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

1. Prove wall/axis/reaction sign with a short motion probe.
2. Estimate effective stiffness and choose a conservative dimensionless gain.
3. Clamp actuator velocity and record target, reaction, error, gain and velocity.
4. Require equilibrium/inertia checks before accepting a stage.
5. Calibrate in sequence: elastic response, strength/interface, time dependence, then post-peak/rutting behavior.
6. Use the two-target solver only when exactly two active levers produce a well-conditioned local response matrix.
7. For larger problems use DOE with real PFC5 runs; treat regression/surrogates as proposal tools only.
8. 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.