Revision as of 12:37, 13 May 2017

A Comparison of two frameworks for kinematic hardening in hyperelasto-plasticity

Knut A. Meyer${\displaystyle ^{*}}$, Magnus Ekh

Abstract

In this work we compare two frameworks for thermodynamically consistent hyperelasto-plasticity with kinematic hardening. The first was formulated by Dettmer and Reese (2004), inspired by Lion (2000), and has been used to model sheet metal forming. The second, formulated by Wallin et al. (2003), has been used to model large shear strains and cyclic ratcheting behavior of pearlitic steel (Johansson et al. 2006). In this paper we show that these frameworks can result in equivalent models for certain choices of free energies. Furthermore, it is shown that the choices of free energy found in the literature only result in minor differences. These differences are discussed theoretically and investigated numerically.

keywords Hyperelasto-plasticity, Finite strains, Kinematic hardening

1 INTRODUCTION

Large strains in metals during room temperature occur in many technical applications, often during manufacturing, such as sheet metal forming. Some components are also subjected to large strains during service, for example in the surface layer of railway rails and wheels (see e.g. [1, 2]). Experiments have shown that the Bauchinger effect, often modeled with kinematic hardening, is pronounced in many metals. Kinematic hardening can be modeled with different thermodynamically consistent hyperelasto-plastic frameworks found in the literature, and two of them are considered here. The first framework is based on rheological models with an Armstrong-Frederick (AF) type of kinematic hardening, and was proposed by Lion [3] and further developed by Dettmer and Reese [4]. The second framework, introduced by Wallin et al. [5], also features an AF type of kinematic hardening, and has been used to model the Swift effect [6] and large deformations in railway applications [7, 8]. In this paper we compare these frameworks, both theoretically and numerically.

2 DESCRIPTION OF FRAMEWORKS

In this section, the modeling frameworks are presented in terms of their assumed kinematics and thermodynamics. The common parts are presented in Subsections 2.1 and 2.2, followed by a description of how the frameworks differ in Subsections 2.3 and 2.4. Three specific models are defined in Subsections 2.5 and 2.6, which are compared in the numerical examples in Section 3. For the clarity of the presentation, only linear kinematic hardening is considered in the current section. In Section 4 we investigate nonlinear kinematic hardening, e.g. of Armstrong-Frederick type.

2.1 Kinematics and notations

Figure 1 shows the different configurations used in both [4] and [5]. Dettmer and Reese [4] introduce the inelastic plastic deformation gradient ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }}$ connecting the fictitious kinematic configuration ${\textstyle {\bar {\bar {\Omega }}}}$ to the initial configuration ${\textstyle \Omega }$. This connection is not introduced in Wallin et al. [5], but otherwise the same configurations and remaining deformation gradients are present in both frameworks.

Tensors on the current configuration ${\textstyle \omega }$ are denoted with lower case letters and no bars, e.g. ${\textstyle {\boldsymbol {b}}}$. Tensors on the intermediate ${\textstyle {\bar {\Omega }}}$ and kinematic ${\textstyle {\bar {\bar {\Omega }}}}$ configurations are denoted by one bar, e.g. ${\textstyle {\bar {\boldsymbol {C}}}_{\mathrm {e} }}$, and two bars, e.g. ${\textstyle {\bar {\bar {\boldsymbol {C}}}}_{\mathrm {ke} }}$, respectively. The following decompositions of the deformation gradients and definitions of the deformation tensors will be used:

The velocity gradients on the intermediate and kinematic configurations are defined as

and hence the velocity gradient ${\textstyle {\boldsymbol {l}}}$ on the current configuration can be written as

2.2 Thermodynamics

The free energy ${\textstyle \Psi }$ is introduced with the additive split according to

whereby the dissipation inequality (see e.g. Simo (1998) [9]) becomes

where ${\textstyle {\boldsymbol {\tau }}}$ is the Kirchhoff stress. Using the requirement of zero dissipation during elastic loading and Equation 5, the reduced dissipation inequality becomes

where the Mandel stress ${\textstyle {\bar {\boldsymbol {M}}}}$ is defined as

In a standard fashion, we adopt an associative evolution of the plastic deformation gradient in this paper:

where the functional dependence of the yield function ${\textstyle \Phi \leq 0}$ will be specified later. Up until this point the two frameworks are identical. We first describe the framework by Dettmer and Reese [4], before proceeding with the framework by Wallin et al. [5].

