PFC5 Servo And Calibration
Use this skill for compaction pressure, confining/load control, Marshall/rutting
reaction control, Burger fitting, or multi-target PFC5 calibration.
Required Inputs
- controlled axis, actuator, sign convention, target and tolerance;
- effective stiffness estimate, timestep/update interval, velocity limits;
- active micro-parameters with units/bounds;
- macro targets, tolerances, experimental provenance and run budget;
- baseline save, seed policy and evaluator output schema.
Workflow
- Prove wall/axis/reaction sign with a short motion probe.
- Estimate effective stiffness and choose a conservative dimensionless gain.
- Clamp actuator velocity and record target, reaction, error, gain and velocity.
- Require equilibrium/inertia checks before accepting a stage.
- Calibrate in sequence: elastic response, strength/interface, time dependence,
then post-peak/rutting behavior.
- Use the two-target solver only when exactly two active levers produce a
well-conditioned local response matrix.
- For larger problems use DOE with real PFC5 runs; treat regression/surrogates as
proposal tools only.
- Confirm the final parameters on an independent seed or loading condition.
Use scripts/servo_gain.py for a bounded proportional step and
scripts/dual_target_solver.py for a guarded local 2x2 solve. Read
calibration-contract.md before campaigns.
Working Rules
- Do not prescribe reaction force by writing a read-only contact-force quantity.
- Do not tune more parameters than the data can identify.
- Reject near-singular local solves; do not hide them with huge parameter jumps.
- Every proposed parameter set needs a true PFC5 confirmation run.
Output Contract
Return controller equations/signs, gain and clamp, histories, parameter/target
tables, run manifest, conditioning diagnostics, confirmation result and runtime status.
Local Contents
references/calibration-contract.md — servo and campaign gates.scripts/servo_gain.py — bounded controller-step calculator.scripts/dual_target_solver.py — conditioned two-target update.agents/openai.yaml — interface metadata.
1---2name: pfc5-servo-calibration3description: Design and audit PFC 5.0 wall servo, loading-rate control, equilibrium checks, micro-to-macro calibration, two-target local solves, DOE campaigns, and independent confirmation for asphalt models.4---56# PFC5 Servo And Calibration78Use this skill for compaction pressure, confining/load control, Marshall/rutting9reaction control, Burger fitting, or multi-target PFC5 calibration.1011## Required Inputs1213- controlled axis, actuator, sign convention, target and tolerance;14- effective stiffness estimate, timestep/update interval, velocity limits;15- active micro-parameters with units/bounds;16- macro targets, tolerances, experimental provenance and run budget;17- baseline save, seed policy and evaluator output schema.1819## Workflow20211. Prove wall/axis/reaction sign with a short motion probe.222. Estimate effective stiffness and choose a conservative dimensionless gain.233. Clamp actuator velocity and record target, reaction, error, gain and velocity.244. Require equilibrium/inertia checks before accepting a stage.255. Calibrate in sequence: elastic response, strength/interface, time dependence,26 then post-peak/rutting behavior.276. Use the two-target solver only when exactly two active levers produce a28 well-conditioned local response matrix.297. For larger problems use DOE with real PFC5 runs; treat regression/surrogates as30 proposal tools only.318. Confirm the final parameters on an independent seed or loading condition.3233Use `scripts/servo_gain.py` for a bounded proportional step and34`scripts/dual_target_solver.py` for a guarded local 2x2 solve. Read35[calibration-contract.md](references/calibration-contract.md) before campaigns.3637## Working Rules3839- Do not prescribe reaction force by writing a read-only contact-force quantity.40- Do not tune more parameters than the data can identify.41- Reject near-singular local solves; do not hide them with huge parameter jumps.42- Every proposed parameter set needs a true PFC5 confirmation run.4344## Output Contract4546Return controller equations/signs, gain and clamp, histories, parameter/target47tables, run manifest, conditioning diagnostics, confirmation result and runtime status.4849## Local Contents5051- `references/calibration-contract.md` — servo and campaign gates.52- `scripts/servo_gain.py` — bounded controller-step calculator.53- `scripts/dual_target_solver.py` — conditioned two-target update.54- `agents/openai.yaml` — interface metadata.
Echo-aloha