Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Full Paper

www.mts-journal.de

Simplification of Hyperelastic Constitutive Model and


Finite Element Analysis of Thermoplastic Polyurethane
Elastomers
Yingzhu Wang, Weiang Luo,* Junwen Huang, Chaohua Peng, Hongchao Wang,
Conghui Yuan, Guorong Chen, Birong Zeng, and Lizong Dai*

equipment, shape memory materials, and


In this work, the simplified hyperelastic constitutive model and finite ele- so on. Wang et  al.[4] selected TPU as the
ment analysis of thermoplastic polyurethane elastomers (TPUs) under base plate of strain sensor of wearable
uniaxial tension are studied. Based on a series of classical strain energy equipment for its excellent flexibility and
large specific surface area. At the same
density function of hyperelastic materials, a “one-step” transformation time, high tensile property, machinability,
algorithm from standard test data to the simplest constitutive model is and environmental adaptability of TPU are
proposed and a three-parameter simplified constitutive model is obtained. required for wearable equipment. Chien
The simplified model has a high accuracy on the experimentally determined et  al.[5] prepared the shape memory poly­
load–displacement curve with fitting errors of 0.95%, 0.81%, and 0.98% to urethane elastomer with excellent biocom­
patibility. So the prediction of its tensile
three TPUs comprising different hard segment contents of 15.19, 22.54, and
state by means of engineering simulation
38.51 wt%, respectively. By combining the simplified hyperelastic constitutive is very conducive to guide the practical
model with the finite element analysis, the mechanical behavior of TPUs in engineering application.
uniaxial tensile state is predicted accurately. The simplification algorithm and Generally, the molecular skeleton struc­
simulation method may be applicable to other hyperelastic materials with ture of TPU includes alternating soft
more complex shapes and working conditions and have certain practical engi- and hard segments.[6] The hard segment
has a high glass transition temperature
neering guiding significance for the subsequent material selection, design, (Tg) or a high melting temperature (Tm),
and production. which endows TPU with excellent per­
formance such as high rigidity, hardness,
and strength. In contrast, the soft segment
1. Introduction has a lower Tg that determines the low-temperature behavior
of TPU, such as flexibility and elasticity.[7,8] Therefore, TPU
Thermoplastic polyurethane elastomer (TPU) has become shows excellent processability and hyperelasticity. Moreover,
one of the most widely used materials because of its excellent the micro-phase separation caused by the polarity difference
mechanical, chemical, and optical properties, which can be between hard and soft regions exhibits a remarkable combi­
used in impact resistance composites, elastomer foam, porous nation of elasticity and inelasticity, which can be adjusted by
materials, adhesive coatings, biomedical devices, etc.[1,2] TPU varying the ingredient or content of soft and hard segments.[1,9]
is also regarded as an important raw material in electrospin­ Therefore, the soft and hard segments of TPU have great influ­
ning, additive manufacturing, and 3D printing.[3] Therefore, ence on its mechanical properties.
researchers have carried out in-depth study on all aspects of the The mechanical behavior of materials can be described by
physical and chemical characteristics of TPUs under different the constitutive model. The stress–strain curve of the elastomer
working conditions. has very significant nonlinear behavior in a large strain range
The tensile mechanical properties of TPU are very impor­ and does not meet the Hooke’s law, making it more compli­
tant in practical applications, such as strain sensor of wearable cated to obtain a complete mechanical behavior model.[10] Over
the years, researchers have proposed several theoretical models
Y. Z. Wang, Dr. W. A. Luo, J. W. Huang, C. H. Peng, H. C. Wang,
for hyperelastic materials. Typically, van der Waals model[11]
Prof. C. H. Yuan, G. R. Chen, Prof. B. R. Zeng, Prof. L. Z. Dai and Arruda–Boyce model[12] are based on thermodynamic
Fujian Provincial Key Laboratory of Fire Retardant Materials statistics, while Mooney–Rivlin models,[13,14] Gent models,[15]
College of Materials, Xiamen University and Ogden models[16] accord to continuum mechanics theory.
Xiamen 361005, China Recently, other theoretical models are proposed,[17] but the con­
E-mail: luoweiang@xmu.edu.cn; lzdai@xmu.edu.cn
stitutive models used commonly are still based on continuum
The ORCID identification number(s) for the author(s) of this article
can be found under https://doi.org/10.1002/mats.202000009.
mechanics theory, and many constitutive models of hyper­
elastic materials are derived from the generalized Mooney–
DOI: 10.1002/mats.202000009 Rivlin model for its universality.[18–21] These theoretical models

