predicting the behavior of microstructured
fects is to modify a classical flow curve expression by introducing
liquids are developed with basis on features of the microstructure
the dependence on the structure parameter. Taking the Herschel–
and its interactions with the main flow. Other theories are based
Bulkley equation as an example, the idea is to allow the flow curve
on mechanical analogs.
parameters to depend on the structuring level:
The resulting model is then checked with respect to its ability to
predict the trends observed experimentally in different flow situa-
s ¼ sy ðkÞ þ KðkÞc_ nðkÞ ð15Þ tions, including steady shear flow and other rheological flows. If
the trends are correct, then the model parameters can in principle
It is worth noting that, in most applications that use this type of
be determined, acquiring the status of properties of the material.
equation [e.g. 24], K and n are taken as constants, so that the
In the case of thixotropy modeling, an alternate approach for
dependence on the structuring level appears in the first RHS term
developing a constitutive theory is to use the flow curve as an in-
(sy(k)) only.
put for the model [8,11]. Since this curve can in principle be ob-
The underlying physics of Eq. (15) is not clear. Due to the histor-
tained in laboratory, it is a material property that contains
ical use of the symbol sy in association with the yield stress, one is
valuable information about the material behavior in shear, being
readily invited to interpret sy(k) as a yield stress that depends on
analogous to the viscosity of the Newtonian fluids. Hence, to use
the structure parameter. And in fact this is the usual interpretation
this information in the construction of the model seems to be a
in a vast majority of articles. However, while the classic Herschel–
natural choice.
Bulkley equation represents a flow curve, Eq. (15) does not. Actu-
A similar approach was used by Schunk and Scriven [25] and
ally, Eq. (15) represents a family of curves parametrized by k (see
others [e.g. 26–30] in the context of predicting viscoelastic behav-
Barnes [2] and Fig. 4 therein). ior in complex flows.
The flow curve is the relation between stress (or viscosity)
One immediate advantage of this approach is the obvious fact
and strain rate that is observed in steady flow. Each point of that the flow curve is predicted exactly. Other advantages are dis-
the flow curve is associated with a different value of k, i.e. a dif-
cussed in detail in the following sections. It will be shown that the
ferent structuring level. Therefore, the flow curve is intercepted importance and convenience of using the flow curve as an input for
by each and every curve obtained from equations of the type
a thixotropy model cannot be overemphasized.
of Eq. (15). This means that, if an equation such as Eq. (15) is
employed, then there will be a different value of sy(k) associated
with each point of the flow curve, even in the range of very high
strain rates. While the material is flowing, no yielding process—
i.e. dramatic change in the microstructure—occurs that could
possibly justify interpreting sy(k) as a yield stress. Therefore it
becomes quite clear that the only resemblance between sy(k)
and the concept of yield stress originates from the form of
Eq. (15).
It is worth emphasizing that the discussion above is directed to-
wards physical meaning and nomenclature issues only. There is no
implication regarding the appropriateness of employing stress
equations of the form of Eq. (15). A straightforward way to pre-
serve the concept of yield stress and to preclude further nomencla-
ture confusions is to use different symbols to denote the structure
dependent parameter in Eq. (15):

sðk; c_ Þ ¼ AðkÞ þ BðkÞc_ CðkÞ : ð16Þ

The parameter A(k) is given by AðkÞ ¼ limc_ !0 sðk; c_ Þ, and it


should be seen just as a convenient quantity for fitting constant k
curves, as suggested by Mewis and Wagner [4]. Fig. 3. The flow curve as the equilibrium locus of the dynamical thixotropic system.
