graintrace.taylor#

Uniaxial Taylor crystal-plasticity forward + calibration model (NEML2 v3).

A mixed-control NEML2 NonlinearSystem (cpfe_base/neml2_cpfe_calibration.i) is wrapped by neml2.pyzag.NEML2PyzagModel and driven through pyzag.nonlinear.solve_adjoint() for analytic parameter gradients.

class graintrace.taylor.UniaxialTaylorModel(factory, axial_index=2, nchunk=5, rtol=1e-06, atol=1e-08, linesearch_iter=5)[source]#

Bases: Module

Differentiable uniaxial Taylor forward via a NEML2 mixed-control eq_sys.

Wraps a neml2.pyzag.NEML2PyzagModel and integrates the strain history with pyzag.nonlinear.solve_adjoint. Uniaxial loading is expressed with NEML2’s MixedControlSetup: control selects the axial component as strain-controlled (prescribed = rate); the remaining components are stress-free, and a global constraint forces target_cauchy_stress to equal the grain-mean per-crystal stress.

Parameters:

  • factory (NEML2PyzagModel)

  • axial_index (int)

  • nchunk (int)

  • rtol (float)

  • atol (float)

  • linesearch_iter (int)

forward(orientations, strain, rate, initial_strains=None, return_state=False)[source]#

Integrate the uniaxial history and return the macro stress trajectory.

Parameters:

  • orientations (Tensor) – (n_grains, 3) NEML2 MRP orientations.

  • strain (Tensor) – (ntime,) axial strain grid (mm/mm), typically starting at 0.

  • rate (float) – assumed strain rate (1/s); t = strain / rate.

  • initial_strains (Tensor | None) – optional (n_grains, 6) initial elastic strain.

  • return_state (bool) – also return per-grain elastic strain history.

Returns:

xx, yy, zz, yz, xz, xy); if return_state, also (ntime, n_grains, 6) elastic strain.

Return type:

macro cauchy stress (ntime, 6) (SR2 Voigt

class graintrace.taylor.TaylorModel(neml2_path, neml2_model_name='model_with_stress', axial_index=2, assumed_rate=0.0001, npoints=500, nchunk=5, equation_system='eq_sys', exclude_parameters=None, compile=False, device='cpu')[source]#

Bases: BaseMaterialApproximationModel

Taylor calibration model backed by the NEML2 v3 + pyzag adjoint engine.

  • Loads and preprocesses experimental data (per-stress-level CSVs).

  • Manages the six calibration parameters as NEML2/torch parameters.

  • Runs differentiable uniaxial stress-strain simulations for optimization.

Parameters:
DEFAULT_EXCLUDE_PARAMETERS = ['slip_rule_gamma0', 'slip_rule_n']#
get_params()[source]#

Current values of the six calibration parameters (detached).

Return type:

Tensor

set_params(values)[source]#

Assign the six calibration parameters from a 1-D iterable.

Return type:

None

load_experiment_data(data_dir, strain_stress_file, full_field_strain_units=None, strain_stress_file_units=None, straintype='eFab', npoints=None, max_strain=None, max_stress=None, n_grains=None, seed=0)[source]#

Load experiment data (macroscopic stress-strain plus per-file CSVs).

Parameters:

  • data_dir (str)

  • strain_stress_file (str)

  • full_field_strain_units (str | None)

  • strain_stress_file_units (str | None)

  • straintype (str)

  • npoints (int | None)

  • max_strain (float | None)

  • max_stress (float | None)

  • n_grains (int | None)

  • seed (int)

simulate(params, d=None, assumed_rate=None, experiment_data=None, return_state=False, initial_strains=None)[source]#

Run a differentiable uniaxial stress-strain simulation.

Parameters:

  • params – optional 1-D tensor of the six calibration values. If given, they are assigned to the factory (no grad) before the solve; use this for plotting a specific parameter set. Pass None to run with the factory’s current (possibly optimizer-owned) parameters so gradients flow (used during calibration).

  • d (Tensor | None) – ignored (kept for API compatibility; the axial direction is set by self.axial_index).

  • assumed_rate (float | None) – strain rate; defaults to self.assumed_rate.

  • experiment_data – dict from load_experiment_data().

  • return_state (bool) – also return the per-grain elastic strain history.

  • initial_strains – optional (n_grains, 6) initial elastic strain.

Returns:

stress_hist (npoints, 6), or (stress_hist, state_hist) where state_hist is (npoints, n_grains, 6) elastic strain.