Macromol. Theory Simul. 2020, 2000009 2000009  (1 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

are reflected in different strain energy density function, which parameters quickly and accurately. Finally, the simplified con­
are usually expressed by a series of material constants Ci. These stitutive model is applied to FEM to predict the mechanical
models can explain the large deformation behavior of materials behavior of materials under real working conditions. This work
from a macro perspective.[22] can provide useful information for reducing calculation proce­
It is generally known that the constitutive model parameters dure and predicting materials behavior, which has important
vary with the material composition and load situation. The guiding significance for practical engineering applications.
mechanical behavior of materials under certain conditions can
be predicted by importing the constitutive model parameters
into the finite element model (FEM).[23–25] Bouaziz et  al.[26,27] 2. Experimental Section
used a thermo-visco-hyperelastic  model for shape memory
poly­mers to investigate the effect of the presence of nanotubes 2.1. Materials
on the mechanical parameters. They also proposed a constitu­
tive mechanical model to describe the thermo-mechanical cycle Diphenylmethane diisocyanate (MDI) and 1,4-butylene glycol
of a kind of semicrystalline shape memory polyurethane. Then (BDO) were obtained from J&K Scientific Ltd. Polybutylene adi­
the proposed thermo-visco-hyperelastic  model and thermo- pate 1,4-butylene glycol diol (PBA, molecular weight: 2028 g mol−1)
mechanical model were implemented in the finite element was obtained from industrial production (Zhejiang Huafeng New
software COMSOL Multiphysics. Their study showed a good Material Co., Ltd). All chemicals were used as received.
prediction of the mechanical behavior during shape memory
tests. Kim et al.[28] calibrated hyperelastic and hyper foam consti­
tutive models for rigid polyurethane foam under compression; 2.2. Specimen Preparation
the indentation event was simulated via finite element analysis
with the calibrated models. In addition, the authors expounded The specimens were prepared by bulk thermal polymerization
the shortcomings of their own research: the hyperelastic and with no catalyst. First, under the protection of nitrogen, poly­
hyper foam constitutive models had medium prediction ability urethane prepolymer was synthesized with the vacuum-
with the fitting errors of 13.6% and 11.5%, respectively. There­ dehydrated PBA and MDI by mechanical stirring for 2 h. After
fore, a prediction ability recalibration of the model is needed the prepolymer was vacuum defoamed, diol BDO was added
to meet a more complex load situation. Fan et al.[29] calculated under stirring, and then the resulting mixture was poured into
the tensile stress–strain relationship in a typical four-step a mold, which was placed at 120 °C in a vacuum oven for 24 h to
thermal shape memory process based on a neo-Hookean hyper- complete curing. After furnace cooling, the standard test speci­
viscoelastic constitutive equation, which could be used for cali­ mens of TPUs were obtained after partial stress relieving. The
brating the parameters of an epoxy shape memory polymer and specimens for uniaxial tension test were placed at room tem­
verifying the numerical method with FEM to predict the shape perature for 1 month after the internal stress was eliminated.
memory process under different conditions. Xiang et  al.[30] The molar ratio of hydroxyl group to isocyanate group was 1:1.
developed a viscoelastic constitutive model based physics; the A series of TPUs with different hard segment (HS) contents
model could be used for describing mechanical behavior and were prepared by fixing MDI at 1.0  mol and the total amount
predicting the mechanical response of soft materials. of BDO and polyol at 1.0  mol, while changing the molar ratio
Through the previous researches, the combination of consti­ between BDO to polyol from 1:3 (15.19 wt% HS, marked as
tutive model exploration and simulation has a good engineering TPU-1), 1:1 (22.54 wt% HS, marked as TPU-2) to 3:1 (38.51 wt%
guidance value for practical material application. However, the HS, marked as TPU-3).
constitutive models of some materials are too complex and a
longer engineering simulation time is needed for complex geo­
metric models. Furthermore, the universality of the models 2.3. The Mechanical Properties of TPU
remains unconfirmed. Therefore, on the premise of ensuring
the accuracy of engineering simulation, it is significant to Due to the special structure of hard and soft segments and
simplify the material constitutive model as much as possible, the multiple hydrogen bond structure between the molecular
so as to improve the calculation efficiency. At the same time, chains, mechanical properties of TPU are also very special
the simplified method should be applicable to the constitutive compared with ordinary rubber, which mainly represents that
models construction of various hyperelastic materials, avoiding TPU has high strength, wide range of hardness, better abrasion
the trouble caused by the variation of material composition and resistance, toughness, and flexibility. One of the raw materials
speeding up the process of engineering simulation. of TPU—MDI— has the rigid structure of benzene ring, which,
In this article, the mechanical behavior of TPUs with dif­ together with BDO as the hard segment, improves the mechan­
ferent hard segment contents under uniaxial tension is taken ical properties of TPU such as rigidity, strength, and hardness.
as examples. First, the influence of hard segment content on While, the other raw material polyester polyol PBA exists as
the mechanical properties of TPU is investigated by uniaxial a soft segment in the TPU, which determines the toughness
tension and scanning electron microscope (SEM). Afterwards, and elasticity of TPU. When the isocyanate group content
based on the fitting results of several classical constitutive (MDI) is fixed, TPU with different mechanical properties can
models of hyperelastic materials, a simplification algorithm of be obtained by adjusting the molar ratio of chain extender BDO
constitutive model is constructed, which can match the sim­ and PBA, including from soft elastomer to hard plastic, which
plest constitutive model and obtain the corresponding model can meet different practical requirements.

