Materials Letters 63 (2009) 877–880
Contents lists available at ScienceDirect
Materials Letters
journal homepage: www.elsevier.com/locate/matlet
A hyperviscoelastic constitutive model for polyurea
Chunyu Li ⁎, Jim Lua
Global Engineering and Materials Inc, Princeton, NJ 08540, USA
article info
Article history:
Received 18 November 2008 Accepted 13 January 2009 Available online 26 January 2009
Keywords:
Hyperviscoelasticity Constitutive modeling Polyurea Ratedependent material High strain rate
abstract
This letter presents a new constitutive model for polyurea by superposing the hyperelastic and viscoelastic behaviors of polyurea. The Ogden model is used for the hyperelastic part and its parameters are determined from curve ﬁtting of quasistatic test data. A nonlinear viscoelastic model is employed for describing the viscoelastic behavior and its relaxation time is obtained based on the test data of shear relaxation modulus. A
special form of Zapas kernel for the damping function is found to be very effective to capture the viscoelastic behavior of polyurea subjected to wide ranges of strain rate. Both the versatility and accuracy of the model are examined via virtual testing.
© 2009 Elsevier B.V. All rights reserved.
1. Introduction
Polyurea is a product from the chemical reaction between an isocyanate and an amine. It has been widely used in the coating industry, because of its extensive beneﬁts over existing epoxy adhesives and rubber linings in terms of impact, abrasion and corrosion resistance. With the development of polyurea spray coat ings technology, speciﬁcally formulated polyurea can be directly and efﬁciently sprayed on the surface of structural components to enhance the mechanical strength and durability of civil and military structures. Previous experimental studies [1] have revealed that polyurea exhibits elastic and nearly incompressible behavior to volumetric deformations and its stress–strain behavior depends on strain rate, temperature and pressure [2,3]. Different from the rubbery behavior under low strain rates, polyurea displays a distinct leathery behavior [4] at high strain rates. Polyurea can be used in a wide range of temperature (from − 50 °C to 150 °C) and has shown a high heat resistance. However, its shear modulus decreases signiﬁcantly with increasing temperature. The glass transition temperature T _{g} for polyurea is roughly − 50 °C [1]. Given the stiffening behavior of polyurea material with both increasing strain and strain rate, it has been used either as a protection coating on a metallic structure or an inserted layer between the outer facesheet and the foam core in a blasttolerant sandwich structure. Both material certiﬁcation and performance evaluation of polyurea coated structural components under hostile environment by tests will be prohibitively expensive and time consuming. Development of a high ﬁdelity constitutive model of polyurea is imperative to perform an optimal design of its coated components subjected to a combined dynamic and thermal loading. Currently, there are only two models have
⁎ Corresponding author. Email address: chunyulee@hotmail.com (C. Li).
0167577X/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
been developed speciﬁcally for polyurea although some similar models have been proposed for other hyperelastic or viscoelastic materials [5,6]. One is an experimentallybased linear viscoelastic constitutive model proposed by Amirkhizi et al. [2]. It incorporates the classical Williams– Landel–Ferry (WLF) timetemperature transformation and pressure sensitivity, in addition to a thermodynamic energy dissipation mechan ism. The model can reproduce experimental results for conﬁned polyurea tests but has limited capability in simulating the unconﬁned test data. The other is a more complex model proposed by Elsayed [7] which consists of an elastoplastic network acting in parallel with several viscoelastic networks. Quasiincompressible Ogdentype potentials were used and the number of model parameters is a function of the number of active Ogden terms and relaxation mechanisms, thus a procedure based on genetic algorithms was employed to calibrate model parameters based on existing experimental data. Although this model is relatively successful in reproducing experimental data, it seems to be too complicated for engineering practice. A constitutive model for polyurea should cover a wide range of strain rates from static to high speed impact and capture both hyperelastic and viscoelasticmaterial behavior. A rational approach based on a combination of hyperelasticity theory and viscoelasticity theory is needed in deriving the constitutive model. Also, for its practical applications, the model should be comparatively simple to be implemented in a general purpose ﬁnite element code while its parameters can be determined experimen tally from the standard tests. To meet the model requirements, this letter presents a new hyperviscoelastic constitutive model for polyurea.
2. Hyperelastic model for low strain rates
Given the compliant rubberlike behavior of polyurea from quasi static tests, a hyperelastic constitutive model is selected to characterize its rubbery stress–strain behavior at low strain rates. The constitutive
878
C. Li, J. Lua / Materials Letters 63 (2009) 877–880
Table 1
Fitting parameters
12 
3 
4 
5 
6 

