For the extension of the eigenerosion algorithm to ductile crack propagation, the rate-independent finite strain
\(J_2\) elasto-plasticity formulation with isotropic hardening from [
37‐
39] is applied. For the numerical implementation, see Miehe et al. [
19] and Klinkel [
13]. In this material model, the deformation gradient is multiplicatively decomposed as
$$\begin{aligned} {\varvec{F}}={\varvec{F}}^{\mathrm {e}}{\varvec{F}}^{\mathrm {p}} \end{aligned}$$
(12)
into an elastic and a plastic part,
\({\varvec{F}}^{\mathrm {e}}\) and
\({\varvec{F}}^{\mathrm {p}}\), respectively, cf. also Kröner [
14,
15]. Following this, the elastic left Cauchy–Green tensor and its spectral decomposition result in
$$\begin{aligned} {\varvec{b}}^{\mathrm {e}}={\varvec{F}}^{\mathrm {e}}{\varvec{F}}^{\mathrm {eT}}=\sum \limits _{i=1}^3(\lambda _i^{\mathrm {e}})^2\,{\varvec{n}}_i \otimes {\varvec{n}}_i \end{aligned}$$
(13)
with the square root of the principal elastic stretches
\(\lambda _i^{\mathrm {e}}\) obtained as eigenvalues of
\({\varvec{b}}^{\mathrm {e}}\), and their principal directions expressed by the eigenvectors
\({\varvec{n}}_i\). With the application of the principal logarithmic elastic strains
\(\epsilon _i^\mathrm {e}=\log (\lambda _i^\mathrm {e})\), the elastic strain tensor
$$\begin{aligned} \varvec{\varepsilon }^{\mathrm {e}}=\sum \limits _{i=1}^3 \epsilon _i^{\mathrm {e}}\,{\varvec{n}}_i \otimes {\varvec{n}}_i \end{aligned}$$
(14)
is obtained in logarithmic representation so that the additive split of the strain
\(\varvec{\varepsilon }=\varvec{\varepsilon }^{\mathrm {e}}+\varvec{\varepsilon }^{\mathrm {p}}\) into an elastic and plastic part
\(\varvec{\varepsilon }^{\mathrm {e}}\) and
\(\varvec{\varepsilon }^{\mathrm {p}}\) can be considered. The strain energy density
$$\begin{aligned} \psi =\psi ^{\mathrm {e}}(\varvec{\varepsilon }^{\mathrm {e}})+\psi ^{\mathrm {p}}(\alpha ) \end{aligned}$$
(15)
is divided into an elastic part
\(\psi ^{\mathrm {e}}(\varvec{\varepsilon }^{\mathrm {e}})\) only depending on the elastic strains
\(\varvec{\varepsilon }^{\mathrm {e}}\) and the plastic part
\(\psi ^{\mathrm {p}}(\alpha )\) depending on the equivalent plastic strain
\(\alpha \) serving as internal variable. For the elastic part, the quadratic elastic strain energy density
$$\begin{aligned} \psi ^\mathrm {e}=\frac{\kappa }{2}\, \mathrm {tr}(\varvec{\varepsilon }^{\mathrm {e}})^2+\mu \, \mathrm {dev}(\varvec{\varepsilon }^{\mathrm {e}})\cdot \mathrm {dev}(\varvec{\varepsilon }^{\mathrm {e}}) \end{aligned}$$
(16)
with the compression modulus
\(\kappa \) and the shear modulus
\(\mu \) is taken into account. The Kirchhoff stress tensor can then be identified as
$$\begin{aligned} \varvec{\tau }=\kappa \,\mathrm {tr}(\varvec{\varepsilon }^\mathrm {e})\,{\varvec{I}}+2\,\mu \, \mathrm {dev}(\varvec{\varepsilon }^\mathrm {e}). \end{aligned}$$
