Materi

© All Rights Reserved

2 views

Materi

© All Rights Reserved

- Pressure Vessel [Design]
- 31968138 Pressure Vessel Design
- PKM-Pengantar Material Komposit
- tugas korosi 1
- WEEK 7- Defleksi Beban Merata n PUNTIRAN
- Doc 5
- Assignmen Ujian Ujian Terhadap Konkrit
- Pembuatan Soda
- Welding
- Hukum Hooke
- Mesin Dan Peralatan
- (Sifat Mekanik)
- Batang Tarik 2013
- K3 Types Fracture
- Composite 21
- Len Turan
- Evaluation of a Leaf Spring Failure Ok
- Chapter4DesignforFlexure [Compatibility Mode].pdf
- (6) Lrfd Stell Design-komponen Tarik & Tekan
- Kuliah 7 Penuangan & Sifat Material(1)

You are on page 1of 34

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/320335507

crystallizing rubber-like materials

DOI: 10.1016/j.jmps.2017.10.007

CITATIONS READS

0 245

2 authors:

Stanford University Stanford University

1 PUBLICATION 0 CITATIONS 64 PUBLICATIONS 844 CITATIONS

Some of the authors of this publication are also working on these related projects:

Systems View project

All content following this page was uploaded by Christian Linder on 21 November 2017.

Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

journal homepage: www.elsevier.com/locate/jmps

rubber-like materials

Reza Rastak, Christian Linder∗

Department of Civil and Environmental Engineering, Stanford University, Stanford, California, 94305, USA

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

Article history: Crystallization can occur in rubber materials at large strains due to a phenomenon called

Received 25 May 2017 strain-induced crystallization. We propose a multi-scale polymer network model to cap-

Revised 2 October 2017

ture this process in rubber-like materials. At the microscopic scale, we present a chain

Accepted 11 October 2017

formulation by studying the thermodynamic behavior of a polymer chain and its crystal-

Available online 13 October 2017

lization mechanism inside a stretching polymer network. The chain model accounts for the

Keywords: thermodynamics of crystallization and presents a rate-dependent evolution law for crystal-

Strain-induced crystallization lization based on the gradient of the free energy with respect to the crystallinity variables

Rubber-like materials to ensures the dissipation is always non-negative. The multiscale framework allows the

Micromechanics anisotropic crystallization of rubber which has been observed experimentally. Two differ-

Homogenization ent approaches for formulating the orientational distribution of crystallinity are studied. In

Maximal advance path constraint the ﬁrst approach, the algorithm tracks the crystallization at a ﬁnite number of orienta-

tions. In contrast, the continuous distribution describes the crystallization for all polymer

chain orientations and describes its evolution with only a few distribution parameters. To

connect the deformation of the micro with that of the macro scale, our model combines

the recently developed maximal advance path constraint with the principal of minimum

average free energy, resulting in a non-aﬃne deformation model for polymer chains. Vari-

ous aspects of the proposed model are validated by existing experimental results, including

the stress response, crystallinity evolution during loading and unloading, crystallinity dis-

tribution, and the rotation of the principal crystallization direction. As a case study, we

simulate the formation of crystalline regions around a pre-existing notch in a 3D rubber

block and we compare the results with experimental data.

© 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Many polymeric materials such as natural and synthetic rubber, polyethylene, polyethylene terephthalate, and polylactide

are subject to polymer crystallization. This phenomenon is observed under certain conditions in vulcanized and unvul-

canized polymers well above the glass transition temperature. Extreme stretchability and the possibility to combine with

carbon nanotubes or conjugated polymers to create stretchable conductors or semiconductors makes these materials perfect

for future applications in devices such as epidermal electronics and implantable sensors (Wang et al., 2017; Xu et al., 2017).

The unique characteristics of rubber-like materials originate from their underlying polymer chain network. The covalent

bonds between monomers lead to long ﬂexible polymer chains and create a soft structure for the material while crosslinks

∗

Corresponding author.

E-mail address: linder@stanford.edu (C. Linder).

https://doi.org/10.1016/j.jmps.2017.10.007

0022-5096/© 2017 Elsevier Ltd. All rights reserved.

68 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

and chain entanglements provide elasticity and load carrying capability. The random distribution of chains in all orientations

results in an isotropic amorphous phase at room temperature. However, the polymer chains can undergo phase transforma-

tion at either lower temperatures or the application of a large amount of stretch. The latter mechanism is referred to as

strain-induced crystallization(SIC) and is the central focus of this work.

For SIC to occur, the chemical composition of the material must enable crystallization. Segments of the macromolecules

have to possess a mutually conforming atomistic structure so that they can arrange themselves into a regular crystalline or-

der. This transition turns an amorphous solid into a partially crystalline solid and is accompanied by volume change, which

is a ﬁrst-order change. Unlike conventional crystallization, this transition is not rapid and does not cause any phase separa-

tion. In fact, amorphous and crystalline fractions coexist and the ratio between them changes gradually. This phenomenon

is possible because of the microscopic structure, which possesses certain gas-like and liquid-like features. Speciﬁcally, the

long ﬂexible macromolecules are quite similar to gases and liquids, because they have a substantial internal freedom to

perform random thermal motion, and they interact by means of volume-exclusion intermolecular forces. Thus, the amor-

phous to semi-crystalline solid-solid transition locally at the microstructural level occurs as a transition from gas/liquid to

solid. The thermodynamics of this process is governed by the latent heat of fusion, and the entropic free energy change in

the amorphous fraction. The relation between these two factors indicates whether crystallization is thermodynamically ad-

missible. Accordingly, crystallization can be triggered through inﬂuencing this relation either by thermal action (decreasing

the temperature) or by mechanical action (stretching the material). Mechanical deformation is the key to SIC of rubber-like

materials. Regardless, crystallization is a kinetic process, for it is a result of the molecular motion. Under most conditions it

has a non-equilibrium nature and leads to a transient history-dependent material behavior with distinct hysteresis effects.

The process of SIC was ﬁrst described by Katz (1925). This discovery was followed by a number of experimental ob-

servations of the structure of crystals formed by SIC (Gehman and Field, 1939; 1944; Luch and Yeh, 1973; Mitchell, 1984).

Most experimental studies perform tensile tests on rubber and use a variety of X-ray diffraction methods to quantify the

crystallization (Candau et al., 2014; Toki et al., 2003; Tosaka et al., 2004). In this process, the intensity of the diffracted

light is measured to compute the degree of crystallinity in the rubber and its directional distribution. The results show

that the process of crystallization during uniaxial tension creates a hysteresis in both the stress response and the evolution

of crystallinity (Toki et al., 20 0 0). Moreover, the results demonstrate that at high stretches, the crystalline regions are ori-

ented parallel to the stretching direction. In addition to uniaxial tests, researchers have focused on different aspects of SIC

such as rate dependence, temperature dependence (Candau et al., 2015; Marchal, 2006; Toki et al., 2013), multiaxial loading

(Pannier et al., 2011), crystalline morphology (Andrews, 1964), and the effect of chemical additives (Ren et al., 2015) and

ﬁllers (Fu et al., 2015; Rault et al., 2006b).

Interestingly , this phenomenon has a dramatic effect on fracture properties of rubber (Brüning et al., 2013; Mars and

Fatemi, 2004; Rublon et al., 2013; 2014; Toki, 2014). Rubber materials fracture at very high local deformation levels which

means that SIC precedes and accompanies their failure. In particular, SIC occurs in the process zone in front of the crack tip,

where the loaded material undergoes signiﬁcant tension in the direction perpendicular to the crack plane. Polymer chains

align correspondingly along this direction, which leads to the formation of oriented crystallites. In-situ X-ray diffraction

measurements fully conﬁrm this picture and reveal a highly crystalline zone surrounding the crack tip with size and mor-

phology controlled by the crack geometry and the loading conditions. This anisotropic semi-crystalline microstructure has

an improved resistance to tearing, hence SIC increases the strength of rubbery polymers. Furthermore, as crystallization is

a time-dependent process, it affects transient aspects of fracture like the crack propagation velocity. Whereas a fast crack

has the chance to pass through the material with no or little crystallization, a slow or static crack will encounter material

reinforced by SIC to the full extent.

While the macroscopic responses of polymers can be evaluated experimentally, a theoretical understanding of the prop-

erties of rubber requires an in-depth study of the microscopic elements of the material and the upscaling of those to the

continuum scale. Early models for polymer elasticity were purely phenomenological. Those models used empirical equa-

tions for the description of strain energy and calibrated their parameters to experimental data in order to simulate various

behaviors of polymers. The subsequent models directly analyzed the polymer chains and their properties using statistical

mechanics leading to the development of the micromechanically-based models. Early examples include the 3-chain model

(James and Guth, 1943; Wang and Guth, 1952), which uses principal stretch values and computes the response of polymer

chains in the three principal directions. While the 3-chain model assumes an aﬃne deformation, the 4-chain model (Flory

and Rehner Jr, 1943; Treloar, 1946) and the subsequent 8-chain model (Arruda and Boyce, 1993; Boyce et al., 1994) introduce

a network model that can simulate non-aﬃne deformations of the material. In this context, an aﬃne deformation is under-

stood as one where the microscopic elements of the network deform with the same deformation mapping that occurs on the

macroscopic scale, whereas a non-aﬃne deformation allows the polymer chains to perform reorientation to redistribute the

stress in a more ﬂexible manner. In other words, the non-aﬃne microscopic deformation seeks to ﬁnd a lower energy state

than the aﬃne deformation while satisfying the kinematic compatibility conditions. One of the most recognized non-aﬃne

models for polymer networks is the microsphere model (Miehe et al., 2004), which uses the notion of isotropic orientation

of similar chains in a continuous orientation space to impose a constraint onto the polymer network deformation. An al-

ternative constraint, accounting for the network functionality, is proposed in our previous work (Tkachuk and Linder, 2012).

Whereas the former model is expanded to inelastic responses of polymers, including SIC in Mistry and Govindjee (2014),

the latter model so far only accounts for elastic effects, and has so far not been applied to three-dimensional (3D) applica-

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 69

tions. In this study, we propose a SIC model that extends the non-aﬃne elastic formulation of Tkachuk and Linder (2012) to

account for inelastic effects and apply it to 3D problems.

Early theoretical investigations of SIC using statistical mechanics were performed by Flory (1947). He considered the

partially crystalline polymer chains as containing freely rotating segments representing the amorphous part and a com-

pletely rigid portion as the crystalline portion. Using Gaussian statistics and simplifying assumptions, he developed a re-

lationship between temperature, elongation and amount of crystallinity in equilibrium condition, which was veriﬁed with

experimental observations (Arlman and Goppel, 1951). Later models extended Flory’s approach to include the effect of folds

in crystallized polymer chains (Gaylord, 1976; Roe et al., 1961) and the effect of the misorientation of the crystals with

respect to the polymer chain (Smith Jr., 1976). The extension of the model to include non-Gaussian statistics is performed

by Smith Jr. et al. (1964). The crystallization process during stretch is not in an equilibrium condition which motivated

the study of the kinematic aspects of SIC (Gent, 1954). Recent SIC models incorporate multiscale computational techniques,

which combine the existing micromechanical elasticity models with the description of crystallization at the polymer chain

scale. These models are able to explain stress hysteresis and the crystallinity evolution. Kroon (2010), who proposed one of

the ﬁrst fully micromechanical models of SIC, classiﬁes polymer chains based on their state as either fully amorphous or

partially crystalline. Using the thermodynamics of crystal nucleation, Kroon incorporates an Arrhenius-type evolution law for

the change in the ratio of the partially-crystalline chains. The calculations for the chains are done at a large number of dis-

crete orientations and the model is able to capture induced anisotropy during SIC. Mistry and Govindjee (2014) developed a

novel SIC formulation based on the non-aﬃne microsphere approach (Miehe et al., 2004). They provide a constitutive model

for both Gaussian and non-Gaussian statistics and a rate-independent evolution equation to simulate the non-equilibrium

crystallization. Other notable recent contributions in modeling SIC include Guilié et al. (2015) and Dargazany et al. (2014a;

2014b).

To further our understanding of SIC, we have developed a non-aﬃne micro-macro approach to gain a mechanics based

interpretation of the entropic and enthalpic processes of polymer chains within amorphous and crystalline domains during

SIC and have created an eﬃcient ﬁnite element based homogenization scheme to investigate its inﬂuence at the experi-

mentally measurable macroscopic scale. Some of the important features of the model is the ability to simulate anisotropic

crystallization of the rubber and the newly formulated continuous crystallinity distribution which helps us describe the

crystallinity state of the microstructure with a reduced number of parameters. Furthermore, we will show that the pro-

posed model is thermodynamically consistent on both, the microscopic and the macroscopic scale. The outline of the paper

is as follows. Our micro-macro approach to strain induced crystallization is presented in Section 2. We start with a macro-

scopic description of SIC in Section 2.1, then outline microscopic aspects of our model in Section 2.2, and bridge the scales

for deformation and crystallinity in Sections 2.3 and 2.4, respectively. The resulting multiscale constitutive model can then

be implemented using the numerical algorithms explained in Section 3. Those are used in Section 4 to perform ﬁnite ele-

ment simulations. The presented results in Section 4 are compared with existing experimental data for uniaxial and three

dimensional settings. Finally, after the conclusions in Section 5, some of the advanced derivations of the equations used in

this work are explained thoroughly in the appendices.

Despite the underlying polymer chain network structure of rubber-like materials, they display an almost homogeneous

mechanical behavior at the continuum scale. Therefore, we will develop a continuum mechanics based formulation for these

materials and will link the microscopic polymer chain deformation and crystallinity to the macroscopic continuum deforma-

tion and a corresponding continuum crystallinity. Using a continuum model will allow us to study larger physical systems at

a lower computational cost compared to molecular dynamics or discrete network simulations. At the same time, the macro-

scopic behavior is linked to the microscopic properties of the individual chains and reﬂects the transformations that occur

in the micro-structure of the material.

The classiﬁcation of state variables into macroscopic and microscopic groups as well as their relationships are illustrated

in Fig. 1. The deformation gradient F and the macroscopic crystallinity variables constitute the macroscopic state variables.

The former represents the amount of strain and rotation in the material and the latter describes the amount of crystallinity

in the material and the orientational distribution of the crystals, from which a conjugate stress tensor P and the evolution

rate of crystallinity

˙ can be obtained. The details about the macroscopic formulation are expressed in Section 2.1. The state

variables of the micro-scale are λ, the stretch of the polymer chain, and ω, its degree of crystallinity (DOC). In the following

sections the relationship between these variables are obtained. As shown in Fig. 1, our multiscale formulation consists of

three main components. In Section 2.2, a mathematical chain formulation is introduced that explains the inﬂuence of λ and

ω on the thermodynamic properties of a chain and the calculation of its internal force f and on ω˙ , the rate of change of the

DOC. Subsequently, a homogenization model for deformation, based on the maximal advance path theory, is introduced in

Section 2.3, connecting the deformation of the continuum material F with its micro-deformation counterpart λ and eventu-

ally obtaining an expression that relates the internal force f of the chains to the macroscopic stress measures P. Finally, in

order to track the anisotropic creation of crystals in the material, the macroscopic crystallinity parameters is introduced

in Section 2.4 and its connection to the micro-crystallization variable ω is studied, leading to an evolution equation for the

rate of change of the crystallinity internal variables

˙.

70 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 1. The different elements of the non-aﬃne strain-induced crystallization model and its relationship with micro and macro scales are shown.

