Build B-06
A worker who is struck by a vehicle or caught in an accident needs to be detected even when nobody sees it happen.
| GPIO | Sensor | Signal | First sketch |
|---|---|---|---|
| 2 | Flame sensor | Digital output | GPIO4 |
| 5 | MQ-7 carbon monoxide | Sensor output | not yet wired |
| 18 | DHT11 | Temperature and humidity data | same |
| 19 | Shock sensor | Digital output | GPIO16 |
| 21 | Knock sensor | Digital output | GPIO17 |
| 22 | Heart-rate sensor | Pulse edges, best-effort BPM | same |
| 23 | HC-SR501 PIR | Motion output | not yet wired |
The fixed pin map of the ESP32 body sensor node. All sensors run on 3V3. Drawn from the final pin assignment; a photograph of the wearable is still to come.
Not documented yet.
The node is an ESP32 DevKit V1 worn on the body as part of a mine worker safety system. It started as a single Arduino sketch printing flame, shock, knock, temperature, humidity and a pulse count to the serial port every five seconds, and grew into a wearable with seven sensors and a browser dashboard.
The hard problem has not been reading sensors. It has been deciding what counts as an accident. A shock sensor that fires on every step is useless, and one that only fires on a violent collision misses a fall. Three firmware versions have landed on either side of the right sensitivity, which is why this page says Testing and not Stable.
The architecture is deliberately split. The ESP32 stays close to the sensors and does as little as possible; a separate gateway Raspberry Pi (not the camera Pi of the vision link) will do the heavier work of normalising each worker’s baseline and spotting anomalies.
No measurements recorded yet.
Source: No calibrated measurements have been recorded yet. Sensitivity has been judged by testing on the hardware.
Flame, shock and knock sat on GPIO4, 16 and 17.
Moved to GPIO2, 19 and 21 in the final pin map, which is now fixed.
The Web Serial dashboard worked, but shock and knock were far too sensitive.
Alerts re-targeted at a worker being hit by a vehicle or in an accident, not every bump.
On real hardware it still flagged impacts constantly and escalated them to severe.
Impact detection became too insensitive.
The target is now a medium level between v3 and v3.2.
The PIR sensor false-triggered and the CO sensor seemed stuck warming up.
Sensor polarity is to be established by testing rather than assuming HIGH or LOW means detected.
A working wearable that reads all seven sensors and streams them to a live dashboard, with impact detection not yet trustworthy.
TESTING
Impact detection is still being tuned on real hardware.
Settle impact sensitivity, move the transport to Wi-Fi and MQTT with the gateway Pi, then try a model running on the ESP32's own inputs to cut end-to-end lag. LoRa later.
Firmware and dashboard are on the board and the bench, not in a public repository yet.