Macromol. Theory Simul. 2020, 2000009 2000009  (2 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 1.  Schematic diagrams of the specimens (unit: mm).

2.4. Microstructure and Uniaxial Tensile Measurements respectively, and λ1, λ2, and λ3 are the stretch ratios in the three
principal directions, respectively.
As an important parameter of TPU, the change of hard seg­ Generally, hyperelastic materials are considered to be incom­
ment content affects the mechanical properties of TPU. In this pressible, that is, the volume does not change after deforma­
article, the mechanical properties of TPU were characterized by tion. In this case, the deformation tensor C is identical to the
the microstructure and uniaxial tensile test. In order to obtain loading direction, and the deformation gradient F can be easily
the variation of material toughness, the morphology of TPU expressed by selecting along the global axis as follows
brittle fracture in liquid nitrogen was tested by SEM (Hitachi
SU-70, 5 kV), while the properties of hyperelasticity were char­  λ1 0 0   λ12 0 0 
acterized by uniaxial tensile test. F= 0 λ2 0  ⇒ C = FT F =  0 λ22 0  (5)
Uniaxial tension experiments are widely used as the data    
 0 0 λ3   0 0 λ32 
resources of modeling.[31] The uniaxial tensile test was carried
out with a dumbbell specimen using an electronic universal
testing machine (AGS-X 5 KN) according to ISO 527-2-2012 ∂x
F= (6)
(International Standard, Plastics-Determination of tensile prop­ ∂X
erties-Part 2: Test conditions for molding and extrusion plas­
tics). The specific shape and size of the specimens are shown J = det F (7)
in Figure  1. In order to meet the actual engineering applica­
tion scenario of TPU and to avoid the influence of tempera­ where x and X represent the coordinates of the same point
ture on its mechanical properties, the tensile test was carried before and after deformation; J is the volume ratio of TPU
out at 25 °C and 50% relative humidity. The tensile speed was before and after deformation. For a uniaxial tensile test, if direc­
10  mm min−1 and the corresponding strain rate was around tion 1 (axis direction) is the tensile direction and the material is
3 × 10−3 s−1, which belongs to the category of quasi-static incompressible, the following equation relationship is derived
loading. The force–displacement curves of the specimens can
be obtained by uniaxial tensile test. J = λ1λ2λ3 = 1(8)

I3 = ( λ1λ2λ3 )2 = 1(9)
3. Constitutive Model
λ1 = λ (10)
3.1. Strain Energy Density Function
λ22 = λ32 = λ −1 (11)
The constitutive relationship of hyperelastic materials is very
complex, which is different from the physical meaning of
λ = 1 + ε (12)
simple linear elastic and elastic-plastic materials. In general,
the hyperelastic constitutive model is represented by the strain
I1 = (1 + ε )2 + 2(1 + ε )−1(13)
energy density function W as follows[32,33]

W = W (I1 ,I 2 ,I3 )(1) I 2 = 2(1 + ε ) + (1 + ε )−2(14)

I1 = λ12 + λ22 + λ32 (2) σt


σ1 = σ = (15)
λ
I 2 = ( λ1λ2 )2 + ( λ2λ3 )2 + ( λ3λ2 )2 (3) σ 2 = σ 3 = 0 (16)

I3 = ( λ1λ2λ3 )2(4) σ t2 = σ t3 = 0(17)


where I1, I2, and I3 are the first, second, and third basic where σ, ε, and σt are nominal stress, nominal strain, and true
invariants of the right Cauchy–Green deformation tensor C, stress, respectively. For pure uniform strain, the relationship

Macromol. Theory Simul. 2020, 2000009 2000009  (3 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

between stress, strain, and strain energy density function is where Ci0 are material constants; Di is a material parameter,
derived as follows[16] representing the compressibility of the material. The parameter
N can generally take values up to 6. The neo-Hookean model
t1 − t2  ∂W ∂W  and Yeoh model can be viewed as a special case of the reduced
= 2 + λ32 (18)
λ12 − λ22  ∂I1 ∂I 2  polynomial with N = 1 and N = 3, respectively.[36]
Ogden proposed that the strain energy density function can
By substituting Equations  (9–17) into Equation  (18), the be expressed by the principal stretch ratios λi instead of the
constitutive relation between hyperelastic material and strain strain tensor invariants, as shown in Equation (22)[16]
energy density function W (Equation (19)) can be obtained
N
2µi α i N
1
W =∑ ( λ1 + λ2α i + λ3α i − 3) + ∑ ( J − 1)2i (22)
 ∂W ∂W  αi2
i =1 Di
σ = 2[(1 + ε ) − (1 + ε )−2 ]  + (1 + ε )−1 (19) i =1