P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15 13

6.2. The flow curve as the set of all attractor points For evolution equations like Eq. (18), which possess a break-
down term that is a function of the strain rate, it is seen that the
The flow curve is the equilibrium locus of the dynamical thixo- keq value that is the attractor point for the (non-equilibrium) value
tropic system. Therefore, the evolution equation for the structure of k at time t depends on the value of the strain rate at the same
parameter must be such that these equilibrium points are reach- time t, say c_ t .
able and linearly stable. This is better understood with the aid of Fig. 3, where we can
Considering viscometric flows, the flow curve is thus the set of see that a constant strain rate experiment starting at time t = t0
attractor points of non-equilibrium states on the Cartesian plane is a vertical trajectory c_ ¼ c_ t0 on the c_  s plane. During the exper-
c_  s. In other words, if at some instant of time t0 the point iment, both the function gðc_ Þ and the equilibrium structure param-
ðc_ t0 ; st0 Þ on this plane does not belong to the flow curve (Fig. 3), eter keq remain fixed at gðc_ t0 Þ and k1 =½k1 þ k2 gðc_ t0 Þ, respectively.
then it represents a non-equilibrium state of the microstructure, The point of the flow curve that plays the role of the attractor of
and hence there is a potential that drives this point along a trajec- the dynamical system is a fixed point corresponding to the pair
tory that ends at some point on the flow curve. ðc_ t0 ; sðc_ t0 ÞÞ, where the function sðc_ Þ gives the value of s of the
Therefore, it becomes clear that introducing the flow curve as (steady-state) flow curve corresponding to c_ .
input into the thixotropy model defines a priori the correct locus A constant shear stress experiment starting at time t = t0 would
of atractor points, which constitutes a piece of information that lead to a horizontal trajectory s ¼ st0 on the c_  s plane. However,
is crucial to the predictive capability of the model. due to the dependence of the function g on c_ , the equilibrium struc-
It remains to determine which point on the flow curve is the ture parameter keq is not constant. Since keq is a function of the cur-
attractor point for the current state. This issue is not as straight- rent value of the strain rate, this quantity varies along the constant
forward as it may seem at first glance. For example, in a constant stress trajectory. Therefore, the attractor point on the flow curve
strain rate experiment, the steady-state situation which corre- also changes with time. At the current time t, it corresponds to
sponds to the current state is defined by the intersection between the point ðc_ t ; sðc_ t ÞÞ on the flow curve, where c_ t0 6 c_ t 6 s1 ðst0 Þ,
the vertical line of the current strain rate and the flow curve. On and the function s1(s) is the inverse function of sðc_ Þ, i.e. it gives
the other hand, in a constant stress experiment the steady-state the value of c_ on the flow curve corresponding to s.
point corresponding to the current state is defined by the inter- In summary, equations such as Eq. (18) implicitly assume that
section between the horizontal line of the current stress and the changes of structuring level are dictated by the strain rate.
the flow curve. In these two flows, it may appear that this stea- A totally different response is obtained if instead we employ the
dy-state point is the only attractor point throughout the flow, following form of evolution equation for k:
but it may be not.
As it is explained in the following section, in fact the attractor
dk 1 c1
point is defined by the characteristics of the evolution equation ¼ ½c1 ð1  kÞ  c2 kf ðsÞ ) keq ðsÞ ¼ ; ð21Þ
dt t eq c 1 þ c 2 f ð sÞ
for the structure parameter used by the model. Therefore, these
characteristics must bear the correct physics, under penalty of
unrealistic predictions. where teq, c1 and c2 are non-negative constants and f(s) is a non-
negative function of the stress. An equation similar to Eq. (21)
6.3. The implicit choice of the attractor point was proposed by De Kee and Fong [9], for a kinetic equation for
the number of structural bonds.
We now analyze a typical kinetic equation whose form is repre- For evolution equations like Eq. (21), in a constant strain rate
sentative of several evolution equations for the structure parame- experiment (c_ ¼ c_ t0 ) starting at t = t0, the attractor point on the
ter found in the literature, for the case that k is chosen to vary flow curve changes with time, while in a constant stress experi-
between 0 (fully unstructured state) and 1 (fully structured state). ment (s ¼ st0 ) starting at t = t0, the atractor point is fixed, corre-
This typical equation is written as sponding to keq ðst0 Þ. Therefore, equations such as Eq. (21)
implicitly assume that the changes of structuring level are dictated
dk 1
¼ ½k1 ð1  kÞ  k2 kgðc_ Þ; ð18Þ by the stress.
