H2-1 · What this page covers (and what it doesn’t)

Intent: lock the scope to “sensor → AFE → ADC/MCU → alarms” for avionics-bay environment & vibration, without drifting into BIT, power, or data-bus deep dives.

Avionics bays fail in very practical ways: heat hotspots push electronics out of margin, moisture drives condensation and leakage, and vibration/shock loosens connectors or excites resonances. This page focuses on the measurement chain that turns those physical stressors into robust, low-false-alarm alerts—from sensor placement to analog conditioning and alarm logic.

• What is covered (measurement objects) Temperature (ambient/hotspots), humidity & condensation risk (dew-point margin), vibration (RMS/band energy), and shock (peak/impulse capture).
• What is covered (signal chain) Sensor interfaces (incl. IEPE accelerometers), analog front-end choices (excitation, coupling, gain, filtering), ADC sampling/trigger capture, and alarm decisions (threshold, hysteresis, debounce).
• Outputs the design must produce Discrete alarms or minimal status messages (severity bucket + key metrics), with predictable latency and hardened behavior under EMI and shock.
• What is intentionally not covered (to avoid overlap) No system-level BIT/BIST architecture, no mission-computer logging pipelines, no power-rail/hold-up design, and no AFDX/PTP/SyncE protocol deep dive.

The core promise is simple: the reader should be able to translate an environmental or vibration concern into a concrete set of chain parameters—range, noise floor, bandwidth, dynamic range, sampling rate, and alarm rules—and then verify the result with repeatable tests.

H2-2 · Environment & vibration requirements: what actually matters

Intent: translate “DO-160 / bay conditions / vibration severity” into chain parameters that can be designed, purchased, and verified.

Requirements are only useful when they answer a design question: what must be measured, over which bandwidth, with what dynamic range, and how the result becomes an alarm. For avionics bays, the most common failure mechanisms map cleanly to a small set of measurement primitives—temperature/humidity (condensation risk) and vibration/shock energy.

• Temperature: accuracy is not enough Range sets survivability. Drift decides long-term trust. Response time (thermal τ) is dominated by mounting and airflow. Treat “what is actually being thermally coupled” as a first-class requirement.
• Humidity & condensation: dew-point margin is the actionable metric High RH alone does not guarantee condensation. The actionable quantity is dew-point margin (T − Td) at the coldest relevant surface. Include sensor hysteresis, contamination drift, and placement (cold spots vs airflow) in the requirement.
• Vibration & shock: use bandwidth + dynamic range + recovery Vibration severity depends on frequency band (resonance vs random), noise density (small motion visibility), range (shock headroom), and overload recovery (how fast readings become valid after an impulse).
• System constraints that can break the design Long cables and EMI can mimic shock spikes. Power limits constrain continuous sampling. Mounting stiffness shifts resonant response. Treat these as requirements because they directly set filter corners, thresholds, debounce windows, and excitation headroom.

A robust requirement statement should always end with a verification intent: “This condition should trigger, this condition should not, and the decision must remain stable for X seconds under noise/EMI.” That framing naturally yields the three alarm styles used throughout this page: instant peak (shock), time-gated RMS (vibration), and dew-point margin with hysteresis (condensation risk).

H2-3 · Temperature sensing chain (RTD/NTC/IC sensor) and placement

Intent: answer “RTD vs NTC in an avionics bay”, “where to place the sensor”, and “why drift/response time dominate field usefulness”.

Temperature monitoring only becomes actionable after the measurement target is defined. “Bay air temperature”, “enclosure surface temperature”, and “PCB hotspot temperature” can differ by tens of degrees during transient loads. The sensor technology choice is secondary to thermal coupling and time constant (τ): mounting and airflow often dominate what the system actually reports.