 ∂I1 ∂I 2 
where µi (infinitesimal shear modulus) and αi (stiffening para­
Equation  (19) shows that the most suitable constitutive meter) are material parameters; Di is a material parameter, rep­
model for hyperelastic materials can be obtained by matching resenting the compressibility of the material. The parameter
different strain energy density functions, which can effectively N can generally take values up to 6, but the complexity of the
predict the mechanical behavior of hyperelastic materials. model will increase obviously with the increase of N value. In
essence, the Ogden model is still the deformation of the con­
stitutive equation of the generalized Mooney–Rivlin model. If
3.2. Classical Constitutive Models N = 2, α1 = 1, α2 = –2, the first-order generalized Mooney–Rivlin
model is obtained. If N = 1, α1 = 2, Ogden model degenerates to
At present, based on a large number of experimental data and the neo-Hookean model.
corresponding mathematical expressions, researchers have In the constitutive model based on thermodynamics statis­
constructed many classical constitutive models of hyperelastic tics, two most typical models are Arruda–Boyce model and van
materials, which can be divided into two types: one is based on der Waals model. Arruda–Boyce model is also known as the
the theory of continuum mechanics, and the other is based on eight-chain model since it starts with a representative volume
the theory of thermodynamics statistics. element, in which eight springs radiate from the center of
For the theory of continuum mechanics, the strain energy the cube to its corners.[12] The equation of Arruda–Boyce model
density function is represented by one or more of the three is expressed as
invariants I1, I2, and I3. Among them, the most classical strain
energy density function is the polynomial form, which is N
Ci 1  J2 − 1 
expressed as follows[14] W = µ∑ (I1i − 3i ) +  − ln J  (23)
i =1 λm
2i − 2
D 2 
N N
1
W= ∑ C (I ij 1 − 3)i (I 2 − 3) j + ∑
Di
( J − 1)2i (20) 1 1
C1 = , C 2 = , C3 =
11
, C4 =
19
, C5 =
519

i + j =1 i =1 2 20 1050 7000 673750

where Cij are material constants; Di is a material parameter, where the values of C1-C5 are calculated by thermodynamic sta­
representing the compressibility of the material. According tistical method, and have corresponding physical significance;
to Equation  (8), for incompressible materials, J  = 1. The para­ λm, µ, and D are the locking stretch, shear modulus, and bulk
meter N can generally take values up to 6; however, values of N modulus, respectively. As can be observed in Equation (23), the
greater than 2 are rarely used when considering both the first Arruda–Boyce model is essentially a higher-order polynomial
and second invariants.[34] The above equation is also known as form of the first invariant I1.
the generalized Mooney–Rivlin model, and many other classical Van der Waals model is also known as Kilian model, in
constitutive models of hyperelastic materials have evolved from which the interaction between molecules of rubber materials
it, including reduced polynomial model and Ogden model. is regarded as van der Waals force between gases.[11] The name
The equation of reduced polynomial model can be obtained “van der Waals” is an analogy in the thermodynamic interpreta­
by removing the term that depends on the second invariant tion of the equation of state of rubber materials and gas. The
from the generalized Mooney–Rivlin model, as shown in equation of van der Waals model is expressed as
Equation (21).[32] According to the relevant literature, when the
strain is more than 20%, the sensitivity of strain energy den­  2 I − 3  2  1  J 2 − 1
3

sity function to the change of the second invariant is gener­ W = µ −( λm2 − 3)[ln(1 − η ) + η ] − a  + − ln J
3  2   D  2 
ally much smaller than that of the first.[35] Therefore, in order  
to reduce the unnecessary calculation process and calibration (24)
parameter time, the reduced polynomial model is widely used
to describe the medium and large deformation behavior of I = (1 − β )I1 + βI 2 (25)
hyperelastic materials
N N
1 I −3
W = ∑ Ci 0 (I1 − 3)i + ∑ ( J − 1)2i (21) η= (26)
i =1 i =1 Di λm2 − 3

Macromol. Theory Simul. 2020, 2000009 2000009  (4 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 2.  Model development process.