dt teq
It is thus clear that these two possible kinds of evolution equa-
where teq, k1 and k2 are non-negative constants and gðc_ Þ is a non- tion lead to totally different predictions for the dynamic behavior
negative function of the strain rate. Eq. (18) assumes that the build- of the thixotropic system. The recognition of this fact has not re-
up term is proportional to the ‘‘distance’’ (1  k) from the present to ceived the deserved attention and is not an issue commonly dis-
the fully structured state, and that the breakdown term is propor- cussed in the literature of thixotropy. A notable exception is
tional both to the ‘‘distance’’ (k  0) from the present to the fully found in Yziquel et al. [7], where three possible causes of structure
unstructured state and to a function of the strain rate, gðc_ Þ. The destruction are analyzed: the level of strain rate, the first invariant
equilibrium (steady-state) value of the structure parameter, keq is of the stress, and the level of dissipation. Comparing with experi-
given by mental results, they conclude that the assumption that dissipation
is the main cause of breakdown was the better choice. However
k1 they did not test the level of stress (represented by the second
keq ðc_ Þ ¼ ð19Þ
k1 þ k2 gðc_ Þ invariant of the stress tensor).
In the next section we discuss the underlying physics of the two
keq ðc_ Þ is the value of the structure parameter corresponding to the options, namely to assume that breaking of the microstructure is
flow curve at the strain rate c_ . governed by the stress or by the strain rate.
It is observed that the equilibrium states as given by Eq. (19) are Before concluding this section, however, we describe how to
linearly stable, since, for a small departure from the equilibrium use the flow curve as an input of the thixotropic model, irrespec-
state Dk, the corresponding derivative of k acquires the sign that tive of the type chosen for the evolution equation or for the stress
will drive k back to its equilibrium value: equation (Type I or Type II).
Because the structural viscosity is given as a function of k (one
dk k1 þ k2 gðc_ Þ of the assumptions needed in any model), then, if we apply this
¼ Dk; ð20Þ
dt t eq function to the equilibrium state and invert it, then we obtain keq
14 P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15

as a function of the equilibrium structural viscosity, which is given inclined plane. Coussot et al. [31] report very interesting experi-
by the flow curve. mental results for a related problem.
The function gðc_ Þ (or the function f(s), depending on the type of Before discussing model predictions, it is worth describing the
the evolution equation) is readily determined by writing it as a physical behavior expected. It is clear that, if the inclination is
function of the equilibrium structure parameter keq. Then it can not large enough to generate stresses above the yield stress, then
be eliminated in the evolution equation in favor of keq: the material will remain fully structured and at rest forever, by def-
  inition of yield stress. For a large enough inclination, the stress le-
k1 1  keq ðc_ Þ dk k1 1  keq ðc_ Þ vel in the material is above the yield stress, and hence breakage of
gðc_ Þ ¼ ) ¼ ð1  kÞ  k ð22Þ
k2 keq ðc_ Þ dt teq keq ðc_ Þ the microstructure occurs. Therefore, the lump will eventually
start moving and deforming down the plane. If thixotropy is negli-
or gible, the flow starts immediately, while, if not, there will be a de-
  lay between the moment that the inclination is applied and the
c1 1  keq ðsÞ dk c1 1  keq ðsÞ moment that the structuring level becomes low enough to cause
f ðsÞ ¼ ) ¼ ð1  kÞ  k ð23Þ
c2 keq ðsÞ dt teq keq ðsÞ the viscosity to decrease dramatically, which in turn causes a sud-
den flow to take place—the avalanche effect.
Note that this procedure reduces by one the number of con-
Since the initial strain rate is zero in this case (neglecting elas-
stants of the model (k2 and c2 cancel out), and embeds in the for-
ticity), models that assume that microstructure breakage is due to
mulation the correct attractor points of the dynamic thixotropic
the strain rate will predict no breakage1 of the microstructure,
system by using the flow curve to determine keq.
meaning that the lump is predicted to remain at rest and fully struc-
This procedure, firstly proposed by de Souza Mendes [8], seems
tured forever, regardless of the level of stress! On the other hand, mod-
more reasonable than the usual approach, which consists of arbi-
els that assume that microstructure breakage is due to the stress will
trarily guessing a function gðc_ Þ (or f(s)) and then obtaining a keq
perform as expected. The breakdown term will be zero while in the
which will be a direct function of this arbitrary guess. The resulting
elastic regime, and will be an increasing function of the stress in the
thixotropy model is usually unable to give acceptable predictions
case that the stress level is above the yield stress. Moreover, the ava-
not even for the flow curve, and, more important, the thixotropic
lanche effect is nicely predicted, the delay time being determined by
dynamical system is governed by incorrect driving potentials.
both the stress level and the equilibrium time teq.