Since rubber-like materials are soft and can easily be stretched to several 100% strain, we develop the constitutive model

in a large deformation setting. To do so, we consider the rubber-like body to be a collection of material points. At time

t ∈ R+ , those occupy the spatial conﬁguration S ⊂ Rndim in terms of the space dimension 1 ≤ ndim ≤ 3, where the position

of an individual material point of the body at time t is given as x ∈ S. The changes these material points experience during

the deformation are described relative to a ﬁxed reference conﬁguration B ⊂ Rndim which for instance can represent the

conﬁguration occupied by the body at the instant time t0 in which the material point is located at position X ∈ B. To avoid

any explicit reference to the body itself, the nonlinear deformation map ϕ (X ) : X −→ x = ϕ (X; t ) is introduced which maps

positions X ∈ B onto positions x ∈ S. Key kinematic quantities are the local deformation gradient F = ∇X ϕ (X; t ), represent-

ing the linear map between tangent vectors in the reference and the spatial conﬁguration, as well as the covariant Cartesian

metric tensors G = δAB and g = δab of those conﬁgurations written in terms of the Kronecker symbol δ .

Since rubber is assumed to be nearly incompressible, a decoupled volumetric-isochoric formulation based on the decom-

position of F into volumetric Fvol and isochoric F̄ parts is assumed as

F = Fvol F̄ , where Fvol = J 1/3 I and F̄ = J −1/3 F (1)

with J = det F > 0 as the Jacobian, given as the ratio of the current local volume to its initial volume. We use the volumetric-

isochoric formulation of Miehe (1994) and Miehe et al. (2004).

Since the process of SIC is an inelastic mechanism, we introduce a set of internal variables to track the evolution of

crystallinity. This set, called the macro-crystallinity , represents the level of crystallization and the orientation of the crys-

tal. Two different methods for computing its evolution ˙ will be proposed in Section 2.4. Using the concept of macro-

crystallinity, we deﬁne the free energy of the material as

(g, F , ) = U (J ) + (

¯ g, F̄ , ) (2)

by separating the volumetric and isochoric responses of the material, and assuming that the crystallization only affects the

isochoric part of the free energy. With the free energy speciﬁed in (2), the Kirchhoff stress tensor is computed as

τ = 2∂g = pg−1 + τ̄ : P (3)

consisting of an isotropic and a deviatoric part. The former is given in terms of the pressure p = JU (J ) and the spatial metric

tensor g, while the deviatoric part of the expression is represented by τ̄ = 2∂g ¯ and is given in terms of the projection

tensor Pi j kl = 12 (δ i k δ j l + δ i l δ j k ) − 13 δ i j δkl . The spatial consistent tangent C represents the rate sensitivities of the Kirchhoff

stress tensor τ and the associated Lie derivatives of the spatial metric as £ v τ = C : 12 £ v g, and can be written as

2

2

C = 4∂gg

2

= ( p + κ )g−1 g−1 − 2 pI + PT : C̄ + (τ̄ : g)I : P − PT : τ̄ g−1 + g−1 τ̄ : P (4)

3 3

where the bulk modulus κ = J 2U (J ) and the isochoric tangent C̄ = 4∂gg

2 ¯ are used. Although the above equations describe

the model in general curvilinear coordinates, only the Cartesian coordinates (gi j = δi j ) will be considered in the subsequent

sections.

In this work, we consider a phenomenological model for the volumetric part as

1

U (J ) = B ( J − 1 )2 (5)

2

where B is chosen to be several orders of magnitude larger than the shear modulus of the nearly incompressible material.

The isochoric response of rubber is related to the underlying structure of the material. The micromechanically motivated

derivation of the mechanical response is a key contribution of this work. The full expression for this free energy is explained

in Section 2.3.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 71

Fig. 2. The physical parameters of a (a) fully-amorphous and of a (b) semi-crystalline polymer chain.

Polymer chains are considered as axial elements, meaning that their state can be determined primarily by their end-to-

end distance. As mentioned above, a partially crystallized chain is described by its stretch λ and its degree of crystallinity

ω. Stretch is the kinematic variable of the chain, whereas the degree of crystallinity will be used as the internal variable of

the chain controlling the transformation of the chain from its amorphous to crystalline state.

Most of the mechanical properties of rubber originates from the oscillation of molecules inside the long polymer chains

between the crosslinks. To study the statistical mechanics properties of these chains it is useful to assume that a polymer

chain consists of a series of rigid segments connected with freely rotating joints (Treloar, 2005). The number of these seg-

ments between two crosslink points and the length of each segment are denoted by N and l, respectively. Due to thermal

ﬂuctuations, all

√ polymer chains are coiled in a random conﬁguration and their end-to-end distance has the root-mean-square

value of r0 = Nl. Using this deﬁnition, the stretch of a chain λ = r/r0 represents the ratio between the current end-to-end

length of the chain and its original average value. These microscopic variables for a typical non-crystallized chain are shown

in Fig. 2(a).

The thermal oscillation of the molecules in the polymer chain pulls the endpoints of the chains towards each other. The

non-Gaussian theory (James and Guth, 1943; Kuhn and Grün, 1942) for a ﬂexible fully-amorphous, i.e. ω = 0, chain states

that the averaged force required to keep a chain at a certain end-to-end length is equal to

kB T

F = β (6)

ω=0 l ω=0

where β is the Langevin parameter corresponding to the stretch of the material, i.e.

ω=0

r λ

= L (β ) = coth β − β −1 = = √ . (7)

ω=0 ω=0 ω=0 ω=0 L N

In the above equation L(· ) is the Langevin function and L = Nl is the total contour length of the chain, which determines

its maximum possible elongation. Therefore, the ratio = r/L is the current end-to-end length of the chain divided by its

maximum value. The non-linear effect of limited extensibility can be seen in the non-Gaussian chain theory. If the length of

the chain gets close to the maximum length, i.e. r → L, then the relative stretch approaches 1, and based on (6) and (7),

the force resisted by the chain increases unboundedly as F → ∞.

Based on these well known results, we now propose a statistical representation of a partially crystalline polymer chain.

After stretching the rubber, some of the chains may form small crystals together with other surrounding chains. In a partially

crystalline chain, we therefore assume that a part of the chain belongs to a crystal and therefore does not ﬂuctuate due to

temperature. Then, the degree of crystallinity of the polymer chain ω = Nc /N = Lc /L will be deﬁned as the ratio of the

number of crystallized segments Nc to the total chain segments N of the material and Lc = Nc l represents the length of

this crystallized section. This deﬁnition of ω based on the length of the crystalline region is similar to the deﬁnitions of

Flory (1947), Mistry and Govindjee (2014), and Guilié et al. (2015). The DOC belongs to the unit interval [0, 1]. It is initially

zero but increases as the material crystallizes. The value of ω is directly related to the crystallinity internal variables

based on the distribution function ω = ω (, λ0 ), explained in Section 2.4. In line with Flory (1947) or Kroon (2010), we

proceed by assuming that the existence of a crystalline phase competes with the existence of the amorphous phase and

thereby changes the overall thermodynamic condition of the chain. We deﬁne the effective end-to-end length ra and the

total length La of the amorphous portion of the chain as ra = (r − Lc )/2 and La = (L − Lc )/2, respectively, also illustrated in

Fig. 2(b). Using these deﬁnitions we can ﬁnd an expression for the force similar to (6) and (7) as

√

kB T β ra r − Lc λ/ N − ω

F= with = L(β ) = coth β − β −1

= = = . (8)

l La L − Lc 1−ω

It is observed that is a decreasing function of ω and crystallization relaxes the amorphous parts of the polymer chain, as

validated by the experiments of Marchal (2006). Note that in Fig. 2(b) the amorphous sections on the right and left of the

crystalline region do not necessarily have to have the same length. In fact, (8) holds as long as we assume the ratio ra /La

72 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

is equal on both sides. One of the main assumptions in this chain formulation is that the crystalline segment lies in the

line connecting the ends of the chain. Although many models (Flory, 1947; Kroon, 2010) make similar assumptions, more

complicated representations can be obtained for a general location and orientation of the crystalline region (Smith Jr., 1976;

Mistry and Govindjee, 2014).

Since the amount of force is the ﬁrst derivative of the free energy, i.e. F = ∂ ψ /∂ r, we obtain the free energy of the chain

ψ by integrating (8) as

r

ψ (λ, ω ) = ψa (λ, ω ) + ψc (ω ) = F dr + ψc ( ω ). (9)

0

The ﬁrst term is the amorphous energy of the partially-crystalline chain, which is computed by integrating the force with

respect to the length of the chain. The sign distinguishes the integration variable with the upper bound of the deﬁnite

integral. The constant of this integration ψ c (ω) is called the crystalline free energy. This free energy is independent of the

entropy of the system and depends only on the DOC of the chain. It is assumed to take the form

ω ω

ψc ( ω ) = c ω − ξ + ln(1 − ) . (10)

ωmax ωmax

This expression consists of a linear term, similar to Flory (1947), Smith Jr. (1976), or Mistry and Govindjee (2014), related to

the change of enthalpy during crystallization and a penalty term, which prevents crystallinity from growing unboundedly.

The linear term cω represents the energy of fusion and the interface energy of the crystalline phase. If all the chains were

completely parallel, they could crystallize up to ω = 1. However, in reality, entanglements and the effects from other chains

in other orientations prohibit any single chain to crystallize more than a speciﬁc maximum value ωmax < 1 called the satura-

tion crystallinity. This effect from the surrounding chains is incorporated using a differentiable logarithmic penalty function

in the crystalline free energy expression (10). The use of a barrier function and the saturation DOC can also be seen in the

model of Guilié et al. (2015). The penalty parameter ξ controls how strong the higher values of ω are penalized. At ω = 0,

the penalty term equals zero with a zero derivative.

Substituting (8) and (10) into (9), we obtain the full form of the free energy as a function of the micro-mechanical

variables, λ and ω, as

ω

β ω

ψ (λ, ω ) = kB T N (1 − ω ) β + ln +cω−ξ + ln(1 − ) . (11)

sinh β ωmax ωmax

This expression is plotted for various levels of crystallization in Fig. 3(a). When the chain is not stretched (λ is small),

the energy is minimized for ω = 0. However, when the chain is stretched (λ > 2), the free energy of the chains with zero

crystallinity exceed that of a partially crystallized chain, making crystallization favorable. The conjugate thermodynamic

variable with respect to the chain stretch is the normalized force

∂ψ √ √

f = = l N F = kB T N β (12)

∂λ

which is later used to calculate the stress tensor at the macroscopic scale. The force response of the chain is linear at small

stretches, as seen in Fig. 3(b). Nonetheless, due to limited extensibility of the chain, the force quickly increases for higher

stretches. Note that crystallization allows the chain to achieve higher length at any constant level of force. Therefore, the

partially crystallized chains show lower amount of force for an equal amount of elongation. In addition to this quantity, we

can deﬁne the driving force of crystallization as

∂ψ β ξω

fw = − = kB T N β + ln −c− (13)

∂ω sinh β ωmax (ωmax − ω )

describing the tendency of the material to crystallize. Hence, it is directly used to calculate the rate of crystallization of the

polymer chain. For each stretch λ, there is an equilibrium degree of crystallinity ωeq , shown in Fig. 3(c), which corresponds

to fw = 0. When the chain is at its equilibrium crystallinity, it does not have any tendency to crystallize or lose crystallinity.

Although various evolution laws, such as the Arrhenius equation used by Kroon (2010), the plasticity-like rate-independent

evolution equation of Mistry and Govindjee (2014), and the method similar to viscoplasticity proposed by Guilié et al. (2015),

can be used for this chain model, here we choose a linear relationship between the crystallization force and the rate of

crystallization as

ω˙ = q fw . (14)

This simpliﬁed evolution law allows us to capture rate-dependence of SIC but we note that for high-strain rates more ad-

vanced formulations containing both, rate dependent and rate-independent components, are needed. This expression for

the crystallization rate is demonstrated in Fig. 3(d). If the chain is not crystallized and the stretch is more than a certain

threshold, the chain will start to crystallize (ω˙ > 0). Conversely, for a chain already in the crystallized state, if the elonga-

tion λ is reduced, the rate of crystallization becomes negative and the chain begins to melt. Note that the crystallinity rate,

by construction, always has the same sign as the crystallization driving force, ensuring consistency with the second law of

thermodynamics, as it can be seen in the expression for the polymer chain dissipation

Dchain = f λ˙ − ψ˙ = fω ω˙ ≥ 0 . (15)

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 73

Fig. 3. The constitutive response of a polymer chain as a function of elongation λ and DOC ω based on the proposed model is shown in the form of (a) the

free energy of the chain, (b) the thermodynamic force of the chain (c), the equilibrium crystallinity ratio for each stretch value, and (d) the crystallization

rate of the chain.

In order to simplify the computation of chain quantities, the parameters and thermodynamic variable of chains are normal-

ized by the number of chains per volume n. The normalized quantities are shown using a bar on top of the symbol. We

deﬁne

ψ̄ = nψ , f¯ = n f , f¯w = n fw (16)

as the normalized thermodynamic variables. Additionally, the nomalized material parameters

1

μ = nkB T , c̄ = nc, ξ¯ = nξ , q̄ = q (17)

n

are used, where μ is the shear modulus of the polymer. The plots (a) and (b) in Fig. 3 make use of these normalized

variables along the vertical axis.

Alternative representations of partially crystalline polymer chains with amorphous and crystalline domains have been

recently proposed (Kroon, 2010; Mistry and Govindjee, 2014; Guilié et al., 2015), all of which are not fully compatible for

our bridging concept of micro- and macroscopic deformation outlined next in Section 2.3. The reason is that our framework

requires the computation of stretch λ as a function of the variable pair (f, ω) which is the inverse of the force-stretch

function (12). The force response in existing models cannot be inverted eﬃciently and accurately, whereas the proposed

chain model can use the Langevin function to obtain λ from f and ω in closed form.

As mentioned earlier, the isotropic response of the presented model is derived directly from the micro-structure of rub-

ber. This material consists of long crosslinked polymer chains that can resist deformation using their one-dimensional tensile

74 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 4. The difference between the macroscopic deformation ϕ(X) and the corresponding microscopic mapping λ(λ0 ) results in a non-aﬃne deformation

of the rubber-like material undergoing strain induced crystallization. The physical reference and current conﬁgurations are shown in the top row with

corresponding boundary conditions in the displacements on ∂X B and tractions on ∂t B and loading t̄ and zooms into the material points in the reference

conﬁguration (no crystallization) and current conﬁguration (crystallization). The bottom row shows the orientation spaces corresponding to the physical

conﬁgurations and the microscopic mapping relating the stretch vectors λ0 and λ.

stiffness. Since tracking the response of every polymer chain in a three dimensional random network is not computation-

ally feasible, the quantities of the network are homogenized by making simplifying assumptions. Polymer chains generally

have different orientations and end-to-end lengths both in the undeformed and the deformed conﬁguration. By assuming

identical length, stiffness and other properties for all the chains in the network, the orientation of the chains in the refer-

ence conﬁguration becomes the main parameter that distinguishes different chains from each other. If we denote r0 as the

undeformed end-to-end vector of the chains, the reference orientation vector λ0 = r0 /|r0 | describes the orientation of the

chains in the reference conﬁguration. All chains with the same reference orientation are assumed to have identical defor-

mation and thermodynamic variables including internal force, degree of crystallinity, and free energy. These assumptions,

visualized in Fig. 4, are similar to those presented by Wu and Van Der Giessen (1993), Miehe et al. (2004), Kroon (2010),

and Tkachuk and Linder (2012). Because rubber behaves isotropically in the undeformed state, it is reasonable to assume

that the polymer chains exist in all undeformed orientations with equal probability. This allows us to study the response of

the random network statistically by assigning each polymer chain orientation to a point on a surface of a unit sphere called

the orientation space (see Fig. 4). Therefore, using the area of the unit sphere S0 , the uniform probability distribution P (λ0 )

of the chain orientation has the form

1 1

P ( λ0 ) = = , P (λ0 )|dλ0 | = 1. (18)

|S0 | 4π S0

Once external deformation is applied to the material, the chains are stretched and rotated in order to provide compatibility

between the micro and the macro scales. The deformation state of the chain is shown by the end-to-end vector r in the

current conﬁguration and the stretch vector λ = r/|r0 |. The chains do not experience equal amounts of stretch, so the end-

to-end vectors of the chains form an ellipsoid in the deformed state. The isochoric free energy of the deformed network,

introduced in Section 2.1, can be written as

¯ g, F̄ , ) = n
ψ (λ, ω ) =
ψ̄ (λ, ω ) = 1

( P (λ0 )ψ̄ λ(λ0 ), ω (λ0 ) |dλ0 | = ψ̄ λ(λ0 ), ω (λ0 ) |dλ0 |. (19)

S0 |S0 | S0

This energy is computed by taking the average, given by the average operator
·, of free energies ψ of all the polymer

chains in the vicinity of a material point. Since

¯ is expressed as energy per unit volume of rubber, the number of chains

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 75

per volume n is used in (19). As described in more detail in Section 2.2, a partially crystallized chain is described on the

micro-scale by λ, its stretch, and by ω, signifying its degree of crystallinity. It is assumed that crosslinking points in the

polymer do not undergo any thermal motion, therefore the total free energy only depends on the thermodynamics of the

chains between the crosslink points.