• Step 1 — Define the measurement target (what is being protected) Ambient/bay trend: representative airflow point.
  Hotspot reliability: heat sink, power device area, or known thermal bottleneck.
  Condensation relevance: coldest local surface (ties directly to dew-point margin in H2-4).
• Step 2 — Choose sensor type by boundary conditions (not by slogans) NTC: fast and low-cost; best for threshold-style alarms and relative changes; manage non-linearity and batch variation.
  RTD: better linearity and long-term stability; best when drift control and calibration traceability matter; requires controlled excitation and wiring.
  Digital temperature IC: simplest wiring and integration; still requires good placement—digital output does not remove thermal-path error.
• Step 3 — Wiring strategy (2/3/4-wire) is an error-budget decision 2-wire: acceptable only when lead resistance/temperature drift is negligible vs allowed error.
  3-wire: practical compromise; relies on lead matching to cancel most lead error.
  4-wire: preferred when long leads or tighter accuracy targets make lead error dominant.
• Step 4 — Thermal coupling decides response and “what temperature is seen” PCB attach: fast PCB tracking; not equal to air temperature.
  Bolt/clip to metal: best for enclosure/hotspot surfaces; stable but can be slower depending on interface.
  Thermal pad/adhesive: interface resistance shifts τ and steady-state offset; treat as part of calibration.
• Common pitfalls (field failures that look like “bad sensors”) Self-heating (excitation too high), lead resistance drift (long harness), using air temperature as a proxy for hotspot temperature, and calibrating the sensor but not the installed system (thermal path + mounting).

H2-4 · Humidity, condensation & dew point: turning RH into actionable alarms

Intent: answer “condensation detection”, “dew point alarm logic”, and “why RH alone causes false alarms or missed events”.

Relative humidity (RH) is not a direct condensation risk indicator. Condensation occurs when a relevant surface temperature falls below the air’s dew point. An avionics-bay alarm must therefore be built around an actionable metric: dew-point margin (surface temperature minus dew point), combined with hysteresis and time qualification to prevent boundary chatter.

• Actionable metric: dew-point margin (not RH) Inputs: T and RH → compute dew point (Td) → compute Margin = Tsurface − Td. Risk rises sharply as the margin approaches zero. The most important “temperature” is the coldest relevant surface, not a random airflow point.
• Sensor selection boundary: polymer RH vs integrated T+RH module Separate RH + T: workable when both measurements represent the same air mass and are close enough to avoid spatial mismatch.
  Integrated T+RH: easier integration and consistent pairing, but still sensitive to placement, membrane lag, and contamination drift.
• Hardening for long life: contamination, hysteresis, and response lag Dust/oil films and conformal-coating outgassing can bias RH sensors and increase hysteresis. Protective membranes reduce contamination but add response delay—alarm logic must be stable under slow sensor recovery and real airflow transients.
• Alarm design: bands + hysteresis + debounce Use Watch / Warn / Fault bands on dew-point margin. Add hysteresis to avoid toggling at the boundary and time qualification (debounce) so short spikes do not trigger maintenance actions.
• What makes alarms credible in the field The decision should be reproducible under repeated environmental transitions. If the alarm cannot be verified with controlled temperature/RH steps and a defined cold-surface proxy, the requirement is incomplete.

H2-5 · Vibration monitoring overview: what you can and cannot infer

Intent: cover “RMS vs peak vs crest factor”, “band energy metrics”, and practical sampling strategies without drifting into predictive maintenance or root-cause diagnosis.

Vibration monitoring in an avionics bay is most reliable when treated as a severity monitor. It can produce repeatable numeric indicators (energy, peaks, event counts) and stable alarms. It should not be used to directly claim a specific mechanical root cause without additional sensors, models, and operating context.

• Metrics that are reliable (and verifiable) RMS (overall energy), Peak and Peak-to-peak (impulse sensitivity), Crest factor (peak/RMS), Band energy (frequency-selective severity), and Shock count (event counting with debounce).
• What should not be inferred directly Avoid turning a single sensor’s severity metrics into conclusions like “bearing failure” or “structural crack confirmed”. A single-point measurement is shaped by mounting stiffness, structural transfer paths, and local resonances; it is best used to say “severity increased” or “a band is elevated”, not to label a component-level fault.
• Three alarm styles that work well in the field RMS over a time window (stable, random vibration), band energy with persistence (resonance-like elevations), and instant peak with debounce (shock/impulse). Add hysteresis to prevent boundary chatter.
• Sampling strategy: continuous vs triggered capture Continuous low-rate monitoring supports low-power trending of slow dynamics and low-frequency bands. Triggered high-rate bursts capture short shock events and higher-frequency content without running the ADC at full rate all the time. Trigger sources may be RMS/peak thresholds or an AFE-level comparator (concept-level).
• Design check: keep metrics credible under EMI and overload Peak and shock counters are sensitive to clipping and EMI spikes. The front-end must define overload recovery and include spike rejection rules, otherwise “peaks” become wiring/EMI artifacts rather than motion.

