🐝 TinyML Hive Anomaly Detector

About this simulation

A coin-sized microcontroller clipped inside a beehive can listen to the colony continuously — something a beekeeper visiting once a week never can. This simulation places a tiny sensor board (MEMS microphone + accelerometer) inside a cutaway hive and streams a simulated signal through a miniature on-device neural network, exactly as real TinyML devices such as Arduino Nano 33 BLE or ESP32-based hive monitors do in the field.

How it works

• Signal synthesis — a healthy hive hum (~200–300 Hz wingbeat frequency) is mixed with random noise scaled by colony activity.
• Varroa signature — mite feeding and irregular brood-cell capping introduce a higher-frequency, bursty vibration pattern; its strength tracks the infestation slider.
• On-device inference — the animated node layers represent a quantized tiny CNN running directly on the microcontroller's flash-resident weights, lighting up in proportion to feature activation.
• Anomaly score — a smoothed 0–100 score is compared against the sensitivity threshold; crossing it flips the model verdict and lights the alert beacon red.

Controls

• Colony activity — overall hive noise floor from bee movement and fanning.
• Varroa infestation — strength of the mite-related vibration signature injected into the stream.
• Detection sensitivity — how low the anomaly score must stay before the model calls it "normal"; higher sensitivity flags mites earlier but risks more false alarms.
• Sensor feed — switch between the microphone and accelerometer channel; the on-board waveform screen and active sensor LED update accordingly.

Did you know?

Real TinyML hive monitors run models as small as a few hundred kilobytes on microcontrollers drawing only milliwatts, letting solar or battery-powered sensor nodes report mite risk for months between visits — instead of waiting for visible colony collapse.