BME688 AI Application fails

[bsec]

HI,

I created 2 AI models, one for coffee similar to the youtube video and one for foul vegetables. Both algorithms produce probabilities > 95%. I checked the configuration file output from AI Studio with 3 different applications: Arduino ESP32 (exact the ino example in BSEC 2.0), ESP32 in eclipse development environment and on a STM32 nucleo platform, all linked with libalgobsec.a (depending of the MCU used). In every test I get the result Gas Estimate 1 (normal air -> 1.0). Whatever I do I never get any probability > 0.0 in Gas Estimate 2, even if probe within the Coffee glas or in the fouled vegetable container. Any experience with BME688?

For sampling I used BME 688 dev kit board (8 sensors) on Adafruit HUZZAH32

BR

[BSTRobin]

Hello bsec,

With the new config file you trained, have you finished test?

[bsec]

Hi,

I use basic_config_state.ico, linked with the newest BSEC_2.0.6.1, (Arduino IDE, / eclipse / a modified .ino with STM32 MPU with STM Cube IDE) use the .h and .c file from the AI package and I always get "normal air", whatever I take as second class, coffee or foul vegetables. I have seen that other users here in the forum have similar problems. So could you supply a working example from getting the probes to running the application. The Youtube video stops after creating the .h and .c files

BR

[RuiMPCosta]

Good afternoon,
I have exactly this problem. Have you had any results?
I use the BME688 feather with all eight sensors. The results are exactly what you quoted.
Regards

[bsec]

Hi,

no reply from Bosch so far, no solution

BR

Bernd

[BSTRobin · Accepted solution]

Hello bsec,

Sorry for the delay reply.

Suggest you collect raw data for at least 30 minutes at every step. And check if gas resistance changes or not, in case that sensor doesn't work.

Please take this video as reference.