1. Introduction
Hyperelastic polymeric materials, which possess exceptional hyperelastic properties, have been widely utilized in various fields such as medical devices [
1], flexible electrodes [
2], and soft robots [
3]. The widespread applications of hyperelastic polymeric materials have also spurred a lot of research into characterizing their hyperelastic properties [
4]. In particular, carrying out the analysis of the 3D stress–strain of complex elastic components relying on finite element analysis (FEA) has become an indispensable part in the process of product design in recent decades. However, the reliability and accuracy of results strongly depend on the performance of the constitutive model predicting the mechanical behavior of the material. Therefore, a constitutive model that can more accurately predict the mechanical behavior of material is the key to making the analytical results based on finite element analysis more realistic.
Currently, constitutive models characterizing hyperelastic properties are usually built by the strain energy density function. In light of the different ways of constructing the strain energy density function, the hyperelastic constitutive models can be divided into two overall categories: the first category encompass statistical mechanics models based on the idealized network structure of molecular chains, while the other category comprise phenomenological models based on continuum mechanics.
For statistical mechanical models, it is usually constructed based on the assumption of the network structure of molecular chains inside the polymer [
5]. Although the parameters of these models usually have practical physical context, the statistical mechanical models with better representational ability often have a more complex form. So, they are not very good for providing an analytical solution, and they are not good for numerical solutions [
6], as well as having a large corresponding calculation [
7]. In addition, some statistical mechanical models are not well suited for dealing with some important issues such as irreversible deformation and inelastic volume expansion [
8].
For phenomenological models, they are usually constructed based on the strain invariants or principal stretches. Although the parameters of these phenomenological models often do not have practical physical context, they have been widely studied and applied because of their relative advantages (including easy-to-obtain parameters, relatively high computational efficiency, no requirement for understanding the microstructure of the materials, and wider material applicability). The Mooney model is the earliest constitutive model used to represent the mechanical behavior of polymeric rubber. It has a linearized form, and it usually has better performance in the case with small deformation [
9]. Taking into account the limited capabilities of the Mooney model, Rivlin extended it in the form of polynomial series to obtain a generalized polynomial model [
10]. Because the polynomial model has higher-order terms of the strain invariants, it is more suitable for the case with large deformation. However, the model with higher-order terms usually requires more parameters, which easily make the model unstable. In line with the structure of polynomial model, researchers explored different orders and combinations of strain invariants, and derived many other models [
11,
12,
13,
14]. In particular, when the second strain invariant in the polynomial model is ignored, the reduced polynomial model is obtained [
15]. Even though the polynomial model and reduced polynomial model can well characterize the hyperelastic properties of filled and unfilled rubber by retaining higher-order terms [
16], they may have difficulties in solving numerical problems [
17]. Therefore, the second-order polynomial model or the third-order reduced polynomial model is generally used. Except for some of the above constitutive models on account of the polynomial form of strain invariants, there are also some constitutive models based on the logarithm or exponential form of strain invariants. Notwithstanding, comparative studies show that these models do not have a particular advantage in characterizing hyperelastic properties [
18,
19]. Besides, the eigenvalues of the strain tensor are related to the principal stretches, so there are also some constitutive models constructed directly based on the principal stretches [
20,
21,
22]. These models are usually composed of special functions related to the principal stretches, and they show good performance in characterizing the hyperelastic properties.