The numerical method for integrating a symmetric function over the unit sphere, described by Bažant and Oh (1986), is

incorporated to compute the average of various quantities over the orientation space. The number of quadrature orientations

nint affects the accuracy of the results as well as the computation cost of the algorithm. Although Bažant and Oh (1986) use

21 × 2 and 37 × 2 integration schemes for describing small deformations, Verron (2015) showed that more than 300 quadra-

ture points are needed to obtain less then 5% error in computing the various invariants during large deformations. Using

this numerical integration scheme, we can approximate (19) as

nint

nint

(

¯ g, F̄ , ) = ψ̄l Wl = ψ̄ (λl , ωl )Wl (20)

l=1 l=1

where Wl is the weight coeﬃcient for the lth integration orientation (λ0 )l , and λl represents the stretch value corresponding

to chains that are oriented along (λ0 )l in the reference conﬁguration.

While the continuum deformation describes the mapping ϕ between the reference material vector dX and the deformed

material vector dx, the deformation of the micro-structure is deﬁned here as the mapping between the uniformly distributed

reference chain vector λ0 and the deformed chain vector λ. The aﬃne full-network model (Wu and Van Der Giessen, 1993),

used for SIC by Guilié et al. (2015), assumes the linear mapping λ = ϕ (λ0 ) = F̄ λ0 . However, this assumption is not always

applicable (Boyce and Arruda, 20 0 0) and hence we incorporate a non-aﬃne model. In order to obtain the non-aﬃne de-

formation of the chains, we ﬁnd the microscopic mapping ϕ that minimizes the free energy of the network described in

(19). This method is denoted as the principle of minimum average free energy for random networks (Miehe et al., 2004).

In addition to this minimization principle, the microscopic deformation must be kinematically compatible with the external

large scale deformation. Thus, here we use the kinematic constraint equations known as the maximal advance path con-

straint (MAPC) proposed by Tkachuk and Linder (2012). This constraint depends on the functionality, the number of chains

connected to each crosslinking point, of the network. For networks with the functionality of four, which is a reasonable

assumption for any crosslinked polymer network, the tensorial micro-macro compatibility equation takes the form

1 1

λ λ0 = λ(λ0 ) λ0 |dλ0 | = F̄ (21)

|S0 | S0 3

and leads to the constrained optimization problem

1

Minimize
ψ̄ Subject to
λ λ0 = F̄ . (22)

λ 3

The deformed stretch vector λ is the main optimization variable and the crystallization of the chains is also dependent on

this variable. This optimization expression for the free energy allows us to obtain the deviatoric stress tensor

∂ ¯

P̄ = =
f¯ λ0 (23)

∂ F̄

as function of the internal forces arising in the polymer chains. The normalized force vector of a chain f¯ is the vector

parallel to λ with the magnitude of f¯. The algorithm for solving this optimization problem and obtaining the stress tensor

and material tangent is explained thoroughly in Section 3.

Although the degree of crystallinity of individual polymer chains ω is a key quantity for studying SIC, it cannot be directly

measured experimentally because of the incredibly small size of polymer chains. However, many experimental procedures

such as X-ray diffraction provide information about the size and orientation of the crystals. We refer to these experimentally

measurable quantities as macroscopic crystallinity variables, one of which is the crystallinity ratio ω̄ of the material. This

quantity represents the mass or volume ratio of the crystallized regions to the total mass of the material and it is obtained

experimentally by measuring the intensity of the X-ray diffraction at various angles and performing a numerical integration

similar to the methods shown by Toki et al. (2003). In our micromechanical constitutive model, we consider the crystallinity

ratio ω̄ to be proportional to the average degree of crystallinity ω of the chains

nint

ω̄ = α
ω = α ωl Wl (24)

l=1

by introducing the proportionality factor α and incorporating the integration weights Wl . The SIC model of Mistry and

Govindjee (2014) considers α = 1 and Kroon (2010) uses α = Nc /N because of the unique way of deﬁning ω. Guilié

et al. (2015) also employ a non-unit proportionality coeﬃcient α = 1 in computing ω̄. Here, α is calibrated as a material

76 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

parameter. Its value is on the order of 1 but may generally deviate from it due to possibly inconsistent assumptions be-

tween experimental measurements and theoretical assumptions.

Next, we will discuss two separate approaches for tracking the crystallinity ω of the chains in the orientation space.

Commonly (Guilié et al., 2015; Kroon, 2010), the crystallinity values are evaluated at discrete orientations over the orien-

tation space resulting in a discrete crystallinity distribution. Alternatively, we propose in this work a continuous distribution

function for ω, resulting in a memory eﬃcient continuous crystallinity distribution.

Experiments (Toki et al., 2003; Tosaka et al., 2004) show that crystallization mostly occurs in chains which are parallel to

the stretching direction, resulting in an anisotropic crystallization. Rubber is originally isotropic, but the formation of crystals

causes it to behave anisotropically. Nevertheless, it becomes isotropic once the crystals melt during unloading. Using the

MAPC approach in our model results in an anisotropic distribution of λ and subsequently an anisotropic distribution of ω in

the polymer chains. This ability to represent anisotropic crystallization should be contrasted with the non-aﬃne microsphere

approach (Miehe et al., 2004) in which the amount of stretch for all the chains are uniform in the orientation space.

In order to keep track of the state of crystallinity, we can store the DOCs only for the chains that are oriented along the

quadrature orientations. This method, which is used by Kroon (2010) and Guilié et al. (2015), represents the orientational

distribution of DOC by specifying it at a number of representative orientations. In other words, the macro-crystallinity pa-

rameter introduced in (2), which represents the current state of crystallization in the material, is deﬁned as

T

= ω1 ω2 . . . ωnint (25)

a column vector containing the DOCs of the chains oriented along the integration directions. Hence, the distribution function

for the discrete formulation

ωl λ0 = (λ0 )l for l = 1, . . . , nint

ω (, λ0 ) = (26)

undeﬁned otherwise

is only deﬁned for the integration directions. This deﬁnition is suﬃcient for calculating the mechanical and crystallinity

response of the material using numerical integration methods. Thus, the number of internal variables is the same as the

number of integration orientations. The memory required for the history variables can be very large if we choose to increase

the accuracy of the integration over the sphere. For example using 770 quadrature orientations as in Guilié et al. (2015),

more than 4.5 GB of the program memory is required in order to store the history variables for a 3D simulation with 105

elements.

Although in our approach the stretches of the chains are not independent of each other due to the MAPC formulation,

the evolution of crystallinity is calculated independently using the rate of crystallization of individual chains as seen in (14).

We use an implicit time integration for the evolution of each chain as

Since the crystallization rates can be negative for small stretches, we enforce the constraint ω (tn+1 ) ≥ 0 using the max { · }

function while solving (27). During each iteration of our MAPC algorithm, the chain force of the current step f (tn+1 ) and

the DOC of the previous step ω(tn ) are used to compute the new stretch λ(tn+1 ) and DOC ω (tn+1 ). If the value of the chain

quantities such as λ or ω are needed for chains that are not parallel to any integration orientation, these quantities can be

interpolated to obtain a continuous distribution over the orientation space for postprocessing purposes.

Since the evolution of chains are independent, we can observe the change in DOC of a single chain as it undergoes axial

loading and unloading, allowing us to understand the microscopic chain formulation of SIC. The force-elongation relation-

ship for our proposed chain is plotted in Fig. 5(a). The dotted line represents the non-crystallized ω = 0 response of the

chain similar to the solid curve of Fig. 3(b). The solid curves in Fig. 5(a) correspond to two simulations for different stretch

rates λ˙ . Initially, the curve matches the non-crystallization result. However, it softens around λ = 1.7 when the chain starts

to crystallize. This crystallization is visualized in Fig. 5(b), showing that lower stretch rates lead to higher levels of crystal-

lization. Since crystallization has a softening effect on the material, the lower stretch rate simulation has a lower level of

internal force in the chain. When the strain rate approaches zero, i.e. λ˙ → 0, the chain follows the equilibrium crystallinity

path (dashed line) in both force and crystallinity response. The equilibrium crystallinity corresponds to the value of ω that

minimizes the free energy of the chain for a given value of λ, resulting in f w = 0 and ω˙ = 0. It can be seen in Fig. 5(b)

that when the amount of crystallinity is less than the equilibrium value, ω increases. Conversely, when DOC of the chain is

higher than equilibrium, the crystals melt and ω decreases. The unloading curve for ω is maximized at its intersection with

the equilibrium crystallinity curve.

The material parameters for our chain model are μ = 0.37 MPa, N = 18.5, α = 0.37, c̄ = 6.9 MPa, ξ¯ = 1.0 MPa, ωmax =

0.78, q̄ = 0.018 MPa−1 min−1 . These parameters and the parameters used to calculate the response of other chain models

are obtained by calibrating each model with experimental data from Toki et al. (2003) for sulfur-cured natural rubber (see

Section 4.1). For comparison, the similar graphs are plotted for other notable SIC chain models including those by Mistry and

Govindjee (2014) and Kroon (2010) in Fig. 5(c–d) and (e–f), respectively.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 77

Fig. 5. Independent evolution of crystallinity in an individual chain for various chain models and strain rates. Left ﬁgures (a,c,e) show the force-elongation

response while the right ﬁgures (b,d,f) represent the evolution of crystallinity. The current model is presented in (a,b), while the chain model by Mistry and

Govindjee (2014) and the one by Kroon (2010) are shown in (c,d) and (e,f), respectively.

Remark 1. The model proposed by Mistry and Govindjee (2014) uses the non-aﬃne microsphere formulation by

Miehe et al. (2004) to calculate the microscopic stretches λ =
F̄ λ0 P from the deformation gradient, where P is the aver-

aging exponent. The stretch values of the chains computed with this method are uniform across all orientations. As a result,

the chain force and the chain DOC is also identical for all chains. Although this representation does not take into account

anisotropic crystallization in rubber, it is computationally very eﬃcient both in terms of memory and cpu time because only

a single history variable is needed to store ω and the computations for the chains are performed only for one set of input

78 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

parameters. In addition to a Gaussian formulation, the model by Mistry and Govindjee (2014) also provides expressions for

the non-Gaussian theory. Focusing on the Gaussian model in this comparison, the resulting linear force response of the

chains can be expressed as

3μ √

f¯ = λ − Nω . (28)

1−ω

The evolution law of Mistry and Govindjee (2014) is a rate-independent formulation similar to plasticity using the yield

function

√ √ √

∂ ¯ ∂ ¯ T 3μ 1 2 N (λ − ω N ) ( λ − ω N )2

g= − (gc + γ ω ) ≤ 0 with = −μd N 1 − 0 − + −

∂ω ∂ω Tm 2 1−ω 1−ω ( 1 − ω )2

(29)

where μd is the heat of fusion and T and Tm0 )

are the current temperature and the crystallization temperature of the rubber.

It can be noted that the term −μd N (1 − T /Tm0 ) is equivalent to c̄ in our proposed model. Since this chain model is rate-

independent, changing the stretch rate does not change the loading and unloading curve in both plots. The topological

tube constraint arising from restricted chain movements due to surrounding chains is also considered in the free energy

expression of this model. However, this effect is not visible in Fig. 5(c) because the tube constraint free energy is not a direct

function of λ. Ignoring the tube constraint, Fig. 5 shows that the force response of the ﬂexible chain starts to decrease after

the crystallization starts for the given set of material parameters. We note that since the MAPC model requires a one-to-

one non-decreasing force-elongation response of the chains, this chain model is not compatible with the SIC chain model of

Mistry and Govindjee (2014). The parameters used to create the plots shown in Fig. 5(c–d) are those calibrated by Mistry and

Govindjee (2014) for sulfur-cured natural rubber, i.e. μ = 0.62023 MPa, N = 195.15, μd = 0.110 MPa, gc = 18, γ = −65.

Remark 2. In the polymer chain formulation proposed by Kroon (2010), the main crystallinity variable ω represents the

ratio of partially-crystallized chains to the total number of chains for a particular orientation. The free energy expression

provided in the model includes a number of terms such as the viscous response of chains, a topological constraint, and a

non-aﬃne penalty term. However, for comparing the chain formulation with our model, Fig. 5(e–f) only show the ﬂexible

chain response of the material. The normalized chain force is computed as

∂ ψ̄ √ √ ∂λ √

f¯ = = μ ( 1 − ω ) N β + 2ω N β

= μ N (1 − ω )β + ωβ (30)

∂λ ∂λ

√ √

where N = (N − Nc )/2 and λ = (λ N − Nc )/(2 N ) are the number of √ segments and the stretch of the amorphous portion

of a partially-crystalline chain, respectively. Additionally, β = L−1 (λ / N ) is the Langevin parameter for the stretch of the

amorphous segments. This force response is coupled with the crystallinity evolution equation

λ β

ω˙ = H (λ )gc (1 − ω ) exp(η ) − ga ω exp(−η ) where η = Nn Nc √ β + ln (31)

N sinh β

to obtain the behavior of the chain during loading and unloading. The parameter gc is the rate at which the chains crystallize

and ga is the rate at which the crystals melt into the amorphous phase. This evolution law includes the Heaviside function

H(λ ) that ensures the chains have been stretched enough. Note that in this chain model all the partially-crystalline chains

have√ the same constant number of crystallized segments. Therefore, Nc is constant and the chains stretched less than λmin =

Nc / N are physically impossible to crystallize. This condition is enforced using the Heaviside function, creating noticeable

sharp kinks in plots for both the chain force and the chain DOC. It can also be visualized by a large jump in the value of the

equilibrium crystallinity. This non-smoothness in the response makes it diﬃcult to combine this chain model with the MAPC

homogenization method. One challenge is the convergence of the model after the ﬁrst chain starts to crystallize. In addition,

the sudden changes in f¯ and ω lead to distinctive kinks in the homogenized macroscopic response of rubber if a small

number of integration directions are used. It can be noted that the hysteresis in the force response is rather small within

Kroon’s model so that an additional viscoelastic response should be added to it to produce the amount of hysteresis seen

in experimental data. The parameters used to create the plots in Fig. 5(e–f) are μ = 0.6 MPa, N = 22.4, Nc = 11.0, Nn = 0.1,

gc = 0.051min−1 , ga = 0.48 min−1 , which are identical to the calibrated results in Kroon (2010).

Due to the large memory requirement when using a discrete representation of crystallinity, a continuous representation

is proposed in this work. The DOC of a chain is calculated using the crystallinity distribution function ω(, λ0 ) which

depends on a small number of crystallinity parameters and on the reference orientation of the chain. This distribution

function must be continuous and even-symmetric (ω (, −λ0 ) = ω (, λ0 )) over the unit sphere. Since crystals created by

SIC are oriented mostly along a preferred direction, the function must be axisymmetric. The chains closer to the axis of

rotation possess the highest amount of crystallinity, whereas the chains perpendicular to that axis will not crystallize. The

axis of symmetry is close to the maximum principal stretch direction, but it can be different if non-proportional loading is

applied.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 79

Fig. 6. Left: Orientation of the crystal and its angle θ with respect to the chains in the reference conﬁguration. Right: The distribution of crystallinity as a

function of the angle θ between the chain and the crystal orientation is shown. The inﬂuence of the parameters a and b on this distribution can be seen,

where a1 > a2 and b1 > b2 .

