BME 688 Training/Test data in lab

[2borN2b]

Hi Everyone, 

Going thru the BME688 documents to generate the AI algorithm, I see the need for good training data for the sensor. This allows for a higher accuracy of the AI algo. Field or test data can then be analyzed versus the AI algo for detection of the compounds trained for. 

From my notes from the BME documentation - 

Two types of recording data needed - 

1. Controlled or Lab - To perform initial training of the sensor. 
2. Uncontrolled or Field - Test with gas or compounds that are not controlled as available in a lab. 

Question here for the community is, is anyone familiar with a Laboratary that can provide the controlled data and has worked with the BME688 or similar sensor? We're looking to work with them to train the sensor to a higher accuracy. 

Thanks in advance for any inputs or suggestions! Cheers 

[BSTRobin]

Hi 2borN2b,

Could we know what type of gas do you want to detect?

[2borN2b]

Hi BSTRobin, 

Thank you for responding. We're interested in being able to train the sensor to detect various types of vape liquid smoke and cigarette/cigar smoke to a plausible degree of accuracy. If we can accomplish a plausible degree of accuracy, we're also looking for cooking gas (Hydrogen sulphide). 

Intent is to use an array of BME688 sensors for monitoring air inside a house or apartment. 

[2borN2b]

Bumping this up if there's any information folks can share. Thank you!

[BSTRobin]

Hi 2borN2b,

Before algorithm training, you can try to manually adjust the data splitting in BME AI-Studio to improve the accuracy.