μ _{i} (MPa) 40.9183 − 16.6532 − 34.7554 
27.7561 

α _{i} 
− 7.6257 3.9473 
3.4994 
− 3.9430 

G _{i} (MPa) 15.8879 28.3111 
38.1695 
44.0718 
66.9582 
86.1898 

ω _{i} (ms) 
301.61 
0.13146 
1.0556 ×10 ^{−} ^{3} 4.3933 ×10 ^{−} ^{5} 2.4659×10 ^{−} ^{6} 1.6715×10 ^{−} ^{7} 
law for an isotropic hyperelastic material is decomposed into volumetric
and deviatoric parts. The deviatoric part is deﬁned by an equation
relating the strain energy density of the material to the three invariants
of the strain tensor. By assuming the incompressible behavior of the
rubberlike material, the strain energy density W becomes a function of
the ﬁrst two invariants, i.e., W = W(I _{1} , I _{2} ), where I _{1} = tr(B), I _{2} = [I ^{2} − tr(B ^{2} )]/
1
2 and B is the left CauchyGreen deformation tensor. Here, we only
consider the uniaxial loading case given the availability of experimental
data. For a specimen under uniaxial loading, the principal stretches can
be expressed as λ _{1} = λ _{u} , λ _{2} = λ _{3} = λ _{u} ^{1}^{/}^{2} , where λ _{u} stands for
−
the stretch in
the loading direction and λ _{u} =1+ ε _{u} with ε _{u} denotes the engineering
strain in the loading direction. The left Cauchy–Green deformation
2
tensor B is then B = diag[λ , λ ^{1} , λ ^{1} ] (the right Cauchy–Green
u
−
u
−
u
deformation tensor C = B). There are several speciﬁc forms of the strain
energy density, such as by ArrudaBoyce [8], MooneyRivlin [9] and Neo
Hookean [10]. Here we select the Ogden model [11], given by
W =
n
∑
i = 1
2μ _{i}
α
2
i
λ ^{α} ^{i} + λ ^{α} ^{i} + λ ^{α} ^{i} −3 ^{} :
1
2
3
ð1Þ
h
Thus, the Cauchy (true) stress σ _{1}_{1} derived from the Ogden model is
n
σ _{1}_{1} ^{h} = −p ^{h} + λ _{u} σ ^{e}
11 ^{=} ^{∑}
i = 1
2μ _{i}
α i
λ ^{α} ^{i} −λ
u
−α _{i} =2
u
;
ð2Þ
e
where σ _{1}_{1} represents the engineering stress. The hydrostatic pressure
p ^{h} representing the volumetric part can be obtained by posing the
h
boundary condition associated with unconﬁned tests σ _{2}_{2}
h
= σ 33
= 0, i.e.,
n
p ^{h} = ∑
i = 1
2μ _{i}
α i
λ
−α _{i} =2 _{−}_{λ} α _{i} =4
u
u
:
ð3Þ
The ﬁnal form of the Cauchy stress in the loading direction is thus
written as
n
σ ^{h}
11 ^{=} ^{∑}
i = 1
2μ _{i}
α i
λ ^{α} ^{i} −2λ
u
−α _{i} =2 _{+} _{λ} α _{i} =4
u
u
:
ð4Þ
To determine the parameters μ _{i} and α _{i} , the experimental data from
Sarva et al. [4] for polyurea under very low strain rate (0.0016 s ^{−} ^{1} ) is
used for the nonlinear least square ﬁtting. The ﬁtted parameters are
Fig. 1. Fitting the Ogden hyperelastic model.
listed in Table 1. Fig. 1 shows that the curveﬁt results are in good
agreement with the experimental data.
3. Nonlinear viscoelastic model for high strain rates
Having established the rateindependent hyperelastic model for
polyurea under static loading conditions, we are now focusing on the
ratedependent characteristic of polyurea. A viscoelastic constitutive
law is a good choice for capturing its ratedependent behavior. For
viscoelastic materials, the stress state depends on the strain and strain
rate histories. Here, we assume the material is isotropic and subjected
to small strains under an isothermal condition. But different from a
linear viscoelastic model used by Amirkhizi et al. [2] an assumption of
nonlinear viscoelasticity is made. The general constitutive relation
ship can be expressed as [12]:
t
σ ^{v} = −p ^{v} δ + Fðt Þ R _{−}_{∞} fCðτ ÞgF ^{T} ðt Þ:
τ =
ð5Þ
where σ ^{v} is the Cauchy stress tensor, p ^{v} is the pressure in the viscoelastic
material, δ a Kronecker tensor, R a matrix functional that describes the
effect of strain history on stress, and F represents the deformation
gradient (F =diag[λ _{u} , λ _{u} ^{−} ^{1}^{/}^{2} , λ _{u} ^{1}^{/}^{2} ] for uniaxial loading). There have been
−
numerous approximations to represent the functional for viscoelastic
solids and ﬂuids [12,13]. Since the KBKZ integral form [14] and its
variants are wellaccepted, we adopt the form shown below
t R _{−}_{∞} fCðτ Þg = ∫ τ = 
−∞ ^{/} ^{I} ^{0} t 
_{1} ; I 
: 
ð6Þ 

