BME688 gas classification working with devkit but not with single sensor

[dmalcorps]

Hello,

I am using the bme688 devkit to gather data for 2 gas classes. After using ai-studio to generate the bsec config file, I am testing the gas classification with the example application provided under https://github.com/BoschSensortec/Bosch-BSEC2-Library, which works. But if I am trying to use the same configuration with one of our own sensors connected over i2c the classification does not work. I tried three different sensors but none showed the correct gas class. In fact the classification output does not change at all, regardless of the air around the sensor. I hardly believe that all three sensors are somehow defective, given that every other metric, be it temperature, humidity, raw gas resistence, does change according to changes in the air. Does a custom bsec config only works with the dev board? What could be the error?

Kind regards,

Dennis

[Mila]

Thanks for your reply. Yes, i can import the data also from the single board sensor, but seems like the data is not recognised correctly. Probably i'm doing something wrong, in this case could you advise the correct way to collect or export the data? here is what i have on AI Studio. Even though i've collected the data for more then 2 hours the graph show less then one hour of data. And i can't select the spacements correctly, becouse the timeframe i was collecting them resulted with no data at all.

 

[BSTRobin]

Hi Mila,

You could check if the data format is correct before you export data log to BME AI-Studio?

[Mila]

Hi, it seems to be correct, but it's obviosly different from the one, generated by 8 sensors, i have just one. The timestamp is present for all elapsed time frame, the values as well, how can i check if the formati s correct or not. It seems to be just cropped by one hour slot. Any suggestion would be appreciated. I can share the log as well if it's appropriate.

[BSTRobin]

Hi Mila,

For the tested sensors and the trained sensors, BME AI-Studio relies more on the trained sensors. There are some problems with the support of the tested sensors. The sensor aging model function of AI Studio is in the development stage. Currently the best way to tackle sensor aging is by including data from both boards(old and new sensors) for training.
If you record data with sensor(old sensor) ,do the training with BME AI-Studio and generate configuration file. The configuration file is good for the current sensor(old sensor), but the accuracy will vary when it is used for other sensors(new sensors). Because the old sensor is aged, the new sensor is not aged.