Softening Law Implementation
Introduction
Softening Laws are used to implement damage mechanics in softening materials.
Softening Law Definition
Implementation of softening laws starts by defining a model that gives strength as a function damage parameter δ. The law is defined in the form
[math]\displaystyle{ F(\delta,s) = \sigma_{init}f(\delta,s) }[/math]
Because the initiation stress is provided by the choosen initiation law, the softening law depends only on dimensionless [math]\displaystyle{ f(\delta,s) }[/math] that is equal to 1 when [math]\displaystyle{ \delta=0 }[/math] and [math]\displaystyle{ s }[/math] is a scaling function to connecting total energy released to particle volume and crack orientation:
[math]\displaystyle{ s = {A_c \over V_p \sigma_{init}} }[/math]
Note that although softening laws are traditionally monotonically decreasing function, that is not a thermodynamic requirement. The only limitation on softening laws is that is derivative must satisfy:
[math]\displaystyle{ \frac{\partial f(\delta,s)}{\partial \delta} \le \frac{f(\delta,s)}{\delta} }[/math]
If you plot [math]\displaystyle{ f(\delta,s) }[/math] as a function of [math]\displaystyle{ \delta }[/math] means no line from [math]\displaystyle{ \bigl(\delta,f(\delta,s)\bigr) }[/math] back to the origin can cross over the [math]\displaystyle{ f(\delta,s) }[/math]. Thermodynamically, it means that dissipated energy is nonnegative and that stiffness is monotonically decreasing. The function [math]\displaystyle{ f(\delta,s) }[/math], however, my increase to a peak before decreasing to failure.
Writing Softening Law Code
The first steps are to create source files for the new softening law class, to give the law a unique name and number, and to allow the main NairnMPM code to instantiate this softening law when it is needed.
Class Source Code
The parent class for all softening laws are in NewMaterial.cpp and NewMaterial.hpp files. Normally, you will create new files that subclass this class (which can be done by starting with new files or duplicating files from an existing softening law). You can pick any unique name for the softening law source files.
Editing Core Code
Almost all coding will be done in the new softening laws class files, but for that law to be recognized as an option in NairnMPM, one source file in the core code has to be edited first. These should be the only changes needed outside the new softening laws's class files.
- Include the new softening law's header file at the top of NairnMPM/src/Materials/MaterialBaseMPM.cpp:
#include "Materials/MySofteningLaw.hpp"
or the appropriate relative path to the new law's header file.
- In void MaterialBase::SetSofteningLaw(char *lawName,int mode), instantiate you new law based on user input of a name or a number. Pick any name and number that does not conflict with currently implemented softening laws. The new code will look like:
else if(strcmp(lawName,"MyFavoriteLaw")==0 || strcmp(lawName,"9")==0) { sLaw = new MySofteningLaw(); }
Implementing Required Methods
Once source code is created and made available to core code, the next steps are to implement all the required methods for softening laws. The required methods are:
- const char *MySofteningLaw::GetSofteningLawName(void)
Return a string with name for the law (to be displayed in output files). It should be unique name among all implemented softening laws. - double MySofteningLaw::GetFFxn(double delta,double s) const
Return [math]\displaystyle{ f(\delta,s) }[/math]. - double MySofteningLaw::GetDeltaMax(double s) const
Return [math]\displaystyle{ \delta_{max} }[/math] or the value of [math]\displaystyle{ \delta }[/math] for which [math]\displaystyle{ f(\delta_{max},s)=0 }[/math]. Some softening laws (like Exponential Softening) may never reach zero, but are advised to implement a critical value where [math]\displaystyle{ f(\delta_{max},s)={\rm min} }[/math] or where the softening law decreases to some small value (specified in min). - double MySofteningLaw::GetFpFxn(double delta,double gScaling) const
Return derivative of the softening law or [math]\displaystyle{ \partial f(\delta,s)/\partial\delta }[/math]. - double MySofteningLaw::GetGToDelta(double delta,double gScaling) const
Return normalized energy released for material point that has evolved to [math]\displaystyle{ \delta }[/math]. It is found from:
[math]\displaystyle{ \Omega(\delta,s) = \int_0^\delta f(u,s) du - \frac{1}{2}\delta f(\delta,s) }[/math] - double MySofteningLaw::GetGoverGc(double delta,double gScaling) const
