Time waveform analysis is an ideal tool when diagnosing a range of fault conditions, including rolling element bearing faults, faults associated with gears, cavitation, rubs, looseness and more - any time the vibration source may include impacts, modulation, beats, rubs, transients, and random bursts of energy, time waveform analysis is the best data to view.
Time waveform analysis is useful because the waveform is a recording of exactly what happened in the machine from one moment to the next. With every impact between a ball and a crack, the waveform will have a 'spike'. With every tooth that comes into mesh inside the gearbox, the time waveform will have a ‘pulse' (or a spike when a tooth is damaged). Every time there is a rub, or a burst of energy from cavitation, or an impact during looseness, the time waveform will capture the event.
The spectrum (FFT) is designed for simple periodic vibration. When the faults listed above occur, the spectrum may have harmonics, sidebands, and/or raised noise floor, which you must 'reverse engineer' to determine what happened inside the machine. The time waveform, on the other hand, will reveal exactly what happened.
The challenge with time waveform analysis is to make sure you have a good time waveform to analyze. The ideal setting, in my opinion, is to have two time waveforms: one in acceleration showing just 6-10 revolutions of the shaft, and a second waveform in velocity showing many more rotations; perhaps 5 seconds in length.
The shorter waveform will reveal gear damage, bearing damage, looseness, rubs and other conditions that have a high frequency and occur frequently.
The longer time waveform will reveal beats, cavitation, and less frequent impacts in the gearbox. And if it is in velocity, you can see patterns unique to unbalance, misalignment, etc.
Of course, if your software can convert the waveform between acceleration and velocity, then you can record both waveforms in acceleration - but many systems cannot perform that conversion in the software.
In both cases you should have at minimum 4096 samples, which corresponds to 1600 lines of resolution in the spectrum.
Many analyzers/software programs require you to specify the Fmax and Lines of resolution (LOR) when setting up the time waveform measurement. Use the following equation to set the Fmax, and select LOR = 1600 lines of resolution, or 3200 lines if you have plenty of memory.
Fmax = RPM x LOR / # revs
1. # revs = number of revolutions of the shaft
2. RPM and Fmax must be in the same units (Hz or CPM) - Divide CPM by 60 to get Hz.
If you can only collect one time waveform, select one with 6-10 revolutions. If LOR = 3200, and # revs = 10, then Fmax = 320 x RPM
I hope this helps you to diagnose faults with greater accuracy.
Tip provided by Jason Tranter of the Mobius Institute.
To view a short presentation on Time Waveform Analysis, go to http://ilearninteractive.com/eLearn/index.html#