It is also important to emphasize that, even when elasticity ef-
6.4. What breaks the microstructure, is it the strain rate or the stress? fects are not present, it is important to assume that the breakdown
term is a function of the stress rather than the strain rate, as dis-
In the previous section we showed that Eqs. (18) and (21) rep- cussed next.
resent rather different descriptions of the evolution of the thixotro- To illustrate this, it is worth analyzing the Lagrangian motion of
pic dynamical system. Therefore, it is important to decide what is a material particle that occupies a given position in a given instant
the choice that gives the best description of the physics involved. of time, in a complex flow of an inelastic thixotropic yield-stress
In what follows we argue that it seems much more physically material. If at this instant of time the material particle is fully
sound to assume that breaking of the microstructure occurs due structured (k = 1), then its instantaneous strain rate must be zero
to the stress level. This is because the microstructure involves (because zero elasticity is assumed). If the breaking agent is chosen
bonds between structural units, and bonds consist of interparticle to be the strain rate, then the breakdown term will always be null.
forces. Thus, to break a bond it takes some external force, provided This means that the microstructure of this material particle is pre-
by the stress level. On the other hand, there is no physical argu- dicted to never break, regardless the level of stress to which it is ex-
ment to support the strain rate as the agent that causes bond posed. Of course this is unacceptable. On the other hand, it is clear
breaking. This line of arguments in favor of the stress as the break- that the correct prediction is again obtained when the breakdown
ing agent had already been adumbrated by de Souza Mendes term is assumed to be a function of the instantaneous local stress.
[8,11], but the evolution equation for k proposed therein is not Finally it is observed that the two choices for the breaking agent
quite consistent with this viewpoint. The correct formulation can discussed above become equivalent in the particular case of yield-
be found in de Souza Mendes and Thompson [12]. stress materials that possess no thixotropy (teq = 0), because k is al-
It is possible to imagine situations in which the stress is zero ways equal to keq in this case. In this connection, it is interesting to
but the strain rate is arbitrarily large, and, since there is no stress, observe that, in this limit of no thixotropy, Type II models reduce to
no bond breaking can occur. For example, we can imagine a fully elasto-viscoplastic models that look like Maxwell or Jeffreys mod-
structured material initially stress-free and at rest, to which at els, with the structural elastic modulus and structural viscosity
time t = 0 a constant shear rate c_ is applied. The stress is zero at being functions of the equilibrium structure parameter keq. The
t = 0 and increases linearly with time during the elastic regime, un- resulting models are thus aligned with the standard formulation
til it reaches the yield stress. No breaking of the microstructure is used to represent this type of behavior. On the other hand, in the
expected during the elastic regime (by definition of elastic regime same limit of teq = 0 (no thixotropy), Type I models reduce to mod-
and yield stress). els that have no resemblance with the usual formulations, and
Since the strain rate is constant in this case, models that assume whose physical interpretation is not quite clear.
that microstructure breakage is due to the strain rate will predict
the highest breaking rate exactly at t = 0 (because k = 1 and gðc_ Þ 7. Final remarks
is constant), when the stress is zero and hence no breaking what-
soever can occur. On the other hand, models that assume that In this paper the different aspects involved in elasto-
microstructure breakage is due to the stress will perform as ex- viscoplastic thixotropic models are discussed in detail.
pected, because the breakdown term will be zero all the way dur- To describe thixotropy, most models rely on the introduction of
ing the elastic regime (since k = keq = 1 and hence there is no a structure parameter, which is a measure of the structuring level
driving potential).