where λm, µ, and D are the locking stretch, shear modulus, 3.4. The Finite Element Model
and bulk modulus, respectively; a is the global interaction para­
meter; and β represents a linear mixture parameter combining The uniaxial tension test was modeled and simulated by the
both invariants I1 and I2 into I. commercial finite element software (Abaqus). In order to
improve the convergence and efficiency of the simulation, sym­
metric model was used in the FEM. According to the loading
3.3. Constitutive Model Simplification Process and geometry characteristics of the TPU specimens, the sym­
metric FEM model adopted was about the symmetry of the XY
Usually, construction of a constitutive model of hyperelastic plane (Figure  3). As shown in Figure  4, the shape and size of
material should meet the following four requirements[37] the FEM were consistent with the experimental dumbbell spec­
imen. In order to improve the calculation accuracy, the element
1. The constitutive model should accurately reproduce the “S”
grid was refined in the gauge length region. The hyperelastic
type response of elastomer;
material model used here is an incompressible model, so the
2. The material parameters of the constitutive model should be
element type was set to C3D8H,[34] and the total number of ele­
as few as possible;
ments was 1720. The properties of the specimens were defined
3. The parameters in the constitutive model should be obtained
by the subsequent simplified hyperelastic model. The loading
by simple test methods;
4. The strain energy density function of the constitutive
model should be relatively simple and easy to realize
numerically.
In this article, the constitutive model simplification process
of TPU is shown in Figure 2.
Step 1: Preparation of the standard specimens. The force–
displacement curve of the selected material was obtained
through standardized mechanical experiments. Step 2:
Matching the constitutive model. Import the experimental data
from step 1 into MATLAB, then the optimized fitting model can
be achieved by fitting various classical hyperelastic models. Step
3: Simplifying model. The optimal model obtained in step 2 is
further simplified through a simplification algorithm program.
Step 4: Model application. The simplified model was applied to
the finite element analysis to obtain the mechanical response
under the setting conditions. All simulations were performed
on the desktop computer with Intel Xeon CPU E5-1660 v4
(3.20 GHz) and 64 GB memory. Figure 3.  The loading condition and boundary condition of the FEM.

Macromol. Theory Simul. 2020, 2000009 2000009  (5 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 4.  The layout and size of elements in FEM.

Figure 5.  a) The force–displacement curves and b) the nominal stress–strain curves for TPU-1, TPU-2, and TPU-3.

condition and boundary condition of the model are also shown where σ and ε are the nominal tensile stress and strain of the
in Figure  3, and the displacement of the reference point was specimen, respectively; F and L are the force and displacement
controlled. data obtained from uniaxial tensile test, respectively; L0 and A0
are the gauge length and cross-sectional area of the specimen
in gauge length part, respectively.
4. Results and Discussion Figure  5b shows the nominal stress–strain curves for
TPU-1, TPU-2, and TPU-3 under uniaxial tensile test. The
4.1. Effect of Hard Segment Content on the Mechanical stress increases correspondingly with the increase of hard
Properties of TPUs segment content under the same deformation, and all curves
show nonlinear “S” type response characteristics of hyper­
For the quasi-static mechanics study, the force–displacement elastic materials. With the increase of hard segment con­
curves of TPUs determined from the uniaxial tensile test tent, the deformation resistance, residual stress, and loss of
under large deformation condition were obtained (Figure  5a). materials also increase correspondingly, which leads to the
In order to analyze the constitutive model of TPUs, the force– increase of hardness and rigidity.[1] In addition, observed
displacement curves need to be transformed into the nominal from the cross-sectional diagrams of three TPUs (Figure  6),
stress–strain curves by the following equations with the increase of hard segment content, the cross section
becomes more and more smooth, indicating the decrease of
F toughness. In conclusion, the change of TPU hard segment
σ= (27)
A0 content has a significant effect on its mechanical properties,
L which means that its constitutive relationship also changes
ε= (28) accordingly.
L0

Figure 6.  SEM images of TPU fracture surfaces.

Macromol. Theory Simul. 2020, 2000009 2000009  (6 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 7.  Fitting results of experimental data and classical models: a) reduced polynomial model, N = 1–3; b) reduced polynomial model, N = 4–6;
c) Ogden model, N = 1–3; d) generalized Mooney–Rivlin model (N = 1, 2); e) Arruda–Boyce model; f) van der Waals model.

4.2. Comparison of Classical Constitutive Models In order to evaluate the fitting degree between the consti­
tutive model and the experimental data, the fitting error was
The force–displacement data of TPU-1 obtained from experi­ quantified using the following equation[28,38,39]
ment were transformed into the nominal stress–strain data, and
εm
fitted with various classical hyperelastic models in MATLAB
(Figure  7), including generalized Mooney–Rivlin (Equa­

Fitting error ( % ) = 0
σ sim − σ exp d ε
× 100 (29)
εm
tion  (20)), reduced polynomial model (Equation  (21)), Ogden ∫0 σ exp d ε
model (Equation  (22)), Arruda–Boyce model (Equation  (23)),
and van der Waals model (Equation (24)). The material parame­ where σsim and σexp are simulated and experimental nominal
ters involved in the constitutive model can be obtained via least- stress, respectively; ε is the uniaxial nominal strain, and εm is
square fitting of the model to the nominal stress–strain curves. the maximum nominal strain.
The optimal model that is suitable for TPU-1 was preliminarily As can be observed in Equation  (21), the reduced poly­
explored. nomial model is simplified by the generalized Mooney–Rivlin

Macromol. Theory Simul. 2020, 2000009 2000009  (7 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 8.  Parameters simplification flowchart (E: fitting error).

