I have a couple of AirGradient ONEs. In the last couple of weeks, one of them has started consistently showing rapid spikes in PM2.5. It does this every minute or so, seemingly 24 hours a day, irrelevant of actual air quality in the room. The other AirGradient ONE is stable.
Here is a video of the spike happening: https://www.youtube.com/shorts/nhnpa30d8sI
I thought it was a faulty sensor, but what is interesting is this is a PMS5003_20240826 sensor, which requires a custom calibration formula, and this behavior ONLY happens when a custom calibration formula is applied.
To be more explicit:
- When using raw data, or only the EPA formula, the reading is stable. But, it is too low when compared to a reference PM2.5 sensor I have.
- When I apply the special calibration formula for the PMS5003_20240826 batch, the cycling behavior starts almost immediately.
- It ran for a few months with the calibration formula without this rapid spiking behavior.
- I have tried briefly using different special batch formulas to see if there’s a difference, but the cycling still happens. I’ve also tried the PMS5003_20240826 formula with and without the EPA formula on top, but the cycling still happens.
- My other AirGradient ONE has a sensor that’s not part of a batch that needs a custom calibration formula. However, if I temporarily set it to use one, it shows the same rapid-spiking behavior!
- I need to use the special batch formula because it does correct the PM2.5 reading. In the example video above, my reference PM2.5 sensor shows (a stable) 4 µg/m3. The AirGradient shows this same value, does the spiking, then returns to the correct value.
- So, I’m currently stuck between a stable, but incorrect reading, or a correct-but-rapidly-spiking reading.
- One other data point, I have outdoor Open Air model too. This is has a sensor that is also part of a special batch and requires a custom calibration formula, but does not display the rapid-spiking behavior when it’s applied. Although, of course, there are no LEDs nor a screen on this device, and the data uploaded to the dashboard is smoothed out, so maybe it is happening too, but I can’t see it.