Various expressions satisfy these requirements for the crystallinity distribution function. In particular, the function

is used to perform our numerical simulations in Section 4. The crystallinity internal variables

T

= φ1 φ2 a b (33)

include the spherical orientation of the crystal (φ 1 , φ 2 ), the intensity of crystallization a, and the spread (width) of crys-

tallinity b. The plot of this distribution as well as the effects of a and b can be seen in Fig. 6. The angle θ between the

crystal orientation and the reference orientation of the chain λ0 , which is shown on the left of Fig. 6, directly determines

the amount of crystallinity in a chain. Note that regardless of the number of quadrature directions, we only need to track

four internal variables in our constitutive model, resulting in a more memory-eﬃcient method. The calculation of the chain

formulation is still performed at every integration orientation, but ω is not stored for every chain. Instead, the crystallinity

internal variables store the distribution information at every time step. The integration orientations are used to calculate

the micromechanical and the evolution of crystallinity, but they are not used to store crystallinity values. The accuracy of

the simulation can be improved by using adaptive quadrature rules, which is only possible if we incorporate the continuous

distribution of crystallinity. In contrast, attempting to use an adaptive quadrature with discrete representation of crystallinity

leads to interpolation challenges and possible inaccuracies. Using the rate of crystallization of the chains, obtained from the

chain formulation, we compute the evolution of the macroscopic crystallinity parameter by performing a homogenization

algorithm. The rate computed from (14) describes the true rate ω˙ of crystallization of the chain. On the other hand, if we

take the time derivative of (32) as

T

∂ ∂ω (, λ0 ) ∂

ω˜˙ = ω (, λ0 ) = (34)

∂t ∂ ∂t

we obtain the so called apparent rate ω

˜˙ of crystallization in the chain. The value of ∂ /∂ t in (34) is the only unknown value

which can be computed by equating the true rate and the apparent rate for all the chains. These rates cannot be exactly

equal but we can minimize the square of the difference between them, leading to a least square optimization problem

T

∂ 1 1 ∂ ∂

= arg min (ω˙ − ω˜˙ ) + γ

2

. (35)

∂t ∂ / ∂ t 2 2 ∂t ∂t

The objective function of the minimization computes the differences of the apparent rate and the actual crystallization

rate of the chains. Additionally, it contains a regularization term which ensures that the system has a minimum required

convexity to achieve reliable numerical results. The value of the regularization constant is chosen as a small positive number.

80 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 7. The evolution of crystallinity is shown as a function of the angle θ between each chain and the crystallinity direction and the difference between

the actual rate and the apparent rate of crystallization is presented.

T

∂ ∂ω ∂ω ∂ω

˙ = = A−1 ω˙ where A = + γ I4 (36)

∂t ∂ ∂ ∂

since the apparent rate is a linear function of the unknown. Here, the matrix I4 represents the 4 × 4 identity matrix. Com-

puting the local dissipation during isochoric deformation in the material, we get

T T

¯˙ = P̄ : F̄˙ − ∂ ¯ ∂

¯ ∂ ¯ ∂ ψ̄ ∂ω T −1 ∂ω

Disochoric = P̄ : F̄˙ − : F̄˙ + ˙ = − ˙ = − A ω˙ (37)

∂ F̄ ∂ ∂ ∂ω ∂ ∂

T

nint

∂ω

nint

∂ω

= ( f¯w )l Wl A −1

Wm ω˙ m (38)

l=1

∂ l m=1

∂ m

resulting in Disochoric ≥ 0 assuming the evolution relationship of the chains is linear as shown in (14) and q̄ ≥ 0. This in-

equality is proved using the symmetry of (38) and the positive deﬁniteness of A. Therefore, the macroscopic crystallinity

evolution model is consistent with the second law of thermodynamics.

The process of evolution of is best explained by illustrating the distribution of crystallinity at various stages of loading

and unloading as shown in Fig. 7. The solid curve and the dashed curve represent the continuous distribution of ω in the

reference conﬁguration at the current time step tn and the next time step tn+1 , respectively. The height of the vertical lines,

corresponding to the true rate of crystallization at the integration points, is equal to ω˙ t where ω˙ is computed using (14).

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 81

Although often a small time step t is chosen for obtaining accurate results, a rather large time step is used for creating

Fig. 7 to show an exaggerated, recognizable change in ω between tn and tn+1 . As seen in Fig. 7, we are only able to compute

the value of the true rate of crystallinity at the discrete set of integration points. On the other hand, the difference between

the solid line and the dashed line ω ˜˙ t is proportional to the apparent rate, which is computed as a continuous function of

θ based on (34). The value of ˙ is chosen to make the true rate and the apparent rate as close as possible using the least

squares method of (35). At the start of loading, see Fig. 7(a), only a small number of chains that are close to the stretch

direction crystallize and consequently, the value of crystallinity becomes non-zero by increasing the parameter a. At later

stages of loading, see Fig. 7(b), crystallinity continues to increase and more chains contribute to crystallization. This time in

addition to an increase in a the value of b decreases to widen the distribution function. When the unloading stage starts,

see Fig. 7(c), the chains close to the stretch direction face a decrease in the stretch while their ω is very high, therefore

they face a reduction in crystallinity ω˙ < 0. However, some of the chains in other orientations have not crystallized to their

equilibrium crystallinity value and they continue to crystallize. Therefore, using the averaging algorithm, the height of the

distribution function a remains constant, while the width of the distribution function increases by decreasing b. Finally, the

unloading continues, see Fig. 7(d), and all of the chains decrease their ω, resulting in a decrease in a. Note that the rotation

of the crystals, not considered in this ﬁgure but responsible for the change in φ 1 and φ 2 , is studied in Section 4.3.

3. Numerical algorithm

In order to create a numerical implementation based on the theory presented in Section 2, the MAPC equations describ-

ing the micro-structure and the equations representing the evolution of crystallinity must be solved. These expressions are

highly non-linear and we have to solve them numerically using the Newton–Raphson scheme. Although it is possible to

solve both of these equations simultaneously, we chose to solve them in a nested conﬁguration in order to ensure numeri-

cal stability. During each iteration of the Newton loop for the MAPC process, the evolution equations are solved iteratively,

generally converging in less than 5 iterations. After updating the micro-structure and moving to the next iteration of MAPC,

the rate of crystallization is modiﬁed and the evolution equation must be solved again. The process of crystallinity evolution

differs for discrete and continuous representations of crystallinity. The algorithm for independent crystallization of chains

is discussed in Section 3.1 for the case of discrete crystallinity distribution. Additionally, the process of updating the crys-

tallinity internal variables for the continuous distribution of crystallinity is explained in Section 3.2. Using the results of the

MAPC algorithm we can compute the spatial stress tensor and consistent tangent (explained in Section 3.3).

As mentioned in Section 2.3, in order to obtain the microscopic deformation of the network, we need to minimize (19) by

imposing the constraint (21). Solving the equality constrained minimization is equivalent to ﬁnding the stationary point of

the Lagrangian form of the problem, written as

1

L[λ, ν] =
ψ̄ (λ, ω ) − ν :
λ λ0 − F̄ . (39)

3

Once solved, the Lagrange multiplier ν is shown (see Appendix C or Tkachuk and Linder (2012)) to be proportional to the

isochoric ﬁrst Piola–Kirchhoff stress tensor

1

P̄ = ν. (40)

3

Furthermore, the normalized force vector f¯ of any chain can be computed based on

f¯ = n f = 3P̄ λ0 (41)

as a function of P̄ . Consequently, the stretch vector, which directly depends on the force vector, can be represented as a

function of P̄ . Using this dependency, we can write the MAPC constraint as a residual

1

Rλ (P̄ ) = F̄ −
λ(P̄ ) λ0 (42)

3

only dependent on the stress tensor. Setting Rλ = 0, we can solve this non-linear tensorial equation to obtain the stress

tensor P̄ .

The computer algorithm solves (42) using the Newton–Raphson method. At the beginning of the ﬁrst iteration, an initial

value is assigned to P̄ . This initial tensor can be the stored value from the previous time step, or the linear small-strain

approximation P̄ ≈ μF̄ . Using the initialized stress tensor, we ﬁnd the set of force vectors along the quadrature orientations

using (41). The orientation of the force vector deﬁnes the orientation of λ and the force magnitude is used in the chain

formulation (12) to compute λ. The set of stretch vectors are then combined to form the residual (42). Unless the residual is

close to zero, the stress tensor P̄ is updated using the MAPC tangent operator. This iterative process of updating the stress

tensor is demonstrated in Table 1.

Note that the amount of force in polymer chains is used to compute its stretch and thus it requires inverting the rela-

tionship between stretch and force. In contrast, the force equation in most existing chain models is not easily invertible and

82 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Table 1

Algorithm of the constitutive model with a discrete crystallinity representation.

0. Given: isochoric deformation gradient F̄ at time tn+1 , the DOC of the chains (ω1 , ω2 , . . . , ωnint )n at time tn

1. Initialize the isochoric ﬁrst Piola Kirchhoff tensor from previous step or from the linear approximation P̄ 0 = μF̄

2. Perform the MAPC Newton iteration loop (iteration k) to compute the stress tensor P̄n+1

a) Compute the normalized force and the deformed unit vector for the chains in all quadrature orientations

f¯lk = 3P̄ k (λ0 )l , f¯lk = f¯lk and λ

ˆ k = f¯ k / f¯k for l = 1, 2, . . . , nint

l l l

b) Compute the Langevin parameters for each chain

√

βl = fl /(μ N ) and l = coth βl − 1/βl

k ¯ k k k k

d) For every orientation (index l) of chains, perform Newton iteration (iteration m) to obtain the crystallinity ratio ω at time tn+1

(i) Compute the crystallization rate of the chain

k,m

( f¯w ) = μN β k + ln(β k / sinh β k ) − c̄ − ξ¯ ωk,m /(ωmax (ωmax − ωk,m )) and ω˙ k,m = q̄ ( f¯w )k,m

l l l l l l l l

(ii) Calculate the crystallinity evolution residual (rw )k,m

l

= −ωlk,m + (ωl )n + t ω˙ lk,m

(iii) Check

the tolerance and make sure ω k,m

l

≥ 0

if (rw )k,m

l

≤ tol go to 2e

if ω˙ l < 0 and ωl = 0 go to 2e

k,m k,m

(kw )k,m

l = 1 + t ξ¯ q̄/(ωmax − ωlk,m )2

(v) Update the DOC of the chains ω√lk,m+1 = max{ωlk,m + (rw )k,m

l

/(kw )k,m

l

, 0}

e) Compute the stretch λkl = N ωlk + (1 − ωlk ) kl for l = 1, 2, . . . , nint

f) Compute MAPC residual

int k ˆ k

Rλk = 13 F̄ − nl=1 λl λl (λ0 )l Wl

g) Check the tolerance: if ||Rλk || ≤ tol exit the algorithm

h) Compute the Jacobian matrix Klk and the stiffness k̄kl of the chains using (48) and (50)

i) Formulate the MAPCtangenttensor

λkl λk ˆ k

K k = 3 nint

λ 1+ 1 − l λ λ ˆ k ((λ0 )l (λ0 )l )Wl where (AB )abcd = Aac Bbd

l=1 f¯lk k̄kl f¯lk l l

−1

j) Update the stress tensor P̄ k+1 = P̄ k + Kλk : Rλk

may require additional Newton–Raphson iterations. Our proposed model provides a closed-form expression to compute the

stretch value as a function of the force value and the DOC by combining (12) and (7) in the form

√ 1

f¯

λ = N ω + (1 − ω )(coth β − ) where β = √ . (43)

β μ N

Before calculating the chain stretch, it is necessary to update the DOC of the chains during the time step. At the beginning

of the simulation the DOC of the chains are set to zero. When the stretch is small, the driving force of crystallinity and

consequently the rate of crystallization are negative. In this case, since we have to ensure ω ≥ 0, the chain remains fully

amorphous. However, once fw > 0, the chain starts to crystallize and we have to update ω. During unloading, the driving

force becomes negative and decreases the DOC until it reaches zero. The algorithm makes sure that ω does not become

negative at any point. This evolution of crystallinity for each chain orientation is performed by solving (27) iteratively using

the following residual and tangent, namely

rw = −ω (tn+1 ) + ω (tn ) + t ω˙ (tn+1 ) (44)

and

∂ rw q̄ξ¯ t

∂ω f =cst

kw = − =1+ . (45)

(ωmax − ω )2

This chain crystallization tangent can also be computed from the 2 × 2 chain Jacobian K as

K12 K21

kw = 1 − t K22 − . (46)

K11

The chain Jacobian matrix K provides the derivatives of the force and rate of crystallization with respect to the thermody-

namic variables of the chain, stretch and DOC. The value of used in the following equation is calculated from β using

(7) and its derivative can be calculated from L (β ) = d /dβ = β −2 − (sinh β )−2 . Thus,

⎡ ⎤

∂ f¯ ∂ f¯

K11 K12 ⎢ ∂λ ∂ω ⎥

K= =⎢ ⎥ (47)

K21 K22 ⎣ ⎦

∂ ω˙ ∂ ω˙

∂λ ∂ω

with

μ √ √ q̄ξ¯

K11 = , K12 = N ( − 1 )K11 , K21 = −q̄K12 , K22 = N ( − 1 )K21 − . (48)

( 1 − ω )L ( β )

(ωmax − ω )2

After updating the crystallinity of the chains and computing the stretch vectors, the MAPC iteration is performed by cal-

culating the MAPC residual (42) and the associated tangent operator. The MAPC tangent required for updating the stress

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 83

tensor P̄ depends on the stretch λ, force f¯, and stiffness k̄ = ∂ f¯/∂λ of the chains, as well as the unit vector oriented along

the deformed chain λˆ = λ/λ = f / f, and can be written as

∂ (Rλ )iJ λ 1 λ ˆ ˆ

(Kλ )iJkL = − = 3
δik + − λi λk (λ0 )J (λ0 )L . (49)

∂ P̄kL f¯ k̄ f¯

This tangent operator is a fourth order tensor with 81 entries. Nevertheless, in the computer implementation, P̄ and Kλ are

stored as a 9 dimensional column vector and a 9 × 9 matrix, respectively. The chain stiffness

K11 = μ/ (1 − ω )L (β ) ω˙ = 0

k̄ = (50)

K11 + tK12 K21 (1 − tK22 )−1 ω˙ = 0

equals K11 when the chain is fully amorphous. However, if the chain is crystallizing, we have to take into account the

softening due to crystallization. The details about calculating the MAPC tangent Kλ and the chain stiffness k̄ is thoroughly

explained in Appendix C.

The steps required to compute the microscopic deformation ﬁeld, the stress tensor and the evolution of crystallinity for

the discrete representation of the chains are summarized in Table 1. For simplicity, the quantities that correspond to time

tn+1 are written without any time subscript whereas the quantities of time tn are shown with subscript n.

The numerical algorithm used to calculate the evolution of crystallinity and the state of stress for a continuous crys-

tallinity distribution has similarities to the previous case (discrete distribution of crystallinity). In contrast, in the continuous

crystallinity formulation, the state of crystallinity is solved for all the chains at the same time instead of using independent

evolution equations. At each step, instead of ω, is updated iteratively and as a result, the state of each chain has higher

impact on the other chains in the material.