Return ratio of energy released to the total toughness. It is defined by
[math]\displaystyle{ \frac{\Omega(\delta,s)}{\Omega(\delta_{max},s)} }[/math]
Although this result could be found from
GetGToDelta(delta,s)/GetGToDelta(GetDeltaMax(s),s)
that approach is less efficient (probably for any softening law). As a result, this implementation requires implementation of a separate method to do the calculation. - double MySofteningLaw::GetMaxSlope(double gScaling) const
Return the maximum value for the negative slope or
[math]\displaystyle{ \max\left(-\frac{\partial f(\delta,s)}{\partial\delta}\right) }[/math]
This calculation is used to verify cell sizes are sufficently small for stable calculations. - double MySofteningLaw::GetEtaStability(void) const
Return dimensionless stability factor, [math]\displaystyle{ \eta }[/math] [1]. It is given by s*Gc*GetMaxSlope(s), but the scaling factor s is not provided as a parameter. None of the current laws need s to find [math]\displaystyle{ \eta }[/math]. If that ever changes for a future softening law, this method will change to include s as an argument. - double MySofteningLaw::GetPhiFxn(double delta,double s) const
Return the energy dissipation function [1] defined by
[math]\displaystyle{ \varphi(\delta,s) = f(\delta,s) - \delta\frac{\partial f(\delta,s)}{\partial\delta} }[/math]
Note that the only requirement for valid softening laws is that [math]\displaystyle{ \varphi(\delta,s)\ge0 }[/math] for all [math]\displaystyle{ \delta }[/math]. - double MySofteningLaw::GetRdFxn(double delta,double s,double ei) const
Return function that relates evolution of damage variable [math]\displaystyle{ D }[/math] to damage variable [math]\displaystyle{ \delta }[/math] defined by:
[math]\displaystyle{ \frac{dD}{d\delta} = \frac{\varepsilon_i\varphi(\delta,s)}{\bigl(\delta + \varepsilon_i f(\delta,s)\bigr)^2} }[/math]
where [math]\displaystyle{ \varepsilon_i }[/math] is initiation strain provided in a parameter.
Implementing Option Methods
The required methods in the previous section are sufficient to implement a new softening law. The parent class, SofteningLaw.cpp implements calculations that work for any softening law that provides these methods. Some of those calculations, however, use numerical solutions to key problems. If any specific law can find closed-form solutions instead, implementing one or more of the following options methods can improve calculation speed when uses you custom softening law. you can refer to Linear Softening or Exponential Softening coding to see examples of overriding these methods.
- double MySofteningLaw::GetDeltaFromDamage(double d,double s,double ei,double deltaGuess)
The damage variable [math]\displaystyle{ D }[/math] is related to the damage variable [math]\displaystyle{ \delta }[/math] by [1]:
[math]\displaystyle{ D = \frac{\delta}{\delta + \varepsilon_i f(\delta,s)} }[/math]
The purpose of this function is to invert this expression to find [math]\displaystyle{ D }[/math] in terms of [math]\displaystyle{ \delta }[/math] and s. The general code inverts it numerically using Newton's method. If it can be inverted analytical for your softening law, override this method and implement the analytical solution. - double MySofteningLaw::GetDDelta(double de,double ei,double delta,double D,double s) const
A core calculation for finite amount of damage evolution is to solve:
[math]\displaystyle{ d\varepsilon = d\delta + \varepsilon_i \bigl(f(\delta+d\delta,s) - f(\delta,s)\bigr) }[/math] for [math]\displaystyle{ d\delta }[/math] where [math]\displaystyle{ d\varepsilon }[/math]=de is an input strain increment. The generic code solves this numerically using Newton's method with bracketing. The initial brackets are:
[math]\displaystyle{ D\varepsilon \le d\delta \le d\varepsilon + \varepsilon_if(\delta,s) }[/math]
The lower limit is found because [math]\displaystyle{ \varphi(\delta,s)\ge0 }[/math]. The upper limit is if failure occurs such that [math]\displaystyle{ f(\delta+d\delta,s)=0 }[/math]. If [math]\displaystyle{ d\delta_n }[/math] is the Newton's method sequence approaching the solution, the next increment by Newton's method is derived to be:
[math]\displaystyle{ d\delta_{n+1} = d\delta_n - {{d\delta-d\varepsilon\over \varepsilon_0f(\delta,s)} + {f(\delta+d\delta,s)\over f(\delta,s)}-1\over {1\over \varepsilon_0f(\delta,s)} + {f'(\delta+d\delta,s)\over f(\delta,s)}} }[/math]
When the next increment is outside the current brackets, a custom bracketing calculation is used to make the solution always stable.
