Now that the main source of non-stationary noise has been removed (71927), we already have another candidate.
A DARM spectrogram shows bursts of noise below 60 Hz, happening from time to time. A better view is provided by a spectrogram of DARM whitened with the median.
There are two interesting features:
Robert had me investigating something similar to this, specifically around the input power changes (4/7 and 6/21). I found that the noise seemed to be strongest after the power increase on 4/7. But it's been hard to tell how it changed after the decrease on 6/21. (I have tons of other reference images but I've attached one from each time period - before the power increase, during, and after.)