Scope lock (what this page covers / excludes)

Covered: compressor inverter drive, fan/drum drives, (optional) heater safety chain, temperature/RH sensing chain, energy metering, and event/fault logging with EMC/ESD/surge evidence points.

Excluded: generic heat-pump textbook derivations, whole-home platforms, cloud/app tutorials, and protocol-stack deep dives. Everything here must map to dryer control signals, waveforms, or logs.

Domain map (domain → typical failures → evidence to collect)

• HV AC input domain (line, surge, EMI filter, inrush): symptoms like random restart, nuisance trips, “works idle, fails under load.”
  Evidence: input sag timing vs load steps; surge/EFT event log; leakage/ground reference anomalies if present.
• PFC / DC bus domain (rectifier/PFC, bus cap, pre-charge): symptoms like compressor start fails, frequent stop/start, abnormal noise during ramp.
  Evidence: DC bus droop at start; ripple shape under steady load; brownout counters aligned to bus events.
• Inverter domain (compressor) (gate driver, power stage, phase current sense): symptoms like OC/OT latch, stall/lose-sync, “runs then faults.”
  Evidence: driver fault flags + timestamp; phase current waveform at ramp; bus current vs commanded speed.
• Motor drives (fan + drum) (BLDC/step/triac depending design): symptoms like long dry time, overheat trips, unstable RH curve, airflow collapse.
  Evidence: fan tach/RPM; drum motor current/torque proxy; airflow/pressure sensor if available.
• LV control domain (3V3/5V/12V, MCU/ADC/AFE): symptoms like “ghost faults,” RH jumping, metering spikes, intermittent resets.
  Evidence: reset reason register; rail dip capture; ADC outliers correlated with inverter/SSR edges.
• Safety chain domain (heater / interlocks) (SSR/relay + thermal cutouts + door lock): symptoms like over-temp stop, heater stuck-on/off, non-recoverable shutdown.
  Evidence: which protection triggers first; heater current signature; relay/SSR command vs feedback mismatch.
• Coupling risk layer (cross-domain): inverter switching noise and heater dv/dt can inject into sensor/MCU.
  Evidence: timing alignment — when a fault appears, does it coincide with inverter PWM edges, heater switching, or bus transients?

First 2 measurements (fast discriminator, minimal tools)

Rule: every later chapter must reference at least one of these signals so diagnosis stays evidence-based.

Why this model exists (so every fault can fall back to evidence)

Dry time and energy are not “feelings.” They are the result of a measurable chain: dehumidification capacity × effective airflow × control consistency.

Energy per cycle can be expressed as an accounting problem:

• kWh ≈ compressor work + fan work + drum work + (optional) heater work + defrost penalty + losses (leaks/blocks/sensor drift-driven overrun)
• Any “longer cycle” must show up as a change in at least one curve: T/RH slope, fan RPM, compressor power, defrost frequency, or Wh.

Cycle phases (A→D) and what each phase should look like

• Phase A — Spin-up & ramp: bus stabilizes, compressor ramps, airflow established.
  Red flag: repeated ramp attempts (energy rises) with little RH improvement.
• Phase B — High dehumidification: RH should drop with a stable slope; airflow should be steady.
  Red flag: RH slope flattens while compressor power stays high → airflow/icing/heat-exchange issue likely.
• Phase C — Tail / end-dry: RH approaches low range; control should avoid “overrun” from sensor lag/drift.
  Red flag: temperature rises but RH does not change meaningfully → control may be chasing a biased sensor.
• Phase D — Defrost / de-ice events: heat-pump dryers pay an energy tax when frosting occurs.
  Red flag: defrost becomes frequent/long → Wh per cycle jumps and dry time stretches.

These phases are not a textbook story. They are a template for curve interpretation used in later debug and validation steps.

Minimum Evidence Pack (the smallest dataset that still closes the loop)

Collect these signals with timestamps aligned (even a 1–10 s sample rate is enough for field triage):

• T_in / T_out (air-path in/out) — shows heat delivery and transfer effectiveness.
• RH_out (or chamber RH) — shows dehumidification progress.
• P_comp (compressor power or equivalent current proxy) — shows effort/cost.
• Fan RPM (tach) — the hidden root cause of “slow dry.”
• Defrost event log (count + duration) — explains sudden Wh spikes.
• Wh per cycle (and if possible stage split) — makes inefficiency quantifiable.