model. Compared with the generalized Mooney–Rivlin model, 4.3. Curve Fitting and Simplification
its strain energy density function is simpler, but according to
the fitting results (Figure 7a,b), the coincidence degree between The simplification algorithm flowchart is shown in Figure 8. The
the experimental curve and the model curve is obviously lower. five parameters of the generalized Mooney–Rivlin model (N = 2)
Even when N  = 6, the reduced polynomial model still cannot are split one by one to get five subexpressions, each subexpres­
obtain a good fit with the experimental curve. Therefore, the sion has one parameter. The algorithm starts from a single-
simplification of constitutive model is not simply to reduce one parameter model to fit the experimental data. When the fitting
item, but to simplify according to practical application. degree meets the requirement (fitting error ≤ 1%), it indicates
As shown in Figure  7, the hyperelastic constitutive models of that the model with this parameter is a simplified model, and the
Ogden model (N = 3) (Figure 7c) and generalized Mooney–Rivlin program derives this model expression. If fitting error > 1%, two-
model (N = 2) (Figure 7d) have the best fitting degree (fitting error parameter model is needed to fit the experimental data, and so
≤ 1%). Under the same fitting degree, Ogden model (N = 3) con­ on, until the simplest model with fitting error ≤ 1% is found. If
tains six material parameters for uniaxial tension, while the gen­ the simplified model is not found after the algorithm operation,
eralized Mooney–Rivlin model (N = 2) only contains five material it indicates that the generalized Mooney–Rivlin model (N  = 2)
parameters. Obviously, the generalized Mooney–Rivlin model cannot be further simplified. The algorithm takes the advan­
(N  = 2) has fewer parameters. Moreover, many classical hypere­ tage of the universality of the generalized Mooney–Rivlin model
lastic models are derived from this model, including Ogden model. and can transform the experimental data into the simplest con­
It indicated that the generalized Mooney–Rivlin model has strong stitutive model on the premise of satisfying the accuracy of the
applicability. In the following work, we will further simplify the model. However, the simplification algorithm in this work also
generalized Mooney–Rivlin model (N = 2), so as to explore the sim­ has its deficiency that complexity of the simplification process
plest constitutive model suitable for the prepared TPU materials. will increase rapidly with the number of items in the model.

Macromol. Theory Simul. 2020, 2000009 2000009  (8 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 9.  Fitting results of the simplified model and experimental data of a) TPU-1, b) TPU-2 and c) TPU-3.

Through the simplification algorithm, three dif­ It is noteworthy that simplified model is very similar to the
ferent force–displacement data of TPU-1, TPU-2, and trinomial model (Equation  (31)) proposed by Isihara et  al.,[40]
TPU-3 obtained from the experiment were imported into except the strain tensor invariants in the quadratic term.
MATLAB, and the simplified strain energy density function By transforming trinomial model into constitutive model
(Equation (30), hereinafter referred to as “simplified model”) (Equation  (19)) and fitting with experimental data. As shown
was successfully obtained. From the simplified results, the in Figure  10, the fitting degree of trinomial model is not very
corresponding simplified model of the three TPUs is the high (all fitting errors >4.5%), and the fitting error of TPU-2
same. Even though the mechanical properties of TPU change reaches 5.67%, which is obviously larger than those of the sim­
with the different hard segment content, the simplification plified model. It can be inferred that the main reason for this
algorithm can also match their most simplified constitutive difference lies in the different mathematical definitions of the
model quickly. Obviously, the simplified strain energy density strain tensor invariants in the quadratic terms. In material
function parameters are reduced from 5 to only 3. Moreover, constitutive model, strain determines stress, and strain tensor
as shown in Figure  9, the material constitutive model repre­ invariant is an important parameter of the model. As shown
sented by the simplified model has a high fitting degree with in Equations  (32) and (33), the mathematical definitions of I1
the experimental data (fitting errors of TPU-1, TPU-2, and and I2 are different, and the function curve of their relation­
TPU-3 are 0.95%, 0.81%, and 0.98%, respectively), indicating ship with nominal strain is also different in Figure  11. With
that simplified model can effectively predict the uniaxial ten­ the increase of nominal strain, the curve of (I1-3)2 grows expo­
sile mechanical behavior of TPU materials with different nentially, while the curve of (I2-3)2 grows slowly, and the latter
hard segment contents. Through the simplification algo­ basically conforms to the nominal stress in the growth trend
rithm, the simplified model contains three terms, and cannot of experimental data. The difference between these two curves
be further simplified, which is consistent with the conclusion becomes especially obvious when the strain >100%. Therefore,
of Tschoegl’s that the strain energy density function of an “S” the quadratic term (I2-3)2 of the strain energy density function
type stress–strain curve should contain at least three terms for the prepared TPU is more consistent with the actual mate­
for an accurate fitting[32] rial deformation trend

W = C10 (I1 − 3) + C01 (I 2 − 3) + C02 (I 2 − 3)2 (30) I1 = λ 2 + 2λ −1(32)

W = C10 (I1 − 3) + C01 (I 2 − 3) + C20 (I1 − 3)2 (31) I 2 = 2λ + λ −2(33)

Macromol. Theory Simul. 2020, 2000009 2000009  (9 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 10.  Fitting results of the trinomial model and experimental data of a) TPU-1, b) TPU-2 and c) TPU-3.

