System Architecture

From buzz to browser.

Data flows in one direction — sensors, edge inference, sensor fusion, MQTT, dashboard — with commands flowing back the other way. Every step is designed for low power and weak-signal environments.

Sensors

Microphone, BME280, load cell sample the hive continuously.

Edge AI

MFCC + NN runs on-device. Outputs class probabilities.

Fusion

Weighted score combines audio confidence + environment.

MQTT

JSON payload published over WiFi to HiveMQ broker.

Dashboard

React app subscribes, animates and alerts in real time.

Sensor fusion formula

score =
  0.35 × audio_normal_confidence × 100 +
  0.25 × (100 - |temp - 35| × 18)     +
  0.20 × (100 - |hum  - 57.5| × 3)    +
  0.20 × clamp((weight - 20) × 4, 0, 100)

status = score ≥ 75 ? Healthy
       : score ≥ 50 ? Warning
       : Critical

MQTT topics

pub

beeguardian/hive-01/state

Full telemetry JSON every 30s

pub

beeguardian/hive-01/alert

Emitted when anomaly detected

pub

beeguardian/hive-01/health

Heartbeat + battery + solar

sub

beeguardian/hive-01/cmd/sleep

Set deep-sleep duration (sec)

sub

beeguardian/hive-01/cmd/calibrate

Trigger tare of load cell

sub

beeguardian/hive-01/cmd/reboot

Soft reboot ESP32