Practical discriminators (curve-based, not guesswork)

• Airflow vs heat-transfer: fan RPM drops → airflow root cause first. Fan RPM stable but RH slope collapses → heat-exchange/icing evidence first.
• Power vs progress: compressor power high while RH barely changes → “energy without removal,” prioritize airflow, icing, or sensor bias.
• Defrost tax: rising defrost frequency with a stable load → check evaporator temperature placement/reading and airflow constraints.
• Control overrun: RH reads “still wet” but temperature rises and Wh increases at tail → sensor lag/condensation on RH element is a prime suspect.

Each discriminator is designed to map directly to the domain map in H2-1, so the next troubleshooting step is always targeted.

Goal: capture all “trip / freeze / reset / metering drift” power causes

Power-related failures in a heat-pump dryer are rarely random. They align with load steps (compressor ramp, heater switching, fan/drum transients) and show up as a measurable chain: AC input margin → DC bus stability → LV rail integrity → protection/latched events.

Rule: treat “software-like” symptoms as power-evidence problems until Vbus and LV rails prove otherwise.

Power tree breakdown (what each layer must guarantee)

• AC Front-End (EMI/surge + NTC/inrush + bypass relay): must avoid nuisance trips and prevent deep input sag during load steps.
  Common pitfall: bypass relay timing creates a brief current spike that looks like a fault to upstream protection.
• Rectifier/PFC + DC Bus (bulk cap + bus sense): must keep Vbus stable across compressor ramp and heater transitions.
  Common pitfall: aged/undersized bulk capacitance increases ripple and makes “start then fault” repeatable.
• Low-Voltage Rails (3V3/5V logic + 12V/driver rails): must keep MCU/AFE/metering references clean under inverter dv/dt and relay edges.
  Common pitfall: LV dips that do not fully reset the MCU still corrupt ADC/metering, causing drift and false alarms.

Protection & verification (turn protection into diagnosable evidence)

• OVP/OCP/OTP: define what is recoverable (retry allowed) vs what is latched (requires clear + root-cause fix).
• Brownout/UVLO policy: log the exact threshold crossing and cause (Vbus droop vs DC-DC collapse vs coupling spike).
• Bypass relay health: verify command vs actual current signature to detect timing issues or contact degradation.
• Heater cutout chain (if present): treat it as a safety domain—detect relay/SSR stick or mismatch by comparing command to measured heater current.

First 2 measurements (minimal tools, maximum separation)

A power root cause is proven only when the voltage event and the fault/reset event share the same timeline.

Why the compressor inverter is the “center of gravity”

The compressor drive dominates both performance and risk: it creates the largest load steps (Vbus stress) and the strongest switching noise (coupling into sensors and MCU rails). A useful inverter chapter must therefore be evidence-first: state timeline + fault latch timing + I_phase(t) + Vbus(t) + thermal points.

Drive lifecycle (what to verify at each state)

• Ready / pre-drive: driver supply stable, faults cleared, bus above enable threshold.
  Evidence: driver “fault pin” idle state; enable timing vs Vbus.
• Start / ramp: establish rotation and limit phase current while Vbus stays within margin.
  Evidence: I_phase(t) at ramp; Vbus droop; first fault flag timestamp.
• Steady: maintain stable current/torque with predictable thermal rise.
  Evidence: current ripple pattern; bus ripple; hotspot temperature trend (power device / heatsink).
• Stop / restart: controlled shutoff without backfeed spikes or repeated latches.
  Evidence: latch persistence; restart count; bus recovery time.

Key risks → discriminators (what proves each root cause class)

• Start failure: repeated attempts with minimal progress.
  Discriminator: does Vbus droop precede the first fault, or does I_phase spike first?
• Over-current (OC): immediate fault on torque demand.
  Discriminator: I_phase spike timing vs switching edges; compare to fault latch time.
• Under-voltage (UV): drive drops out when load increases.
  Discriminator: UVLO events aligned to Vbus droop vs LV rail dip (tie back to H2-3).
• Over-temperature (OT): fault appears after minutes.
  Discriminator: thermal point rises steadily; OT threshold crossing precedes latch (power device vs heatsink).
• Lose-sync / stall: abnormal noise and efficiency loss, sometimes followed by OC.
  Discriminator: asymmetric / distorted I_phase waveform with relatively stable Vbus suggests a mechanical or control mismatch.

