# How to read a vibration spectrum: 1x, 2x and harmonics

> The vibration meter reads 8 mm/s. One number does not tell you what to do next: balance the rotor, align the shafts, or order a bearing. A spectrum breaks the vibration down by frequency and answers the question "where from." Below we walk through the patterns you will see on screen, what each one means, and how to check your guess before you pick up a tool.

**In short:** Convert the frequency of every peak into a multiple of running speed. If 1x dominates and overall vibration sits close to 1x, that is unbalance, and balancing makes sense. A clear 2x plus a rise in axial vibration usually points to shaft misalignment, which needs shaft alignment, not balancing. A comb of 1x, 2x, 3x with half-order harmonics and a raised noise floor means mechanical looseness, cracks or rubbing. Peaks higher up the frequency range that are not multiples of running speed are usually bearings. The spectrum gives you a hypothesis; the diagnosis is confirmed with the time waveform, phase, repeatability and history.

Source: https://axiline.pt/en/articles/how-read-vibration-spectrum/  
Publisher: AXILINE · Vila Nova de Gaia, Portugal · +351 931 831 229 · axilinegeral@gmail.com

## The spectrum shows what your vibration is made of

A familiar situation: a fan starts humming after the impeller gets cleaned. You hold a sensor against it, see 9 mm/s, and stop there. What you need next is the spectrum.

The instrument records the signal from an accelerometer on the bearing housing, and FFT — Fast Fourier Transform, the label you will usually see on the instrument — converts that recording from time into frequency. On screen you see a set of vertical lines. A line's position along the horizontal axis is its frequency; its height is amplitude. One sine wave turns into one line.

From here on, think in multiples of running speed, not in hertz. The rotation frequency is 1x. A fan at 1450 RPM gives 1x = 24.2 Hz, 2x = 48.3 Hz, 3x = 72.5 Hz. Convert hertz into multiples straight away, or the spectrum stays just a set of numbers for you.

While you are at it, compare the two numbers most instruments show side by side: overall vibration and the 1x running-speed component. Balancing only brings down 1x. An overall reading of 9 mm/s with 1x = 8 mm/s tells you the rotor is driving the vibration. An overall reading of 9 with 1x = 2 tells you balancing will not help — the cause is something else.

> A simple sorting rule: divide the peak's frequency by the running-speed frequency. Land on 1.0 / 2.0 / 3.0, and you are looking at the rotor, the coupling, the mounting. Land on 3.7 or 6.4, and you should be looking at bearings, blades, gear teeth, a belt.

## Four patterns you should recognise at a glance

### 1x dominates

The peak at running speed stands several times higher than every other line, the spectrum looks almost empty, overall vibration sits close to 1x, and the 1x phase (the angle showing where in the turn the vibration reaches its peak) stays stable from one reading to the next. This is unbalance. Balancing is the right call and usually gets there in a single visit.

### A clear 2x

2x is comparable to 1x or exceeds it, axial vibration has risen, and the phase on either side of the coupling differs by around 180°. This is what shaft misalignment or a worn coupling looks like. Balancing will not remove it — the machine needs shaft alignment.

### A comb of harmonics

1x, 2x, 3x, 4x and on up, often with half-order harmonics at 0.5x and 1.5x, and the noise floor — the flat background between the peaks — sits higher than usual. This means mechanical looseness, a cracked foot, a poured foundation gone soft, or rubbing. A spectrum like this gets confirmed with a wrench and a walk-round inspection, not with software.

### Peaks high up, not tied to running speed

Lines at 3.7x, 5.2x, 8.4x, blurred at the edges, growing month after month. This is most often rolling-element bearings. Balancing will not touch them — plan for a replacement.

