Hive Telemetry and Honey House

Sampling, noise, alert delay and the slowest stage of the line.

Switch to 2D

Presets

Playback

Sensors and sampling

How the scale and its radio sample the hive, and the interfering vibration.

Alert rules and the hive

Block averaging, thresholds, the swarm and the nectar flow.

Honey house

Hives to harvest and the speed of each stage in the line.

Actions

Camera

Quality

Layers

Model checks

The science

What is measured

A hive stands on four load cells, so the scale reports the whole colony mass: bees, comb, stores and the hive itself. Here the true mass starts at 42 kg, gains nectar through a half-sine foraging bump between 08:00 and 18:00 (the slider sets the net gain per day), and loses 0.03 kg per hour to evaporation and upkeep. A swarm leaving is a sudden step down. A second probe in the brood nest reads about 34.5 C, which a healthy colony holds within roughly 32-36 C; after a swarm the smaller cluster holds it a little less well for a while. Numbers are plausible orders of magnitude for a teaching model, not data from a real apiary.

Sensor noise and averaging

Each sample is mass + vibration + e with Gaussian noise e ~ N(0, sigma^2). Averaging a block of m samples cuts the noise to sigma / sqrt(m), at the price of waiting m / fs hours for the block.

Sampling theorem and aliasing

A signal containing frequencies up to f can be recovered only if the sampling rate satisfies fs > 2f (the Nyquist limit is fs/2). A frequency f above that limit is not lost: it folds down to the apparent frequency |f - round(f/fs) fs| and masquerades as a slow wobble that no later averaging can remove, because it is already in the numbers. The cure is an analogue low-pass filter before the converter (the "anti-alias filter" switch, a first-order filter with its corner at fs/4), or a higher fs. The vibration in the lab stands for any periodic disturbance such as a ventilation fan or a wind-driven sway of the stand.

Alert rule, false alarms and delay

The detector compares the mean of the latest block with the mean of the block before it and raises an alert when the mass fell by more than a threshold th. The difference of two block means has noise sigma_d = sigma sqrt(2/m), so with no real event the chance of a false alert per check is Q(th / sigma_d) where Q is the normal tail probability. A real drop D is caught with probability Q((th - D) / sigma_d) by the first complete block after it, never sooner than one block length. Lowering the threshold catches smaller events but multiplies false alarms; longer blocks improve both at the cost of delay. The trade-off graph plots exactly that curve. The brood-temperature alarm is a simple level rule with hysteresis (one alert per crossing).

Honey house line

Three stages in series: the uncapper (frames per hour), the extractor (a load of N frames every cycle of load + spin + unload minutes, so N * 60 / cycle frames per hour) and the bottler (jars of 340 g per hour, converted to frames with the honey per frame). Whatever the buffers do, the steady throughput is the minimum of the three capacities, and speeding up any other stage changes nothing. The tank (400 kg) and the frame buffer only smooth the flow. An apiary's harvest of hives x 10 frames then takes frames / throughput hours.

Simplifications

Independent Gaussian noise, one swarm event per run, an idealised foraging curve, no robbing, queen loss or disease signatures, and a detector that is deliberately simple (real monitors use trends, several sensors and day/night baselines). Nothing here is a product specification or an approval of any equipment; frame counts, speeds and jar sizes are adjustable example values. Time runs at 3 simulated hours per second at normal speed, so a week lasts under a minute.

What to try
  1. Leave the defaults: the fan vibration (5.5 per hour) is faster than the Nyquist limit (3 per hour), so graph 3 shows a false 0.5 per hour wobble. Press "Fast enough: 20 samples per hour" or switch the anti-alias filter on and the wobble disappears.
  2. Look at the mass graph: the red lines are alerts. Press "Hair-trigger alert" and count the extra false alarms per week; then read how little earlier the real swarm is caught.
  3. Press "Swarm leaves now" and compare the measured alert delay with the block length m / fs.
  4. Move m from 1 to 20 on the trade-off graph: the curve rises toward the top-left corner (more catches, fewer false alarms).
  5. In the honey house press "Bottleneck: slow bottler", then raise the extractor load: the throughput does not move until the bottler is faster than the other stages.
Loading Blender model...
Hive mass: raw samples, true mass, block means and alerts
Brood-nest temperature and its alarm level
Sampling theorem: the true vibration against what the samples show
Alert trade-off: chance to catch the swarm against false alarms per week
Honey line: capacity of each stage and the measured output
Honey line: tank honey and frames waiting over time
Made with 🟠 Blender · Three.js