Without aligned timestamps, “cause” and “coincidence” cannot be separated in the field.

Bridge rule (tie back to the power chapter)

Why airflow + drum behavior explains “doesn’t dry / takes too long / overheats”

In a heat-pump dryer, many complaints look like “heat-pump efficiency” or “control logic,” but the root cause often sits in hidden mechanics: airflow and drum load. When airflow collapses, moisture removal stalls and the system compensates with longer runtime and higher energy. When drum motion degrades, clothes stop tumbling effectively and heat exchange becomes uneven, which can amplify overheat events.

Fan drive: discriminate blockage vs degradation vs PWM/EMI side-effects

• Airflow reduction (filter/duct blockage)
  Proof: for the same drive command (duty/target RPM), fan current rises or RPM drops; if a pressure sensor exists, ΔP indicates restriction. Drying time grows while moisture removal slows.
• Fan mechanical degradation (bearing / rubbing / imbalance)
  Proof: startup current increases; current shows a repeating ripple; RPM becomes unstable under humidity/temperature extremes.
• PWM noise / EMI coupling into sensing and logs
  Proof: tach/RPM glitches align to switching edges; nuisance faults appear at the same phase of PWM or inverter activity. The fan may be fine, but measurement and control are polluted.

Drum motor drive: separate load variation from friction/slip from protection behavior

• Normal load variation
  Proof: drum current shows periodic modulation synchronized with tumble dynamics; speed tracks target with mild, repeatable ripple.
• Belt slip / friction increase / mechanical bind
  Proof: current rises but speed fails to follow; speed ripple grows; retries become frequent. If a door-open event forces stop/restart, compare restart current—abnormally high restart current strongly indicates friction/bind.
• Over-current protection sensitivity / inconsistent retries
  Proof: OC triggers cluster during a specific motion phase; fault timestamps align to torque spikes. Confirm whether the system reduces torque, changes speed, or hard-stops (a “pattern” matters more than a single trip).

First 2 measurements (minimal proof set)

A root cause is “locked” when speed/current signatures and event timing are consistent across repeated cycles.

Engineering boundary: measurement quality sets the ceiling

Control loops and field diagnostics cannot outperform their measurements. In heat-pump dryers, wrong sensor placement, slow response, or drift can create false “plateaus,” nuisance trips, and oscillations that look like logic errors. The sensing chain must therefore be treated as a first-class subsystem with provable dynamics.

Temperature points: what each location represents (and the trap)

• Inlet / outlet air: represents drying progress and thermal balance.
  Trap: placement lag creates delayed feedback; small airflow changes can look like large temperature changes.
• Evaporator temperature (T_evap): critical for frosting evidence.
  Trap: condensation contact errors can fake a “flat platform” or add bias.
• Condenser / exhaust: relates to delivered heat and recovery quality.
  Trap: sensor sees local hot spots, not bulk air, if mounted poorly.
• Heater-adjacent safety point (if present): must align with the safety chain.
  Trap: slow thermal coupling causes the hardware chain to trip first.

RH sensing: saturation, contamination, condensation, and slow response

• Saturation: RH clamps near the upper range and loses information.
  Proof: measure the time to saturate and the time to recover after humidity drops (P1).
• Contamination: lint and residue slow the sensor and create hysteresis.
  Proof: recovery time grows over cycles; readings lag behind temperature changes.
• Condensation influence: droplets create step-like jumps or long bias tails.
  Proof: RH anomalies align with drainage/defrost events; post-event baseline shifts.
• Slow response → control oscillation: delayed RH produces over/under corrections.
  Proof: duty changes precede RH response by an excessive lag; oscillation period matches the sensor time constant.

AFE/ADC: sampling, filtering, and drift containment (only what changes evidence)

• Sampling vs filtering: heavy filtering can create a fake plateau; too-slow sampling hides transitions.
  Proof: temperature step response and RH recovery time change significantly when filter settings change (P2).
• Noise coupling: PWM/inverter activity can modulate readings.
  Proof: reading noise aligns to switching activity or tach frequency; treat it as correlation evidence.
• Self-test and open/short logs: sensors must be diagnosable.
  Proof: open/short detection, out-of-range flags, and “stuck reading” detection are recorded as event logs (P3).