## Table: what you see, what it probably means, how to check

| What you see in the spectrum | Likely cause | How to check |
| --- | --- | --- |
| 1x dominates, overall ≈ 1x, 1x phase stable | Rotor unbalance | A trial weight of known mass at a known radius. A valid trial run changes 1x amplitude by 20–30% or phase by 20–30° |
| 1x and 2x are comparable, axial vibration has risen | Shaft misalignment, a worn or over-tightened coupling | Compare the phase on both sides of the coupling. Then remove the coupling half and check shaft alignment with dial indicators or a laser system |
| A comb of 1x, 2x, 3x… plus 0.5x and 1.5x, raised noise floor | Loose mounting, a crack, the rotor rubbing against the housing | Check the torque on the foundation bolts, look for a soft foot, play in fits and clearances. Look at the time waveform: looseness shows up as impacts and an unstable phase |
| A peak at twice the mains frequency (100 Hz on a 50 Hz supply) | Air-gap eccentricity, a loose stator | A power-off test. An electromagnetic line disappears instantly; a mechanical one decays gradually as the machine coasts down |
| Symmetric sidebands around 1x, spaced a few hertz apart | Broken or cracked rotor bars in an induction motor | This needs fine frequency resolution plus stator current analysis. It is a job for the electrical team, not an on-site balancing visit |
| Many lines higher up the frequency range, not multiples of 1x, blurred, with a growing noise floor | Rolling-element bearing defects | An acceleration envelope spectrum in the 2–10 kHz band (processing that isolates impact pulses), bearing-housing temperature, trend history, a look at the lubrication |
| A subsynchronous line at 0.4–0.5x (below running speed) on a machine with sleeve bearings | Oil-film instability (oil whirl) | Not a job for case-mounted sensors. You need non-contact shaft probes (proximity probes) and orbit analysis — tracking the path of the shaft centre |
| Peaks at blade-pass frequency (number of blades × 1x) plus broadband noise | Hydraulics: cavitation, flow separation, blockage | Inlet pressure, the actual operating point, the condition of the impeller |
| Amplitude climbs sharply over a narrow speed range, 1x phase shifts by around 180° | Resonance in the structure or supports | Controlled start-up and coast-down with amplitude and phase recorded, plus a bump test. You cannot balance out a resonance — the result will not hold |

> The table gives you a hypothesis, not a conclusion. The same spectrum pattern can come from two different causes, and the reverse is also true: looseness can raise 1x, and shaft misalignment can add harmonics.

## Frequency resolution: Δf = Fmax / N

A spectrum is made of discrete lines. The spacing between neighbouring lines works out as Δf = Fmax / N, where Fmax is the top of the spectrum's frequency range and N is the number of lines. The same formula has a second half people tend to forget: Δf = 1 / T, where T is the length of the recording.

There is one practical takeaway. To separate two close peaks, you need a longer recording. Want Δf = 1 Hz? Record for one second. Want Δf = 0.1 Hz? Record for ten seconds. No amount of averaging, smoothing or nice-looking rendering gets around this.

Let's work through a real example. An induction motor at 1480 RPM, 1x = 24.67 Hz. You are looking for a broken rotor bar, which produces sidebands spaced at 2·s·P, where s is the slip frequency in hertz and P is the number of pole pairs. At 1.3% slip and two pole pairs, the spacing is 1.3 Hz. So you need to separate the lines at 24.67 and 23.37 Hz, which calls for a Δf of around 0.3 Hz and a recording no shorter than 3.3 s. Setting N = 6400 at Fmax = 2000 Hz gives Δf = 0.3125 Hz and T = 3.2 s. You will see the sidebands.

The opposite mistake happens even more often. You set Fmax = 10 kHz "just to see everything," leave the number of lines the same, and the resolution gets ten times coarser. Close lines merge into one fat peak, and unwanted high-frequency noise creeps into the reading. Set Fmax to match the highest-frequency defect you are actually looking for.

- Δf = Fmax / N: more lines at the same Fmax means finer resolution.
- T = 1 / Δf: fine resolution always costs recording time.
- On a slow-turning machine, the recording time grows. A roll at 60 RPM needs seconds, not milliseconds.
- The running speed has to hold steady over the recording. A speed drift of more than a quarter of Δf smears the peak across several neighbouring spectral lines, the amplitude drops, and you conclude the defect has gone away.

> On a drive with a VFD and a floating load, you will not get fine resolution: the speed drifts over the recording. There, the approach is order analysis — a spectrum built in multiples of running speed, referenced to a once-per-turn mark.

## Sidebands: look at the spacing, not the height

A sideband is a line sitting next to a large peak, mirrored on its left and right. It signals modulation: the amplitude or frequency of the carrier vibration is changing periodically. The spacing between sidebands equals the modulating frequency, and it is that spacing that points to the culprit.