2.3 1st framework [4]

In the first framework proposed by Lion [3] and further developed by Dettmer and Reese [4], the plastic deformation gradient ${\textstyle {\boldsymbol {F}}_{\mathrm {p} }}$ is multiplicatively decomposed into an elastic part ${\textstyle {\boldsymbol {F}}_{\mathrm {ke} }}$ and a plastic part ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }}$. The physical motivation is that ${\textstyle {\boldsymbol {F}}_{\mathrm {ke} }}$ represents local elastic deformations on the microscale caused by dislocations and ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }}$ represents irreversible displacements in the slip systems. The assumption is that development of ${\textstyle {\boldsymbol {F}}_{\mathrm {ke} }}$ results in linear kinematic hardening and the development of ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }}$ reduces ${\textstyle {\boldsymbol {F}}_{\mathrm {ke} }}$, hence causing saturation (dynamic recovery) of the kinematic hardening. This is illustrated using a rheological model by Lion [3]. The multiplicative split of ${\textstyle {\boldsymbol {F}}_{\mathrm {p} }}$ results in the following additive split of the plastic velocity gradient ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {p} }}$

For the case of purely linear hardening (${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }={\boldsymbol {I}}}$), the reduced dissipation, Equation 8, can be written as

where the kinematic hardening stress of Mandel type (also denoted back-stress) is defined as

This motivates that the driving force for plastic flow is ${\textstyle {\bar {\boldsymbol {M}}}-{}\mathrm {^{1}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} }$ and thereby a yield criterion expressed as ${\textstyle \Phi ({\bar {\boldsymbol {M}}}-{}\mathrm {^{1}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} )}$.

2.4 2nd framework [5]

In the second framework, proposed by Wallin et al. [5], the deformation gradient ${\textstyle {\boldsymbol {F}}_{\mathrm {ke} }^{-1}}$ is introduced to model the deformation of the crystal lattice, due to the residual micro stresses responsible for the Bauchinger effect. From this deformation gradient the kinematic hardening stress of Mandel type is defined as

which yields that the reduced dissipation inequality 8 is

Using the standard interpretation of ${\textstyle {}\mathrm {^{2}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} }$ as a back-stress that reduces the driving force for plasticity, motivates the yield function ${\textstyle \Phi ({\bar {\boldsymbol {M}}}-{}\mathrm {^{2}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} )}$. This gives, by the postulate of maximum dissipation, the kinematic relation

and the same reduced dissipation inequality as in Equation 12 is obtained:

2.5 Specific formats for free energy

The elastic and kinematic free energies (with the third invariant ${\textstyle I_{\mathrm {3\bullet } }=\det(\bullet )}$) proposed by Vladimirov et al [10] are

The part of the elastic free energy, corresponding to Lamé's second parameter ${\textstyle \Lambda }$, is thus not included in the kinematic free energy.

A similar split is introduced in [5], but the free energy in that work is decomposed into an isochoric and a volumetric part. The formulation for the volumetric part is not the same in [5] and [7, 8], and here we use the formulation from [7, 8]. This difference only affects the bulk elastic response and the influence on the numerical results studied in this paper is therefore negligible.

From the discussion so far, there seem to be several differences between the frameworks: (1) the definition of the Mandel back-stress (${\textstyle {}\mathrm {^{1}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} }$ or ${\textstyle {}\mathrm {^{2}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} }$), (2) the variable of which the kinematic free energy depends on (${\textstyle {\bar {\bar {\boldsymbol {C}}}}_{\mathrm {ke} }}$ or ${\textstyle {\bar {\boldsymbol {c}}}_{\mathrm {ke} }}$) and (3) what part of the elastic free energy formulation that is used to formulate the kinematic free energy. The third of these can be investigated by taking the format of free energy from the second framework, but using the definitions and variables from the first framework to obtain model ${\textstyle \mathrm {C} }$:

2.6 Stresses for each model

We have now described both frameworks, the first by Dettmer and Reese [4] and the second by Wallin et al. [5]. By letting ${\textstyle \Psi _{\mathrm {kin} }}$ depend on ${\textstyle {\bar {\bar {\boldsymbol {C}}}}_{\mathrm {ke} }}$ or ${\textstyle {\bar {\boldsymbol {c}}}_{\mathrm {ke} }}$ we can use 13 and 14, respectively, to obtain the Mandel back-stresses for the two frameworks:

The Mandel stresses for model ${\textstyle \mathrm {A} }$, ${\textstyle {}\mathrm {^{A}} {\bar {\boldsymbol {M}}}\mathrm {} }$ and ${\textstyle {}\mathrm {^{A}} {\bar {\boldsymbol {M}}}_{\mathrm {k} }\mathrm {} }$, are found using Equations 9 and 24 with the free energies in Equations 18 and 19:

If ${\textstyle {}\mathrm {^{B}} {\bar {\boldsymbol {c}}}_{\mathrm {ke} }\mathrm {} ={}\mathrm {^{C}} {\bar {\boldsymbol {b}}}_{\mathrm {ke} }\mathrm {} }$ then clearly model ${\textstyle \mathrm {B} }$ and ${\textstyle \mathrm {C} }$ are equivalent. Assuming that is the case for some point in time, we also have ${\textstyle {}\mathrm {^{B}} {\bar {\boldsymbol {L}}}_{\mathrm {p} }\mathrm {} ={}\mathrm {^{C}} {\bar {\boldsymbol {L}}}_{\mathrm {p} }\mathrm {} ={\bar {\boldsymbol {L}}}_{\mathrm {p} }}$. Since model ${\textstyle \mathrm {B} }$ is using the second framework, and model ${\textstyle \mathrm {C} }$ is using the first, we also have ${\textstyle -{}\mathrm {^{B}} {\bar {\boldsymbol {L}}}_{\mathrm {ke} }\mathrm {} ={}\mathrm {^{C}} {\bar {\boldsymbol {L}}}_{\mathrm {ke} }\mathrm {} ={\bar {\boldsymbol {L}}}_{\mathrm {p} }}$, hence

As ${\textstyle {}\mathrm {^{B}} {\bar {\boldsymbol {c}}}_{\mathrm {ke} }\mathrm {} ={}\mathrm {^{C}} {\bar {\boldsymbol {b}}}_{\mathrm {ke} }\mathrm {} ={\boldsymbol {I}}}$ initially, the statement ${\textstyle {}\mathrm {^{B}} {\bar {\boldsymbol {c}}}_{\mathrm {ke} }\mathrm {} ={}\mathrm {^{C}} {\bar {\boldsymbol {b}}}_{\mathrm {ke} }\mathrm {} }$ is true for all points in time under the assumption that ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {p} }}$ is symmetric. If the free energy is isotropic, the Mandel stresses ${\textstyle {\bar {\boldsymbol {M}}}}$ and ${\textstyle {\bar {\boldsymbol {M}}}_{\mathrm {k} }}$ are symmetric, and hence ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {p} }}$ becomes symmetric for the associative choice of ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {p} }}$ in Equation 10. This leads to the conclusion that model B and C are equivalent, which is verified numerically later. Furthermore, this proof leads to the interesting conclusion that the frameworks can give exactly the same model with proper choices of free energy. The possibility to formulate an unsymmetric ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {p} }}$ for isotropy is discussed in e.g. Wallin et al. [5] and Wallin and Ristinmaa [6], but is not investigated in this paper.

3 NUMERICAL RESULTS

In this section we evaluate the response of the material models for uniaxial loading and simple shear loading. The models are implemented using a standard Backward Euler integration scheme, for which Vladimirov et al. [10] noted that the accuracy suffers at large time steps. To avoid these accuracy problems, approximately ${\textstyle 4\cdot {10}^{4}}$ and ${\textstyle 5\cdot 10^{4}}$ load steps are used for the uniaxial and simple shear loading, respectively.

