This morning I installed temporarily a second QPD to monitor the input beam. The goal was to understand where the vibrations at frequencies below 2kHz couple from. As shown in the photo, the second QPD was close to the first one.
The signals in the two QPDs were quite different, and the coherence between them wasn't great. So I concluded that the main coupling path is not through input beam of QPD vibration, but more likely real motion of the disk.
I removed the additional QPD and restored the setup to its nominal configuration. The readout infrastructure is still in the model.