In addition to an initial value for P̄ , as discussed in 3.1, we also need to provide an initial value for the crystallinity

internal variables. Since there are no crystals in the material at the beginning of the simulation, we can set a = 0 and

therefore, based on (32), the DOC is initially zero for chains in all orientations regardless of the value of φ 1 , φ 2 , and b. The

polar angles φ 1 and φ 2 , representing the orientation of the crystals, are set to arbitrary numbers initially and their value

are irrelevant as long as no crystal has been formed. When the material is deformed and starts to crystallize for the ﬁrst

time, we set the polar angles to the orientation which has the largest value of chain stretch in the micro-structure. The

orientation with the maximum stretch of the chain also has the maximum chain force and according to (41), its direction is

obtained by computing the singular-value decomposition of the stress tensor as

3

P̄ = σζ u ζ vζ (51)

ζ =1

where the vector vζ corresponding to the largest singular value σ ζ is chosen as the initial crystal orientation in the reference

conﬁguration. During this initial calculation of the spherical coordinates φ 1 and φ 2 of the crystal orientation, we can choose

any coordinate axis (X, Y, or Z) as the main axis for spherical coordinates. For example, the Z axis is used in Fig. 6 as

the main spherical axis and φ 1 is deﬁned according to the Z axis. After this initialization, the orientation of the crystal is

updated through the evolution law described in (36) and the singular value decomposition method is not used anymore.

During each iteration of the MAPC algorithm, the crystallinity internal variables must be updated, which is performed by

an additional Newton–Raphson algorithm nested inside the main MAPC iteration loop. Along this process, after calculating

the force vectors using (41), the crystallinity (DOC) of the chains are calculated from (32). The unit orientation vector of the

crystal

uˆ = sin φ1 cos φ2 e1 + sin φ1 sin φ2 e2 + cos φ1 e3 (52)

is obtained from the two polar coordinates of the macroscopic crystallinity. Substituting cos θ = uˆ · λ0 into (32), we get

ω = ω (, λ0 ) = a exp −b(ln uˆ · λ0 )2 (53)

which allows us to calculate the DOC of the chains. Hence, the DOC of the chains are not stored as a history variable and

the simulation only keeps track of 4 numbers included in , thus avoiding an otherwise needed large storage requirement

when using a large number of quadrature orientations. The gradient of this quantity with respect to the crystallinity internal

variables

⎡ ⎤ ⎡ ⎤

∂ ω/∂ φ1 −2ab (ln uˆ · λ0 /uˆ · λ0 ) exp −b(ln uˆ · λ0 )2 (∂ uˆ · λ0 /∂φ1 )

∂ω ⎢

⎢∂ ω/∂ φ2 ⎥ ⎢−2ab (ln uˆ· λ0 /uˆ ·λ0 ) exp −b(ln uˆ · λ0 )2 (∂ uˆ · λ0 /∂φ2 )⎥

=⎣ ⎥

∂ ω/∂ a ⎦ ⎣exp −b

= 2 (54)

∂ ⎦

(ln uˆ · λ0 )

∂ ω/∂ b −a(ln uˆ · λ0 )2 exp −b(ln uˆ · λ0 )2

is essential for calculating the evolution of the crystallinity. Here, the derivative of uˆ · λ0 with respect to the polar coordi-

nates can be written as

84 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

∂ uˆ · λ0

= sign(uˆ · λ0 )[cos φ1 cos φ2 (λ0 )1 + cos φ1 sin φ2 (λ0 )2 − sin φ1 (λ0 )3 ] (55)

∂φ1

and

∂ uˆ · λ0

= sign(uˆ · λ0 )[− sin φ1 sin φ2 (λ0 )1 + sin φ1 cos φ2 (λ0 )2 ]. (56)

∂φ2

In addition to the gradient [∂ ω/∂ ], also the Hessian [∂ 2 ω/∂ ∂ ] must be calculated, whose closed form expression is

omitted here for the sake of brevity. Note that based on (56), we need to make sure that φ 1 > 0 at all times in order to

obtain a non-zero gradient for the variation of φ 2 . Thus, we try to choose the spherical axis that is not parallel to the

crystal orientation.

Using the chain parameters f¯ and ω, the stretch λ and the rate of crystallization ω˙ of each chain are calculated directly

using (43) and (14). Now, using (36) and (54) the rate of change of the crystallinity internal variables ˙ can be calculated

and the crystallinity Jacobian ∂

˙ /∂ is obtained from

2 T T 2

∂ ˙ ∂ω K12 K21 ∂ω ∂ ω ˙ ∂ω ˙ ∂ ω

=A −1

(K22 − ) − +
ω˙ − . (57)

∂ ∂ K11 ∂ ∂ ∂ ∂ ∂ ∂

These calculated quantities allow us to update the value of the crystallinity internal variables using a Newton–Raphson

scheme. The evolution of the crystallinity internal variables can represent the growth or shrinking of the crystal (updating

a), the rotation of the crystal (updating φ 1 , φ 2 ), and the change in the angular width of the crystalline region (updating b).

The residual R and the tangent K for updating the crystallinity internal variables are calculated as

R = −(tn+1 ) + (tn ) + t (

˙ tn+1 ) (58)

and

∂ R ∂ ˙

K = − = I4 − t (t ). (59)

∂ (tn+1 ) ∂ n+1

The calculation of the MAPC residual and its tangent is performed after the crystallinity internal variables are converged.

Although the MAPC residual is obtained identical to (42), the expression for the tangent must be modiﬁed to include the

softening effect of the evolution of the internal variables. The full expression of the MAPC tangent

∂ Rλ 4

∂ Rλ ∂ i

Kλ = − + (60)

∂ P̄ =cst i=1 ∂ i P̄=cst ∂ P̄ R =0

is similar to (49), but it includes a new term representing the effect of the evolution of on the tangent operator. The ﬁrst

term corresponds to the derivative of the residual for constant and it is equal to (49) with k̄ = K11 . However, the second

term explains how the MAPC tangent changes due to the change in the crystallinity internal variables while the stress tensor

is kept constant. The expressions to compute these quantities are presented as

∂ Rλ K12 ∂ω ˆ

=
λ λ0

∂ i P̄=cst

(61)

K11 ∂ i

and

4

∂ i 4

K ∂ω ˆ

= 3t (K )−1 A−1
21 λ λ0 . (62)

∂ P̄ R =0 11 ∂ k

ij jk K

j=1 k=1

The complete algorithm for the continuous crystallinity distribution is summarized in Table 2 through a pseudo-code and

the proofs for some of the equations are provided in Appendix B and C.2

The deviatoric part of the ﬁrst Piola–Kirchhoff stress tensor obtained from the MAPC algorithm can be transformed into

the spatial deviatoric stress tensor

τ̄ = P̄F̄ T (63)

and be used to compute the total stress from (3) as

τ = p1 + τ̄ : P (64)

assuming the coordinate system is Cartesian (gi j = δi j ). Furthermore, the MAPC tangent operator can be transformed to the

spatial tangent tensor corresponding to the deviatoric stress using

∂ P̄iA ∂ P̄ 1 −1

C̄i jkl = F̄lB F̄jA − δik τ̄ jl where = K (65)

∂ F̄kB ∂ F̄ 3 λ

which must be combined with (4) to represent the total spatial tangent operator.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 85

Table 2

Algorithm of the constitutive model with a continuous crystallinity distribution.

0. Given: isochoric deformation gradient F̄ at time tn+1 , the crystallinity internal variables n at time tn

1. Initialize the isochoric ﬁrst Piola Kirchhoff tensor from previous step or from the linear approximation P̄ 0 = μF̄

2. Perform the MAPC Newton iteration loop (iteration k) to compute the stress tensor P̄n+1

a) Compute the normalized force and the deformed unit vector for the chains in all quadrature orientations

f¯lk = 3P̄ k (λ0 )l , f¯lk = f¯lk and λ

ˆ k = f¯ k / f¯k for l = 1, 2, . . . , nint

l l l

b) Compute the Langevin parameters for each chain

√

βl = fl /(μ N ) and l = coth βl − 1/βl

k ¯ k k k k

T

d) Perform Newton iteration (iteration m) to obtain the crystallinity internal variables = φ1 φ2 a b at time tn+1

(i) Compute the DOC of the chains for all orientations

ωlk,m = ω (k,m , (λ0 )l ) = ak,m exp −bk,m (ln uˆ k,m · (λ0 )l )2

(ii) Compute the crystallization rate of the chains for all orientations

k,m

( f¯w ) = μN β k + ln(β k / sinh β k ) − c̄ − ξ¯ ωk,m /(ωmax (ωmax − ωk,m )) and ω˙ k,m = q̄( f¯w )k,m

l l l l l l l l

(iii) Calculate the gradient [∂ ωl /∂ ]k, m using (54) and the Hessian [∂ 2 ωl /∂ ∂ ]k, m of the crystallinity distribution function at all of the chain orientations

(iv) Calculate the rate of change of the crystallinity internal variables

k,m k,m T k,m

nint ∂ωl ∂ωl nint ∂ωl

Ak,m = Wl + γ I4 and ˙ k,m = (Ak,m )−1 Wl ω˙ lk,m

l=1 ∂ ∂ l=1 ∂

(v) Calculate the crystallinity evolution residual R k,m

= −k,m + n + t

˙ k,m

(vi) Check the

k,m tolerance and make sure a k, m

≥ 0

if R ≤ tol go to 2e

if ˙ k,m = a˙ k,m < 0 and 0 ≤ k,m = ak,m ≤ tol go to 2e

3 3

k,m

˙ /∂

(vii) Compute the Jacobian matrix Klk,m of the chains and the crystallinity Jacobian ∂ using (48) and (57).

(viii) Calculate the crystallinity tangent

k,m

K k,m = I4 − t ∂

˙ /∂

(ix) Update the crystallinity internal variables k,m+1 = k,m + αline (K ) R

k,m −1 k,m

√ that guaranties the crystallinity magnitude (3 = a) always remains non-negative

e) Compute the stretch for all orientations λl = N ωl + (1 − ωl ) l

k k k k

f) Compute the MAPC residual Rk = 1 F̄ − nint λk λ

ˆ k (λ0 )l Wl

λ 3 l=1 l l

g) Check the tolerance: if ||Rλk || ≤ tol exit the algorithm

h) Formulate the MAPC

tangent

tensor

int λkl λk ˆ k

Kλk = 3 nl=1 ¯k 1 +

1

( K )k

− ¯kl λ λˆ k (λ0 )l (λ0 )l Wl

fl

l

11 l l f

l k

k

4 nint (K12 )kl ∂ωl 4 4 nint ( K )k ∂ωl

−3t λˆ kl (λ0 )l Wl (K )−1 A−1 21 l

λˆ kl (λ0 )l Wl

i=1 l=1 ( )

K11 kl ∂ i j=1 p=1 l=1 ij jp (K )k 11 l ∂ p

−1

i) Update the stress tensor P̄ k+1 = P̄ k + Kλk : Rλk

4. Numerical results

To calibrate the material parameters for our model, we investigate the uniaxial loading and unloading experiments of

Toki et al. (2003) for sulfur vulcanized natural rubber (NR-S). The macroscopic tensile stretch λ̄ is applied with the rate of

0.4/min through the deformation gradient

⎡ ⎤

λ̄ 0 0

F = F̄ = ⎣0 λ̄−1/2 0 ⎦ (66)

0 0 λ̄ −1/2

ensuring an isochoric deformation. The total Piola stress tensor P = τ F −T is obtained from the value of τ deﬁned in (64) for

a certain value of the pressure p which corresponds to the stress free lateral boundary condition τyy = τzz = 0. Fig. 8 shows

the experimental data for natural rubber alongside our simulation results for both discrete and continuous representations

of crystallinity. As observed from the ﬁgure, our numerical simulation is able to capture the fundamental behaviors of the

SIC process. Initially, the material has no crystals and deforms elastically with the applied load. When loading beyond λ̄ =

3 is applied, the material starts to crystallize and the slope of the crystallinity ratio changes accordingly. Since both the

chain model and the crystallinity evolution are described as differentiable mathematical expressions, the crystallinity ratio

changes smoothly in the numerical simulation. Maximum stress is seen at λ̄ = 6 after which the specimen is unloaded.

The crystallinity grows slightly in the initial unloading stage because SIC is a rate dependent process and it is not fully

equilibrated at the maximum-stretched state. When the amount of stretch goes below λ̄ = 5.5, the degree of crystallinity

decreases until it almost vanishes at the end of the unloading stage. The degree of crystallinity is less than the equilibrium

crystallinity during loading. In contrast, it becomes larger than the equilibrium crystallinity in the unloading stage (except for

λ̄ > 5.5), thus forming a hysteresis in the plot of the degree of crystallinity. Since higher levels of crystallinity correspond

to lower mechanical stresses, the rubber material experiences lower amounts of stress during unloading. Therefore, the

86 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 8. Uniaxial loading simulation (the solid lines) compared with experimental data (the dashed lines) for sulfur vulcanized natural rubber (Toki et al.,

2003). The stress-stretch relationship is plotted in (a) and (c) and the evolution of crystallinity during the test is demonstrated in (b) and (d). The top Figs.

(a,b) refer to the discrete crystallinity representation while the bottom graphs (c, d) represent the continuous crystallinity distribution.

hysteresis in the crystallinity response corresponds to a hysteresis in the amount of stress in the material. It is assumed that

no other dissipative effects such as viscoelasticity are present in this uniaxial test and the hysteresis is solely the result of

crystallization. Experimentally, Marchal (2006) shows that the hysteresis in the stress response of rubber is primarily due to

DOC whereas viscous effects are very small. The simulations are performed for both discrete and continuous distributions

of crystallinity. The calibrated parameters for these two cases are exactly the same as the values mentioned in Section 2.4.1.

In the case of continuous crystallinity formulation, the initial value for the width parameter is chosen as b = 2.0. For all

simulations presented here, 37 × 2 integration orientations are incorporated.

Next we examine the experimental observations of Marchal (Marchal, 2006; Rault et al., 2006a). The material is sulfur

vulcanized natural rubber with 1.2 g of sulfur per 100 g of rubber and it is stretched to various maximum elongations

for which the unloading paths are measured in the experiment. Fig. 9 shows the experimental data as well as simulation

results we obtained with our model. Material parameters, calibrated for the maximum elongation λ̄max = 6.5, are μ = 0.203

MPa, N = 23.8, α = 0.388, c̄ = 7.22 MPa, ξ¯ = 0.187 MPa, ωmax = 0.7833, q̄ = 0.0215MPa−1 min−1 for the discrete crystallinity

distribution, and μ = 0.22 MPa, N = 23.1, α = 0.33, c̄ = 7.6 MPa, ξ¯ = 0.051 MPa, ωmax = 1.0, q̄ = 0.019 MPa−1 min−1 for the

continuous crystallinity distribution model. The size of the specimen used by Rault et al. (2006a) is 2 × 5 × 30 mm and the

displacement is applied with a rate of 0.035 mm/s. The plots show reasonable agreement between the experimental and

simulation results. During initial portion of unloading, the proposed model predicts higher levels of crystallinity and larger

decrease in the uniaxial force. This is due to the rate-dependent evolution law of (14).

Various softening and hardening stages of SIC can be seen in Fig. 9(c) with the notation deﬁned in Albouy et al. (2005).

These transition points are named A, B, C, D, and E. From λ̄O to λ̄A , rubber behaves elastically. From λ̄A to λ̄B , crystallization

starts and the stress remains relatively constant, after which the hardening region between λ̄B and λ̄C occurs. In this part

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 87

Fig. 9. Uniaxial loading simulation (the solid lines) compared with experimental data (dots) for sulfur vulcanized natural rubber (Rault et al., 2006a). The

stress-stretch relationship is plotted in (a) and (c) and the evolution of crystallinity during the test is demonstrated in (b) and (d). The top Figs. (a,b) refer

to the discrete crystallinity representation while the bottom graphs (c, d) represent the continuous crystallinity distribution.

Fig. 10. The evolution of the half-width angle hf during uniaxial tension simulation (solid curves) in comparison with the experimental results of material

NR-04 (dashed lines) described by Tosaka et al. (2004). (a) the discrete crystallinity distribution approach (b) the continuous crystallinity distribution

method.

