Quick identification (what to check first)

• Wiring/label: 2-wire / 3-wire / 4-wire; presence of FG/tach; dedicated PWM or 0–10V control pin.
• Control ownership: ECM fan (internal controller) vs raw BLDC/PMSM (external 3-phase drive required).
• Supply type: AC mains / 24Vac / 24Vdc determines switching method and dominant EMI coupling paths.

Bucket A — ECM / BLDC (commanded speed, optional feedback)

• Control variables: PWM frequency/window (avoid audible and sensor aliasing), 0–10V output impedance and reference, and FG/tach input conditioning (Schmitt, RC, or time-window debouncing for low-speed pulse dropout).
• Common failures (must map to evidence): low-speed whine (PWM band vs mechanical resonance), start failure (UVLO / rail droop), speed hunting (loop instability or noise injection), EMI exceed (fast edges / return path).
• Evidence points (minimum set): Command (PWM or 0–10V) + Feedback (FG/tach or ΔP response) + one power integrity trace (LV rail droop/ripple).

Bucket B — AC fan (relay steps / triac phase / SSR)

• Relay steps: simple and robust, but contact arcing can couple into sensor/MCU rails; evidence is reset counters aligned to switching events.
• Triac phase control: enables speed shaping, but creates sharp current edges; evidence is EMI peaks aligned to firing angle and sensor noise correlated to phase edges.
• SSR: quieter switching, but check leakage and thermal rise; evidence is residual current (off-state) and temperature drift during continuous operation.
• Failure signatures: low-speed buzz, intermittent start (insufficient torque at small conduction angles), EMI-driven dropouts, and nuisance resets.

Actuator families (and the best evidence per family)

• 24Vac two-wire on/off: evidence = coil current presence + action timing + interlock contact (if provided).
• DC motor + limit/Hall: evidence = current ramp + back-EMF signature + limit transitions consistency.
• Stepper: evidence = phase current + travel time consistency + end-stop/limit confirmation (missed-step shows drift).

Failure chain A — Jam / stiction (won’t move)

• Symptom: command issued but position evidence does not change.
• Electrical proof: drive current rises quickly and hits limit, while back-EMF stays low.
• Discriminator: current high + no movement → mechanical jam; current near zero → wiring/drive/power missing.

Failure chain B — Backlash / rebound (moves then slips back)

• Symptom: end-position reached but then “unreaches”; airflow/ΔP becomes unstable.
• Electrical proof: limit signal toggles twice or ΔP drops after a stable command.
• Discriminator: command steady + evidence retreats → mechanical rebound/seal force, not firmware logic.

Failure chain C — Stepper missed steps (position drifts over cycles)

• Symptom: repeated moves gradually shift the effective end-point or travel time increases.
• Electrical proof: phase current remains nominal but end-stop timing drifts; occasional “late limit” events.
• Discriminator: normal current + drifting timing → torque margin/installation friction; abnormal current → drive limit/rail droop.

Minimum measurement template (fast field proof)

• First 2 measurements: (1) actuator drive output (voltage or current), (2) position evidence (limit/Hall or ΔP response).
• Third proof when needed: LV rail droop during actuation (separates power weakness from mechanics).
• First fixes: add supply margin/return path cleanup, tune current limit/soft-start, improve harness contacts, reduce seal friction or adjust end-stops.

Interface map (what must be nailed first)

• UART: define voltage domain and framing; log freshness (timestamp per sample) and retry counters.
• I²C: verify pull-ups, bus idle level, and noise margin; log NACK/timeout events and recovery attempts.
• PWM output: measure frequency and duty stability; decode duty over a fixed window to avoid jitter artifacts.
• Power sensitivity points: verify CO₂ module VDD ripple during fan/actuator events; keep a local ground return.

Drift & bias sources (hardware-verifiable)

• Temperature-related drift: bias changes with enclosure temperature; verify with local temperature reference and consistent drift direction.
• Optics aging / dust: slow long-term offset; verify via status/health flags (if exposed) and increasing deviation vs a reference point.
• Update rate / averaging: “slow response” can be an output cadence artifact; verify by reading sample interval and moving-average window behavior from logs.

Field symptoms → discriminators (prove the root cause)

• Always high / always low: check reference point deviation (outdoor or well-ventilated baseline) + long-term trend.
• Slow response: compare sensor update interval vs the expected ventilation time constant; confirm cadence before suspecting “bad sensor”.
• Step jumps / spikes: correlate spikes with fan PWM edges, actuator switching, or radio bursts; then check CO₂ VDD ripple alignment.

