Public-material validation: source gate and execution plan

Status: source audited; no calibrated Alloy 617 simulation or held-out prediction has passed. Do not confuse the existing synthetic/analytical verification suite with experimental material validation.

Selected public source

ORNL/TM-2022/2517 (July 2022) is a useful candidate: Alloy 617 at 950 °C, tensile holds, ordinary strain control and separately identified elastic follow-up. PDF pages 15 and 19 (printed pages 5 and 9) were visually checked, not only text-extracted. Table 2 gives A-6/F-5: strain range 1%, hold 180 s, lives 350/450 cycles; A-13: 1%, 600 s, 300 cycles. These are ranges, not amplitudes. Keep repeats and test-family labels separate. Table 4 and Equation 8 include negative softening parameters and damage-dependent evolution. They cannot be copied into AgentFEM's current nonnegative-saturation Chaboche model as an equivalent law. The report also separates cavity-growth and time-fraction assumptions.

ASTM E2714-26's public catalogue identifies the creep-fatigue test-method family. The complete normative text has not been reviewed; neither a dataset audit nor a passing calculation establishes ASTM or component design-code compliance.

Admission decisions

1. Retain the source version, specimen/test-family identifiers, temperature, strain range versus amplitude, rate, hold direction/duration, failure definition and runouts. Reconcile duplicated records against their original reports before joining datasets; never silently overwrite a discrepancy.
2. Extract stress–strain loops and hold-relaxation histories with explicit digitization uncertainty, or obtain numerical author data. A life table alone cannot identify a unique deformation model.
3. Freeze a compatible local law and its validity domain before fitting. If cyclic softening is needed, use the existing external material boundary and verify its tangent/state/work first. Do not remove hardening-model safeguards to make coefficients fit or add a material-specific core solver.
4. Split by complete loading conditions/specimens, not time samples from the same curve. Reserve at least one strain/hold condition for prediction. Retain replicate scatter; do not tune against the held-out result.
5. Run material point → homogeneous structure → hold/reversal → restart. Compare loop stresses, relaxation shape and strain partition before life. Use the new co-located history channel, not unrelated field extrema.
6. Only then assess a declared rupture/interaction model within its calibration domain. Separate numerical error, fit error, prediction error and design acceptance. Publish failures and remaining parameter/data uncertainty.

Present executable evidence and next composition

Existing Norton relaxation, elastic follow-up and manufactured nonuniform creep/plastic/thermal fields remain the numerical oracles. The NAFEMS R0027 thick-cylinder gate is a separate public structural reference. None supplies an Alloy 617 fit.

The first composition in this slice is nonuniform accepted temperature plus co-located histories, with existing regional pure creep and homogeneous joint creep/plasticity tested separately. The serial thermal v1 restart identity bug was found by this acceptance work and corrected. New regional joint plasticity, temperature-dependent generic finite-strain user materials, contact and MPI admission remain separate tasks. Priority after source reconciliation: the smallest real local law and its temperature/state contract, not all four combinations at once.