P92 long-cycle response: retain discrepancies, do not refit¶
Scope and reproducibility¶
Reuse the already executed Radau/BDF campaigns of the fixed published equations.
Read five experimental peak-stress square markers in each Figure 10 panel of
Metals 10 (2020) 307, printed page 11.
The pinned PDF hash, exact Poppler crop command, axis ticks and marker pixels
are retained in evidence/creep/p92-cyclic-figure-points.json. A five-pixel
coordinate sensitivity covers manual reading of the dense marker band; it is
not a confidence interval. Only peak stress is sampled, not blue relaxation
markers or red predicted curves. No experimental raw-data precision is claimed.
python tools/p92_published_recovery_reference.py --published-campaign --output campaign.json
python tools/p92_cyclic_figure_screen.py campaign.json
Reuse an existing campaign rather than launching the first command again. The plot's horizontal coordinate is cycle divided by observed half-life. The observed lives 1303 and 675 are normalization inputs, not model outputs. Fractional cycle coordinates use linear interpolation of the computed peaks; reading intervals must lie entirely within the run, with no endpoint clamping. Seven inexpensive tests verify axes, scope and inconsistent input rejection.
Results with unchanged coefficients¶
| Tensile dwell at 600 C, range .008 | Peak sample mean absolute relative error | Maximum absolute error |
|---|---|---|
| 60 s | 4.03% | 15.32 MPa |
| 600 s | 1.70% | 9.19 MPa |
Several discrepancy intervals exclude zero. Radau and BDF predictions differ by at most 5.26e-6 MPa at these sample positions, much smaller than experimental figure reading sensitivity or response discrepancy. Tightening numerical tolerances is not a demonstrated remedy. Do not tune to these points and then reuse them as validation. Sparse long-cycle agreement complements the existing first-cycle dwell screen; it does not establish a full cyclic experimental history, multiaxial response, fracture or engineering allowable life.
Life criterion and next data gate¶
Figure 2's 10% peak-stress loss is relative to the extrapolated intermediate softening trend shown in the schematic, not 10% below the first-cycle peak. The rapid initial cyclic softening must not be mistaken for a failure clock. The published reference does not contain a validated crack-initiation law.
NIMS data sheet 48B has authoritative T92/P92 creep-rupture data (catalogue), but the old public PDF link now redirects to a service migration notice. The current CDS access page requires registration/application and sets reuse conditions. No authenticated data were acquired or copied. Obtain authorized data with heat identity, temperature, stress and rupture definition; do not substitute another P92 heat or optimized Grade 92 silently.
Remaining engineering gate: mutually compatible fatigue, creep and interaction evidence, independent specimens/loads and a justified uncertainty allowance. The software composition and research screens can advance without pretending that missing experimental qualification is a coding defect.