Evidence pack (minimum set for field triage)

• Status word / error code: self-test, “measurement ready”, weak-signal flags, sensor fault flags.
• Sample interval + timestamp: detect stale samples and buffering artifacts (freshness must be explicit).
• Power proof: CO₂ VDD ripple/rail droop during fan/actuator switching; log brownout/reset counters.
• Environmental reference: a stable baseline point (outdoor / high-ventilation) used only to validate bias direction and magnitude.

First fixes (evidence-driven)

• If spikes align to VDD ripple: improve local decoupling, separate return path, and avoid reading during switching bursts.
• If freshness is wrong: add timestamps/sequence counters, enforce update timeout, and record retries/errors.
• If bias grows over months: confirm optics/dust suspicion with a reference check; add a maintenance flag rather than “blindly recalibrating”.
• If slow response is cadence-driven: reduce averaging window or expose update interval to logs and UI for correct interpretation.

Heater drive (the main disturbance source)

• PWM heater: edges inject ground bounce and supply ripple; verify by aligning ADC noise with PWM edges.
• Constant-current heater: reduces ripple but still requires thermal headroom; verify ripple reduction and consistent warm-up behavior.
• Return-path separation: keep heater current loop away from AFE/ADC reference return; verify by reduced code jitter under identical PWM.

AFE chain (make the small signal measurable)

• Divider / resistive readout: simplest, but sensitive to ground/reference noise; keep source impedance and ADC sampling consistent.
• Bridge / differential readout: improves immunity to common-mode noise; verify by lower variance under the same heater drive.
• Instrumentation / low-noise stage + RC anti-alias: pushes PWM/radio energy out of band; verify via reduced high-frequency code components.
• RC anti-alias: set to attenuate PWM harmonics before ADC; confirm with FFT or time-domain jitter reduction.

Sampling discipline (avoid edges and bursts)

• Sampling window: sample only in a stable region between PWM edges; keep a fixed timing relationship.
• Radio burst coupling: if VOC noise increases during Wi-Fi/Thread activity, avoid sampling during bursts or isolate rails/returns.
• Fan PWM coupling: if VOC noise rises when fan speed changes, correlate VOC code variance with fan PWM edges and supply ripple.

Field symptoms → evidence discriminators

• VOC “random jumping”: check if jumps are periodic and phase-locked to heater PWM or fan PWM edges.
• Worse during wireless activity: check if jitter increases during radio bursts; confirm via timestamp correlation.
• Humidity/temperature driven misreads: if VOC follows RH/T trends without electrical correlation, treat as cross-sensitivity (do not blame the ADC first).

ΔP sensor practical boundary (what matters in field)

• Interface: analog / I²C / SPI — prioritize stable readings and error visibility over rich features.
• Range: low range improves sensitivity; avoid saturation during boost mode or high duct resistance.
• Zero drift: temperature and stress shift baseline; require a logged “zero check” event to keep bias visible.
• Hose/condensation: water in tubes causes lag, hysteresis, and false offsets; treat as a primary field failure mode.

Fan-curve evidence (turn symptoms into discriminators)

• Same PWM, higher ΔP → likely restriction increased (filter clog / duct resistance).
• Same PWM, lower speed → load increased or power margin reduced; validate with current/ripple snapshot.
• Same speed, lower ΔP → leakage/short-circuit airflow or damper position mismatch.
• Non-monotonic trio (PWM↑ but speed/ΔP not trending) → verify sensor validity and mechanical consistency before blaming firmware.

Physical consistency rules (fast validity checks)

• Monotonic trend: in a fixed duct configuration, PWM↑ should not repeatedly produce ΔP↓ over steady windows.
• Lag: ΔP should settle within a reasonable time constant; excessive lag points to hose water or blockage.
• Hysteresis: up/down ramps differing strongly suggests condensation or mechanical sticking, not “random noise”.

Common duct issue signatures (evidence-first)

• Filter clog: ΔP baseline rises over time; speed drops or current rises at the same PWM.
• Restriction: boost mode fails to increase airflow; ΔP saturates high while speed cannot climb.
• Condensation: ΔP shows slow response + large hysteresis, worse in humid/cold conditions.
• Leak: speed normal but ΔP low; improves after sealing or damper correction.

Inputs (only what the device can measure)

• ΔP: strongest direct evidence, but must guard against condensation and hose artifacts.
• Speed vs PWM deviation: same PWM producing lower speed indicates higher load/resistance; requires reliable FG/tach.
• Power/current: supportive evidence; rising current at comparable operating points suggests increasing resistance or mechanical drag.
• Run hours: stable fallback; never used as the sole trigger when strong evidence exists.

Simplified model options (field-verifiable)

• Threshold + slope: detect baseline shift (ΔP↑) and trend (ΔP slope↑) at stable operating points.
• State machine: Clean → Warning → Replace, with clear entry/exit conditions and a recorded trigger reason.
• 2-of-3 evidence gate: reduce false positives by requiring agreement (ΔP + speed deviation + power).