H2-6 · IEPE accelerometers: the electrical model engineers actually design for

Intent: cover “IEPE interface”, “constant-current excitation”, “compliance voltage”, “bias monitoring”, and AC coupling / HPF selection.

An IEPE accelerometer is best designed from an electrical viewpoint as: a two-wire sensor powered by a constant current source, producing an AC signal riding on a DC bias level. The front-end must preserve useful low-frequency content, avoid clipping under large acceleration, and provide bias-based health checks to detect open/short faults and out-of-range operating points.

• IEPE mental model (what the chain must support) Constant current excitation feeds the sensor over two wires. The sensor presents a bias node with an AC vibration signal superimposed. Any design that ignores the bias level, compliance headroom, or overload recovery will produce misleading “peak” and “shock” metrics.
• Key design knobs that determine success Excitation current (Iexc), compliance voltage (Vcomp), bias monitor window, and open/short detection. These parameters decide maximum signal swing, clipping behavior, and whether the chain can detect missing sensors or harness faults.
• Bias monitoring as a health signal The bias level can be tapped and digitized (or thresholded) to identify out-of-range conditions. This supports “sensor present / wiring healthy / operating in range” checks without requiring any system-level diagnostic framework.
• AC coupling and HPF corner (fc) selection AC coupling removes the DC bias but introduces a high-pass corner (fc). If fc is too high, useful low-frequency vibration content is lost. If fc is too low, overload recovery and baseline settling can become slow. fc should be selected based on the lowest frequency of interest and the required post-shock recovery time.
• Overload and recovery define peak credibility Under large shocks, clipping and recovery time directly impact peak detection and shock counts. The chain must define a “valid-after-overload” behavior so peak metrics are not dominated by saturation artifacts.

H2-7 · IEPE AFE design: protection, filtering, gain, and dynamic range

Intent: cover “IEPE AFE schematic”, “anti-alias filtering”, “gain planning”, and “overload recovery” as the factors that make vibration metrics credible.

A robust IEPE analog front-end (AFE) must survive transients, preserve low-noise small-signal fidelity, avoid clipping under shock, and return to valid measurement quickly after overload. The design focus is not a single gain value—it is a dynamic-range budget tied to protection leakage/capacitance, analog filtering, ADC full-scale, and recovery behavior.

• Input protection without corrupting the signal ESD/transient protection is required, but leakage and capacitance must not shift the bias window or dominate the noise floor. Series impedance must be controlled so it does not create excess thermal noise or attenuate the band of interest.
• Gain planning is a dynamic-range budget The chain must resolve small vibration while keeping shock events below clipping. Practical approaches include switchable gain (low/high ranges) or dual paths (one low-gain for shock headroom, one high-gain for low-noise RMS). The goal is stable RMS and credible peaks.
• Anti-alias filtering: analog does the blocking, digital does the shaping Analog low-pass filtering must limit out-of-band energy so it cannot fold into the band of interest. Digital filtering can then shape windows and band-energy calculations. If analog anti-alias is weak, “band energy” becomes alias artifacts.
• Overload recovery defines peak credibility Shock events can saturate the PGA/ADC and disturb the AC-coupling baseline. The system must define when metrics become valid again, otherwise peak and shock counts are dominated by saturation/recovery artifacts rather than motion.
• Practical design-check list (chain-level) Compliance headroom and bias window are maintained under cable drop; protection leakage/capacitance is bounded; LPF corner matches sampling rate; and post-shock recovery time meets the intended trigger-capture windows.

H2-8 · Sampling & conversion: choosing ADC, rate, and trigger capture

Intent: cover “vibration ADC sampling rate”, “effective dynamic range (ENOB)”, and trigger capture with ring buffers and pre/post windows.

Sampling choices should start from the highest frequency of interest and the analog anti-alias roll-off, then confirm that the effective noise floor (AFE + ADC) supports the smallest vibration level that must be resolved. Trigger capture is best implemented with a ring buffer so the record includes both pre-trigger context and post-trigger recovery.