Although there are already many classical constitutive models to characterize the hyperelastic property, there are still continuously improved and completely new models emerging. The reason for this may be that the deficiencies of existing models still exist [
23] and the importance of hyperelastic models in designing engineering components still motivates researchers to develop more general and robust constitutive models [
24,
25]. As presented in some reviews [
7,
26], the performances of these existing models are different from each other, and not all models can effectively characterize the multiaxial deformation of hyperelastic material based solely on a single set of model parameters, and most of these models are applied to a specific type of hyperelastic material. Furthermore, many models with relatively few parameters cannot reliably predict the whole range of strain and different modes of deformation. Considering that hyperelastic materials often exhibit multiaxial states of deformation in practical applications, it is unquestionable that the ability of a model to characterize multiaxial deformation should be evaluated first. However, evaluating the ability of a model to characterize multiaxial deformation often requires the simultaneous use of experimental data from three modes of homogenous deformation (including uniaxial tension, equibiaxial tension, and pure shear) to calibrate the parameters of the model. This may be prohibitively expensive and even infeasible in practice [
15]. Due to limited hardware conditions or limitations in the tensile strength of the target hyperelastic material, some pure deformation mode data of materials are simply not measurable, especially for the equibiaxial tension. So, there are also some researchers evaluating the performance of existing models from another dimension. As presented in the comparative studies in [
15,
27], the parameters of these models were calibrated only using data from a single pure mode of deformation (such as uniaxial tension). The corresponding comparison results highlight the effective characterization ability of Yeoh model [
28] for untested deformation modes (exceeding the well-known Ogden model [
21]). Although some studies believe that it may not be reliable for a model only calibrated through the data of a single mode to describe the multiaxial mode of deformation [
6,
29], as an auxiliary evaluation method, it is also being pursued by researchers [
18,
30,
31,
32]. Especially in [
3], researchers refer to the model’s ability to effectively predict the untested mode of deformation as the property of deformation-mode independency. As described above, this property is particularly useful for real-world engineering applications where the available experimental data are limited [
32].
Considering the pursuit of a universal and robust hyperelastic constitutive model, this study aims to improve the Yeoh model to compensate for its shortcoming in performance and applicability. The idea of focusing on improving the Yeoh model is not arbitrary, but rather based on its unique advantages in characterizing untested modes of deformation as explained above. The reason why we want to take advantage of this advantage is because we have indeed encountered the situation in practice where the equibiaxial deformation mode of a used polymer material is not measurable. With the widespread application of different soft polymer materials in engineering design, we believe that the probability of this situation occurring will be higher. So, the actual demand also drives us to carry out this work. Furthermore, it is well known that the Yeoh model will underestimate the equibiaxial mode of deformation when characterizing multiaxial state of stress; hence, it should be further improved in order to have more accurate characterization results. Hitherto, there are several sporadic improved Yeoh models. Earlier, Yeoh proposed to improve the fitting accuracy of the Yeoh model to the data from simple shear by adding an additional exponential term related to the first principal invariant [
33]. The comparative studies confirmed that the improved model has a poorer performance for predicting the biaxial behavior of natural rubber (using Treloar’s data), and the comprehensive performance of the improved model is not as good as the original Yeoh model [
34]. This may be because the improved model is still not related to the second principal invariant or the added additional exponential item is inappropriate. Relevant study has confirmed that the model containing the second principal invariant is important for improving the prediction accuracy of the model, especially for the equibiaxial deformation [
35]. Based on this, a recently improved Yeoh model adds the square root of the second principal invariant as a correction term to the original Yeoh model [
30]. The research results confirm that the modified model has significantly improved predictive performance for equibiaxial tension, but the entire modification process is based on uniaxial data, which may lead to insufficient improvement in the model’s actual predictive ability for multiaxial deformations. In addition, it has not yet imposed effective constraints on the parameters of the modified model to ultimately make the modified model convex and stable. In addition to these two modified Yeoh models mentioned above, Hohenberger et al. also replaced all the determined orders in the Yeoh model with undetermined coefficients to expand it into a generalized Yeoh model, so as to describe the low and high strain nonlinearity of highly filled high damping rubber [
36]. This generalized Yeoh model has been proven to have improved prediction accuracy for uniaxial tension and compression, but its characterization accuracy for multiaxial deformation has not been determined. To sum up, there is still a certain distance between the current modified Yeoh models and the universal and robust model; so, it is still meaningful to improve the Yeoh model again.