It is also worth examining the performance of models employ- 1
Here and in the remainder of this text it is assumed that the function gðc_ Þ
ing each of the options in the case of a lump of thixotropic elasto- appearing as a factor in the breakdown term is null when c_ ¼ 0, as it is invariably
viscoplastic material, initially at rest and fully structured, on an postulated.
P.R. de Souza Mendes, R.L. Thompson / Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15 15

of the microstructure. This structure parameter is governed by an [6] K. Dullaert, J. Mewis, A structural kinetics model for thixotropy, J. Non-
Newtonian Fluid Mech. 139 (2006) 21–30.
evolution equation that associates the rate of change of the struc-
[7] F. Yziquel, P. Carreau, M. Moan, P. Tanguy, Rheological modeling of
turing level with a competition between the processes of buildup concentrated colloidal suspensions, J. Non-Newtonian Fluid Mech. 86 (1999)
and breakdown of the microstructure. 133–155.
In addition to the evolution equation for the structure parame- [8] P.R. de Souza Mendes, Modeling the thixotropic behavior of structured fluids, J.
Non-Newtonian Fluid Mech. 164 (2009) 66–75.
ter, all models possess a stress equation and expressions that relate [9] D. De Kee, C. Fong, Letter to the editor: a true yield stress?, J Rheol. 37 (1993)
the structural elastic modulus and the structural viscosity with the 775.
structuring level. [10] D. Quemada, Rheological modelling of complex fluids: IV: Thixotropic and
‘‘thixoelastic’’ behaviour. Start-up and stress relaxation, creep tests and
If the stress equation is algebraic and originated from the hysteresis cycles, Eur. Phys. J. AP 5 (1999) 191–207.
Bingham model (Type I models), then an additional equation is [11] P.R. de Souza Mendes, Thixotropic elasto-viscoplastic model for structured
needed to describe the elastic strain ce. For models whose stress fluids, Soft Matter 7 (2011) 2471–2483.
[12] P.R. de Souza Mendes, R.L. Thompson, A unified approach to model elasto-
equations are differential and originate from the classic viscoelastic viscoplastic thixotropic yield-stress materials and apparent-yield-stress fluids,
models (Type II models), no additional equation is needed. Rheol. Acta. submitted for publication.
It is argued that the framework of Type II models is more robust, [13] F. Moore, The rheology of ceramic slips and bodies, Trans. Brit. Ceram. Soc. 58
(1959) 470–492.
because [14] J.G. Oldroyd, A rational formulation of the equations of plastic flow for a
Bingham solid, Proc. Cambridge Philos. Soc. 43 (1947) 100–105.
 the underlying physics of Type II models are clearly described [15] G. Marrucci, G. Titomanlio, G.C. Sarti, Testing of a constitutive equationfor
entangled networks by elongational and shear data of polymer melts, Rheol.
by a mechanical analog, and thus the elastic strain ce—a variable
Acta 12 (1973) 269–275.
pertaining to the microscopic description only—does not appear [16] D. Acierno, F.P. La Mantia, G. Marrucci, G. Titomanlio, A non-linear viscoelastic
explicitly in the formulation; model with structure-dependent relaxation times: I. Basic formulation, J. Non-
 Type II models can be seen as general models that reduce nicely Newtonian Fluid Mech. 1 (1976) 125–146.
[17] D. Acierno, F.P. La Mantia, G. Marrucci, G. Rizzo, G. Titomanlio, A non-linear
to usual formulations for a wide range of special cases of inter- viscoelastic model with structure-dependent relaxation times: II. Comparison
est, such as inelastic thixotropic viscoplastic materials, and sev- with LD polyethylene transient stress results, J. Non-Newtonian Fluid Mech. 1
eral types of non-thixotropic materials like elasto-viscoplastic (1976) 147–157.
[18] D. Acierno, F.P. La Mantia, G. Marrucci, A non-linear viscoelastic model with
materials, and viscoelastic solids and fluids; structure-dependent relaxation times: III. Comparison with LD polyethylene
 The formulation adopted by Type II models is consistent with creep and recoil data, J. Non-Newtonian Fluid Mech. 2 (1977) 271–280.
the yield stress concept in its classical sense. [19] P. Coussot, A.I. Leonov, J.M. Piau, Rheology of concentrated dispersed systems
in a low molecular weight matrix, J. Non-Newtonian Fluid Mech. 46 (1993)
179–217.
