We are observing an extremely rare behavior: Sometimes it happens that several measured values are completely wrong under normal conditions (temperature > 70°C, pressure > 1100mbar, humidity delta > 10%). We get similar values for all subsequent measurements, so this issue it is not present for one measurement (forced mode) only.
A power cycle solves the problem and the measured values are plausible again. The calibration data is retrieved only once after booting - just like adviced in the API (https://github.com/BoschSensortec/BME280_driver). In our case, the waiting time after the soft reset ("dev->delay_ms(2);") is much longer, approx. 15ms. This works very well in the lab. In production, however, we had the problem described above on several devices (estimated 1 out of 1000).
When we examined the affected parts for analysis, we could not detect any more problems - since a power cycle happened. Thus it can be assumed that the calibration data was possibly sent or received incorrectly. Is anybody aware of such case, or could it happen under certain circumstances that the startup time is significantly longer than 2ms or even 15ms?
Please provide the sensor's raw data ,timestamp and trimming value to us. The information will be sent back to the internal team for analysis.
The problem is, that the issue happens very seldom an we did not capture the request data yet, so I can't provide right now. The question was, if anybody experienced similar problems. Anyway, V3.3.7 of BME280 API writes as release notes "Added a wait until the NVM copy is complete", so I assume that the startup times of 2ms is not always guaranteed?