Isotropic Phase Field Softening Material

Constitutive Law

In phase field fraction modeling, sharp crack surfaces are replaced by diffuse phase fields. The growth of these diffuse cracks is controlled my energy flow through the material and the material's fracture toughness. This material implements phase field fracture methods using methods described in Miehe and extends it in a few areas.

Phase Field Methods

In phase field fracture model of small-strain elastic materials, the total strain energy is partitioned into damaging, [math]\displaystyle{ \Psi^{(+)} }[/math], and nondamaging, [math]\displaystyle{ \Psi^{(-)} }[/math], terms by

     [math]\displaystyle{ \Psi = g(\phi)\Psi^{(+)} + \Psi^{(-)} }[/math]

where [math]\displaystyle{ g(\phi) }[/math] is a damage or softening law that depends on damage parameter [math]\displaystyle{ \phi }[/math] given by a phase value tracked on each material point that varied from 0 for undamaged to 1 for complete damage.

Using a viscous regularization (see Miehe ), the phase field evolves by the equation

     [math]\displaystyle{ \eta \frac{d\phi}{dt} = G_c \ell \nabla^2\phi - \frac{G_c}{\ell}\phi - g'(\phi)\mathcal{H} }[/math]

where [math]\displaystyle{ G_c }[/math] is toughness, [math]\displaystyle{ \ell }[/math] is a length scale (describing width of diffuse cracks), and [math]\displaystyle{ \mathcal{H} }[/math] is history variable given by [math]\displaystyle{ \max(\Psi^{(+)}) }[/math]. This equation is a diffusion equation that describes evolution of damage driven by damaging energy. The first two terms lead to diffusive cracks that decay exponentially from complete damage state with exponential decay distance determined by [math]\displaystyle{ \ell }[/math]. The last term is a source term to causes crack propagation. In other words, crack propagation of controlled by the methods used to partition energy into [math]\displaystyle{ \Psi^{(+)} }[/math] and [math]\displaystyle{ \Psi^{(-)} }[/math].

Simulations using viscous regularization must couple solution of phase field diffusion equation to the material's mechanics calculations. For that to work, all simulations must include a Diffusion command and choose (style) of "fracture. If needed, you can set phase value on the grid using the "fracture" option to define initial damage states. Although you can set phase flux on particle surfaces, such conditions may have few applications with physical meanings. Note that MPM solutions of transport equations can develop oscillations that prevent output of viable phase field descriptions of cracks. This problem can be solved bad using FMPM(k) for transport tasks with the PeriodicXPIC Custom Task

Energy Partitioning

Final results depend on the method chosen to partition energy into [math]\displaystyle{ \Psi^{(+)} }[/math] and [math]\displaystyle{ \Psi^{(-)} }[/math]. A common method for isotropic materials is to use an eigenstrain analysis and partition energy such that [math]\displaystyle{ \Psi^{(+)} }[/math] depends only on tensile principle strains and tensile trace of the strain tensor . The physical interpretation is that only tensile stress cause fracture while compression does not. A weakness of this approach is that shear loading may not respond as expected.

An alternative methods is to partition energy in the pressure and deviatoric strains. Now [math]\displaystyle{ \Psi^{(+)} }[/math] depnds on hydrostatic tension (negative pressure) and deviatoric strain while [math]\displaystyle{ \Psi^{(-)} }[/math] depends only on positive pressure and does not promote fracture.

This material supports both these options and can be selected with the partition material property.

Phase Field Damage Law

Final results also depend on the damage (or softening) law used for the phase field material. The vast majority of phase field fracture papers set [math]\displaystyle{ g(\phi) = (1-\phi)^2 }[/math] under the misunderstood concept that [math]\displaystyle{ g'(1) }[/math] needs to be zero to stop dissipating energy or to prevent [math]\displaystyle{ \phi }[/math] from exceeding 1. In dynamic codes, however, any [math]\displaystyle{ g(\phi) }[/math] can be used provided [math]\displaystyle{ g(\phi)=1 }[/math], [math]\displaystyle{ g(1)=k }[/math], and [math]\displaystyle{ \phi }[/math] is prevented from ever decreasing or ever exceeding 1. Here [math]\displaystyle{ k }[/math] is a small (much less then 1) stability factor to prevent stiffness from dropping to zero.