(17)
The convex plastic dissipation potential
$$\begin{aligned} \psi ^{\mathrm {p}}=y_0\,\alpha + (y_{\infty }-y_0) \,\left[ \alpha +\frac{\exp (-h^{\mathrm {exp}}\,\alpha )-1}{ h^{\mathrm {exp}}}\right] +\frac{1}{2}\,h^{\mathrm {lin}}\,\alpha ^2 \end{aligned}$$
(18)
is chosen as the superposition of linear- and exponential-type hardening, cf. Voce [
40]. Herein, the hardening parameters have the following meaning:
\(h^{\mathrm {exp}}\) specifies the degree of exponential hardening,
\(h^{\mathrm {lin}}\) defines the slope of the superimposed linear hardening,
\(y_0\) denotes the initial yield strength, and
\(y_{\infty }\) describes the plastic yield strength at the transition to the almost purely linear hardening. Hence, we obtain the hardening function
$$\begin{aligned} \beta =\frac{\partial \, \psi ^{\mathrm {p}}}{\partial \, \alpha }=y_0+(y_{\infty }-y_0)\,\left[ 1-\exp (-h^{\mathrm {exp}}\,\alpha )\right] +h^{\mathrm {lin}}\,\alpha . \end{aligned}$$
(19)
We consider plastic incompressibility, which is typical for metal plasticity, so that the flow condition becomes the von Mises type
$$\begin{aligned} \phi =||\mathrm {dev}\,\varvec{\tau }||-\sqrt{\frac{2}{3}}\,\beta \le 0. \end{aligned}$$
(20)
Together with the plastic variable
\(\lambda ^{\mathrm {p}}\ge 0\), the Kuhn–Tucker condition
$$\begin{aligned} \phi \,\lambda ^{\mathrm {p}}=0 \end{aligned}$$
(21)
is fulfilled, where
\(\lambda ^{\mathrm {p}}\) is connected to the evolution of the norm of plastic strains
$$\begin{aligned} ||\dot{\varvec{\varepsilon }}^{\mathrm {p}}||=\lambda ^{\mathrm {p}} \end{aligned}$$
(22)
and to the evolution of the equivalent plastic strains
$$\begin{aligned} {\dot{\alpha }}=\sqrt{\frac{2}{3}}\,\lambda ^{\mathrm {p}}. \end{aligned}$$
(23)
It is numerically computed from
\(\phi = 0\), see Eq. (
20), using the Newton scheme. In the case of ductile crack propagation, the imposed energy is proposed to be
$$\begin{aligned} -\Delta U_K=\int \limits _{\Omega _K}\left( \psi ^{\mathrm {e}}+\psi ^{\mathrm {p}} \right) \,\mathrm {d}V, \end{aligned}$$
(24)
which represents an extension of the elastic strain energy density by the plastic potential
\(\psi ^{\mathrm {p}}\). This approach is conceptually in line with the idea in Irwin [
9] who extended the Griffith energy release rate for brittle materials to a Griffith-type energy release rate for ductile materials by adding the energy that dissipates into plasticity near the crack tip before the crack propagates. Therefore, the Griffith-type energy release rate
\(G_c\) has to be large enough, in order to trigger plastic deformations before the crack propagation. Note that this parameter
\(G_c\) for ductile crack propagation should not be confused with the classical Griffith constant, which is connected to brittle fracture.
