Journal

HVAC Predictive Maintenance: IoT Monitoring of Air Conditioner Health and Compressor Before Breakdown

HVAC Predictive Maintenance: ใช้ IoT วิเคราะห์สุขภาพเครื่องแอร์และ Compressor ก่อนเสียหาย

May 12, 2026 · 1 min read
HVAC Predictive Maintenance: IoT Monitoring of Air Conditioner Health and Compressor Before Breakdown

An air conditioner failing in Bangkok during summer (35°C+) is a household nightmare, and it almost always results from problems that accumulated gradually before the breakdown. IoT Predictive Maintenance detects warning signs 2–4 weeks in advance.

Vibration Sensor: Compressor Health Monitoring

SW-420 vibration sensor or ADXL345 3-axis accelerometer mounted on the outdoor compressor unit, connected to ESP32 with ESPHome firmware, sends data to Home Assistant via MQTT.

Normal vibration for a 1.5-ton Inverter compressor: 1–5 mm/s RMS. Values above 8 mm/s indicate compressor wear or imbalance. Values above 15 mm/s indicate serious damage requiring immediate service.

Power Consumption Anomaly Detection

A watt-measuring smart plug (Shelly 1PM or Tasmota-flashed Sonoff Pow R2) at the AC circuit breaker monitors consumption. A 1.5-ton Inverter at full load normally draws 800–1,200W. Sustained consumption above 1,400W may indicate: clogged filter (increased compressor load), refrigerant leak (compressor working harder), or dirty condenser fins.

Home Assistant uses anomaly detection (Holt-Winters forecasting or Z-Score) to flag power consumption deviating more than 20% from seasonal baseline → Yellow alert.

Filter Dirtiness Sensing

A differential pressure sensor (Honeywell ABPDRRT, THB 300–800) measures air pressure before and after the AC filter. Rising pressure differential indicates filter clogging. When pressure drop exceeds baseline by 30% → LINE notification to change filter.

Budget alternative: a Wi-Fi camera (Wyze Cam, Reolink) photographs the filter every 2 weeks. HA Image Processing with brightness analysis checks filter colour (white = clean, grey/black = replace).

Refrigerant Leak Detection

A refrigerant leak sensor (R32/R410A compatible, such as Figaro TGS2611-E00) near the indoor unit monitors for Freon leaks. On detection of R32/R410A above 500 ppm → LINE alert, automatic AC shutdown via smart plug, and ventilation fan activation.

Maintenance Dashboard in Home Assistant

Dashboard displays: last filter change date, compressor runtime hours (typical compressor life 50,000–80,000 hours), 30-day average power consumption versus today, vibration level, and upcoming maintenance prediction.

Automation: monthly LINE summary report to homeowner with maintenance recommendations and estimated costs.

Predictive Maintenance ROI

Total sensor system cost: THB 2,000–5,000. Emergency compressor repair: THB 5,000–15,000. Preventing one emergency failure recovers the entire hardware investment immediately. Clean filters and fins also reduce electricity consumption 10–15% year-round.

Questions & answers

Can vibration sensors be installed on outdoor compressor units exposed to rain and sun?
Only sensors with IP65 rating or higher are suitable for outdoor installation. A standard SW-420 has no IP rating and must be housed in a waterproof junction box. ADXL345 in a waterproof enclosure (THB 50–100) is more appropriate. The ESP32 for MQTT transmission also requires a waterproof enclosure.
Do Inverter and non-inverter ACs show different vibration patterns?
Very differently. Inverter compressors change speed continuously with load, creating constantly varying vibration — requiring separate baselines for low, medium, and high load levels. Non-inverter units produce more consistent vibration (on/off only), making anomaly detection simpler.
If an AC brand already shows error codes on the remote, are additional sensors still needed?
Error codes indicate failures that have already occurred (reactive), not predictions (predictive). Sensors detect gradual degradation before it reaches the threshold that triggers an error code. Both systems are complementary — sensors predict, error codes confirm.

Related reading