Reading a Vibration Spectrum (FFT)

A vibration spectrum is the FFT of a machine's vibration, the raw time signal broken into the individual frequencies inside it. You read it by seeing where the peaks land relative to the machine's running speed, called 1x. Each fault puts its energy at a characteristic frequency, so the location of a peak, expressed as a multiple of running speed, names the fault. Imbalance sits at 1x, misalignment at 2x, looseness across many harmonics, and bearing defects at their own non-synchronous frequencies.

This matters because the overall vibration level tells you a machine is in trouble but not why, and two different faults can produce the same overall number. The spectrum is where the diagnosis lives. Reading it well is the difference between “this pump is shaking” and “this pump has a misaligned coupling and an outer-race bearing defect,” which is the difference between a guess and a work order. This guide covers how the FFT works, what the classic peak patterns mean, how bearing tones appear, and how to turn a screen full of spikes into a fault.

What is an FFT and why use it?

The FFT, Fast Fourier Transform, is the math that converts a vibration waveform from the time domain into the frequency domain. A raw vibration signal is amplitude versus time: a wiggly line that mixes every frequency the machine produces into one trace you cannot easily interpret. The FFT decomposes that trace into the separate frequencies that make it up and shows how much vibration sits at each, giving you amplitude versus frequency, the spectrum. What was a tangle becomes a row of distinct peaks.

You use it because faults are frequency-specific but the time waveform hides that. In the spectrum, imbalance, misalignment, looseness, and each bearing defect land at their own predictable frequencies, so they separate into individual peaks you can read one at a time. The time waveform still has its uses, it is better for seeing impacts and transient events, but for identifying steady rotating-machinery faults, the spectrum is the primary tool. It builds directly on the overall-level screening in vibration analysis basics: the overall level finds the sick machine, the spectrum diagnoses it.

What do 1x, 2x, and harmonics mean?

These are the machine's running speed and its multiples, and they are the reference grid you read every spectrum against. 1x is one vibration cycle per shaft revolution, running speed itself. 2x is twice per revolution, 3x three times, and so on; these integer multiples are the harmonics. The first thing an analyst does with a spectrum is find running speed and mark 1x, because a peak's meaning depends entirely on where it sits relative to that. A peak at 1,780 cpm on a 1,780 rpm motor is 1x; the same 1,780 cpm peak on a 3,560 rpm motor is 0.5x and means something very different.

The classic patterns follow from this. A dominant 1x peak points to imbalance, the once-per-revolution force of a heavy spot. A strong 2x, especially with raised axial vibration, points to misalignment, because a misaligned coupling loads the machine twice per turn. A long picket fence of harmonics, 1x, 2x, 3x, 4x and beyond, signals mechanical looseness, and structural looseness can add half-order subharmonics like 0.5x and 1.5x. These are tendencies, confirmed with phase and axial readings, not certainties, but they are the backbone of spectrum reading.

How do bearing defects show up in the spectrum?

Bearing defects show up as non-synchronous peaks, energy at frequencies that are not integer multiples of running speed. This is the key that separates a bearing fault from imbalance or looseness, which sit on the running-speed grid. A rolling-element bearing has four defect frequencies set by its geometry and speed: the ball-pass frequency outer race (BPFO), ball-pass frequency inner race (BPFI), ball-spin frequency (BSF), and fundamental train frequency (FTF), the cage rate. Each corresponds to rolling elements striking a defect on a particular surface, and each falls at a calculable, non-integer multiple of running speed.

When a raceway spalls, impacts at its defect frequency appear in the spectrum, often with harmonics of that frequency and with sidebands spaced at running speed or cage speed from load-zone modulation. Because these early impacts are small and high-frequency, analysts often use envelope detection (demodulation) to pull the repetitive bearing impacts out of the background before transforming to the spectrum. Calculating the four frequencies for a given bearing, and matching spectrum peaks to them, is its own discipline covered in bearing defect frequencies. The presence of non-synchronous peaks is the flag; the exact frequency names the defective surface.

What do sidebands tell you?