4.4. The Prediction Capability of the Simplified Model by FEM was programmed by Fortran, and the model parameters C10,
Analysis C01, and C20 were input into ABAQUS. The working condi­
tions were consistent with the experiments, which were set to
Based on the simplified constitutive model, we can use FEM obtain true stress–strain states of the prepared TPUs. In order
to predict the mechanical behavior of TPU under true engi­ to verify the accuracy of the simplified model in the simula­
neering condition. Through the fitting process of simpli­ tion, the output true stress–strain curves of the element in the
fication algorithm, the model parameters C10, C01, and C20 gauge length part were compared with the experimental data.
required for the simplified model can be obtained, which are Through Equations  (27), (28) and Equations  (34), (35), the
listed in Table 1. The UHYPER subroutine of simplified model force–displacement curves obtained from the experiment can
be transformed into true stress–strain curves

σ t = σ (1 + ε ) (34)

ε t = ln(1 + ε )(35)

where σ and ε are nominal stress and nominal strain, respec­


tively; σt and εt are true stress and true strain, respectively.
As shown in Figure  12, the results show that the curve of
simulation has a perfect fitting with the experimental curve,
which indicates that the simplified model can effectively pre­
dict the mechanical behavior of materials in the actual working

Table 1.  Ci values required for FEM.

Specimen C01 C10 C02


TPU-1 17.4 –11.11 3.134
TPU-2 24.38 –15.4 4.185
TPU-3 26.36 –16.5 4.524
Figure 11.  Curves of (I1-3)2 – nominal strain and (I2-3)2 – nominal strain.

Macromol. Theory Simul. 2020, 2000009 2000009  (10 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Figure 12.  The true stress-strain curves of experiment and simulation of a) TPU-1, b) TPU-2 and c) TPU-3.

conditions through engineering simulation. In addition, when model, the simplification algorithm proposed in this paper
the hyperelastic material element is used in the complex multi­ may be applicated to the construction of constitutive models of
body dynamics analysis program, not only the accuracy of the other hyperelastic materials, and the combination of simplified
program, but also the efficiency of the program should be con­ constitutive model and simulation method has certain guiding
sidered. Therefore, the simplification of the equation is helpful significance for practical engineering applications. However,
to reduce the calculation time of engineering simulation for for simulation of an event under more complicated conditions,
materials with complex shape or working conditions.[37] such as the variation of temperature, the influencing factor of
temperature should be considered in the simplified model algo­
rithm, which is a limitation of this study.
5. Conclusions
In this article, three kinds of TPU hyperelastic materials
(TPU-1, TPU-2, and TPU-3) were prepared by adjusting the pro­
Acknowledgements
portion of hard segment content to soft segment content. Based The work was financially supported by the National Natural Science
on the generalized Mooney–Rivlin model (N  = 2) function Foundation of China (51973179, U1805253), the Natural Science Foundation
structure, a “one-step” transformation algorithm from standard of Fujian Province of China (2018J01079), the Science and Technology
Major Project of the Fujian Province (2018HZ0001-1, 2018HZ0003-1),
test data to the simplest constitutive model was proposed, and the Xiamen Science and Technology Project (3502Z20183001,
which was successfully applied to the prepared TPUs, and their 3502Z20183014, YDZX20193502000004).
simplest constitutive models are obtained. The fitting errors of
the simplified constitutive model curves to the experimentally
determined nominal stress–strain curves of TPU-1, TPU-2, and
TPU-3 are 0.95%, 0.81%, and 0.98%, respectively, indicating Conflict of Interest
that the simplified constitutive model can accurately express The authors declare no conflict of interest.
the mechanical behavior of the material under the actual state,
and the fitting accuracy does not reduce regardless of the vari­
ation of the internal composition of the material itself. Finally,
through FEM analysis, the true stress–strain states of TPUs Keywords
under the actual working condition are successfully predicted. finite element analysis, hyperelastic constitutive model, simplification
Due to the universality of the generalized Mooney–Rivlin algorithm, thermoplastic polyurethane elastomers, uniaxial tension

Macromol. Theory Simul. 2020, 2000009 2000009  (11 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.mts-journal.de

