Thanks for the update and graphs. That is an amazing improvement. In the “after” plot, it looks like any acceleration from the train is well below the noise level of your accelerometer. So, within the limits of your measuring equipment, you’ve effectively eliminated all train vibration. If I were in your place, I would declare success and move on with life! Don’t even bother with foam and rubber feet, because this configuration is working great.

But you could analyze further if you really want; there could be some train signal hiding in all that noise. Since there’s periodic noise in the Z axis, you could take a reading during a still time (computer off, no trains) and see where your spikes are in the frequency domain. Then you could apply a filter (or filters) to cut out that periodic noise.

But unless you’re really into learning about signal analysis, I’d say you could skip it.