88 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 11. The simulation for two-step tension and shear loading using the continuous formulation for different values of initial stretch λ̄y . (a) The schematic

representation of the two-stage simulation (b) the evolution of crystallinity ratio ω̄ (c) rotation of the reference angle of the crystals (d) rotation of the

deformed angle of the crystals together with experimental data from Pannier et al. (2011).

of loading, the value of ω has grown close to the value of ωmax for some of the chains and because of the penalty term in

(10), further crystallization is slowed down. Additionally, Since the elongation of the chain is near its limiting extensibility,

hardening is observed. During unloading (C, D, E, O), the same process is observed in the reverse order. The stress remains

relatively constant between λ̄D and λ̄E .

Since the directional distribution of crystallinity is explicitly monitored in our model, we try to validate this distribution

by experiments. The peaks in the x-ray diffraction intensity plot provide information about the distribution of crystallinity.

The orientation with the maximum intensity shows the orientation of the crystal in a speciﬁc point in the material. This

intensity is reduced when we look at orientations further from the peak. The plot of X-ray diffraction intensity versus angle

can be approximated by a normal distribution function expressed as (Tosaka et al., 2004)

θc2

I = Imax exp − (67)

2w2

where θ c is the angle between the deformed chain orientation and the peak intensity direction. Imax is the peak intensity

and w is the distribution width parameter. This distribution of intensity allows us to deﬁne the half width angle hf as the

angle between two orientations in the deformed state which correspond to the intensity of Imax /2. The half width angle is

calculated as

√

h f = 2w 2 ln 2. (68)

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 89

Fig. 12. The 3D notched specimen: (a) The schematic geometry. (b) The ﬁnite element mesh. (c) A close up of the mesh.

Fig. 13. Distribution of the crystallinity ratio ω̄ in the vicinity of the tip of one of the notches: (a) Numerical simulation contour results (b) experimental

observations from Rublon et al. (2014). (c) Simulation results along the horizontal x-axis (d) simulation results along the vertical y-axis.

In our simulations, in order to calculate the angle of the chains in the deformed state with respect to the main crystalline

orientation, we use the stress tensor as the mapping based on (41). The uniaxial loading is applied in the X direction and

crystals are forming along the same axis. Then the stress tensor takes the diagonal form

P̄11 0 0

P̄ = 0 P̄22 0 (69)

0 0 P̄22

90 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Fig. 14. Three dimensional views of the distribution of crystallinity around the notch corresponding to strain values of (a) ε = 0 (b) ε = 20% (c) ε = 40%

(d) ε = 60% (e) ε = 80% (f) ε = 92%.

and allows us to compute the deformed chain orientation. Assuming that the chain originally exists in the X − Y plane

λ0 = cos θ e1 + sin θ e2 , the deformed orientation is

f 3P̄ λ0 3

λˆ = = = P̄ cos θ e1 + P̄22 sin θ e2 (70)

f ¯f ¯f 11

which makes the angle θc = tan−1 (P̄22 /P̄11 ) tan θ with the crystallinity axis. In addition to computing the orientation of the

chains in the current conﬁguration, we can obtain the degree of crystallization in the deformed state ωc using the concept

of area transformation on a unit sphere

which is also used in Kroon (2010). The X-ray diffraction intensity is assumed to be proportional to ωc , so by ﬁtting ωc (θ c )

to a normal distribution and using (68) we can calculate the half-width parameter.

Tosaka et al. (2004) computed the half width of the crystals inside a rubber specimen at various stages of the uniaxial

loading and unloading. Fig. 10 compares those experimental data for natural rubber NR-4 with the values computed from

our simulations. The displacement rate is 10 mm/min applied to a 24 mm specimen. The material parameters for this simu-

lation are calibrated by the uniaxial data provided by Tosaka et al. (2004). The calibrated parameters are μ = 0.37 MPa,

N = 32, α = 0.25, c̄ = 8.5 MPa, ξ¯ = 1.1 MPa, ωmax = 0.92, and q̄ = 0.015 MPa−1 min−1 for both discrete and continuous

crystallinity simulations. Similar to Section 4.1, the initial value of the width parameter is set to b = 2.0 for the continuous

crystallinity formulation. Two different phenomena contribute to the change of the half-width angle. Increasing crystallinity

ratio causes more chains to crystallize and should cause the half-width to increase, but this effect can only be seen if we

look at the angle of the chains in the reference orientation. However, during uniaxial tension, the crystallized chains are

aligned toward the principal strain direction and this effect decreases the half-width angle in the deformed conﬁguration.

During uniaxial loading simulations, the crystals grow in the direction of the loading and they will not rotate unless

the load changes its direction. In order to examine how the computational model simulates the rotation of the crystals, a

non-proportional loading scheme is used. First the specimen is pulled in Y direction up to a certain stretch λ̄y , after which

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 91

Fig. 15. Orientation of the polymer crystals in the region surrounding the notch in the current conﬁguration. φ c is the positive angle between the polymer

crystal orientation and the vertical axis. (a) Numerical simulation contour results (b) experimental ﬁndings from Rublon et al. (2014). (c) Simulation and

experimental results along the vertical axis (d) schematic view of the deﬁnition of φ c and the coordinate axes.

⎡ −1/2 ⎤

λ̄y γxy λ̄y 0

F = F̄ = ⎣0 λ̄y 0 ⎦. (72)

0 0 λ̄−1

y

/2

The shear strain rotates the principal axis of stretch and consequently, the chains in the ﬁrst and third quadrant of the X − Y

plane crystallize more than the chains in the second and fourth quadrants. Therefore, the crystal axis rotates toward the

ﬁrst and third quadrants by changing the spherical angle φ 2 . This rotation occurs automatically by solving the crystallinity

evolution residual (58).

Fig. 11 shows the results of this non-proportional loading with our continuous crystallinity distribution model. Three

different simulations are performed with different values of initial stretch λ̄y . In the case of λ̄y = 2, the crystallinity is

negligible at the end of the tensile region, and the shear displacement allows the ﬁrst crystals to grow. However, in the

other cases, strain-induced crystallization already happens before any shear is applied. The rotation of the crystals are visible

when looking at the angle between the crystal axis and the Y axis in the reference conﬁguration φ r . When the crystals are

small, they are easier to rotate, thus enabling higher rotation for the simulations with lower λ̄y . However, this difference in

rotation angles is not clearly distinguishable when looking at the crystallinity angle in the current conﬁguration φ c because

the rigid-body rotation coming from the shear strain contributes strongly to this rotation. Moreover, at these large strains,

the deformation of the networks rotates almost all of the chains into the same orientation and the different angles in the

reference conﬁguration are mapped into the same narrow region in the current conﬁguration.

92 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Pannier et al. (2011) performs this combined tensile and shear loading and measures the angle φ c . Those experimental

results are also shown on Fig. 11(d). The tensile and shear loading speed is set to λ˙ y = 0.05/min and γ˙ xy = 0.25/min, based

on the values used in the experiment. Since there is no stress-strain information of the material used in the experiment, the

same material parameters (expressed in Section 2.4.1) are employed except for N = 28 and c̄ = 8 MPa in order to allow large

amounts of tensile stretch. The simulation results are close to the experimental values for the case of λ̄y = 2.0. However,

one of the points of the experiment for λ̄y = 4.0 has a large discrepancy with respect to the simulation curve.

Using our 3D network model allows us to study the behavior of rubber under complex loading conditions, which provides

more insight into the effectiveness of the SIC model. We recreated the experiments of Rublon et al. (2014) as a 3D numerical

simulation to study the effects of SIC in the vicinity of cracks and high-stress regions. Since Rublon et al. (2014) use sulfur

vulcanized natural rubber and do not provide uniaxial test results, we incorporate the continuous crystallinity distribution

model with the same parameters as in Section 4.1 to perform these simulations. When a specimen with a notch is pulled,

we observe strain localization in the vicinity of the notch. This region becomes crystallized due to SIC and this crystalliza-

tion helps reinforce the material, thereby reducing the tendency to form a crack. The geometry and the dimensions of the

simulation are shown in Fig. 12(a). This simulation uses 2436 hexahedral elements and 4020 nodes, reﬁned in the areas

close to the notches as shown in Figs. 12(b–c). The tip of the notches are modeled as half circles with the radius of 0.2

mm in order to avoid stress singularities. The simulation does not consider tearing or the formation of cracks and thus the

length of the notch is constant. The specimen experiences up to 92% strain perpendicular to the notch axis while clamping

the ends of the specimen.

The distribution of crystallinity close to the notch tip can be seen in Fig. 13, showing qualitative and quantitative agree-

ment with the experimental results. In addition to the contour data, the variation of the crystallinity is shown in two

perpendicular axes around the notch. The crystalline region is relatively small and it diminishes as we get further from

high-strain areas. The three dimensional snapshots of the crystallization around the notch are also shown in Fig. 14.

Since our constitutive model stores the orientation of the polymer crystals at every material point, we can plot the orien-

tation of the crystals surrounding the simulated notch. At the beginning of the simulation, the material has zero crystallinity

at every point, the direction of the crystals is undeﬁned. When the crystallization starts, the singular value decomposition

method is used to compute the initial orientation of crystallinity, which continues to rotate afterward based on the evolution

of . Because the stored crystal orientations belong to the reference conﬁguration, we have to consider the ﬁnite rotation of

the material to calculate the orientation of the crystal in the current conﬁguration. The calculated crystal orientation in the

current conﬁguration is performed by considering P̄ as the rotation operator in the microscale based on (41). The results

can be viewed in Fig. 15, which are very close to the experimental ﬁndings.

5. Conclusion

A micromechanical homogenization model for studying strain-induced crystallization is presented which connects the

deformation and phase transformation of polymer chains with the large scale measurable properties of the material such as

the deformation or the X-ray diffraction intensity. A new polymer chain model is introduced based on non-Gaussian statis-

tical mechanics which takes into account the change of the ﬂexible length of the chain during crystallization and provides

an evolution law for crystallization using the gradient of the free energy expression. The MAPC model is incorporated in

the proposed framework to compute the microscopic stretch and rotation of the chains from the large scale local contin-

uum deformation ﬁeld. Furthermore, in accordance with experimental evidence, the anisotropic crystallization of rubber is

observed computationally due to the non-uniform stretch of chains in different orientations. In addition to the traditional

representation of crystallinity values at discrete orientations, a new continuous crystallinity distribution is presented and

the respective mathematical expressions for computing the evolution of crystallinity are derived. The material parameters

can ﬁt experimental uniaxial tension tests with great accuracy. It is shown that the model accurately predicts the rotation of

the crystals as well as the change in the half-width angle. The accumulation of a crystalline phase in the high-strain region

around a notch is simulated using the proposed model and the results agree with the experimental data.

Our model can be improved in the future by incorporating various mechanical phenomena such as viscosity, damage, and

fracture. At high strains, rubber material can be damaged due to the Mullins effect (Göktepe and Miehe, 2005; Mullins, 1948)

causing softer response in subsequent loading cycles, which can have a combined effect with SIC. Therefore, introducing the

Mullins effect in the SIC model is one of the future directions for expanding this model. Incorporating viscous effects based

on Linder et al. (2011) is another direction of future research to be pursued. Additionally, in order to better study the role

of SIC in the vicinity of a crack tip, it is necessary to combine this presented model with fracture formulations such as the

phase ﬁeld approaches (Miehe et al., 2010; Zhang et al., 2016) or the strong discontinuity methods (Linder and Raina, 2013;

Linder and Zhang, 2013; 2014; Raina and Linder, 2015).

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 93

Acknowledgements

Financial support for this research was provided by Samsung Electronics and the National Science Foundation through

CAREER Award CMMI-1553638. This support is gratefully acknowledged.

Based on Fig. 2, in a partially-crystallized chain, the end-to-end length and the total length of the effective random walk

can be calculated as

ra = r − Lc and La = L − Lc . (A.1)

This result is used to calculate the relative stretch of the freely jointed chains compared to the maximum possible stretch

as

√ √

ra λ Nl − ωNl λ/ N − ω

= = = . (A.2)

La Nl − ωNl 1−ω

The variation of this quantity versus the change in λ or ω, given as

√

dλ dω (λ/ N − 1 ) dλ

d = √ + = ( 1 − ω ) −1

√ + ( − 1 ) d ω (A.3)

N (1 − ω ) ( 1 − ω )2 N

is useful for calculating the derivatives of the free energy later in this appendix. The main thermodynamic quantity of a

chain formulation is its free energy. This free energy is deﬁned in (9) based on the force response of a non-Gaussian chain.

The amorphous portion of the free energy can be calculated by substituting (8) into (9), yielding

λ

r

k T β (λ ) √

ψa (λ, ω ) = F ( r )d r = ( B )( Nldλ ). (A.4)

0 0 l

This integral represents the mechanical work done when stretching the chain from zero end-to-end length to its current

length. The star symbol is used to distinguish the integration variable from the upper bound of the deﬁnite integral.

During this stretch, the amount of crystallization and the temperature

√ remains constant. In order to solve the integral we

change the integration variable from λ to by using dλ = N (1 − ω )d from (A.3) (considering dω = 0). We then

replace with β using integration by parts. Thus,

β

√ √

ψa (λ, ω ) = kB T Nβ ( )( N (1 − ω )d ) = kB T N (1 − ω ) ( β − 0 ) − (β )dβ . (A.5)

−ω / ( 1 −ω ) 0

The lower bound of the integral is changed from λ = 0 to its equivalent = −ω/(1 − ω ) leading to a negative relative

stretch. This case refers to the condition that the end-to-end length of the chain is shorter than the crystallized segment.

For calculating this integral, the entropic force at this initial stage is considered to be zero. Thus, we can substitute 0 instead

of −ω/(1 − ω ) as the lower bound of the integral. Now we use the anti-derivative of the Langevin function

1

β

− (β )dβ = − coth β − dβ = −(ln(sinh β ) − ln β ) = ln (A.6)

β sinh β

to obtain the expression in (11). The thermodynamic force and the crystallization force are formed by calculating the differ-

ential of the free energy as

β β

dψ = kB T N (−dω ) β + ln + kB T N (1 − ω )d β + ln + c dω

sinh β sinh β

dω dω ω −1

−ξ − (1 − ) . (A.7)

ωmax ωmax ωmax

From (A.6) we know that d(ln(β / sinh β ) ) = − dβ . Thus, combining it with (A.3), we get

β β β dλ

d β + ln = dβ + β d + d ln = β d = √ + ( − 1 )dω (A.8)

sinh β sinh β (1 − ω ) N

and

β √ ξ ω dω

dψ = f dλ − fw dω = −kB T N β + ln dω + kB T Nβ dλ + kB T Nβ ( − 1 )dω + c dω + .

sinh β ωmax (ωmax − ω )

(A.9)

94 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

Regrouping the terms in (A.9) results in the expression for the chain force and the crystallization force shown in (12) and

(13). These equations can be differentiated once more to formulate the jacobian matrix K, i.e.

√

√ ∂β √ 1 dλ μdλ + μ N ( − 1 )dω

d f¯ = μ N d = μ N (1 − ω )−1 √ + ( − 1 )dω = = K11 dλ + K12 dω

∂ L (β ) N L (β )(1 − ω )

(A.10)

and

q̄ξ¯ dω q̄μN (1 − ) dλ q̄ξ¯ dω

dω˙ = q̄d f¯w = q̄μN (dβ − dβ ) − = √ + ( − 1 )dω −

(ωmax − ω )2 L (β )(1 − ω ) N (ωmax − ω )2

= K21 dλ + K22 dω. (A.11)

As mentioned in Section 2.4.2, the amount of change of the crystallinity internal variable is computed by minimizing the

expression in (35). The crystallinity rates of the chains are considered constants and the optimization variable is the rate of

crystallinity internal variables

˙ . It can be easily seen that we are minimizing a quadratic function of ˙ and the solution

can be obtained by solving a linear system of equations. First we substitute the apparent rate from (34) to get

T 2

˙ = 1
(ω˙ − ω

˜˙ )2 + 1 γ ˙ = 1
∂ω 1

f () ˙ T ω˙ − ˙ + γ ˙ T ˙ (B.1)

2 2 2 ∂ 2

˙ out of the sphere average because it does not depend on the orientation

of the chains. Thus,

T T

˙ = 1 ω˙ 2 + 1 ∂ω ∂ω ∂ω 1

f () ˙T ˙ − ω˙ ˙ + γ ˙ T ˙ (B.2)

2 2 ∂ ∂ ∂ 2

T T

1 ˙T ∂ω ∂ω ∂ω 1

=
+ γ I4 −
ω˙

˙ ˙ +
ω˙ 2 (B.3)

2 ∂ ∂ ∂ 2

T

∂ω ∂ω ∂ω

min f ()

˙ → + γ I4 ˙ = A˙ = ω˙ (B.4)

˙ ∂ ∂ ∂

which can be solved because the matrix A is guaranteed to be positive deﬁnite due to the regularization term. In order

to ﬁnd the tangent for the non-linear evolution Eq. (58), the derivative of the crystallization rate with respect to the crys-

tallinity variables are needed. This can be achieved by taking the derivative of (B.4) with respect to , i.e.

