We have the following accelerometer in our setup
IIS3DHHC - High-resolution, high-stability 3-axis digital inclinometer for industrial applications - STMicroelectronics
Experimental setup:
The board in which the accelerometer is integrated is screwed onto a flat, wooden board to ensure that the accelerometer doesn't change orientation on average over time. A controlled (programmed) frequency sweep between 0Hz and ~4000Hz is then performed with the use of an eccentric motor attached to the same wooden board. While it is hard to get an accurate estimate of the vibration levels due to undersampling in our logging tools we believe that levels rarely exceed 0.1 - 0.2 g in amplitude.
Applied filtering:
The accelerometer is sampled at a rate of 1kHz. The default filtering options are used: FIR filter with bw of 440Hz.
Then a 4th order low pass filter is applied with a cutoff frequency of 15Hz, before downsampling to 50Hz.
Issue:
What we see is that two of the axis pretty consistently shift their dc-values to something corresponding to an offset angle of 20 degrees. As soon as the vibrations stop, the bias disappears. The (filtered) accelerometer readings are shown below. There's a high level of reproducibility in the offset. We always end up at a corresponding angle of around 20deg. The offset is consistently introduced in the frequency range 2800rpm to 4000rpm.
Is this a common phenomenon with MEMS-accelerometers?
What is happening?
How can the issue be mitigated?
Could there be an aliasing issue?
It should be noted that we really aren't interested in frequencies above ~5Hz but we are very much interested in accurate 0Hz readings.
