Gas Scans and Fire Models

After Phase 1 alerts, gas scans are compared to Fire Models to determine if the air quality deterioration is the result of smoldering fires.

A Phase 1 alert is not necessarily an indicator of a fire (under normal conditions). It only indicates a change in IAQ value where IAQ > 50. This alert triggers a process in the BME688 sensor which is a prescribed set of gas scans, each of which is compared to a ML Model in the firmware.

Up to 60 gas scans are run

After a Phase 1 alert is triggered, the BME866 gas sensor begins to run up to 60 gas scans to determine if a fire is the most likely cause for the deterioration of air quality. (1) Since gas scans consume a lot of energy, they are reduced to the minimum required and only start after a Phase 1 alert has been triggered.

1. Each scan consists of individual measurements of the raw resistance values taken at different temperatures. The ML model deployed on the sensor takes these measurements as input to determine if the observed change can be classified as wildfire smoke.

Each gas scan is compared to a Fire Model

For each gas scan, the sensor compares the results of the gas scan with the Fire Model (ML Model) (1) loaded in the sensor's firmware. This procedure determines if the change in air quality is from a fire or from some other source. If during one of the gas scans the sensor determines that the deterioration of air quality has a high probability of being from a smoldering fire, the sensor stops running the gas scans and immediately issues a Phase 2 alert.

1. ML (Machine Learning) Models are pre-trained in the Dryad laboratory and can be fine-tuned / programmed specifically for the species of trees present in a particular deployment. Updated models can be remotely installed in the sensors by Dryad’s support team using FUOTA. Firmware update Over the Air (FUOTA) is a standard for distributing firmware updates using unicast or multicast. It allows firmware updates to be delivered to many devices (Silvanet Sensors) at the same time efficiently and securely.