I have and ESP32 custom board which has BME688 on it. I also have the bosch 8 sensor board for training . I can successfully run the BSEC2 library on my board using the example which uses the supplied "field_air_hand_sanitizer" config file. It will successfully detect hand sanitizer when present.
My issue is the following:
I cannot get my own trained scents to work.
I successfully collect 30 minute samples of smells like field air (class 1), cheese(class 2), ipa beer(class 3).
I successfully import them into AI studio and build a 3 smell algorithm which shows 98+% probability of telling the difference.
I use the same example and replace the hand sanitizer includes with mine.
I assume when you build an algorithm with 3 classes, that class 1 will be gas1, class 2 will be gas2, etc...
I get 3 gas results but the results are odd. For example, when sensor is in clean field air I get 40% gas2, and 56% gas3, 4% gas1
If I move the sensor toward IPA source gas 1 goes to 98%, indicating class 3 is really gas 1.
Introducing cheese that was used for training simply returns to same settings as field air.
Any help to explain the steps after the successfully training of the algorithm and how to integrate that config file properly into your application?
I think I am doing things right and I am getting results but sensor feedback works extremely well for IPA, but sensor seems confused over feild air and cheese with no clear indication.
Solved! Go to Solution.
BME688 is more sensitive to alcohol.
After generating the configuration file from BME AI-Studio, you can run the algorithm in BME688 dev kit and test it. After confirming that the algorithm is correct, replace the configuration file of the hand sanitizer example and test it on ESP32 platform.