In this study, we use a thoughtful correction term to modify the Yeoh model, so as to improve its ability for characterizing multiaxial deformations. This correction term is derived from the corresponding residual strain energy when the Yeoh model predicts the equibiaxial mode of deformation, and its specific form is a composite function based on a power function represented by the principal stretches. The strain energy density function of the modified model is represented by the first strain invariant and the principal stretches, and it contains only five parameters to be identified. In the case of specific parameters, the model can be degraded to the neo-Hookean model, the Mooney–Rivlin model, the Yeoh model, and the Biderman model. Therefore, the modified model can also be regarded as the parent model of these four classical models. In addition, we also introduce a special parameter identification method based on the cyclic genetic-pattern search algorithm in order to improve the predicted accuracy of constitutive models. For demonstrating the modified model’s robustness and generality, the modified model is applied to six different types of hyperelastic materials. During this process, in addition to comparing the performance of the modified constitutive model with some landmark constitutive models in the literatures, the stability and convexity of the modified constitutive model is also verified. Finally, the validity of the introduced parameter identification method is also confirmed by comparing it with four existing parameter identification methods.
4. Conclusions
This paper presents a modified hyperelastic constitutive model based on the first strain invariant and the principal stretches. In order to more accurately identify the parameters of model, a special method of parameter identification based on the cyclic genetic-pattern search algorithm has also been introduced. Combining the experiment data of different rubber materials with the introduced parameter identification method, the performance of the proposed modified constitutive model is fully evaluated. The results show that the proposed modified model not only possesses a significantly improved ability to predict multiaxial deformation, but also has a wider range of material applicability. This advantage is not only reflected in the comparison with the original Yeoh model, but also in the comparison with other improved Yeoh models (such as the modified Yeoh model, the generalized Yeoh model and the Melly model). Even compared with the excellent third-order Ogden model and Alexander model, our proposed modified model with only five undetermined parameters is not necessarily inferior in characterizing the multiaxial deformation of rubber materials. For example, the overall prediction accuracy of the modified model for isoprene vulcanized rubber and poly-acrylamide hydrogel is slightly better than that of the third-order Ogden model and Alexander model. Anyway, for the five rubber materials used in this study, our modified model has a similar level of total prediction accuracy as the two models of the third-order Ogden model and Alexander model (total error is less than 5%). Furthermore, our modified model is also proven to have a good ability to characterize the deformation of human brain cortex tissue. Its prediction accuracy for multiaxial deformations of human brain cortex tissue is not only similar to that of the third-order Ogden model (the third-order Ogden model is considered to be the best model currently characterizing the deformation of the human brain cortex tissue.), but also has no loss of convexity compared to the third-order Ogden model. In addition, the proposed modified model in this study is also proven to hold the improved ability to predict the untested equibiaxial deformation of most rubber materials used in this study. The modified constitutive model calibrated based on different datasets is also verified to have a posteriori Drucker stability and polyconvexity, which lays a foundation for the modified constitutive model to be applied to finite element analysis.
Based on the above conclusions, we believe that the main advantages of the modified constitutive model proposed by us are as follows:
Firstly, although the modified model has five undetermined parameters, it still maintains a relatively simple functional form compared to those models containing logarithmic, exponential or integral expressions, making it easier to perform related mathematical calculations. Having five parameters also ensures that it is neither incompatible with fourth-order weak nonlinear elasticity theory due to having too few parameters, nor does it face difficulties in parameter identification due to having too many parameters.
Secondly, thanks to our proposed correction term based on the principal stretch, this modified constitutive model has significantly improved ability for predicting multiaxial deformation. Moreover, this improved predictive ability has certain universality, that is, the modified model can be applied to various different types of rubber materials (including natural unfilled or filled rubber, silicone rubber and hydrogel). There are currently few studies that apply their model to so many types of material.
Thirdly, this modified constitutive model can, with relative accuracy, predict the multiaxial deformation of human brain cortex tissue (including uniaxial tension, uniaxial compression and simple shear) while ensuring convexity. This gives it the potentiality to characterize the biomechanics of soft biological tissues.
Finally, the modified constitutive model also has an improved capacity to predict untested equibiaxial deformation. This advantage is very useful in the situation where equibiaxial tension cannot be completed under limited hardware conditions.
Compared with other parameter identification methods, the introduced method of parameter identification has been proved not only to take into account different modes of deformation, but also makes models have better performance. The excellent capability of the introduced method of parameter identification benefits from the strong ability of global search of the genetic algorithm and the strong ability of local search of the pattern search algorithm, and the addition of the cyclic structure further weakens its dependence on the initial value, thereby making its uniform convergence better.