The thixotropy phenomenon is described as a dynamic system [20] F. Bautista, J.M. de Santos, J.E. Puig, O. Manero, Understanding thixotropic and
whose equilibrium locus is the flow curve, and the importance of antithixotropic behavior of viscoelastic micellar solutions and liquid
crystalline dispersions. I. The model, J. Non-Newt. Fluid Mech. 80 (1999) 93–
using the flow curve as an input of the model is explained in detail. 113.
Lastly, it is shown that the breakdown term of the evolution [21] K.R. Rajagopal, A.R. Srinivasa, Mechanics of the inelastic behavior of
equation for the structure parameter must be a function of the materials—Part I. Theoretical underpinnings, Int. J. Plast. 14 (1998) 945–967.
[22] K.R. Rajagopal, A.R. Srinivasa, Mechanics of the inelastic behavior of
instantaneous stress (rather than strain rate), under penalty of an
materials—Part II. Inelastic response, Int. J. Plast. 14 (1998) 969–995.
unphysical description of the thixotropy phenomenon. [23] K.R. Rajagopal, A.R. Srinivasa, A thermodynamic framework for rate type fuid
models, J. Non-Newtonian Fluid Mech. 88 (2000) 207–227.
Acknowledgments [24] A.N. Alexandrou, N. Constantinou, G. Georgiou, Shear rejuvenation, anging and
shear banding in yield stress fluids, J. Non-Newtonian Fluid Mech. 158 (2009)
6–17.
The authors are indebted to Petrobras S.A., MCTI/CNPq, CAPES, [25] P.R. Schunk, L.E. Scriven, Constitutive equation for modeling mixed extension
FAPERJ, and FINEP for the financial support to the Group of Rheol- and shear in polymer solution processing, J. Rheol. 34 (1990) 1085–1117.
[26] P.R. de Souza Mendes, M. Padmanabhan, L.E. Scriven, C.W. Macosko, Inelastic
ogy at PUC-RIO. constitutive equations for complex flows, Rheol. Acta 34 (1995) 209–214.
[27] R.L. Thompson, P.R. Souza Mendes, M.F. Naccache, A new constitutive equation
References and its performance in contraction flows, J. Non-Newt. Fluid Mech. 86 (1999)
375–388.
[28] E. Ryssel, P. Brunn, Flow of a quasi-newtonian fluid through a planar
[1] J. Mewis, Thixotropy—a general review, J. Non-Newtonian Fluid Mech. 6 (1979)
contraction, J. Non-Newt. Fluid Mech. 85 (1999) 11–27.
1–20.
[29] E. Ryssel, P. Brunn, Comparison of a quasi-newtonian fluid with a viscoelastic
[2] H.A. Barnes, Thixotropy—a review, J. Non-Newtonian Fluid Mech. 70 (1997) 1–
fluid in planar contraction flow, J. Non-Newt. Fluid Mech. 86 (1999) 309–335.
33.
[30] R. Thompson, P. de Souza Mendes, A constitutive model for non-Newtonian
[3] A. Mujumdar, A.N. Beris, A.B. Metzner, Transient phenomena in thixotropic
materials based on the persistence-of-straining tensor, Meccanica. 46 (2011)
systems, J. Non-Newtonian Fluid Mech. 102 (2002) 157–178.
1035–2045.
[4] J. Mewis, N.J. Wagner, Thixotropy, Adv. Colloid Interface Sci. 147-148 (2009)
[31] P. Coussot, Q.D. Nguyen, H.T. Huynh, D. Bonn, Avalanche behavior in yield
214–227.
stress fluids, Phys. Rev. Lett. 88 (2002) 175501.
[5] M. Houska, Engineering Aspects of the Rheology of Thixotropic Liquids, Ph.D.
Thesis, Czech Technical University of Prague-CVUT, Prague, Czechoslovakia,
1981.

You might also like