Why heater control must be “two-layer” (hardware safety + software proof)

A heater is a high-risk subsystem because it can create unsafe energy even when the MCU is wrong, frozen, or affected by EMI. A robust design therefore needs two independent lines of defense:

• Hardware safety chain: mechanical thermostat and/or one-shot thermal fuse that cuts power regardless of firmware state.
• Software loop + diagnostics: temperature limits, duty/ramp policies, and event logging that detect drift, false triggers, and stuck-on behavior early.

Field truth is not a claim. It is proven by current signature, state readback, and the trigger order of the over-temp chain.

Control elements (what to verify, not textbook theory)

• Switch element: SSR or relay (or triac). Verify command timing and the real conduction state.
• Zero-cross / edge behavior (if used): treat it as an EMI signature tool—compare current spikes and nuisance triggers across switching styles.
• Temperature sensing: NTC/RTD used for the software limit; placement errors show up as a wrong slope or delayed response.
• Hard cut chain: thermostat (resettable) and thermal fuse (one-shot). These define the last line of defense.

Typical failures → discriminators (what proves each class)

• Relay/SSR stuck ON: command OFF but heater still conducts.
  Proof: heater current remains non-zero after OFF command; “state readback” disagrees with command; temperature keeps rising.
• False trigger / EMI-induced misfire: unintended brief conduction.
  Proof: short current bursts aligned to inverter/relay edges; temperature sensor shows spikes aligned to switching events (timing correlation).
• Real overheat (thermal/airflow issue): conduction is correct but heat is not removed.
  Proof: heater current normal, yet temperature slope exceeds expected; hard cut chain triggers after a predictable ramp.
• Sensor or limit error: temperature reading is biased/slow.
  Proof: “temperature vs current” relationship breaks; the same duty results in different slopes between cycles.

Heater Evidence Pack (minimum set to diagnose in the field)

A heater problem is “proven” only when command, current, temperature slope, and trigger events share one timeline.

Why defrost + drainage collapses performance

Heat-pump dryers fail “silently” when moisture removal collapses: drying time grows, energy per cycle rises, and user complaints appear before hard faults. Two drivers dominate field collapse:

• Frost / ice reduces heat exchange and effective airflow.
• Condensate faults (pump/level/overflow) prevent recovery after defrost and can force protective shutdown.

This chapter stays evidence-first: detect frost by discriminators, validate defrost by expected curve response, and prove drainage by pump-current and level-state timelines.

Frost detection discriminators (what proves it is really frosting)

• ΔT + RH slope mismatch: temperature changes without the expected RH decline (dehumidification stalls).
• Evaporator temperature “flat platform”: T_evap stays in a narrow band while performance degrades.
• Airflow proxy drops: fan tach/RPM falls or becomes unstable near the collapse point.
• Energy tax signature: Wh/cycle increases primarily during frequent or long defrost segments.

Defrost control (write only controllable evidence, not heat-pump textbook)

• Trigger: ΔT anomaly, T_evap flat, RPM drop, RH slope stall. Record the 2–5 minute pre-trigger window.
• Action: pause/ramp compressor, adjust fan, optional auxiliary heat. Prove action timing by log timestamps.
• Exit: defined by curve recovery (T_evap changes, ΔT returns, RH slope recovers) or time limit.

Condensate management (pump/level/overflow must be diagnosable)

• Loop: condensate → sump → level sensor → pump → drain path → overflow protect.
• Pump stall / clog: pump current rises or shows abnormal ripple; command ON but level does not fall.
• Drain restriction / backflow: pump current can look normal, yet level drops slowly or rebounds; prove by level-state timeline.
• Level sensor fault: level state toggles inconsistently with pump command/current; overflow protect triggers unexpectedly.

Defrost + drainage Evidence Pack (minimum set)

Goal: separate “real high energy” from “bad measurement”

“High kWh per cycle” can be caused by real inefficiency (longer runtime, more defrost, restricted airflow), or by metering errors (phase, offset, drift, and non-sinusoidal waveforms under PFC/inverter conditions). The metering chain must therefore be auditable: it should produce repeatable Wh/cycle, stage-by-stage energy splits, and a clear error budget that explains when readings can be trusted.

