[QUALITY] Sensor calibration, accuracy validation & compensation algorithms

[farcomiot]

🔬 Sensor Calibration, Accuracy Validation & Compensation Algorithms

Problem

Raw sensor readings from the Enviro+ hat have known biases and limitations:

• BME280 temperature reads ~5-8°C high due to CPU/board heat proximity
• MICS-6814 gas readings are in resistance (kΩ), not calibrated to concentration (ppm)
• PMS5003 accuracy degrades over time without cleaning/verification
• ICS-43432 noise dB values need reference calibration against a certified meter
• No cross-validation against external reference stations

Goal

Implement software compensation algorithms and establish calibration protocols to maximize measurement accuracy for industrial-grade environmental monitoring.

Proposed Improvements

BME280 Temperature Compensation

• CPU heat offset — measure delta between BME280 and reference thermometer, apply linear correction
• Dynamic compensation — factor based on CPU temperature (higher CPU = more offset)
• Running average — smooth temperature readings over 30-second window
• Humidity correction — BME280 humidity accuracy degrades above 80% RH

# Temperature compensation formula
# Measured offset: ~5.5°C at idle, ~7.2°C under load
def compensate_temperature(raw_temp, cpu_temp):
    # Factor derived from empirical measurement
    factor = max(0, (cpu_temp - 40) * 0.05)
    offset = 5.5 + factor
    return raw_temp - offset

MICS-6814 Gas Calibration

• Baseline resistance — measure clean-air reference values after 48h burn-in
• Rs/Ro ratio calculation — convert kΩ to ratio against clean-air baseline
• Approximate ppm estimation — use datasheet sensitivity curves for NO2, CO, NH3
• Temperature/humidity correction — gas sensor sensitivity varies with ambient conditions
• Document limitations — MICS-6814 is qualitative, not certified for regulatory compliance

Gas	Clean Air (Ro)	Alarm Ratio	Approximate ppm
NO2 (oxidising)	~2-20 kΩ	Rs/Ro > 2.0	~0.5 ppm
CO (reducing)	~100-500 kΩ	Rs/Ro < 0.3	~50 ppm
NH3	~200-800 kΩ	Rs/Ro > 2.5	~25 ppm

PMS5003 Accuracy Maintenance

• Periodic zero-check — cover intake, verify readings drop to 0-2 µg/m³
• Cross-reference — compare with nearby government AQI stations (SIMA Monterrey)
• Sensor aging log — track baseline drift over months
• Cleaning schedule — compressed air every 6 months per Plantower recommendation
• Humidity correction — PM readings inflate above 85% RH (hygroscopic growth)

# Humidity correction for PM readings (EPA recommendation)
def correct_pm_humidity(pm_raw, humidity):
    if humidity > 85:
        # Reduce by estimated hygroscopic factor
        correction = 1.0 + 0.25 * ((humidity - 85) / 15)
        return pm_raw / correction
    return pm_raw

ICS-43432 Noise Calibration

• Reference calibration — compare against certified SPL meter at 3 points (40, 70, 94 dB)
• A-weighting filter — apply frequency weighting curve for human-relevant dB(A)
• Frequency band validation — verify low/mid/high split accuracy
• Background noise floor — characterize minimum detectable level

Cross-Validation with External Data

• SIMA Monterrey API — fetch government air quality data for Monterrey metro
• OpenWeatherMap comparison — cross-check temperature, humidity, pressure
• Automated drift detection — alert if readings consistently deviate >15% from references
• Calibration dashboard page — show sensor health and accuracy metrics

Calibration Protocol Document

Create docs/CALIBRATION.md with:

1. Step-by-step calibration procedure for each sensor
2. Required equipment (reference thermometer, SPL meter, clean air environment)
3. Calibration schedule (monthly temperature, quarterly PM, annual gas)
4. Data recording templates for calibration sessions
5. Compensation formula documentation with derivation notes

Implementation Priority

1. BME280 temperature offset (immediate — biggest impact, simplest fix)
2. PMS5003 humidity correction (quick win for rainy season accuracy)
3. MICS-6814 baseline + ratio (moderate effort, significant data quality improvement)
4. Noise reference calibration (requires physical calibration session)
5. External cross-validation (API integration, ongoing monitoring)

Monterrey-Specific Considerations

• High summer temperatures (35-42°C) — CPU heat compensation critical
• Rainy season humidity (70-95% RH) — PM humidity correction essential
• Industrial zone proximity — gas baseline may be elevated vs rural clean air
• Altitude ~540m — pressure readings need no sea-level correction for local use

---

Ref: #1 Roadmap — Data Quality & Calibration