4

∂ 2 ω ∂ω ∂ω ∂ 2 ω 4

∂ ˙ k ∂ ω˙ ∂ω ∂ 2ω

+ ˙ k + Aik = + ω˙ (B.5)

k=1

∂ i ∂ j ∂ k ∂ i ∂ k ∂ j k=1

∂ j ∂ j ∂ i ∂ i ∂ j

or

T

2 T T 2

∂ 2ω ∂ω ∂ω ∂ ω ˙ ∂ ˙ ∂ω ∂ ω˙ ∂ ω

+

˙ +A = + ω˙ . (B.6)

∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂

Using the chain rule we can relate [∂ ω˙ /∂ ] to [∂ ω˙ /∂ω], which is non-trivial to calculate, because we have to assume that

the normalized force f¯ of the chain is constant. The MAPC loop determines the chain force based on the stress tensor and

it remains constant throughout the crystallinity iterations. Consider the chain variables in differential form

d f¯ = K11 dλ + K12 dω and dω˙ = K21 dλ + K22 dω. (B.7)

Applying the fact that the chain force is constant , we can solve for dω˙ and ﬁnd our desired derivative as

K12 K21 dω˙ K12 K21

d f¯ = 0 → dω˙ = K22 − dω → = K22 − . (B.8)

K11 dω f¯=cst K11

Hence we can substitute

∂ω

∂ ω˙ dω˙ ∂ω K12 K21

∂

=

dω f¯=cst ∂

= K22 −

K11 ∂

(B.9)

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 95

As discussed in Section 3.1, solving for the MAPC deformation ﬁeld is the same as ﬁnding the stationary point of the

Lagrangian (39). Setting ∂ L/∂ λ = 0, where λ is the deformed vector of an arbitrary chain orientation, we can ﬁnd an ex-

pression relating the normalized force of the chains and the Lagrange multiplier, i.e.

∂ L ∂ ψ̄

0= = − νλ0 = f¯ − νλ0 → f¯ = νλ0 . (C.1)

∂λ ∂λ

The stress tensor P̄ is obtained by taking the derivative of the macro free energy with respect to the isochoric deformation

gradient to

∂ ¯ ∂ ψ̄ ∂λ

P̄ = =
· . (C.2)

∂ F̄ ∂λ ∂ F̄

Using the result of (C.1) and the MAPC constraint (21), the expression for the isochoric stress tensor can be simpliﬁed as

∂λ ∂λ ∂λ λ0 1

P̄ =
f¯ · =
νλ0 · =
ν : = ν (C.3)

∂ F̄ ∂ F̄ ∂ F̄ 3

as mentioned earlier in (40). The stationary condition of the Lagrangian with respect to ν is equivalent to the MAPC con-

straint. We use this constraint as a residual described in (42) and solve this residual considering P̄ as the independent vari-

able. This residual requires a corresponding tangent that guaranties quadratic convergence of the Newton–Raphson solver.

In the following, this step to obtain the expression for the tangent operator is explained for two separate cases: The discrete

and the continuous crystallinity representation.

We can write the differential form of the residual to observe its dependency on various variables especially the stress

tensor, i.e.

1 1

d Rλ = d F̄ −
λ λ0 = dF̄ −
dλ λ0 . (C.4)

3 3

The change in the stretch vector λ = λλ

ˆ originates from changes to its magnitude and the changes to its orientation. We

calculate each of them and show its dependency on the force vector of the corresponding chain as

d f¯ λˆ · d f¯ ∂ f¯ ∂ 2 ψ̄

dλ = = where k̄ = = (C.5)

k̄ k̄ ∂λ ∂λ2

f¯ f¯d f¯ − f¯ d f¯ d f¯ − λ

ˆ (λ ˆ · d f¯ ) 1−λ

ˆ λ

ˆ

dλ

ˆ =d = = = d f¯ (C.6)

f¯ ¯f 2 ¯f ¯f

λ λˆ λˆ

dλ = λdλ

ˆ +λ

ˆ dλ = 1−λ

ˆ λ

ˆ + d f¯ . (C.7)

f¯ k̄

Next we consider the dependence of the chain force on the stress tensor using (41), i.e.

d f¯ = d νλ0 = 3dP̄λ0 . (C.8)

Now we can substitute the results from (C.7) and (C.8) into (C.4) to obtain the expression for the local tangent tensor. During

the calculation of the microscopic deformation using MAPC, the deformation gradient remains constant during the iterations

(dF̄ = 0). Thus,

1 λ 1 λ ˆ ˆ

d Rλ = dF̄ − 3
1+ − λ λ dP̄λ0 λ0 (C.9)

3 f¯ k̄ f¯

and

∂ Rλ λ 1 λ ˆ ˆ

Kλ = − = 3
1+ − λ λ ( λ0 λ0 ) . (C.10)

∂ P̄ f¯ k̄ f¯

This latter expression can also be written in index notation as mentioned in (49). The stiffness of the chain k̄ represents the

additional amount of force required per unit of additional elongation. When the chain is not crystallizing, its value is given

as

∂ f¯

k̄ = = K11 (not crystallizing) (C.11)

∂λ

96 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

but during the change in crystallization, we must account for the softening due to crystallization. In the MAPC algorithm, the

crystallization occurs while the chain crystallization residual rw remains zero. So replacing drw = 0 in (44) and incorporating

(B.7) we obtain

0 = drw = −dω + t (K21 dλ + K22 dω ) (crystallizing) (C.12)

tK21

dω = dλ (crystallizing) (C.13)

1 − tK22

tK21

d f¯ = K11 dλ + K12 dω = K11 + K12 dλ (crystallizing) (C.14)

1 − tK22

tK12 K21

k̄ = K11 + (crystallizing). (C.15)

1 − tK22

As described in (60) the MAPC tangent for the continuous crystallinity distribution is composed of the tangent with con-

stant crystallinity and the adjustment required for the change in during the iteration. The ﬁrst expression, corresponding

to no change in crystallinity, is the same as substituting K11 for k̄ in (49) because K11 refers to the change in the chain force

for no change in stretch assuming the crystallinity is constant. Thus,

nint

k

∂ Rλ λl 1 λkl ˆ k ˆ k

− =3 1+ − λl λl ((λ0 )l (λ0 )l ) where (AB )i jkl = Aik B jl . (C.16)

∂ P̄ =cst l=1

f¯lk (K11 )kl f¯lk

The second part of the tangent expression requires the calculation of the change in the MAPC residual with change in the

crystallinity parameter ∂ Rλ /∂ i and the evolution of the crystallinity internal variables as a function of the change in the

stress tensor ∂ i /∂ P̄ . The calculation of ∂ Rλ /i must be carried out assuming that the stress tensor is constant because the

variation of the MAPC residual with respect to the stress tensor is already included in (C.16). Thus,

∂ Rλ ∂λ ∂λ ∂ω ˆ

= −
λ0 = −
λ λ0 .

∂ω f¯=cst ∂ i

(C.17)

∂ i ∂ i

Note that since the stress tensor is kept constant, it implies that the chain force is also assumed to be constant during this

calculation, i.e.

∂λ K12

0 = d f¯ = K11 dλ + K12 dω → =− (C.18)

∂ω f¯=cst K11

and

∂ Rλ K12 ∂ω ˆ

=
λ λ0 . (C.19)

∂ i K11 ∂ i

The ﬁnal expression to calculate is ∂ i /∂ P̄ . This parameter is zero when the material is not crystallizing, i.e. when the

crystallinity is zero, a = 0, and the crystallinity rate is not positive, a˙ ≤ 0. In the case when the crystallinity internal variables

are changing, this derivative must be calculated assuming that the crystallinity evolution residual is identically zero during

the time step, i.e.

∂ω

0 = dR = −d + t d A
ω˙ .

−1

(C.20)

∂

∂ω

Both A and do not depend on the value of P̄ . Therefore, this expression can be simpliﬁed as

∂

∂ R ∂ R ∂ ˙ ∂ ω˙ ∂ω

0= d + : dP̄ = − I4 − t d + t A−1
: dP̄ (C.21)

∂ ∂ P̄ ∂ ∂ P̄ ∂

∂ω

−1 ∂ ω ˙

0 = −K d + t A
3λ · dP̄ λ0

ˆ (C.22)

∂ f¯ ω=cst ∂

where the equality d f¯ = λˆ · d f¯ = 3λ

ˆ · dP̄ λ0 is incorporated. The calculation of ∂ ω˙ /∂ f¯ is performed with no change in ω

because this variation is already included in d. Thus,

d f¯ − 0 ∂ ω˙ K21

dω = 0 ⇒ dω˙ = K21 dλ + 0 = K21 ⇒ = . (C.23)

K11 ∂f ¯ ω = cst K11

By substituting (C.23) into (C.22) and solving for ∂ i /∂ P̄ , expression (62) is obtained.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 97

For computing the consistent tangent of the material we ﬁrst ﬁnd the relationship between the change in P̄ and the

change in F̄ , using (C.4). In all steps of the iteration, the MAPC residual is kept identically zero and thus we can incorporate

dRλ = 0 as

1

0 = d Rλ = dF̄ − Kλ : dP̄ (D.1)

3

∂ P̄ 1 −1

= K (D.2)

∂ F̄ 3 λ

∂ P̄iJ 1 λ 1 λ ˆ ˆ

=
δik + − λi λk (λ0 )J (λ0 )L −1 . (D.3)

∂ F̄kL 9 f¯ k̄ f¯

In order to convert ∂ P̄ /∂ F̄ to the isochoric consistent tangent C̄ we have to write the expression for the rate of change of

the isochoric stress

∂ P̄ ˙ T

£¯v (τ̄ ) = τ̄˙ − l¯τ̄ − τ̄ l¯T = P̄˙ F̄ T + P̄ F̄˙ T − l¯τ̄ − P̄ F̄ T l¯T = : F̄ F̄ + P̄ (l¯F̄ )T − l¯τ̄ − P̄ F̄ T l¯T (D.4)

∂ F̄

∂ P̄ ¯

£¯v (τ̄ ) = : (l F̄ ) F̄ T − l¯τ̄ (D.5)

∂ F̄

and extract the velocity gradient tensor l¯ using the index notation

∂ P̄iA ¯ ∂ P̄iA

£¯v (τ̄ ) = lkl F̄lB F̄jA − l¯il τ̄l j = F̄lB F̄jA − δik τ̄l j l¯kl . (D.6)

ij

∂ F̄kB ∂ F̄kB

The 4th order tensor appearing in equation (D.6) can be shown to be symmetric with respect to the indices k and l (the

proof is presented in Remark 3). Thus, we can replace l¯kl with its symmetric component d¯kl = (l¯kl + l¯lk )/2, to get

1

£¯v τ̄ = C̄ : d¯ = C̄ : £¯v g (D.7)

2

where

∂ P̄iA

C̄i jkl = F̄lB F̄jA − δik τ̄l j . (D.8)

∂ F̄kB

Remark 3. To prove the minor symmetry of the isochoric spatial modulus, we can write the spatial modulus as

∂ (P̄ F̄ ) ∂ F̄jA ∂ τ̄

F̄lB − δik τ̄l j = − P̄iA δ jk δAB F̄lB − δik τ̄l j

iA jA ij

C̄i jkl = − P̄iA (D.9)

∂ F̄kB ∂ F̄kB ∂ F̄kB

∂ τ̄i j

C̄i jkl = F̄lB − δ jk τ̄il + δik τ̄l j . (D.10)

∂ F̄kB

The second term in (D.10) has obvious minor symmetry. For proving the minor symmetry of the ﬁrst statement assume

= τ̄i j is the scalar value of one of the components of τ̄ . is a function of F̄ and since τ̄ is objective, is also a function

of C̄ . Thus,

∂ ˙

˙ = : F̄ (D.11)

∂ F̄

and

∂ ˙ ∂ ˙ T ∂ T ˙ ∂ ˙

˙ = : C̄ = : F̄ F̄ + F̄ T F̄˙ = : 2F̄ F̄ = 2F̄ : F̄ (D.12)

∂ C̄ ∂ C̄ ∂ C̄ ∂ C̄

Since F̄˙ is arbitrary, equating expressions (D.11) and (D.12) results in

∂ ∂

= 2F̄ (D.13)

∂ F̄ ∂ C̄

∂ ∂ T ∂ T

F̄lB ek el = F̄ = 2F̄ F̄ → symmetric . (D.14)

∂ F̄kB ∂ F̄ ∂ C̄

∂ τ̄i j

Therefore, F̄lB is symmetric with respect to the indices k and l. As a result, the minor symmetry of C̄ is proved.

∂ F̄kB

98 R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99

References

Albouy, P.A., Marchal, J., Rault, J., 2005. Chain orientation in natural rubber, part i: the inverse yielding effect. Euro. Phys. J. E 17 (3), 247–259. doi:10.1140/

epje/i2004- 10145- 6.

Andrews, E.H., 1964. Crystalline morphology in thin ﬁlms of natural rubber ii. crystallization under strain. Proc. Royal SocLondon A 277 (1371), 562–570.

doi:10.1098/rspa.1964.0040.

Arlman, J.J., Goppel, J.M., 1951. On the degree of crystallinity in natural rubber. Appli Sci Res 2 (1), 1. doi:10.10 07/BF0 0411969.

Arruda, E.M., Boyce, M.C., 1993. A three-dimensional constitutive model for the large stretch behavior of rubber elastic materials. Je Mech Phys Solids 41

(2), 389–412.

Bažant, Z.P., Oh, B.H., 1986. Eﬃcient numerical integration on the surface of a sphere. ZAMM - J. Appl. Math. Mech. Zeitschrift für Angewandte Mathematik

und Mechanik 66 (1), 37–49.

Boyce, M.C., Arruda, E.M., 20 0 0. Constitutive models of rubber elasticity: a review. Rubber Chem. Technol. 73 (3), 504–523. doi:10.5254/1.3547602.

Boyce, M.C., Arruda, E.M., Jayachandran, R., 1994. The large strain compression, tension, and simple shear of polycarbonate. Polymer Eng. Sci. 34 (9), 716–

725. doi:10.1002/pen.760340904. https://doi.org/10.1002/pen.760340904

Brüning, K., Schneider, K., Roth, S.V., Heinrich, G., 2013. Strain-induced crystallization around a crack tip in natural rubber under dynamic load. Polymer 54

(22), 6200–6205. https://doi.org/10.1016/j.polymer.2013.08.045.

Candau, N., Laghmach, R., Chazeau, L., Chenal, J.-M., Gauthier, C., Biben, T., Munch, E., 2014. Strain-induced crystallization of natural rubber and cross-link

densities heterogeneities. Macromolecules 47 (16), 5815–5824. doi:10.1021/ma5006843.

Candau, N., Laghmach, R., Chazeau, L., Chenal, J.-M., Gauthier, C., Biben, T., Munch, E., 2015. Inﬂuence of strain rate and temperature on the onset of strain

induced crystallization in natural rubber. Euro. Polymer J. 64, 244–252.

