- Top Results
Dears,
I'm testing BME688 in initial 150 pcs batch production and compare to previous 10 pcs development sample.
BSEC Software lib version is 2.4 works with BME688 API lib (from github). Ported to running on ATMEL SAMG55.
I have a sample made 1 month ago, and the other 10 pcs made this week. I test it in same environment.
The two batch sample running with my bsec_config_selectivity[1974], that is training to recognize 1 pollution air and 1 clean air.
We put 11 pcs (includes 1 engineering sample(marked as Golden) and 10 pre-MP samples(marked as New)) in a space to test.
With comparing to the training definition, the golden response correct result, but the new 10 pcs report total different. This really confuses us.
The record also shows the two groups get very different results. Here's the chart and attched excel raw data for every min.
Please advise. Thank you in advanced.
Hi brianchen,
1. The resistance output of new sensors is usually relatively small;2. We suggest that you use BME AI-Studio v2.3.2 for training and run the generated configuration on BSEC v2.5. Because latest BME AI-Studio v2.3.2 supports the Generate aged sensor data option which allows you to generate additional training data that will ultimately make your algorithm more robust to sensor aging effects. In addition, this feature can also be useful to improve generalizations related to sensor-to-sensor variance.
Thanks Robin, I'll upgrade later. By the way, I found that when a sensor moved from palce A to place B, it seems need time to learn the environment itself? If so, how long will it need to take for consistence?
As this chart, there're 70 pcs working in a lab "called Lab B" for 3 days, and another single (golden, blue curve) one moved into the Lab B for 16 hours.
The value of gas estimation #1 measured by the golden one, shown in blue curve, is different from the group of 70 pcs.
