4.4 Calibration & Adjustment

Calibration is the final refinement that transforms a functional device into one that delivers precision with confidence. By comparing each unit to a trusted reference under controlled conditions, subtle offsets and drifts are corrected so performance lands safely within specification. This process ties measurement integrity to the serial number, with traceable records of “as-found” and “as-left” states that prove not only compliance but also stability over time. When performed with disciplined environments, reliable references, and straightforward math, calibration becomes less about chasing errors and more about building trust into every product shipped.

4.4.1 What this is (and isn’t)

Calibration compares the unit to a known reference and finds error.

Adjustment changes the unit (coefficients, trims, pot turns) to bring it into spec.

Do both with traceable references, in the right environment, and record as-found/as-left to the serial.

4.4.2 Where it sits in the flow

Program (25.3) → Functional smoke (25.1) → Calibration/Adjustment → Re-verify → Safety (25.2) → Closeout.

If cal is part of FCT, the script must clearly mark cal steps vs verification steps.

4.4.3 Environment & measurement basics

Warm-up: DUT and instruments to thermal steady (typ. 10–30 min).
Ambient capture: log Temp/RH—some specs apply correction.
Test Accuracy Ratio (TAR): reference uncertainty ≤ 25% of the DUT tolerance (4:1 rule; 10:1 ideal).
Instruments in cal: per 18.2; station blocks run if a cert is out-of-date.
Stable hookups: Kelvin for low-ohms; shielded leads for mV/RF; avoid hand heat on sensors.

4.4.4 References & fixtures (pick the right yardsticks)

Electrical: precision DMM, DC sources, electronic loads, voltage/current standards, decade resistors, timebase/OCXO/GPSDO.
Sensors: temperature chambers/blocks, precision pressure/vacuum source, gas standards, light source panel.
RF/Comms: power meter, calibrated attenuators, spectrum/network analyzer (or golden radio + shield box).
Mechanicals: force/weight, linear scales for encoders, torque testers for motors.
Fixtures: short, repeatable paths; mating connectors; guided nests; no clip leads on production.

4.4.5 Common calibration patterns (use the lightest that works)

Offset-only (zero): measure at 0 (or known null), store offset.
Two-point gain/offset: measure at low & high points; solve y = m·x + b.
Multi-point linearization: 3–9 points; fit piecewise linear or poly; ensure monotonic LUT.
Ratiometric/bridge: calibrate excitation first; then sensor span.
Timing: trim RTC/clock via ppm offset against a reference (GPSDO or calibrated counter).
RF power/PA bias: step drive, measure output, build gain table vs channel/temp.

Tip: keep math simple and stable; prefer piecewise linear over high-order polynomials on small MCUs.

4.4.6 Domains & quick recipes

Voltage/Current rails

Source known value(s). 2) Read DUT. 3) Compute gain/offset. 4) Write constants → CRC. 5) Re-read to confirm ≤ spec.

ADC/DAC

ADC: short/known reference (e.g., 0 V & Vref/2), compute offset/gain.
DAC: set codes to low/high, measure with DMM; solve m/b; limit to code range.

Temperature

Two-point (ice/ambient block or chamber steps): store slope/offset; log ambient.
If using thermistors, verify table index matches curve (Beta/R-T table).

Pressure/Flow

Apply two or three known points; zero at vent; store span; run leak check if relevant.

Timing/RTC

Count edges vs reference for ≥10 s; compute ppm trim; write; recheck ≤ spec ppm.

Set known channel/power; measure after attenuator; store cal factor; verify ACL R/EMC sanity in shield box if in scope.

Displays/LEDs

Use light panel or sensor; set white/time-out levels; store per-color factors; eyeball uniformity.

Motors/encoders

Home to hard stop or sensor; store zero offset; verify count per rev.

4.4.7 Adjustment discipline (don’t chase noise)

Average N readings (e.g., 10–32) at each point; use median if spiky.
Guard band: adjust to hit center of spec, not the edge.
Write-protect cal region with CRC/Hash; recipe refuses to run if CRC fails.
As-found vs as-left always captured—even if no adjustment was required.
Limit the wrench: for physical trims/pots, apply torque limits and mark with paint dot after.

4.4.8 Data & traceability (what the record must hold)

Attach to unit SN:

Recipe ID, firmware hash, cal algorithm version.
References/instruments with cert numbers & expiry.
Ambient Temp/RH, warm-up time.
As-found values/errors; as-left values/errors; coefficients/LUT snapshot.
CRC/hash of cal region; operator/fixture IDs; timestamp.
Any retries or anomalies (e.g., chamber ramp slow).

Store plots for long tests (e.g., temp sweep), at least for NPI and sample lots.

4.4.9 Throughput without cheating

Automate: scripted points, auto-calc, auto-write, auto-verify.
Ping-pong fixtures or parallel stations for long soaks/chamber steps.
Pre-cal modules (sensors, radios) at subassembly level when allowed; perform a short verify in box build.
Cache chamber setpoints and test short profiles (two-point instead of five) when the spec permits.

4.4.10 Acceptance cues (fast eyes)

Area	Accept	Reject
References	In-cal, certs valid	Expired/unknown accuracy
As-found	Recorded, within drift policy	Missing; wildly off with no ticket
As-left	Within spec + guard band	At edge or still out of spec
Data store	Coeffs written; CRC OK	CRC fail; defaults restored silently
Repeatability	Re-run gives same within noise	Big jumps between runs
Labeling	Cal date/rev noted (if required)	No trace on unit that needs it

4.4.11 Common traps → smallest reliable fix

Trap	Symptom	Fix
Cal before warm-up	Drift after ship	Enforce warm-up timer; log ambient
One-point “cal” on nonzero slope	Still off at span	Use two-point (gain + offset)
High-order fits on noisy data	Wiggly LUT; fails corners	Use piecewise linear; median-average
Using bench gear “as-is”	Hidden offsets	Run zero/standard check each shift
Manual entry of IDs/constants	Typos in NVM	Script reads back; MES push, no typing
Overwriting customer keys/region	Field returns	Separate cal partition; protect IDs/keys
No as-found record	Can’t prove drift vs damage	Always log as-found before adjustment

4.4.12 Pocket checklists

Before

SKU/Variant scan → cal recipe & limits loaded
Instruments in cal; references warmed; ambient logged
DUT at temp; power stable; fixture latched

Run

Measure as-found; save
Execute points (two/multi) → compute → write coeffs
CRC of cal region OK; lock if required
Verify as-left meets spec with guard band

After

Results (coeffs, errors, plots) to MES by SN
Sticker/metadata updated (if required by customer/spec)
If fail: raise ticket (module swap or MRB), don’t over-tweak

With calibration done properly, devices leave the line consistent, defensible, and ready to perform in the field without surprises. The payoff is precision that lasts, backed by records strong enough to stand in audits and real-world use alike.