Metering chain: I/V sensing → synchronization → power computation

• Current sensing (shunt / Hall / CT)
  What breaks evidence: shunt thermal gradient → zero drift; Hall bias/bandwidth → dynamic error; CT phase shift at low frequency → systematic power bias.
• Voltage sampling (divider / isolation / ADC reference)
  What breaks evidence: divider drift and ADC reference drift change amplitude; isolation delay or filtering adds phase error.
• Synchronization (sample timing and windowing)
  What breaks evidence: small phase mismatch between V and I becomes a large P error, especially when waveforms are non-sinusoidal (PFC harmonics, inverter switching).
• Power computation (P(t), Wh integration, logging granularity)
  What breaks evidence: coarse logging (10 s) can smear or miss startup/step energy; 1 s (or finer) is required for diagnostic truth.

Dominant error sources: phase, offset, drift, and PFC harmonics

• Phase error (timing mismatch / CT delay / filtering delay)
  Signature: systematic high/low power across the cycle; readings shift predictably when filter or timing settings change.
• Offset / zero drift (shunt amp bias, ADC bias, thermal gradients)
  Signature: “power does not return to zero” at very low load; baseline changes with temperature or time-on.
• Temperature drift (sense resistor, divider, AFE gain)
  Signature: cold cycle vs hot cycle shows consistent bias direction at the same operating point.
• PFC harmonics / non-sinusoidal waveforms
  Signature: power or energy jumps only in certain compressor speeds or operating modes; errors correlate with waveform distortion and switching activity.

Stage energy split: make compressor/fan/drum/heater attribution defendable

Stage splits must be driven by events (enable flags, mode transitions, or clear step signatures), not assumptions. The simplest defensible approach is to define a cycle timeline and bind each boundary to a timestamped marker:

• Compressor segments: start spike + steady platform (inverter active window).
• Heater segments (if present): clear step power delta aligned to control state.
• Fan / drum baseline: smaller increments that persist across multiple segments; validate by correlating RPM/current to the baseline change.
• Defrost segments: identified by defrost flags and temperature signature; energy is separated as a “tax” on efficiency.

First 2 measurements (minimal proof set)

When phase, offset, drift, and windowing are controlled, “mystery inefficiency” becomes a reproducible numeric story rather than a debate.

Intermittent failures must be converted into timelines and evidence

“Random resets” and “occasional false alarms” are rarely random. In appliances, intermittent events often result from coupling between fast switching edges (inverter dv/dt, SSR edges), cable injection, and ground bounce. Robustness work starts by building an evidence timeline: what happened, when, and what the system believed (reset reason, fault registers, IO states) right before the event.

Symptom taxonomy: group “random” events by victim type

• System-level: reset / brownout / hang → reset reason register + brownout flag + last-event timestamp
• Nuisance faults: false compressor/fan/door/water-level alarms → fault latch source + input snapshot at trigger time
• False triggers: SSR/relay mis-action, door input glitch → GPIO timestamp + edge counter + actuator current signature
• Metering glitches: power/energy jumps → sampling sync flag + mode transition alignment + log granularity check

High-impact sources: inverter dv/dt, SSR edges, sensor cable coupling, ground bounce

• Inverter switching noise: dv/dt and di/dt drive common-mode currents and ground shifts.
  Victims: MCU rails, ADC references, tach/RH/NTC lines, and fault pins. Evidence: event timing correlates with inverter state transitions.
• SSR / relay edges: fast edges inject into control and sensing, creating false triggers.
  Evidence: GPIO glitches align with actuation edges; actuator current signature shows unintended conduction.
• Sensor cable coupling: long or high-impedance inputs pick up injected transients.
  Evidence: “stuck low / stuck high / bounce” counters spike after EFT/ESD; input snapshots show brief invalid states.
• Ground bounce: shared returns and fast currents shift local ground reference.
  Evidence: multiple unrelated inputs glitch simultaneously; brownout/reset reasons cluster at high-load switching.