Sidebands are pairs of smaller peaks spaced evenly on either side of a larger peak, and they tell you the main signal is being modulated, its amplitude rising and falling at the spacing frequency. The spacing carries the diagnosis. On a gearbox, a gear-mesh peak flanked by sidebands spaced at a shaft's running speed points to a problem on that shaft's gear, an eccentric or cracked gear modulating the mesh. On a bearing, sidebands spaced at running speed or cage speed around a defect frequency confirm the fault and hint at the inner-race or load-zone interaction.

The rule of thumb is that a clean single peak is often benign, while a peak growing a family of sidebands is a fault developing structure, and the sideband spacing tells you which component. Reading them requires enough frequency resolution to separate the sidebands from the main peak, which is a matter of FFT settings. Sidebands are one of the strongest confirmations available in a spectrum, turning “there is energy at the gear-mesh frequency” into “the pinion on the input shaft is at fault.”

What FFT settings matter when collecting a spectrum?

Three settings decide whether the spectrum can show the fault: the frequency span, the resolution, and the parameter. Fmax, the maximum frequency, has to be high enough to include the fault frequencies you care about, set it too low and bearing tones or gear mesh fall off the right edge and are simply not there to read. A common practice is to set Fmax well above the highest expected defect frequency, and to collect a second, lower-Fmax spectrum when you need fine detail around running speed.

Resolution, set by the number of FFT lines, decides whether you can separate close peaks, critically, whether sidebands split from their main peak or blur into one lump. More lines give finer resolution at the cost of a longer sample. The parameter matters too: acceleration for high-frequency bearing and gear content, velocity for the mid-band overall picture, chosen to match the fault, as covered in vibration sensor types. And the mount sets a ceiling on all of it, a magnet-mounted sensor cannot deliver a trustworthy high-frequency spectrum no matter how you set the analyzer, which is why accelerometer mounting methods is inseparable from spectrum quality.

How do you read a spectrum step by step?

Reading a spectrum is a disciplined sequence, not a glance. Following the same order every time keeps you from fixating on the tallest peak and missing the real story.

1. Find and mark running speed. Establish 1x from the shaft rate before anything else. Every peak's meaning is relative to it, so label 1x first.
2. Express peaks as orders of 1x. Convert each significant peak to a multiple of running speed. Is it exactly 1x, 2x, 3x, a half-order, or a non-integer? That classification is most of the diagnosis.
3. Check the synchronous peaks. Look at 1x for imbalance, 2x with axial energy for misalignment, and a run of harmonics for looseness. Note which dominates.
4. Hunt for non-synchronous energy. Scan for peaks that are not integer multiples, the signature of bearing defects. Compare them against the bearing's calculated BPFO, BPFI, BSF, and FTF.
5. Look for sidebands. Around gear-mesh and bearing peaks, measure the sideband spacing and translate it to the modulating component.
6. Confirm with phase, axial, and waveform. Use phase readings to tell imbalance from misalignment, axial amplitude to confirm misalignment, and the time waveform to spot impacts the spectrum flattens out.
7. Judge severity and trend. Compare velocity amplitude against ISO zones and against the machine's own history, then tie the diagnosis to a work order so the reading becomes a repair.

Where spectrum analysis fits your reliability program

Spectrum analysis is the diagnostic engine of a vibration program. Overall-level screening and simple sensors find the machines in trouble; the FFT spectrum is where a trained analyst turns “something is wrong” into a named, located fault with a recommended action. It is a skill that rewards practice, the patterns are learnable, and once you read spectra fluently you can plan a repair weeks ahead of a failure, with the specific parts and labor known in advance.

The bottleneck in most plants is not the analysis but the flow around it: getting good data collected consistently, getting spectra in front of someone who can read them, and connecting the diagnosis to the maintenance and production systems that act on it. A brilliant fault diagnosis that never reaches a work order saves nothing. Pulling vibration data together with the plant's maintenance history, run hours, and schedule into one operational view is where machine monitoring platforms like Harmony fit, so a diagnosed fault lands next to the context and the workflow that turn it into a planned repair. It layers onto the systems you already run, with no rip-and-replace. For the strategy this feeds and the standards behind severity, see predictive maintenance condition-based maintenance and the ISO 10816 vibration standards or read the CLS case study.