where ϕ(I _{1} ',I _{2} ') is a damping function with I _{1} ',I _{2} ' as the ﬁrst two 

invariants of strain tensor C, and the strain rate is deﬁned by : E = 1 2
: F ^{T} F + F ^{T} F : ; 
ð7Þ 
and m(t) is a relaxation function.
The relaxation function is generally represented by a series of simple
exponentials, or socalled Prony series, with its individually deﬁned
relaxation time. In the literature, different number of exponential terms
(e.g.1, 2 or 5), thus different relaxation times were selected driven by the
required ﬁdelity of the curveﬁt model [15–17]. While the use of
combination of relaxation times and other undetermined parameters
can provide a good ﬁt of the experimental data of stress–strain curves, it
has no physical foundation and rational for a speciﬁc combination. In
fact, the relaxation times depend on the microstructure of a material.
Motion within the microstructure determines the relaxation time of the
material and its activation depends on temperatures and corresponding
loading rates. For example, at higher temperatures, the molecular
relaxation processes are more easily and frequently activated due to the
Fig. 2. Fitting of the master curve to ﬁnd relaxation times.
C. Li, J. Lua / Materials Letters 63 (2009) 877–880
879
Fig. 3. Comparison of experimental data and model predictions.
̇
̇
Substitution of the strain rate E
11 = λ u λ
_{u} and the pressure p ^{v} into
Eq. (12) yields the stress for the viscoelastic polyurea in the loading
direction:
σ ^{v}
11 ^{=}
1
2
^{λ}
−1
u
t −2
∫
0 ^{λ}
u
A + A I _{2} −3 i mðt −τ Þ λ _{u} dτ
h
1
2
0
:
+ λ
2
t
u ^{∫}
0 ^{λ} u
A + A I _{2} −3 i mðt −τ Þ λ _{u} dτ :
h
1
2
0
:
ð14Þ
Adding up Eqs. (4) and (14) together, we have the hyper
h
viscoelastic stress for polyurea σ _{1}_{1} = σ _{1}_{1}
v
+ σ 11
. In this total stress
deformation formulation, only two parameters A _{1} and A _{2} are
undetermined and they can be obtained by curveﬁtting of the
experimental data associated with a uniaxial high strain rate testing.
4. Results and discussions
higher thermal energy. In this study, therefore, we propose to determine
the relaxation function based on the shear modulus relaxation ﬁrst
before computing other curveﬁtting parameters.
Knauss [1] conducted relaxation tests on polyurea at various
temperatures. The experimental data is shown in Fig. 2, in which the
data has been shifted according to the timetemperature shift
principle. The applicability of the shift process is assured by the
rather smooth master relaxation curve. The standard WLF shift
function has been given as [1]
aðT Þ = 10 ^{A} ^{T}
ð
^{−}^{T} ref
Þ= B + T −T _{r}_{e}_{f}
ð
^{Þ} ;
ð8Þ
with A = − 8.86, B = − 171 °C and T _{r}_{e}_{f} =0 °C. The master curve can be
approximated by a Prony series with a number of terms:
n 

