Moving air through the system in a repeatable way.

A machine cannot smell well if air reaches the sensor randomly. It needs a controlled breath — a repeatable path from sample to chamber and back to a clean baseline.

What we research

Airflow, chamber shape, pump timing, clean-air purging, and recovery time often decide whether a smell reading is useful at all. Aeralyte treats sampling, sensing, and AI as one system — and puts more early effort into active sampling than into exotic sensors.

Technical terms: active sampling, micro-pumps, valves, sensor chamber, purge cycle, flow control.

The path

Clean-air baselineEstablish a reference before each exposure.

Valve switchSelect clean path or sample environment.

Pump + flow controlDraw the sample at a controlled rate.

Sensor chamberExpose the array; record the response.

Purge & recoveryReturn to baseline; check repeatability.

Findings

We track the sampling context as first-class data — room volume, ventilation, and pump flow. An adversarial check confirmed that context metadata alone cannot stand in for the sniff: a metadata-only model reached just 0.60, well short of the sensor-driven baseline.

room_volume_m332.0

ventilation_ach0.6

pump_flow_lpm1.2

sample_period_s15

Next: characterizing real airflow — pump timing, purge, and recovery — on the bench rig we are bringing up now.

References

01AERALYTE_RESEARCH_LAB_BRIEF.md — pillar 2, controlled sniffing.
02PHASE0_HARDWARE_LAB.md — the ESP32-S3 bench rig & sampling workflow.
03Kimi_Agent · Precision Agriculture SmellTech Strategy — field sampling considerations.
04GitHub — XoAnonXo/aeralyte.