• Sampling rate follows bandwidth plus anti-alias margin Choose the sample rate (fs) to cover the target bandwidth and leave margin for the analog LPF transition band. If fs is chosen too close to the band edge, out-of-band energy will alias into “band energy” metrics and produce misleading severity indicators.
• Resolution is not the same as usable dynamic range ENOB and the combined noise floor of the AFE and ADC determine what small vibration can be measured reliably. Once the AFE noise dominates, adding ADC bits does not improve RMS stability.
• ADC selection boundary (conversion behavior matters) For transient capture and burst triggers, conversion settling and overload behavior can matter as much as nominal resolution. The ADC must pair cleanly with the AFE LPF and the intended trigger-burst strategy.
• Triggered capture: ring buffer with pre/post windows A continuously updated ring buffer supports “always-ready” capture. When a trigger occurs, the system freezes and saves a window that includes pre-trigger history and post-trigger recovery, enabling both severity confirmation and recovery timing checks.
• Windowing ties directly to metrics RMS uses a defined time window; peak/shock events use debounce; band energy uses band-specific windows. These must remain consistent with fs and filtering to keep the dashboard metrics internally consistent.

H2-9 · Event detection: thresholds, hysteresis, and false-alarm hardening

Intent: cover “vibration alarm threshold”, hysteresis/debounce/time-gating, and EMI-spike hardening while staying at chain-level (bias/open/short only).

Reliable alarms come from a layered decision model: instant shock detection, short-window vibration severity, and long-window baseline drift. Each layer must include hysteresis, persistence time, and gating rules so alarm state changes are repeatable across EMI exposure, mounting variations, and overload recovery.

• Three-layer decision model (what each layer is allowed to claim) Shock: instant peak with debounce (event). Vibration: short-window RMS/band energy with persistence (severity). Drift: long-window mean/percentile shift (baseline change). The model reports “elevated severity / increased events / baseline shifted” instead of component-level root-cause claims.
• Mandatory hardening primitives Hysteresis (Th_up / Th_down), debounce (minimum samples or repeats), time gating (startup/mode-switch/overload blanking), and sensor gate (bias out-of-range, open/short). Sensor checks stay at interface level only—no system BIT architecture.
• False-alarm hardening: EMI spike vs real mechanical shock EMI spikes are often single-sample or extremely short with weak band-energy persistence. Real shocks typically show multi-sample duration and/or band-energy elevation consistent with structural response. Practical rules: require minimum duration, require a short follow-up window confirming energy, and ignore “events” during overload recovery.
• State transitions must be explainable and reversible Use persistence timers for escalation (e.g., “RMS > Th for T seconds”) and a separate clear rule (e.g., “RMS < Th_down for T_clear”). This prevents chatter at boundary conditions and makes Watch/Warn/Fault behavior predictable.
• Verification framing (chain-level) Validate with controlled vibration profiles plus injected transients: confirm stable RMS thresholds, bounded false-peak rate under EMI, and deterministic state transitions with blanking and hysteresis enabled.

H2-10 · Low-power MCU alarms: wake strategy, outputs, and minimal telemetry

Intent: cover “wake-on-event vibration monitor”, low-power continuous watch + burst capture, discrete alarms, and a minimal telemetry field set (interface-level only).

Low-power monitoring is best implemented as a layered runtime: a sleep monitor performs continuous low-cost metrics, and an event wake enables short high-rate capture for evidence and recovery timing. Outputs should be simple and robust: discrete alarm lines plus an optional minimal message containing only the fields required for triage.

• Power strategy: low-rate watch + event-driven burst Keep a continuous baseline using low-rate sampling and lightweight RMS/band checks. When thresholds are met, wake into a bounded high-rate capture window (pre/post) to confirm the event and measure recovery time.
• Wake sources and gating (chain-level) Wake can be driven by peak candidate, RMS persistence, or band-energy elevation. Apply blanking during mode transitions and overload recovery to prevent repeated wake storms and alarm chatter.
• Output forms that integrate cleanly Discrete alarms (OK/Watch/Warn/Fault) provide the most robust integration. A minimal bus/register message can be added for context without expanding into protocol stacks or network architecture.
• Minimal telemetry set (keep it small, keep it useful) Recommended minimal fields: alarm level, local timestamp, peak magnitude, persistence/duration, RMS summary, and band tag. This supports triage without becoming a logging system.
• Rate limiting and spam control Limit report frequency and coalesce repeated events. This keeps power predictable and prevents burst loops from dominating system activity.