Gas estimation influenced by alcohol?

[brianchen]

Hi,  I'm using BME688 with BSEC 2.4.0.0. I train the sensor to detect Oolong Tea and Clean Air. In general cases, it can identify that sensor is placed in a tea box or in a open air area. But when I put the sensor outside the box and I use little alcohol to clean desk, the sensor report that it has detected the tea. I'm confused with the performance. I'm using 2 estimation now (clean air and tea), should I add a type called Alcohol in training specimen to improve it?  

[Minhwan · Accepted solution]

Hi brianchen, 

Why that issue happened was Alcohol could be more similar to tea  VOC value than normal air. 

There are three ways to avoid that issue. 

1) Accept when the possibility is over high level. 

- Our app shows tea or air when each value is over 50% probability. In our BSEC library, you can see that we provide probability and range is 0 ~ 1. (1 is 100%). So, in your application, you might show "Tea" if only tea probability is over some level like 0.9 or 0.95. 

2) Train potential gases including alcohol 

3) Apply both suggsetions.

You might have similar issue when other gases are coming. Then, AI studio shows among your specimens. 

Therefore, I recommend 3) Apply both suggestions  option. 

Thank you. 

[brianchen]

Make sense. Thank you!