Vibration Spectrum Analysis: Reading the FFT to Diagnose Faults is the practice of converting a machine's raw vibration waveform into a frequency spectrum so that each mechanical fault can be identified by the frequency at which its energy appears. A rotating machine vibrates at frequencies tied to its running speed and its internal geometry. By decomposing the signal, an analyst can separate unbalance from misalignment, looseness from a failing bearing, and one healthy gearbox from another that is shedding teeth, often weeks before the fault becomes audible or destructive.
A vibration sensor produces a time-domain signal: amplitude plotted against time. On its own it is hard to read. The Fast Fourier Transform (FFT) is an efficient algorithm that resolves the waveform into the individual sine components that add up to make it. The output is a spectrum: amplitude plotted against frequency, where each peak marks a discrete vibration source.
Two settings govern what the spectrum can reveal. Frequency span sets the highest frequency captured, and the number of lines of resolution sets how finely closely spaced peaks can be separated. Because faults are best expressed as multiples of shaft speed, analysts commonly plot the x-axis in orders, where 1x equals running speed in hertz (RPM divided by 60).
The same vibration can be expressed three ways, and the choice determines which faults stand out.
For assessing overall condition against a threshold, see ISO 10816-3 vibration severity, which sets velocity limits by machine class and mounting.
Mass unbalance produces a clean, dominant peak at exactly 1x running speed, radial in direction, with relatively little energy elsewhere. Amplitude rises with the square of speed, and phase is steady. A single high 1x peak with low harmonics is the classic unbalance signature.
Misalignment between coupled shafts typically raises the 2x running-speed peak and, importantly, produces strong axial vibration, because the coupling forces the shafts to flex once and twice per revolution. When axial amplitude at 1x or 2x approaches or exceeds radial amplitude, suspect misalignment or a bent shaft.
Mechanical looseness produces a long series of running-speed harmonics: 1x, 2x, 3x, 4x and beyond. Structural looseness and worn fits can also generate half-order subharmonics such as 0.5x, 1.5x and 2.5x.
Rolling-element bearing defects appear as non-synchronous energy, meaning peaks at frequencies that are not integer multiples of running speed. Each defect has its own tone: the ball pass frequency outer race (BPFO), ball pass frequency inner race (BPFI), ball spin frequency (BSF) and fundamental train frequency (FTF). These start as low-level high-frequency peaks over raised broadband noise; the maths is set out in bearing defect frequencies.
Gears vibrate at gear mesh frequency (GMF), equal to the number of teeth multiplied by that shaft's speed. A healthy mesh shows a modest GMF peak; a developing fault raises GMF and surrounds it with sidebands spaced at the running speed of the defective gear. Sideband growth is a strong indicator of localised tooth damage, as explained in gear mesh frequency.
| Fault | Characteristic frequency | Dominant direction | Spectral clue |
|---|---|---|---|
| Mass unbalance | 1x RPM | Radial | Single high 1x, low harmonics |
| Angular misalignment | 1x and 2x RPM | Axial | High axial 1x or 2x |
| Parallel misalignment | 2x RPM | Radial | 2x often exceeds 1x |
| Bent shaft | 1x and 2x RPM | Axial | Axial dominant, steady phase |
| Mechanical looseness | 1x plus many harmonics; sometimes 0.5x subharmonics | Radial | Long harmonic series |
| Bearing outer race | BPFO (non-integer of RPM) | Radial | Non-synchronous, high-frequency |
| Bearing inner race | BPFI (non-integer of RPM) | Radial | Sidebands at 1x |
| Gear tooth defect | GMF = teeth x shaft RPM | Radial and axial | Sidebands at running speed |
| Oil whirl (journal bearing) | 0.42x to 0.48x RPM | Radial | Sub-synchronous peak |
A single spectrum tells you the current state; a baseline tells you what changed. Record a reference spectrum when the machine is known to be healthy, then trend overall level and individual peaks over time at the same speed, load and measurement point. Consistency is essential: use the same sensor location, orientation and mounting each time, because loose or handheld sensors distort high-frequency content.
This is where route-based collection connects to the wider maintenance workflow. Feeding vibration readings into Fabrico lets teams trend each measurement point against its baseline and trigger a condition-based work order automatically when a peak crosses its alarm limit, so a rising bearing tone becomes a scheduled repair rather than an unplanned breakdown. Book a Fabrico demo to see the route-to-work-order flow.
Velocity is roughly flat in amplitude across the mid-frequency band where most rotating-machine faults live, so a single velocity threshold covers unbalance, misalignment and looseness fairly. That is why severity standards are written in mm/s RMS.
Unbalance sits exactly at 1x running speed and is synchronous. Bearing defects appear at non-synchronous frequencies that are not whole multiples of running speed, along with raised high-frequency broadband energy. If the suspicious peaks do not line up with integer orders, suspect a bearing.
Set the span to capture the highest frequency of interest, typically several multiples of gear mesh or bearing defect frequency, and use enough lines of resolution to separate closely spaced peaks such as GMF sidebands. Too coarse a resolution smears adjacent tones into one peak and hides early faults.
Anmeldung
Blog