This PR adds viscoplastic regularisation to the rheology model of LaMEM (red item):
Previous, we could do plasticity regularisation already by specifying the eta_st viscosity per phase, which changes the minimum viscosity from its global value (eta_min), to a phase-wise value.
Another possibility is by amending the plasticity yield stress directly:
where $\tauy$ is a yield stress, $\eta{\textrm{vp}}$ a regularisation viscosity, and $\dot{\varepsilon}_{II}^{\textrm{pl}}$ the plastic strainrate. This regularisation was used by Duretz et al. 2018 , among others.
This PR adds a new, phase-wise, material parameter to specify this (set to zero by default):
eta_vp = 1e19
Comparison of regularisation methods
The newly added test /tests/t4_Loc/localization_eta_vp_reg.dat was used to compare the different formulations, with $\eta{vp}=10^{21}$ Pas, or $\eta{st}=10^{21}$ Pas. Looking at the number of SNES iterations shows that both regularisations reduce the number of iterations, but that the viscoplastic method is slightly more efficient:
A table with timings (on an M2 MacBook Pro):
method
total time [s]
no regularisation
392
eta_st
82
eta_vp
57
A comparison of the 2nd invariant of the strain rate at the end of the simulation for different resolutions for a case that uses the new method using eta_vp=1e21.
This PR adds viscoplastic regularisation to the rheology model of LaMEM (red item):
Previous, we could do plasticity regularisation already by specifying the
eta_st
viscosity per phase, which changes the minimum viscosity from its global value (eta_min
), to a phase-wise value. Another possibility is by amending the plasticity yield stress directly:where $\tauy$ is a yield stress, $\eta{\textrm{vp}}$ a regularisation viscosity, and $\dot{\varepsilon}_{II}^{\textrm{pl}}$ the plastic strainrate. This regularisation was used by Duretz et al. 2018 , among others.
This PR adds a new, phase-wise, material parameter to specify this (set to zero by default):
Comparison of regularisation methods The newly added test
/tests/t4_Loc/localization_eta_vp_reg.dat
was used to compare the different formulations, with $\eta{vp}=10^{21}$ Pas, or $\eta{st}=10^{21}$ Pas. Looking at the number of SNES iterations shows that both regularisations reduce the number of iterations, but that the viscoplastic method is slightly more efficient:A comparison of the 2nd invariant of the strain rate at the end of the simulation for different resolutions for a case that uses the new method using
eta_vp=1e21
.