G _{r}_{e}_{f} ðt Þ = G _{∞} + ∑ G _{i} e ^{−}^{t} ^{=}^{ω} ^{i} : 
ð9Þ 
i = 1 
Fig. 2 shows our nonlinear regression to the master curve using a
6term Prony series. The ﬁtting parameters are summarized in Table 1.
The value of the goodness of ﬁt R ^{2} , which is computed from the sum of
the squares of the distances of the points from the bestﬁt curve
determined by the nonlinear regression and then normalized to the
sum of the square of the distances of the points from a horizontal line
through the mean of all G _{r}_{e}_{f} values, is 0.9998. This is much close to 1.0
and indicates that the model ﬁts the data better.
Because G _{∞} is not related to time, the relaxation function m(t − τ) in
Eq. (6) can then be assumed as
6
mðt −τ Þ = ∑
i = 1
_{G} i _{e} − ðt −τ Þ=ω _{i} _{;}
_{ð}_{1}_{0}_{Þ}
for isothermal deformations. After some strenuous trials, the damping
function ϕ in Eq. (6) is taken as a special form of Zapas kernel [18],
/ðτ Þ = A _{1} = ð A + I −3Þ;
1
ð11Þ
where I = A _{2} I _{1} ' + (1 − A _{2} )I _{2} ' and A _{1} and A _{2} are undetermined parameters.
Substituting Eq. (11) into Eq. (6) and then Eq. (5) and neglecting the
effect of deformation history prior to loading on the stress, we obtain
σ ^{v} = −p ^{v} δ + F
n
t
∫
0
h
A + A
1
2
I _{2} −3 i mðt −τ Þ E ðτ Þdτ o
0
:
F ^{T} :
ð12Þ
Considering the free boundary condition of the transverse stress
and
:
E 22 = λ 2
:
λ _{2} = −
1
2
^{λ}
−2
u
:
̇
λ _{u} (the stretching rate λ _{u} is equal to the
engineering strain rate ε̇ _{1}_{1} ), the pressure p ^{v} can be obtained as
p ^{v} = −
1
2
^{λ}
−1
u
t −2
∫
0
^{λ}
u
A + A I _{2} −3 i mðt −τ Þ λ _{u} dτ :
h
1
2
0
:
ð13Þ
In this study, the data for the unconﬁned uniaxial compression test at strain rate of 2250 s ^{−} ^{1} from Sarva et al. [4] is used to determine the parameters A _{1} and A _{2} . The parameters determined from the ﬁtting are A _{1} = 0.1435 and A _{2} = 0.1098. Fig. 3 displays the comparison between the test data and our prediction from the present model. The symbols triangle, square, pentagon, hexagon and circle represent, respectively, the test data under the strain rate 0.0016/s, 80/s, 800/s, 2250/s and 6500/s. The color lines represent predictions of our present model with gray, cyan, wine, blue and red stand for the result under the strain rate 0.0016/s, 80/s, 800/s, 2250/s and 6500/s, respectively. It can be seen that the prediction of the present model is in a very good agreement with the experimental data for a large range of strain rates. In addition, the ﬁdelity of the present model is also evaluated via its comparison with prediction from other existing models, as shown in Fig. 4. Clearly, higher accuracy has been achieved from the present model in simulating the polyurea constitutive behavior than the model given by Amirkhizi et al. [2] As shown in Fig. 4, our model captures the early response more accurately than the model developed by Elsayed [7]. Given the simplicity of the present model, it can be used as a strongly competitive alternative to Elsayed's model.
5. Conclusions
This letter presents a new constitutive model for polyurea by
including both its hyperelastic and viscoelastic behavior. The Ogden
model is used for the hyperelastic part and its parameters are curve
ﬁtted using the quasistatic test data. A nonlinear viscoelastic model is
proposed to characterize its viscoelastic part and the relaxation time is
determined based on the test data of the shear relaxation modulus. A
special form of Zapas kernel for the damping function is found to be
very effective in prediction of the viscoelastic behavior of polyurea
subjected to a wide range of strain rates. It is expected that this new
constitutive model can be implemented in general purpose ﬁnite
element code as a userdeﬁned material model to assess and design of
a protective structure subjected to extreme dynamic loading.
Fig. 4. Comparison of the present model with existing models.
880
C. Li, J. Lua / Materials Letters 63 (2009) 877–880
Acknowledgement
This work is supported by the Ofﬁce of Naval Research under
Contract N0001408C0614 with Dr. Roshdy Barsoum as the Program
Manager.
References
[1] W.G. Knauss, Final report to the Ofﬁce of Naval Research. (ONR Grant No. N00014 0310539), CIT, Pasadena, CA. [2] Amirkhizi AV, Isaacs J, Mcgee J, NematNasser S. Philos Mag 2006;86:5847–66. [3] Yi J, Boyce MC, Lee GF, Balizer E. Polymer 2006;47:319–29. [4] Sarva SS, Deschanel S, Boyce MC, Chen W. Polymer 2007;48:2208–13. [5] Qi HJ, Boyce MC. Mech Mater 2005;37:817–39.
[6] Reese S, Govindjee S. Int J Solids Struct 1998;35:3455–82. [7] T.M. Elsayed, PhD dissertation, California Institute of Technology (2008). [8] Arruda E, Boyce MC. J Mech Phys Solids 1992;41:389. [9] Mooney M. J Appl Phys 1940;11(582). [10] Treloar LRG. Proc Phys Soc 1948;60:135.
[11] Ogden RW. Proc R Soc Lond 1972;A326:565–84. [12] Lockett FJ. Nonlinear viscoelastic solids. New York: Academic Press; 1972.
[13]
Carreau PJ, De Kee DCR, Chhabra RP. Rheology of polymeric systems: principles and applications. New York: Hanser Publishers; 1997.
[14] Tanner RI. J Rheology 1988;32:673. [15] Osaki K, Ohta S, Fukuda M, Kulata M. J Polym Sci 1976;14(1701). [16] Wagner MH. Rheol Acta 1979;18(33). [17] Yang LM, Shim VPW, Lim CT. Int J Impact Eng 2000;24:545. [18] Zapas LJ. J Res Natl Bur Stand 1966;70A(525).