The von Mises effective stress is used to define the yield function ${\textstyle \Phi }$ according to

whereby the evolution of the plastic deformation gradient in 10 becomes

From this it follows that the plastic deformation is isochoric: By time differentiation of ${\textstyle \det \left({\boldsymbol {F}}_{\mathrm {p} }\right)}$ and using 2

hence, ${\textstyle \det \left({\boldsymbol {F}}_{\mathrm {p} }\right)=1}$.

We first consider uniaxial stress in Figure 2a, by letting the normal deformation gradient ${\textstyle F_{11}}$ increase from 1 to 10 while keeping the Cauchy stresses ${\textstyle \sigma _{22}=\sigma _{33}=0}$ (${\textstyle {\boldsymbol {\sigma }}={\boldsymbol {\tau }}/I_{3\mathrm {F} }}$). As previously shown theoretically, model ${\textstyle \mathrm {B} }$ and ${\textstyle \mathrm {C} }$ give the same response. Model ${\textstyle \mathrm {A} }$ gives a somewhat stiffer response at large deformations, due to the different choice of free energy.

In the case of simple shear loading, Figure 2b shows that the response of all the models coincide. While this is expected for model ${\textstyle \mathrm {B} }$ and ${\textstyle \mathrm {C} }$, the fact that model ${\textstyle \mathrm {A} }$ and ${\textstyle \mathrm {C} }$ coincide is explained with simple shear being an isochoric process (${\textstyle I_{3\mathrm {F} }=I_{3\mathrm {F} _{\mathrm {p} }}=1\Rightarrow I_{3\mathrm {F} _{\mathrm {e} }}=1}$)

The results in Figure 2 show negligible differences between the different formulations of free energy. From a theoretical point of view, one could argue that the deviatoric dependence of the back-stress is more correct, based on the experimental evidence of volume preservation for metal plasticity.

4 ARMSTRONG-FREDERICK SATURATION

Linear kinematic hardening was considered in Section 2, for which Dettmer and Reese [4] set ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }={\boldsymbol {I}}}$, yielding the reduced dissipation inequality in Equation 12. If the general case with an evolving ${\textstyle {\boldsymbol {F}}_{\mathrm {kp} }}$ is considered, the reduced dissipation inequality, using Equation 11, becomes

From this the kinematic stress of Mandel type on the kinematic configuration ${\textstyle {}\mathrm {^{1}} {\bar {\bar {\boldsymbol {M}}}}_{\mathrm {k} }\mathrm {} }$ is defined as

Saturation is motivated by the rheological model, setting the evolution on ${\textstyle {\bar {\bar {\boldsymbol {L}}}}_{\mathrm {kp} }^{\mathrm {sym} }}$. As the Mandel stresses in [4] are symmetric, this can be written as

${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {ke} }}$ will be symmetric if a modified yield potential ${\textstyle \Phi _{\mathrm {kin} }^{*}}$ exists and we have, as before, symmetric Mandel stresses. This is the case for the considered model with Armstrong-Frederick type of non-linear kinematic saturation. ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {ke} }}$ is also symmetric in the work by Zhu et al. [11], where the first framework was extended to include nonlinear kinematic hardening of Ohno-Wang type. When ${\textstyle {\bar {\boldsymbol {L}}}_{\mathrm {ke} }}$ is symmetric, the same arguments as before relating to Equations 29 and 30 hold true. Hence, for appropriate free energies the two frameworks give the same model also for nonlinear kinematic hardening.

5 CONCLUDING REMARKS

We have shown that the two different frameworks, introduced by [4] and [5], can give equivalent models for isotropic free energies. The major difference between the models used for the different frameworks is the kinematic free energy. To obtain the same model, the same structure of the kinematic free energy must be used, but with a different variable (${\textstyle {\bar {\bar {\boldsymbol {C}}}}_{\mathrm {ke} }}$ or ${\textstyle {\bar {\boldsymbol {c}}}_{\mathrm {ke} }}$). The numerical results confirm these theoretical findings. They further show that the difference between the formulations of free energy has a negligible effect on the material response up to a stretch of 5 for uniaxial loading, and no effect during simple shear.

6 ACKNOWLEDGMENTS

This work has been partly financed within the European Horizon 2020 Joint Technology Initiative Shift2Rail through contract no. 730841. The use of AceGen [12] has been very effective in speeding up the implementation of the material models.

REFERENCES