Evidence pack: make EFT/Surge/ESD events reproducible

• After-event log record: test type (EFT/surge/ESD), time, mode, and counters.
• Reset reason register: brownout vs watchdog vs external reset vs software reset.
• GPIO glitch capture: edge counters + timestamps for 2–3 critical lines (door input, SSR control feedback, key sensor fault pins).
• Waveform proxy: actuator current signature and DC rail snapshot around the event window.

First 2 measurements (minimal proof set)

When events are timestamped and correlated to switching states, “random” becomes an isolate-and-fix workflow.

How this SOP is meant to be used

This chapter converts common field complaints into an evidence-driven workflow. Every symptom follows the same structure:

• First 2 measurements (two concrete points that work as a universal “truth anchor”)
• Discriminator (fast proof to classify root cause: airflow / refrigerant loop / power / sensing / control / EMC)
• First fix (lowest-cost fixes first; each fix must change evidence, not just “feel better”)

Example MPNs below are representative parts commonly used in appliance platforms. Selection must match voltage/current/safety requirements and vendor lifecycle.

Universal First 2 Measurements (use for every symptom)

Measurement #1 — DC bus ripple/sag at the bulk capacitor

• Where to probe: across the DC bus bulk capacitor (or the closest low-inductance test pads to it).
• What to capture: startup sag, steady ripple envelope, and any dips aligned to compressor start/stop or heater switching.
• What it proves: whether power integrity is the primary driver (UVLO/brownout) or a secondary effect.

Measurement #2 — Compressor inverter fault/status timeline

• Where to probe/read: inverter nFAULT/OTW pins (if available) and/or driver status registers; record enable/disable timestamps.
• What to capture: fault type, latch timing, retry counters, and correlation to DC bus events.
• What it proves: whether the “core energy machine” is running as expected or failing early.

Symptom: “Not dry / cycle time extremely long”

First 2 measurements

• M1: DC bus ripple/sag across bulk cap (focus on compressor steady window).
• M2: compressor inverter fault/status + retry counters (confirm stable run vs early faults).
• Optional (high leverage): outlet air T + RH at 1 s logging for the entire cycle segment.

Discriminator (proof-based classification)

• Airflow-limited: fan command looks normal but RPM drops or fan current rises; T/RH curves show slow recovery and weak moisture removal.
• Refrigerant-loop limited: compressor power present but dehumidification is weak; evaporator temperature shows abnormal plateau patterns (often tied to icing).
• Sensing-limited: RH saturates and recovers slowly; temperature step response lags (filter/placement causes control to “see the past”).
• Control-scheduling: mode toggles frequently; defrost enters/exits too often without matching temperature evidence.

First fix (lowest-cost first)

• Increase diagnostic log fidelity to 1 s and record mode transitions with reasons (prevents false conclusions).
• Validate fan RPM feedback and set an “airflow not credible” guard (limit power when airflow evidence degrades).
• Validate RH sensor recovery time; reduce over-filtering; add anti-condensation placement/guarding if evidence shows wetting.

Symptom: “Compressor won’t start / frequent start-stop”

First 2 measurements

• M1: DC bus sag during start ramp (capture the first seconds).
• M2: inverter fault/status latch timing (OC/UV/OT categories if available).
• Optional: phase current proxy (driver current-sense output or shunt amp output) to confirm OC vs mis-detection.

Discriminator

• Undervoltage-driven: bus sag aligns to failure and UVLO flags appear; restart happens after bus recovers.
• Overcurrent / stall-driven: OC or stall flags latch; repeated retries show similar current signature each time.
• Thermal-driven: starts OK when cold, fails when hot; OTW/OTP evidence precedes shutdown.
• Control-limited: shutdown reason is policy (defrost gating, temperature protection), not electrical fault evidence.

First fix

• Record every failed start with fault code + timestamp + bus minimum (turns “random” into a dataset).
• Adjust start ramp / soft-start parameters to reduce the start surge and bus collapse risk.
• If UVLO is proven: validate bulk capacitor health and inrush bypass timing (evidence must show reduced sag after change).

Symptom: “Overheat shutdown / burning odor”

First 2 measurements

• M1: DC bus ripple during the minutes before shutdown (look for load spikes and instability).
• M2: compressor status (confirm whether shutdown is inverter-fault-driven or a separate safety trip).
• Optional: temperature trigger order (which sensor hits threshold first: outlet air / condenser / heater-adjacent).