Although generic code can solve this differential equation for any softening law with provided properties listed above, some softening laws may have more efficient solutions. Any new softening laws can override this method to return a closed-form solution or a more efficient numerical solution. - double MySofteningLaw::GetRelFFxn(double delta,double x,double gScaling) const
This method returns [math]\displaystyle{ f(\delta+d\delta,s)/f(\delta,s) }[/math] that is needed in the Newton's method calculations described for GetDDelta(). If that method is overridden, this method provides no function. If it is not overridden, however, this method might provide some benefit (although to date, is has only been useful for Exponential Softening where finding the ratio is more efficient than evaluating the numerator and denominator separately). - double MySofteningLaw::GetRelFFxn(double delta,double x,double gScaling) const
This method returns [math]\displaystyle{ [\partial f(\delta+d\delta,s)/\partial \delta]/f(\delta,s) }[/math] that is needed in the Newton's method calculations described for GetDDelta(). If that method is overridden, this method provides no function. If it is not overridden, however, this method might provide some benefit (although to date, is has only been useful for Exponential Softening where finding the ratio is more efficient than evaluating numerator and denominator separately). - double MySofteningLaw::GetDDeltaElastic(double delta,double sigmaAlpha,double s,double D,double E1md) const
This methods solves for elastic change in damage variable [math]\displaystyle{ \delta }[/math] when another parameter (such as pressure) affects the damage evolution (it implements some unpublished damage mechanics methods). The numerical task is to solve
[math]\displaystyle{ {\rm E1md*}(\delta +d\delta_e) = {\rm D*sigmaAlpha*}f(\delta+d\delta_e,s) }[/math]
for [math]\displaystyle{ d\delta_e }[/math]. The generic code solves this equation numerically. Any new softening laws can override this method to return a closed-form solution or a more-efficient numerical solution.
Including Additional Softening Law Properties
The parent class SofteningLaw.cpp mains two softening law properties for toughness, Gc, and minimum value for failure, min, used by some laws. New custom laws are likely to need additional parameters. Because softening laws are dimensions, new parameters should be dimensionless as well. To allow users to input the laws, the new class will need the following overrides:
MySofteningLaw::MySofteningLaw() : SofteningLaw()
{ myVariable = 0.;
}
char *MySofteningLaw::InputSofteningProperty(char *xName,int &input,double &gScaling)
{ if(strcmp(xName,"MyParameter")==0)
{ input=DOUBLE_NUM;
return &myVariable;
}
return SofteningLaw::InputSofteningProperty(xName,input,gScaling);
}
void MySofteningLaw::PrintSofteningProperties(double sigmainit)
{ SofteningLaw::PrintSofteningProperties(sigmainit);
MaterialBase::PrintProperty("MyParameter",myVariable,"");
cout << endl;
}
The first methods add a constructor (that was not needed before) and it initializes your laws parameter (which here is a class variable myVariable. The second method overrides method to decode input parameters. If the current one matches the name for you parameter, return a point to the class variable. If not, pass the method onto your methods superclass. The third method will print information about the value used for your parameter in the simulation output file. This override calls the parent class for its parameters and then prints this laws new variable. This scheme can be continues to add any number of parameters with any names.
To support new parameters in input XML files, you have to add multiple definitions for that parameter (because of the way softening laws properties are set). The best approach is to copy definitions for a previous softening law parameter as follows:
- Open the NairnMPM.dtd file and search for all occurrences of "-Gc". They will show up in various forms such as "I-Gc", "II-Gc", etc. (if any).
- For each occurrence of "-Gc", add a corresponding entry for your parameter (e.g., for "I-Gc", add "I-MyParameter", where you use the name used in InputSofteningProperty() above.
Compiling with make Command
To enable compiling code with the new softening law in Linux, Unix, or Mac without using XCode, the new softening law should be added the the makefile. The best approach is to search the make for for an existing softening law (such as LinearSoftening) and use all those entries at templates for adding your new softening law.