sensor-data-handling
/sensor-data-handling
Zamienia surowe odczyty z czujników w wiarygodne dane przez właściwe próbkowanie, filtrowanie, kalibrację i grupowanie ze znacznikami czasu.
- Autor
- Amey-Thakur ↗
- Repozytorium
- AI-SKILLS ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 05.09.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 05.09.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zamienia surowe odczyty z czujników w wiarygodne dane przez właściwe próbkowanie, filtrowanie, kalibrację i grupowanie ze znacznikami czasu.
Oryginalny opis (EN)
Turn raw sensor readings into trustworthy data with correct sampling, filtering, calibration, and timestamped batching. Use when building sensor pipelines or debugging noisy, aliased, or drifting measurements.
Jak działa
- Method
- Boundaries
Podgląd SKILL.md
--- name: sensor-data-handling description: Turn raw sensor readings into trustworthy data with correct sampling, filtering, calibration, and timestamped batching. Use when building sensor pipelines or debugging noisy, aliased, or drifting measurements. --- # Sensor data handling A sensor gives you a noisy, biased, drifting proxy for reality. The pipeline's job is honesty: sample fast enough, filter with intent, calibrate against truth, and never let timestamps or units get ambiguous. ## Method 1. **Sample at the signal's rate, not the firmware's convenience.** Nyquist demands >2x the highest frequency of interest; below that, aliasing folds fast phenomena into plausible-looking slow lies (mains hum becoming a "drift"). Anti-alias in analog or with sensor-side filtering before downsampling digitally; jitter in sampling intervals is itself noise, so sample from timers/DMA, not from "whenever the loop gets there" (see interrupt-safe-code's handoff for the ISR side, rtos-task-design for the rate task). 2. **Filter with a chosen tool, not an accumulated habit.** Moving average for cheap smoothing (know its lag); EMA/IIR for memory-light smoothing with tunable response; median filter for spike rejection (outliers from EMI, loose wires); complementary/Kalman-class fusion when combining sensors (IMU + magnetometer). Every filter trades latency for smoothness: control loops need that trade made explicitly (a laggy filter inside a fast loop oscillates; see rtos-task-design deadlines). 3. **Calibrate against reference, store per-device.** Offset and gain (two-point) at manufacture, stored in NVM with the calibration date; temperature compensation where the datasheet's tempco matters; field re-zeroing flows for drift-prone sensors (gas, load cells). Uncalibrated fleets produce beautiful dashboards of nonsense: per-device calibration is what makes cross-device comparison meaningful (see data-quality-checks' spirit at the source). 4. **Timestamp at acquisition, in one time base.** Stamp when the reading was *taken* (not when transmitted), monotonic locally, mapped to real time at sync points; batching and store-and-forward then preserve ordering and gaps (see clock-skew: the device clock drifts, so record sync offsets rather than rewriting history). Sequence numbers detect loss; explicit gap markers beat silently missing rows (see delivery-guarantees). 5. **Reduce on-device, ship what the question needs.** Raw 1kHz vibration becomes RMS/peak/spectral bands per window; temperature becomes minute min/mean/max: aggregation near the sensor cuts radio cost 100x (see low-power-design, iot-messaging) while keeping the analytics honest: design aggregates *from the downstream question* (see warehouse-modeling's grain thinking), and keep a raw-burst diagnostic mode for the field engineer. 6. **Flag quality at the source.** Out-of-physical-range, stuck-at (no change beyond deadband for too long), rate-of-change violations, sensor self-test failures: attach validity flags rather than dropping or "fixing" silently (see data-quality-checks' quarantine rule): the platform can discard flagged data, but it cannot recover data the device judged and deleted. ## Boundaries - Control-loop feedback filtering and telemetry filtering have different latency budgets; do not share one filtered value between them without checking both masters. - Sensor selection, analog front-end design, and EMI layout decide the noise floor before firmware sees a bit; when the data is garbage at ADC, the fix is hardware review, not filtering heroics. - ML-at-the-edge (anomaly detection on-device) rides on this pipeline's quality; deploy it after the calibration and quality-flag layers exist, not instead of them (see ml-baselines).
Wersje
W bazie nie ma innych skilli o nazwie sensor-data-handling.
Migawka bazy z 13.11.2026 · wersja paczki f7ec23fc7f68 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90esp32-sensor-sampling-synchronization
Synchronizuje próbkowanie wielu czujników podłączonych do ESP32, żeby odczyty były ze sobą porównywalne w czasie. - 100sensor-patterns
Podpowiada wzorce sterowników czujników w C: protokół SPI, przeliczanie termistora NTC, kalibrację i filtr medianowy. - 100sensor-signal-filtering
Diagnozuje zakłócony sygnał analogowy z przetwornika ADC i dobiera filtrację sprzętową oraz programową dla czujników. - 100esp32-pressure-sensor-compensation
Kompensuje odczyty czujnika ciśnienia na ESP32 z uwzględnieniem temperatury i zmian kalibracji. - 96skyeye-sensor-use
Odczytuje listę alertów i statystyki z czujnika ruchu SkyEye Sensor przez prosty interfejs wiersza poleceń. - 100client-sdk-ros2-integration
Pomaga zintegrować strumienie czujników Aria Gen 2 z ROS2: publikowanie danych przez tematy, dekodowanie w węzłach subskrybujących i współdzielenie kalibracji urządzenia.