2.3.1 Regularization of plasticity
In order to avoid mesh dependence, which may already occur during the formation of plastic shear bands before fracture occurs, regularization is applied. Here, the regularization is achieved by extending the elasto-plastic material model to elasto-viscoplasticity of the Perzyna-type, cf. e.g., Perzyna [
29], Junker et al. [
10]. Since here the rate dependency of the material is small and not to be modeled, the viscous part is only introduced for numerical reasons. Instead of computing
\(\lambda ^{\mathrm {p}}\) from
\(\phi = 0\), we assume the plastic parameter to be computed from
$$\begin{aligned} \lambda ^{\mathrm {p}}=\frac{1}{\eta }\,\left\langle \phi \right\rangle _{+} \end{aligned}$$
(25)
with the Macaulay bracket
\(\left\langle (\bullet ) \right\rangle _{\pm }=((\bullet ) \pm |(\bullet )|)/2\) and the viscosity
\(\eta \). Here, the same evolution Eqs. (
22) and (
23) as for the previous elasto-plastic model are considered. Thereby, the plastic deformation becomes more diffuse because in contrast to the unregularized model, the plastic deformation is delayed in the affected elements so that the neighboring elements are subjected to higher loading and have to deform plastically as well. Note that Eq. (
25) may lead to a violation of the yield criterion, i.e.,
\(\phi >0\), if the viscosity
\(\eta \) is chosen large enough. Hence, the parameter
\(\eta \) controls the speed of plastic deformation. In order to adjust the thickness of the shear bands and thus, the diffusive character of the regularized model, the hardening function is modified to
$$\begin{aligned} \beta =y_0\,\exp (-\delta \, \alpha )+(y_{\infty }-y_0)\,\left[ 1-\exp (-h^{\mathrm {exp}}\,\alpha )\right] +h^{\mathrm {lin}}\,\alpha \end{aligned}$$
(26)
with the localization parameter
\(\Delta \) reducing the yield stress with increasing equivalent plastic strain
\(\alpha \). This is necessary to imply additional plastic deformation after plasticity once has taken place in the particular material point, because the viscosity
\(\eta \) strongly smears the plastic deformations over the body and therefore weakens the plasticity on the corresponding elements. Hence, without the parameter
\(\Delta \), no clear shear bands can develop within the viscoplastic material model if the viscosity
\(\eta \) is chosen large enough to regularize. With increasing values of the localization parameter
\(\Delta \), the thickness of the bands of plastic deformations is reduced independently from the spatial discretization except if the thickness of this band is smaller than the element size. Hence, a low localization parameter leads to an increase in the plastic deformations in the elements within the shear bands. Furthermore, the value of this rate-independent parameter has to be chosen in line with the selected viscosity
\(\eta \) in each specific application. For this regularized material model, the resulting plastic dissipation potential is modified accordingly to
$$\begin{aligned} \psi ^{\mathrm {p}}=\frac{y_0}{\delta }\,[1-\exp (-\delta \, \alpha )] + (y_{\infty }-y_0) \,\left[ \alpha +\frac{\exp (-h^{\mathrm {exp}}\,\alpha )-1}{h^{\mathrm {exp}}}\right] +\frac{1}{2}\,h^{\mathrm {lin}}\,\alpha ^2. \end{aligned}$$
(27)
Note that
\(\delta \) has to be small enough to keep the convexity of the plastic dissipation potential
\(\psi ^{\mathrm {p}}\) for the occurring range of the equivalent plastic strains
\(\alpha \). Due to the viscosity, the criterion for ductile crack propagation is extended through the modified imposed energy
$$\begin{aligned} -\Delta U_K=\int \limits _{\Omega _K}\left( \psi ^{\mathrm {e}}+\psi ^{\mathrm {p}} +\int \limits _t {{{\mathcal {D}}}}^{\mathrm {vis}} \,\mathrm {d}t \right) \,\mathrm {d}V \end{aligned}$$
(28)
which additionally takes into account the viscous dissipation
\({{{\mathcal {D}}}}^{\mathrm {vis}}=\eta \,{\dot{\alpha }}^2\) in order to capture the full dissipation of the viscoplastic part of this material model. In order to achieve results comparable to experiments, all material parameters have to be fitted again considering the two additional material parameters
\(\eta \) and
\(\delta \). Note that the special type of viscous regularization considered here can be exchanged by any kind of regularization, including alternative viscous formulations, gradient enhanced formulations or energy-relaxed approaches. The regularization is only required in order to avoid mesh dependency even before the crack evolves. However, it is of course not sufficient to guarantee mesh independence during crack evolution and thus, a suitable crack propagation approach—here the eigenerosion algorithm—has to be considered.
