In a spectrum, if the entire noise floor is raised, it is possible that we have a situation of extreme bearing wear.
If the noise is biased towards the higher frequencies in the spectrum then we may have process or flow problem like possible cavitation, which may be further confirmed by high acceleration measurement (or filtered acceleration measurement) on the pump body on the delivery side (since high frequency waves are always localized).
Smaller “humps” may be due to resonance (possibly excited by anti-friction bearing damage, cavitation, looseness, rubs or impacts) or closely spaced sidebands arising from other defects. A high resolution measurement (or graphical zoom and a log scale) may reveal whether the source is problems that exhibit sidebands or a problem of resonance. If machine speed can be changed, (for e.g.motor connected to VFD drives) the resonant frequency would not move – but the other peaks would. Sidebands will typically be symmetrical around a dominant peak – e.g. 1X, 2X, 2x LF (100 or 120 Hz) etc indicating different faults.
Interestingly, the time waveform would reveal the reason as to why the noise floor has been raised.
We would see signs of looseness, severe bearing wear, rubs, and other sources of impacts in the time waveform. We must make sure that there are 5 – 10 seconds of time waveform if we suspect an intermittent rub (e.g. white metal bearings of vertical pumps or loose electrical connection of motor terminals) or if we suspect flow turbulence or cavitation.
If the time waveform looks normal (making sure there is a high Fmax (following Niquist criteria) and we view the waveform in units of acceleration then increase the resolution in the spectrum to 3200 lines or higher in case we are seeing a family of sidebands (like the sidebands we find around gear mesh frequency or rotor bars).
But if a natural frequency is being excited (necessary condition for resonance) then we have to perform a bump/impact test or a run-up/coast down test to confirm the situation.