Metrology + experiments¶

What you will develop¶

Calibrate, quantify uncertainty, design informative experiments, distinguish process variation from measurement variation and reproduce results.

Recorded prerequisites:

Progress by the work you can do, not by a fixed calendar.

Spine: NIST/SEMATECH e-Handbook

Reading: Ch. 2 Measurement Process Characterization + uncertainty/calibration sections

Know: Separate measurand, instrument, calibration, bias, repeatability, reproducibility and uncertainty budget.

Reconstruct: Rebuild an uncertainty budget from independent sources; derive first-order propagation via local linearization.

Do: Calibrate a cheap sensor against a reference; quantify bias, hysteresis, repeatability and uncertainty.

Defend: What is the difference between resolution, accuracy, precision and uncertainty?

Gate: deliver a calibration certificate-style report with traceable assumptions.

Source: source

Spine: NIST/SEMATECH e-Handbook

Reading: Ch. 4 Process Modeling + Ch. 5 Process/Product Improvement (DOE)

Know: Design experiments that identify causal factors efficiently; use blocking, randomization, factorial thinking and response surfaces.

Reconstruct: Derive main-effect estimate in a 2^2 factorial and explain interaction geometrically.

Do: Run a small factorial experiment on software/robot/sensor parameters; pre-register response, nuisance factors and analysis.

Defend: Why is changing one factor at a time often information-inefficient?

Gate: design a 16-run-or-less experiment for an unfamiliar engineering question and defend identifiability.

Source: source

Spine: NIST/SEMATECH e-Handbook

Reading: Ch. 3 Production Process Characterization + Ch. 6 Monitoring + Ch. 8 Reliability

Know: Distinguish common/special causes, process capability, drift, reliability and validation from a one-off demo.

Reconstruct: Derive standard error of mean and basic reliability of independent series/parallel components.

Do: Take one previous Wave-1 build and specify how it would be monitored at 10,000-unit deployment scale.

Defend: When does more testing fail to reduce epistemic uncertainty?

Gate: Final Wave-1 gate: reproduce a measurement, publish raw data/code, identify uncertainty and one untested failure mode.

Source: source

Closed-book: 120 min: uncertainty budget; calibration model; 2^k factorial effects; process/reliability questions.

Novel problem: Design an experiment with ≤16 runs that separates key factors from nuisance variation.

Artifact: Execute/reproduce a small calibration or factorial study; publish raw data/code and uncertainty.

Defend: Defend measurand, traceability, randomization, blocking, uncertainty and inference boundaries.

Pass criterion: Pass if an independent reader can reproduce the result and identify what remains unknown.

Try these before consulting solutions or asking for the complete answer.

Calibration: Estimate offset/gain/nonlinearity of a sensor against a reference.
Uncertainty: Build a full uncertainty budget and propagate to final measurand.
Repeatability: Separate within-run from between-run variation.
Gauge R&R: Design a small operator/instrument repeatability-reproducibility study.
Factorial: Analyze a 2^3 factorial experiment including at least one interaction.
Blocking: Design an experiment where temperature/day/batch is a nuisance factor and block appropriately.
Randomization: Show how nonrandom run order can confound drift with treatment.
Process capability: Given specification limits and process data, assess capability and caveats.
Monitoring: Design a control-chart/monitoring rule and simulate false alarms vs detection delay.
Reliability: Estimate series/parallel system reliability and identify common-cause failure that breaks independence.