Dargazany, R., Khiêm, V.N., Itskov, M., 2014. A generalized network decomposition model for the quasi-static inelastic behavior of ﬁlled elastomers. Int. J.

Plasticity 63, 94–109. https://doi.org/10.1016/j.ijplas.2013.12.004.

Dargazany, R., Khiêm, V.N., Poshtan, E.A., Itskov, M., 2014. Constitutive modeling of strain-induced crystallization in ﬁlled rubbers. Phys. Rev. E 89, 022604.

doi:10.1103/PhysRevE.89.022604.

Flory, P.J., 1947. Thermodynamics of crystallization in high polymers. i. crystallization induced by stretching. J. Chem.l Phys. 15 (6), 397.

Flory, P.J., Rehner Jr, J., 1943. Statistical mechanics of cross-linked polymer networks i. rubberlike elasticity. J. Chem. Phys. 11 (11), 512–520.

Fu, X., Huang, G., Xie, Z., Xing, W., 2015. New insights into reinforcement mechanism of nanoclay-ﬁlled isoprene rubber during uniaxial deformation by in

situ synchrotron x-ray diffraction. RSC Adv. 5, 25171–25182. doi:10.1039/C5RA02123E.

Gaylord, R.J., 1976. A theory of the stress-induced crystallization of crosslinked polymeric networks. J. Polym. Sci.: Polym. Phys. Ed. 14 (10), 1827–1837.

doi:10.1002/pol.1976.180141008. https://doi.org/10.1002/pol.1976.180141008

Gehman, S.D., Field, J.E., 1939. An x-ray investigation of crystallinity in rubber. J. Appl. Phys. 10 (8), 564–572. doi:10.1063/1.1707343.

Gehman, S.D., Field, J.E., 1944. Observations on the x-ray structure of rubber and the size and shape of rubber crystallites. J. Appl. Phys. 15 (4), 371–379.

doi:10.1063/1.1707441.

Gent, A.N., 1954. Crystallization and the relaxation of stress in stretched natural rubber vulcanizates. Trans. Faraday Soc. 50, 521–533. doi:10.1039/

TF95450 0 0521.

Göktepe, S., Miehe, C., 2005. A micro-macro approach to rubber-like materials. part III: the micro-sphere model of anisotropic mullins-type damage. J.

Mech. Phys. Solids 53 (10), 2259–2283.

Guilié, J., Thien-Nga, L., Le Tallec, P., 2015. Micro-sphere model for strain-induced crystallisation and three-dimensional applications. J. Mech. Phys. Solids

81, 58–74. doi:10.1016/j.jmps.2015.05.0 04. www.sciencedirect.com/science/article/pii/S0 0225096150 0109X

James, H.M., Guth, E., 1943. Theory of the elastic properties of rubber. J. Chem. Phys. 11 (10), 455–481.

Katz, J.R., 1925. Röntgenspektrographische Untersuchungen am gedehnten Kautschuk und ihre mögliche Bedeutung für das Problem der Dehnungseigen-

schaften dieser Substanz. Naturwissenschaften 13 (19), 410–416. doi:10.1007/BF01560952.

Kroon, M., 2010. A constitutive model for strain-crystallising rubber-like materials. Mech. Mater. 42 (9), 873–885.

Kuhn, W., Grün, F., 1942. Beziehungen zwischen elastischen Konstanten und Dehnungsdoppelbrechung hochelastischer Stoffe. Kolloid-Zeitschrift 101 (3),

248–271. doi:10.1007/BF01793684.

Linder, C., Raina, A., 2013. A strong discontinuity approach on multiple levels to model solids at failure. Comput. Methods Appl. Mech. Eng. 253, 558–583.

Linder, C., Tkachuk, M., Miehe, C., 2011. A micromechanically motivated diffusion–based transient network model and its incorporation into ﬁ-

nite rubber viscoelasticity. J. Mech. Phys. Solids 59 (10), 2134–2156. doi:10.1016/j.jmps.2011.05.005. http://www.sciencedirect.com/science/article/pii/

S0 0225096110 01086

Linder, C., Zhang, X., 2013. A marching cubes based failure surface propagation concept for three-dimensional ﬁnite elements with non-planar embedded

strong discontinuities of higher-order kinematics. Int. J. Numer. Methods Eng. 96 (6), 339–372. doi:10.1002/nme.4546.

Linder, C., Zhang, X., 2014. Three-dimensional ﬁnite elements with embedded strong discontinuities to model failure in electromechanical coupled materials.

Comput. Methods Appl. Mechan. Eng. 273, 143–160. doi:10.1016/j.cma.2014.01.021.

Luch, D., Yeh, G.S.Y., 1973. Morphology of strain-induced crystallization of natural rubber . Part II . X-ray studies on cross-linked vulcanizates. J. Macromol.r

Sci. Part B 7 (February), 121–155. doi:10.1080/00222347308212577.

Marchal, J., 2006. Cristallisation des caoutchoucs chargés et non chargés sous contrainte: effet sur les chaines amorphes. Université Paris Sud - Paris XI

Theses.

Mars, W., Fatemi, A., 2004. Factors that affect the fatigue life of rubber: a literature survey. Rubber Chem. Technol. 77 (3), 391–412.

Miehe, C., 1994. Aspects of the formulation and ﬁnite element implementation of large strain isotropic elasticity. Int. J. Numer. Methods Eng. 37 (12),

1981–2004. doi:10.1002/nme.1620371202.

Miehe, C., Göktepe, S., Lulei, F., 2004. A micro-macro approach to rubber-like materials. part i: The non-aﬃne micro-sphere model of rubber elasticity.

Journal of the Mechanics and Physics of Solids 52 (11), 2617–2660.

Miehe, C., Welschinger, F., Hofacker, M., 2010. Thermodynamically consistent phase-ﬁeld models of fracture: variational principles and multi-ﬁeld FE imple-

mentations. Int. J. Numer. Methods Eng. 83 (10), 1273–1311. doi:10.1002/nme.2861.

Mistry, S.J., Govindjee, S., 2014. A micro-mechanically based continuum model for strain-induced crystallization in natural rubber. Int. J. Solids Struct. 51

(2), 530–539. doi:10.1016/j.ijsolstr.2013.10.027. http://www.sciencedirect.com/science/article/pii/S0020768313004162

Mitchell, G.R., 1984. A wide-angle X-ray study of the development of molecular orientation in crosslinked natural rubber. Polymer 25 (11), 1562–1572.

doi:10.1016/0032-3861(84)90148-4. http://www.sciencedirect.com/science/article/pii/0032386184901484

Mullins, L., 1948. Effect of stretching on the properties of rubber. Rubber Chem. Technol. 21 (2), 281–300. doi:10.5254/1.3546914.

Pannier, Y., Proudhon, H., Mocuta, C., Thiaudière, D., Cantournet, S., 2011. In situ multi-axial loading frame to probe elastomers using x-ray scattering. J.

Synchrotron Radiat. 18 (6), 907–911. doi:10.1107/S0909049511024861.

Raina, A., Linder, C., 2015. A micromechanical model with strong discontinuities for failure in nonwovens at ﬁnite deformation. Int. J. Solids Struct. 75–76,

247–259.

Rault, J., Marchal, J., Judeinstein, P., Albouy, P.A., 2006. Chain orientation in natural rubber, part ii: 2h-nmr study. Euro. Phys. J. E 21 (3), 243–261. doi:10.

1140/epje/i20 06-10 064-6.

Rault, J., Marchal, J., Judeinstein, P., Albouy, P.A., 2006. Stress-induced crystallization and reinforcement in ﬁlled natural rubbers: 2h nmr study. Macro-

molecules 39 (24), 8356–8368. doi:10.1021/ma0608424.

Ren, Y., Zhao, S., Yao, Q., Li, Q., Zhang, X., Zhang, L., 2015. Effects of plasticizers on the strain-induced crystallization and mechanical properties of natural

rubber and synthetic polyisoprene. RSC Adv. 5, 11317–11324. doi:10.1039/C4RA13504K.

R. Rastak, C. Linder / Journal of the Mechanics and Physics of Solids 111 (2018) 67–99 99

Roe, R.J., Smith Jr., K.J., Krigbaum, W.R., 1961. Equilibrium degrees of crystallization predicted for “single pass” and folded chain crystallite models. J. Chem.

Phys. 35 (4), 1306–1311. doi:10.1063/1.1732043.

Rublon, P., Huneau, B., Saintier, N., Beurrot, S., Leygue, A., Verron, E., Mocuta, C., Thiaudière, D., Berghezan, D., 2013. In situ synchrotron wide-angle X-ray

diffraction investigation of fatigue cracks in natural rubber. J. Synchrotron Radiat. 20 (1), 105–109. doi:10.1107/S0909049512044457.

Rublon, P., Huneau, B., Verron, E., Saintier, N., Beurrot, S., Leygue, A., Mocuta, C., Thiaudière, D., Berghezan, D., 2014. Multiaxial deformation and strain-

induced crystallization around a fatigue crack in natural rubber. Eng. Fracture Mech. 123, 59–69. doi:10.1016/j.engfracmech.2014.04.003.

Smith Jr., K.J., 1976. Crystallization of networks under stress. Polymer Eng. Sci. 16 (2), 168–175.

Smith Jr., K.J., Greene, A., Ciferri, A., 1964. Crystallization under stress and non-gaussian behavior of macromolecular networks. Kolloid-Zeitschrift und

Zeitschrift für Polymere 194 (1), 49–67. doi:10.1007/BF01499976.

Tkachuk, M., Linder, C., 2012. The maximal advance path constraint for the homogenization of materials with random network microstructure. Philosophical

Mag. 92 (22), 2779–2808. doi:10.1080/14786435.2012.675090. http://www.tandfonline.com/doi/abs/10.1080/14786435.2012.675090

Toki, S., 2014. The effect of strain-induced crystallization (SIC) on the physical properties of natural rubber (NR). In: Kohjiya, S., Ikeda, Y. (Eds.), Chemistry,

Manufacture and Applications of Natural Rubber. Woodhead Publishing, pp. 135–167.

Toki, S., Che, J., Rong, L., Hsiao, B.S., Amnuaypornsri, S., Nimpaiboon, A., Sakdapipanich, J., 2013. Entanglements and networks to strain-induced crys-

tallization and stress-strain relations in natural rubber and synthetic polyisoprene at various temperatures. Macromolecules 46 (13), 5238–5248.

doi:10.1021/ma400504k.

Toki, S., Fujimaki, T., Okuyama, M., 20 0 0. Strain-induced crystallization of natural rubber as detected real-time by wide-angle X-ray diffraction technique.

Polymer 41 (14), 5423–5429. doi:10.1016/S0 032-3861(99)0 0724-7.

Toki, S., Sics, I., Ran, S., Liu, L., Hsiao, B.S., 2003. Molecular orientation and structural development in vulcanized polyisoprene rubbers during uniaxial

deformation by in situ synchrotron x-ray diffraction. Polymer 44 (19), 6003–6011. doi:10.1016/S0 032-3861(03)0 0548-2. In Honour of Ian Ward’s 75th

Birthday. http://www.sciencedirect.com/science/article/pii/S0 0323861030 05482

Tosaka, M., Murakami, S., Poompradub, S., Kohjiya, S., Ikeda, Y., Toki, S., Sics, I., Hsiao, B.S., 2004. Orientation and crystallization of natural rubber network

as revealed by WAXD using synchrotron radiation. Macromolecules 37 (9), 3299–3309. doi:10.1021/ma0355608.

Treloar, L.R.G., 1946. The elasticity of a network of long-chain molecules. III. Trans. Faraday Soc. 42, 83–94.

Treloar, L.R.G., 2005. The physics of rubber elasticity, third edition Oxford university press.

Verron, E., 2015. Questioning numerical integration methods for microsphere (and microplane) constitutive equations. Mech. Mater. 89, 216–228. doi:10.

1016/j.mechmat.2015.06.013.

Wang, M.C., Guth, E., 1952. Statistical theory of networks of non-gaussian ﬂexible chains. J. Chem. Phys. 20 (7), 1144–1157.

Wang, Y., Zhu, C., Pfattner, R., Yan, H., Jin, L., Chen, S., Molina-Lopez, F., Lissel, F., Liu, J., Rabiah, N.I., Chen, Z., Chung, J.W., Linder, C., Toney, M.F., Murmann, B.,

Bao, Z., 2017. A highly stretchable, transparent, and conductive polymer. Sci. Adv. 3 (3). doi:10.1126/sciadv.e1602076.

Wu, P.D., Van Der Giessen, E., 1993. On improved network models for rubber elasticity and their applications to orientation hardening in glassy polymers.

J. Mech. Phys Solids 41 (3), 427–456. https://doi.org/10.1016/0 022-5096(93)90 043-F.

Xu, J., Wang, S., Wang, G.-J. N., Zhu, C., Luo, S., Jin, L., Gu, X., Chen, S., Feig, V.R., To, J.W.F., Rondeau-Gagné, S., Park, J., Schroeder, B.C., Lu, C., Oh, J.Y.,

Wang, Y., Kim, Y.-H., Yan, H., Sinclair, R., Zhou, D., Xue, G., Murmann, B., Linder, C., Cai, W., Tok, J.B.-H., Chung, J.W., Bao, Z., 2017. Highly stretchable

polymer semiconductor ﬁlms through the nanoconﬁnement effect. Science 355 (6320), 59–64. doi:10.1126/science.aah4496.

Zhang, X., Krischok, A., Linder, C., 2016. A variational framework to model diffusion induced large plastic deformation and phase ﬁeld fracture during initial

two-phase lithiation of silicon electrodes. Comput. Methods Appl. Mech. Eng. 312, 51–77. doi:10.1016/j.cma.2016.05.007.

- Pressure Vessel [Design]Uploaded byVictor Rizal Filosofi
- 31968138 Pressure Vessel DesignUploaded byAnton Napitupulu
- PKM-Pengantar Material KompositUploaded byLisna Putri
- tugas korosi 1Uploaded byNina Asriana Andi
- WEEK 7- Defleksi Beban Merata n PUNTIRANUploaded byAria Ardiansyah
- Doc 5Uploaded bysupriyojawoto
- Assignmen Ujian Ujian Terhadap KonkritUploaded byAnne Chioozy
- Pembuatan SodaUploaded byCherryDevil
- WeldingUploaded byYuneo Nurcahya
- Hukum HookeUploaded byMohammad Syaiful Lutfi
- Mesin Dan PeralatanUploaded byArbik
- (Sifat Mekanik)Uploaded bygillangrizky
- Batang Tarik 2013Uploaded byDevie Hikariyue
- K3 Types FractureUploaded byNindy Aulia
- Composite 21Uploaded byFazry Nurokhman
- Len TuranUploaded bydudi nur abdillah
- Evaluation of a Leaf Spring Failure OkUploaded byAgustian Bakhtiar
- Chapter4DesignforFlexure [Compatibility Mode].pdfUploaded byRivanto Sariputra
- (6) Lrfd Stell Design-komponen Tarik & TekanUploaded byGilang Tiar Prakoso
- Kuliah 7 Penuangan & Sifat Material(1)Uploaded byAndre Satria Maulana
- catatanUploaded bydiandianadiana
- AppendixsdaUploaded bylkmcommunity
- 10174-13017-2-446920178325Uploaded byDorado Sb
- Abstrak.docxUploaded byChristian Natanael Simamora
- 03-Rudy_Djamaludin_08-10-07Uploaded byTofandy Yumahira
- COMPOSITE-21.pptxUploaded byPutra Nugroho
- 10.1016@j.msea.2018.08.096.en.idUploaded byYuniarto Hendrik Stiawan
- Gs Tgc Brantas 1234 p3Uploaded byfarel
- Analisis NonlinierUploaded byAhmad Prodigi
- Tranlate Stress-Strain Diagram PresentationUploaded byRichardus Gusti Ngurah Dima