Because [math]\displaystyle{ g(\phi) }[/math] appears in total energy, it implicitly corresponds to stress-strain behavior of the material as a function of damage. The universal us of [math]\displaystyle{ g(\phi) = (1-\phi)^2 }[/math] implies all materials have the same stress-strain law. Reverse engineering that law further implies that all material are ductile. If phase field materials are expected to apply to a range of materials, the modeling should admit other laws including laws with varying ductility. The available [math]\displaystyle{ g(\phi) }[/math] are selected using the gd material property and, depending on law, using the garg material property to pick and argument [math]\displaystyle{ \chi }[/math]. The current damage laws (by gd setting) are:

0. Quadratic Law: this law generalizes the common quadratic into a new function with an argument:
        [math]\displaystyle{ g(\phi) = (1-\phi)(1-(2\chi-1)\phi) }[/math]
   where [math]\displaystyle{ 0\le\chi\le1 }[/math] varies the material's ductility (higher is more ductile). Choosing [math]\displaystyle{ \chi=1 }[/math] is the standard (and very ductile) law used in the literature. When partitioning energy using an eigenstrain analysis, this law is unstable for [math]\displaystyle{ \chi }[/math] less than about 0.5 (which is when [math]\displaystyle{ g(\phi) }[/math] becomes linear.
1. Exponential Law: a trunctated exponential functions to go from 1 to 0 and [math]\displaystyle{ \phi }[/math] goes from 0 to 1:
        [math]\displaystyle{ g(\phi) = \frac{e^{-\chi\phi} - e^{-\chi}}{1-e^{-\chi}} }[/math]
   where [math]\displaystyle{ \chi\gt0 }[/math] increases the ductility. In the limit as [math]\displaystyle{ \chi\to0 }[/math], this law is linear in [math]\displaystyle{ \phi }[/math] but cannot be modeled using this law. To use a linear law, pick quadratic law with [math]\displaystyle{ \chi=0.5 }[/math].
2. Linear Softening Law: if the energy partitioning is pressure and deviatoric stress, one can reverse engineer a linear softening law where stress in linear elastic until failure and then decays linearly to zero as damage increases:
        [math]\displaystyle{ g(\phi) = 1 + \frac{\chi\phi^2}{2} - \phi\sqrt{1+\chi + \frac{\chi^2\phi^2}{4}} }[/math]
   where [math]\displaystyle{ \chi\gt0 }[/math] increases the ductility. The peak stress to initiate damage (if loaded in uniaxial tension) is
        [math]\displaystyle{ \sigma_i = \sqrt{\frac{G_c E}{\ell \chi}} }[/math]
   The area under a linear softening law can be divided into elastic area up to the peak ([math]\displaystyle{ A_1 }[/math]) and energy dissipated as post-peak stress decays to zero. For phase field material using this law
        [math]\displaystyle{ A_2 = \chi A_1 = \frac{G_c}{2\ell} }[/math]
   Note that [math]\displaystyle{ \chi\to0 }[/math] corresponds to [math]\displaystyle{ A_2\to0 }[/math] for an elastic-brittle material. This behavior, however, cannot be modeling with this law because the initiation stress becomes infinite. Although linear softening response was determined by assuming pressure and deviatoric partitioning of energy, the material can use this law even when partitioning by eigenstrain analysis. Simulations show this option provides approximate linear softening response.

Stability Factor

Some phase field literature suggest that letting [math]\displaystyle{ g(\phi) }[/math] reach zero when damage reaches one can cause instabilities. One option to avoid such instability is to revise and damage law to be

     [math]\displaystyle{ g^*(\phi) = (1-k)g(\phi) + k }[/math]

such that [math]\displaystyle{ g^*(1)=k }[/math] where [math]\displaystyle{ k\lt \lt 1 }[/math].

and stability factor [math]\displaystyle{ k }[/math] (set with stability material property).

Material Properties

The isotropic variational mechanics model using a single energy release rate that scales evolution of damage. The critical energy release rate is enter using the base material JIc property. The other needed material properties are as follows:

Property	Description	Units	Default
(Isotropic Properties)	Enter all properties needed to define the underlying isotropic material response	varies	varies
ell	Length scale parameter used in variational fracture mechanics	length units	none
viscosity	Viscosity to use when solving coupled phase field evolution in a diffusion tasks	viscosity units	none
gd	Softening law with options 0 = quadratic, 1 = exponential, 2 = linear softening	none	0
garg	An optional argument for use within the softening law. If not provided, default values depend on gd and are 1, 3, and 4, for gd = 0, 1, or 2, respectively	none	varies
stability	A stability factor thought to stabilize post-failure analysis	none	0
partition	Chose the method used to partition energy into energy that causes fracture and energy that does not cause fracture. The options are 0 = using eigenstrain analysis and 1 = divide into pressure and deviatoric strains	none	1
(other)	Properties common to all materials	varies	varies

The results in Miehe correspond to gd = 1, garg = 1, and partition = 0. These choices give poor results in some problems. This material has extension that can explore different phase field options.

History Variables

This material stores several history variables that track the extent of the damage and evolution of the phase field:

1. Maximum energy history term that provides source terms for phase field evolution
2. Damage state equation to 0 if not failed and 1 if failure (i.e., phase value has reached 1)
3. Current phase field value
4. Change in phase field since the last time step. It is used in constitutive law modeled and is scaled by 0.5 when using USAVG method.