Calibration baseline (the most common deployment failure)

• Post-install baseline: capture ΔP₀, (PWM→Speed)₀, and I₀ after a stable warm-up window.
• Exclude transients: ignore actuation changes and mode switches during baseline capture.
• Light seasonal handling: allow a maintenance or re-baseline hint when long-term offset shifts are detected (keep it device-side).

Logs (must record reason, not only %)

• State: Clean / Warning / Replace
• Trigger reason (enum): DP_HIGH, SPEED_DROP, POWER_RISE, HOURS
• Snapshot: ΔP, PWM, speed, current, timestamp (and optional temp/RH as context)
• Evidence count: how many signals agreed at trigger time (1/2/3)

Validation of false-positive / false-negative

• False-positive test: after installing a new filter, ΔP and speed deviation should return near baseline; if not, suspect duct/condensation, not filter-life logic.
• False-negative test: simulate restriction (controlled airflow block); ΔP↑ + speed deviation↑ should escalate state within defined persistence time.
• Proof pattern: escalation must be reproducible and reversible with evidence changes.

Must be local (real-time, never depend on gateway)

• Fan control loop: PWM/command → speed/ΔP feedback → local regulation and sanity checks.
• Safety interlocks: cover/door, damper limits, over-current/over-temp, sensor fault flags.
• Fail-safe mode: a deterministic offline behavior (hold / degrade / safe) with a logged reason.

Can be deferred via gateway (loss-tolerant)

• Status reporting: modes, setpoints, filter-life state, sensor health summary.
• Alerts: Replace warning, sensor weak-signal, EMC-related reconnect statistics.
• Firmware update: execute only in a safe window; keep basic rollback/failure protection concept-only.

Three classic field failures → required evidence fields

• Mode fallback on drop: offline_enter_ts, offline_duration, mode_before/after, fallback_reason.
• Duplicate commands: cmd_id/cmd_seq, ack_seq, dedup_drop_count, retry_count.
• State desync: state_version, state_gen_ts, last_applied_cmd_seq, resync_success_count.

Offline buffering (device-side, concept-level but testable)

• Command side: accept only sequence-tagged commands; discard stale/expired sequences.
• Status side: buffer snapshots in a ring; flush in timestamp order after reconnection.
• Resync step: align current state version before applying new commands after reconnect.

First fixes (evidence-driven)

• Frequent duplicates: enforce dedup by cmd_id/cmd_seq and make command execution idempotent.
• Frequent desync: add state_version + state_gen_ts, and require resync before new actuation.
• Drop-triggered regressions: define offline policy explicitly and log fallback_reason for service proof.

Risk map (ERV coupling paths)

• Motor PWM → supply ripple / radiated noise → RF throughput drop + sensor readout jitter.
• Valve/relay actuation → EFT-like spikes → I²C/UART errors, brownout resets, mode fallback.
• Long wires/terminals → ESD injection → false alerts, desync, random reconnect loops.

Surge / ESD (power entry & terminals)

• Risk: transients at mains/24V entry and on long I/O terminals.
• Probe: entry-before/after protection, low-voltage rails, reset/brownout counters.
• Pass gate: no uncontrolled reset loops; no persistent sensor fault flags; no corrupted logs.
• Fix: improve entry clamping/return, add staged filtering, isolate sensitive sensor rails.

EFT (valve harness coupling)

• Risk: actuation edges couple into low-voltage domains and comm lines.
• Probe: valve drive waveform + LV rail ripple + interface error counters.
• Pass gate: no burst of NACK/timeouts; no mode fallback triggered by actuation.
• Fix: reduce loop area, add snubbers/return management, schedule sensor sampling away from edges.

EMI (motor PWM harmonics ↔ RF drop evidence)

• Risk: PWM harmonics reduce RF throughput or cause reconnect storms.
• Probe: reconnect_count, retry_count, packet_loss proxy + PWM duty/frequency logs (time-aligned).
• Pass gate: RF counters do not correlate with PWM changes; mode does not revert on RF events.
• Fix: adjust PWM edge control, add filtering, improve ground partitioning, antenna separation.

Sensor integrity under noise (NDIR/MOX coexistence)

• Risk: sensor rails and ADC windows polluted by motor/valve switching or radio bursts.
• Probe: sensor VDD ripple + CO₂/VOC value jumps + sampling timestamp relative to switching events.
• Pass gate: moving the sampling window away from edges reduces jitter; no false VOC/CO₂ spikes during actuation.
• Fix: local rail filtering/LDO, controlled sampling windows, return-path discipline.

General EMC theory (shared subsystem link)

• For cross-product EMC/safety patterns and protection building blocks, jump to the shared page: EMC / Safety & Metering Subsystem (general methods, components, and test taxonomy).