2.3.2 Compression–tension asymmetry
For the simulation of cyclic fatigue, a previous crack may close again. Then, it is assumed that the two sides of a crack shall only transfer compression forces due to their contact and no tensile forces. To account for this, the formulation has to be modified in order to only partly erode elements, i.e., the tensile part of the energies, and keep the compressive part in the cracked elements. Therefore, the split of the elastic energy density into a tensile and a compressive part similar to Miehe et al. [
20] and Ambati et al. [
2] is considered as
$$\begin{aligned} \psi ^{\mathrm {e}}=(1-D)\,\psi ^{\mathrm {e}}_++\psi ^{\mathrm {e}}_-. \end{aligned}$$
(29)
Here, the variable
D is binary and indicates if an element is either intact (
\(D=0\)) or eroded (
\(D=1\)). This variable
D should not be confused with continuous damage variables commonly used in continuum damage mechanics approaches. Considering this split, the tensile and compressive parts of the energy density become
$$\begin{aligned} \psi ^{\mathrm {e}}_{\pm }=\frac{\lambda }{2} \left\langle \mathrm {tr}(\varvec{\varepsilon }^{\mathrm {e}})\right\rangle _{\pm }^2+\mu \, \varvec{\varepsilon }^{\mathrm {e}}_{\pm }\cdot \varvec{\varepsilon }^{\mathrm {e}}_{\pm } \end{aligned}$$
(30)
with the Lamé parameter
\(\lambda =\kappa -2/3\,\mu \). Herein, the strain tensor is assumed to be split into
$$\begin{aligned} \varvec{\varepsilon }^{\mathrm {e}}=\varvec{\varepsilon }^{\mathrm {e}}_++\varvec{\varepsilon }^{\mathrm {e}}_- \quad \mathrm {with}\quad \varvec{\varepsilon }^{\mathrm {e}}_{\pm }=\sum \limits _{i=1}^3 \left\langle \epsilon _i^{\mathrm {e}} \right\rangle _{\pm }\,{\varvec{n}}_i \otimes {\varvec{n}}_i \end{aligned}$$
(31)
as in Miehe et al. [
20]. Accordingly, the Kirchhoff stress tensor results in
$$\begin{aligned} \varvec{\tau }=(1-D)\,\varvec{\tau }_++\varvec{\tau }_- \quad \mathrm {with}\quad \varvec{\tau }_{\pm }=\lambda \,\left\langle \mathrm {tr}(\varvec{\varepsilon }^{\mathrm {e}})\right\rangle _{\pm }\,{\varvec{I}}+2\,\mu \, \varvec{\varepsilon }^{\mathrm {e}}_{\pm }. \end{aligned}$$
(32)
For intact elements, we assume the elastic energy density
\(\psi ^{\mathrm {e}}=\psi ^{\mathrm {e}}_++\psi ^{\mathrm {e}}_-\), the stress
\(\varvec{\tau }=\varvec{\tau }_++\varvec{\tau }_-\), and the material tangent
\({\mathbb {C}}={\mathbb {C}}_++{\mathbb {C}}_-\) needed for the Newton–Raphson scheme, respectively, while for eroded elements, only the compressive parts remain, i.e.,
\(\psi ^{\mathrm {e}}=\psi ^{\mathrm {e}}_-\),
\(\varvec{\tau }=\varvec{\tau }_-\), and
\({\mathbb {C}}={\mathbb {C}}_-\) holds. Hence, the total potential energy to be considered in the fracture criterion becomes
$$\begin{aligned} -\Delta U_K={\int }_{\Omega _K}\left( \psi ^{\mathrm {e}}_++\psi ^{\mathrm {p}} + \int \limits _t {{{\mathcal {D}}}}^{\mathrm {vis}} \,\mathrm {d}t \right) \,\mathrm {d}V. \end{aligned}$$
(33)
The resulting criterion for crack propagation ensures that elements would rather erode under tension loads and than under compression and consistently represents the energy that is dissipated into plasticity and crack propagation. Herein, the term
\({{{\mathcal {D}}}}^{\mathrm {vis}} \) is set to zero for elasto-plastic materials. Note that in eroded elements, no further plastic deformation is considered in order to avoid an unphysical response under further tension loads. The algorithmic implementation of the material law itself including the material tangent
\({\mathbb {C}}\) follows Miehe and Lambrecht [
18], Miehe et al. [
20].