Discriminator

• Airflow collapse: temperature rises quickly while fan RPM evidence degrades (RPM drops or torque rises).
• Heater mis-action (if present): heater current indicates unintended conduction or excessive on-time.
• Sensing/placement issue: measured temperature lags reality; a hardware cut-out triggers before firmware “sees it.”
• Control oscillation: repeated over/under-shoot cycles indicate time-constant mismatch and insufficient hysteresis.

First fix

• Add a power-limiting guard when airflow evidence becomes non-credible (prevents thermal runaway).
• If heater exists: enforce a hard evidence condition (heater enable must correlate with temperature demand; log heater current on every enable).
• Re-tune filtering/hysteresis on safety-critical temperature points (remove oscillation without hiding true overheat).

Symptom: “Severe icing / defrost too frequent”

First 2 measurements

• M1: DC bus behavior during defrost entry/exit (confirm whether defrost is electrically stressful).
• M2: compressor status + mode timeline (defrost events must have timestamps and reasons).
• Optional (high leverage): evaporator temperature curve and defrost counters (count + duration per hour).

Discriminator

• True icing: evaporator temperature shows sustained plateau; airflow evidence degrades; defrost produces short-lived recovery.
• False defrost triggers: defrost enters without matching temperature evidence; sensor placement/filtering is the likely driver.
• Condensate / drain coupling: after defrost, pump current or water-level evidence indicates blockage or overflow that accelerates re-icing.

First fix

• Require at least two independent evidences before defrost entry (e.g., evap temperature signature + airflow proxy).
• Enforce minimum intervals and robust exit conditions to prevent rapid “enter-exit-enter” cycling.
• If drain evidence is abnormal: fix drain first (a stable drain path is often a prerequisite for stable defrost behavior).

Symptom: “Energy spikes / metering abnormal”

First 2 measurements

• M1: DC bus ripple and mode transitions (verify that spikes align to real load events).
• M2: compressor status timeline (separate true compressor cycling from measurement artifacts).
• Optional: metering proof set: zero drift + a single load step error (1 s logging).

Discriminator

• True energy increase: stage split shows larger compressor/heater/defrost Wh; evidence aligns to longer runtime or more defrost.
• Phase/sync error: systematic power bias or mode-dependent jumps under distorted waveforms (PFC/inverter harmonics).
• Offset/drift error: baseline power does not return to near-zero at low load; drift grows with temperature.
• Granularity artifact: 10 s logging hides short spikes; switching to 1 s changes the narrative immediately.

First fix

• Force diagnostic logs to 1 s and store the event timestamps used for stage boundaries.
• Freeze calibration/offset snapshots in logs to end “measurement debates.”
• Only after metering is credible: optimize control (reduce unnecessary defrost/heater assistance).

Symptom: “Intermittent reset / random errors”

First 2 measurements

• M1: DC bus ripple/sag near the time of reset (confirm if brownout is present).
• M2: compressor status at reset time (inverter switching state matters for EMC correlation).
• Required add-on: reset reason register + GPIO glitch timestamp/counter (door input, SSR control/feedback, one critical sensor line).

Discriminator

• Power-driven reset: brownout flags and rail dips correlate with the reset.
• EMC injection: IO glitch counters spike and correlate with inverter/SSR switching edges; multiple inputs glitch together.
• Watchdog/software: watchdog reset with no rail evidence; logs show task stall or timing issues.
• False fault chain: a brief illegal input state triggers protection logic; the reset is a secondary consequence.

First fix

• Make reset reason + last-event timestamp non-optional (store it every time).
• Harden 2–3 most sensitive inputs with debounce/RC/clamp only after evidence proves injection.
• Reduce edge aggressiveness at the source (switch timing adjustments) only if correlation is proven.

Concrete “first fix” checklist (evidence must improve)

• Logging: 1 s diagnosis logs + explicit timestamps for mode transitions, faults, and defrost events.
• Guards: airflow credibility guard (fan RPM/tach sanity), sensor plausibility guard (RH saturation/recovery), and metering credibility guard (zero drift).
• Electrical hardening: debounce/RC/clamp on proven victim inputs; avoid “shotgun” changes without correlation.
• Retest: repeat the same trigger condition; accept fixes only if the same evidence pack improves consistently.