Received: January 31, 2020 [21] A.  Isihara, N.  Hashitsume, M.  Tatibana, J. Chem. Phys. 1951, 19,
Revised: March 26, 2020 1508.
Published online: [22] R. W.  Ogden, Non-Linear Elastic Deformations, Dover, Mineola, NY
1997.
[23] R. Hao, D. X. Li, Adv. Eng. Mater. 2019, 21, 1900095.
[24] M. Knezevic, I. J. Beyerlein, Adv. Eng. Mater. 2018, 20, 1700956.
[1] H. Cho, S. Mayer, E. Poselt, M. Susoff, P. J. in’t Veld, G. C. Rutledge, [25] P. Moseley, J. M. Florez, H. A. Sonar, G. Agarwal, W. Curtin, J. Paik,
M. C. Boyce, Polymer 2017, 128, 87. Adv. Eng. Mater. 2016, 18, 978.
[2] J. A.  Fritz Fidel Rocco, J. E.  Salgueiro Lima, V. L.  Lourenco, [26] R. Bouaziz, K. Prashantha, F. Roger, Mech. Adv. Mater. Struct. 2019,
N. L.  Batista, E. C.  Botelho, K.  Iha, J. Appl. Polym. Sci. 2012, 126, 26, 1209.
1461. [27] R.  Bouaziz, F.  Roger, K.  Prashantha, Smart Mater. Struct. 2017, 26,
[3] Z. H. Li, J. Lambros, Int. J. Solids Struct. 2001, 38, 3549. 055009.
[4] Y. L.  Wang, J.  Hao, Z. Q.  Huang, G. Q.  Zheng, K.  Dai, C. T.  Liu, [28] S. Kim, H. Shin, S. Rhim, K. Y. Rhee, Composites, Part B 2019, 163,
C. Y. Shen, Carbon 2018, 126, 360. 297.
[5] Y.-C.  Chien, W.-T.  Chuang, U. S.  Jeng, S.-H.  Hsu, ACS Appl. Mater. [29] P.  Fan, W.  Chen, B.  Zhao, J.  Hu, J.  Gao, G.  Fang, F.  Peng, Polymer
Interfaces 2017, 9, 5419. 2018, 148, 370.
[6] C. P. Buckley, C. Prisacariu, C. Martin, Polymer 2010, 51, 3213. [30] Y.  Xiang, D.  Zhong, P.  Wang, T.  Yin, H.  Zhou, H.  Yu, C.  Baliga,
[7] C. H. Y.  Chentsai, E. L.  Thomas, W. J.  Macknight, N. S.  Schneider, S. Qu, W. Yang, J. Mech. Phys. Solids 2019, 128, 208.
Polymer 1986, 27, 659. [31] M. Timmel, S. Kolling, P. Osterrieder, P. Du Bois, Int. J. Impact. Eng.
[8] F. Feng, L. Ye, J. Appl. Polym. Sci. 2011, 119, 2778. 2007, 34, 1465.
[9] T. Choi, D. Fragiadakis, C. M. Roland, J. Runt, Macromolecules 2012, [32] N. W. Tschoegl, J. Polym. Sci., Part A-1: Polym. Chem. 1971, 9, 1959.
45, 3581. [33] R. W.  Ogden, G.  Saccomandi, I.  Sgura, Comput. Mech. 2004, 34,
[10] S.  Iniguez-Macedo, R.  Lostado-Lorza, R.  Escribano-Garcia, 484.
M. Angeles Martinez-Calvo, Materials 2019, 12, 1019. [34] K. Hibbitt, Sorensen, Inc., Abaqus Online Documentation 6.14, http://
[11] H. G. Kilian, Colloid Polym. Sci. 1985, 263, 30. ivtabaqusdoc.ivt.ntnu.no:2080/v6.14/index.html (accessed: 2014).
[12] E. M. Arruda, M. C. Boyce, J. Mech. Phys. Solids 1993, 41, 389. [35] O. H. Yeoh, P. Fleming, J. Polym. Sci., Part B: Polym. Phys. 1997, 35,
[13] M. Mooney, J. Appl. Phys. 1940, 11, 582. 1919.
[14] R. S. Rivlin, Philos. Trans. R. Soc. A 1948, 241, 379. [36] O. H. Yeoh, Rubber Chem. Technol. 1993, 66, 754.
[15] A. N. Gent, Rubber Chem. Technol. 1996, 69, 59. [37] G.  Chagnon, G.  Marckmann, E.  Verron, Rubber Chem. Technol.
[16] R. W. Ogden, Proc. R. Soc. London, Ser. A 1972, 326, 565. 2004, 77, 724.
[17] Q. Zhao, H. J. Qi, T. Xie, Prog. Polym. Sci. 2015, 49–50, 79. [38] H.  Shin, J. B.  Kim, S. J.  Kim, K. Y.  Rhee, Comput. Mater. Sci. 2015,
[18] L. Hoss, R. J. Marczak, Mec. Comput. 2010, 29, 2759. 100, 31.
[19] G. Marckmann, E. Verron, Rubber Chem. Technol. 2006, 79, 835. [39] J. B. Kim, H. Shin, Int. J. Impact. Eng. 2009, 36, 746.
[20] C.  Wex, S.  Arndt, A.  Stoll, C.  Bruns, Y.  Kupriyanova, Biomed. Tech. [40] A.  Isihara, N.  Hashitsume, M.  Tatibana, J. Chem. Phys. 1951, 19,
2015, 60, 577. 1508.

Macromol. Theory Simul. 2020, 2000009 2000009  (12 of 12) © 2020 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

You might also like