Pfc Fast Calibration

PFC Fast Calibration

Use this skill to design, explain, or implement rapid micro-parameter calibration for an improved linear parallel-bond model (LPBM / linearpbond) in PFC3D. The route is: 13 micro-parameters -> strong/weak contact grouping -> Weibull damage -> 27-run orthogonal design -> PFC3D UCS/UTS/triaxial tests -> Pearson correlation -> regression formulas -> back-solve micro-parameters from macro targets.

Parent Skill Relationship

pfc-fast-calibration is a child skill of pfc-workflow. It owns the fast-calibration specialist portion, not the full model lifecycle.

Use these handoffs:

• Parent pfc-workflow: owns full planning, case orchestration, solve campaign control, V&V, and delivery.
• Sibling pfc-contact-models: owns contact-law selection and detailed LPBM / CMAT / contact property ordering.
• Sibling pfc-servo-calibration: owns wall-servo, confining pressure, stress-control stability, and loading boundary details.
• Sibling pfc-standard-tests: owns UCS, UTS/Brazilian, triaxial, and other canonical laboratory-test command flows.
• Sibling pfc-mineral-heterogeneity: owns mineral phase grouping, multi-mineral parameter assignment, and mineral-style Weibull damage.
• Sibling pfc-fish: owns reusable FISH function refactoring, callback cleanup, histories, and object traversal helpers.
• Sibling pfc-postprocessing: owns exported curves, summary tables, and figure production after calibration runs.

If an older project mentions pfc-foundations or pfc-modeling-techniques, map them to the current split as follows:

pfc-foundations -> pfc-contact-models + pfc-basics
pfc-modeling-techniques -> pfc-servo-calibration + pfc-standard-tests + pfc-postprocessing

When To Use

Use through pfc-workflow when the task asks to:

• raise a low PFC UCS/UTS ratio from about 3-4 toward realistic rock values such as 10-30
• calibrate many LPBM micro-parameters faster than manual trial-and-error
• run or reproduce a 13-factor, 3-level, 27-run orthogonal calibration campaign
• derive macro-micro regression formulas from PFC numerical experiments
• back-solve improved LPBM micro-parameters from target E, nu, UCS, UCS/UTS, phi, c, and sigma_cd/UCS
• implement strong/weak ball-ball contact grouping plus Weibull damage
• compare traditional LPBM and improved LPBM calibration behavior

Required Inputs

Ask for these if missing:

• PFC version and dimensionality; this source route is PFC3D-oriented.
• Target specimen geometry, particle size range, porosity, and density.
• Macro target values: E, nu, UCS, UTS or UCS/UTS, phi, c, and sigma_cd/UCS if available.
• Whether the task is source reproduction, new calibration, or just skill documentation.
• Parameter bounds or whether to use the source 3-level table.
• Loading paths required: UCS, UTS, triaxial, or all.
• Random seed policy and acceptable error tolerance.

Core Method

The improved LPBM has 13 micro-parameters:

Ebar_star          bonded effective modulus
E_ratio            E_star / Ebar_star
kbar_star          bonded normal/shear stiffness ratio
sigma_c_bar        bonded tensile strength
coh_ratio          c_bar / sigma_c_bar
mu                 friction coefficient
phi_bar            bond friction angle
beta_bar_moment    moment contribution factor
Rf                 weak-contact filling ratio
beta_weibull       Weibull shape parameter
R_sigma            weak/strong strength ratio
R_E                weak/strong bond modulus ratio
R_k                bond stiffness-ratio multiplier

Macro targets:

E, nu, UCS, phi, c, sigma_cd/UCS, UCS/UTS

Default logic:

1. Build a reproducible PFC3D bonded cylinder.
2. Randomly split ball-ball contacts into strong and weak groups using Rf.
3. Assign weak contacts reduced modulus/strength by R_E, R_sigma, and R_k.
4. Apply Weibull random multipliers to bond strength and stiffness.
5. Run UCS, UTS, and triaxial tests for each orthogonal trial.
6. Extract macro values using standard strain/stress definitions.
7. Fit or reuse regression equations.
8. Back-solve micro-parameters from macro targets.
9. Re-run and iterate until target error is acceptable.

Documentation-Backed Rules

PFC command families checked through pfc-mcp or previous PFC 6.0 documentation enrichment are summarized in references/pfc-doc-notes.md.

Relevant command families:

• model random, model clean, model calm, model save, model restore
• contact cmat, contact model, contact method, contact property, contact group
• ball group, wall generate, wall attribute, ball attribute
• fish define, fish callback, fish history, history export
• measure create, measure history
• program call

Formula And Code Migration Rules

When the request asks about formulas, theory, exact regression equations, or source-code migration, load these first:

• references/formulas.md: LPBM, strong/weak grouping, Weibull, orthogonal analysis, macro extraction, Pearson correlation, and regression formulas.
• references/orthogonal-design.md: 13-factor level table, 27-run design, result table, correlation table, and calibration example.
• references/source-code-complete-pfc6.md: staged PFC 6.0-oriented source-code route from specimen generation through UTS.
• scripts/canonical/: reusable templates for orthogonal design generation, regression back-solving, improved LPBM assignment, and metric extraction.

Operating Rules

• Always state whether source regression equations are being reused or recalibrated for a new specimen.
• Treat source regression equations as initial-value predictors, not universal material laws.
• Keep strong/weak contact grouping, Weibull damage, loading tests, metric extraction, and regression analysis as separate stages.
• Fix random seeds for orthogonal comparisons; vary seeds only as a deliberate robustness study.
• Save staged models after packing, consolidation, improved LPBM assignment, and each loading setup.
• Verify sigma_cd/UCS carefully because the source example reports the largest validation error for this metric.
• Use a few confirmation simulations after back-solving; do not stop at algebraic inversion.

Output Contract

A complete fast-calibration handoff back to pfc-workflow should include:

• target macro values and tolerances
• chosen fixed parameters and back-solved parameters
• orthogonal design table or selected trial set
• strong/weak contact grouping rule and seed
• Weibull alpha and beta settings
• PFC staged command files and save-state names
• extracted macro metrics and error table
• regression formula source and fit quality
• downstream routes for standard tests, servo, post-processing, and V&V

Local Contents

• references/overview.md: workflow boundary, use cases, and routing.
• references/formulas.md: complete formula migration.
• references/orthogonal-design.md: 13-factor tables, 27-run design, correlations, and example calibration.
• references/source-code-complete-pfc6.md: complete staged PFC command-flow migration.
• references/pfc-doc-notes.md: PFC 6.0 command documentation notes.
• examples/README.md: how to materialize and validate a calibration campaign.
• scripts/canonical/orthogonal_design_13params.csv: source 27-run design table.
• scripts/canonical/regression_fast_calibration.py: regression/back-solve helper.
• scripts/canonical/improved_lpbm_assign.p3fis: strong/weak contact and Weibull assignment template.
• scripts/canonical/extract_macro_metrics.py: macro metric extraction helper template.
• scripts/canonical/manifest.json: script inventory.