Work out the spacing and match it against the machine's kinematics: the running speed of each shaft, the number of gear teeth, the number of blades, the motor's slip. Until you find a match, it is too early to call it modulation.

- Spacing equals 1x of a shaft: the source is rubbing or getting loaded once per turn of that shaft. An eccentric gear, an uneven load on the bearing housing.
- Sidebands around the gear-mesh frequency, spaced at the running speed of one of the gears: that particular gear is the defective one.
- Sidebands around motor 1x, spaced a few hertz apart: rotor bars.
- Sidebands around a bearing frequency, spaced at 1x: an inner-race defect or an uneven load around the circumference.
- Two close lines with no symmetric sidebands around them: that is not modulation, it is beating between two independent sources. For example, two fans on a shared duct running at 24.5 and 24.9 Hz. Nothing is wrong with the machine — what you hear is a "beat."

> A common mistake: mistaking two lines that happen to sit close together for sidebands, or mistaking coarse resolution for their absence. With a coarse Δf, sidebands simply merge into the carrier.

## A spectrum alone does not make a diagnosis

A spectrum averages. It honestly shows which frequencies are present on average, and it hides one-off events. An early-stage bearing fault produces an occasional impact once every few shaft turns. In the spectrum, that impact smears into the noise; in the time waveform, you see the individual pulses.

- [x] Time waveform. Look for impacts, clipped peaks, asymmetry, modulation. Record for at least 4–6 shaft turns: at 1500 RPM that is 0.2 s, in practice people use 1–2 s, and a roll at 60 RPM needs about 6 s.
- [x] 1x phase. A difference of around 90° between the horizontal and vertical directions points to unbalance. A difference of around 0° or 180° is a sign of a different mechanism.
- [x] Repeatability. Take the reading twice, reposition the sensor, check the magnet and the cable. A jumpy high-frequency peak often turns out to be a mounting resonance from the paint layer, rather than a fault in the machine.
- [x] History. One spectrum shows a state; two spectra taken apart in time show a trend. A steady rise from the baseline matters more than the absolute number.
- [x] Kinematics and inspection. The number of blades, gear teeth, gear ratios, belts. Without this, you cannot tell a blade-pass frequency from a harmonic of running speed.
- [x] The same settings for comparison. The same point, the same direction, the same Fmax, the same number of lines, the same mounting. Do not compare spectra taken with different Fmax values directly.

## What to do about a humming fan: the sequence to follow

1. **Measure overall and 1x at every bearing** — Sensor on the bearing housing, horizontal-radial, the same direction every time you measure. Compare overall vibration with 1x. If overall is several times higher, something other than unbalance is driving most of the vibration, and balancing will not remove it.
2. **Take the spectrum and convert the peaks into multiples** — Divide the frequency of every notable line by the running-speed frequency. Note down what lands on 1x, 2x, 3x, and what is not a multiple of running speed. It takes a minute and narrows down the possible causes straight away.
3. **Check the mechanics by hand** — Torque on the foundation bolts, soft foot, play, fits, the condition of the frame. A comb of harmonics in the spectrum is almost always confirmed with a wrench, not with software. You can balance a machine that has looseness, and 1x might even drop for a while, but the fault stays and comes back.
4. **Rule out resonance** — Run a controlled start-up or coast-down while recording 1x amplitude and phase. A sharp, narrow amplitude peak and a phase shift of around 180° mean the operating speed sits close to a natural frequency. Here the influence coefficient — the machine's response to the trial weight, which the whole calculation is built on — will be unstable, and the balancing result will not hold.
5. **Balance it, if 1x dominates** — A trial weight of known mass and radius, a change in amplitude of 20–30% or in phase of 20–30°, the correction calculated, the weights fitted, a check run. A typical order of magnitude on mechanically sound machinery: a fan at ~12 mm/s drops to ~1.5–2 mm/s. Treat that as a guide, not a promise for your machine.
6. **Judge the result separately from the software** — A "within tolerance" readout means exactly one thing: the residual 1x is below the target you entered yourself. Judge the overall condition of the machine by overall vibration, mm/s RMS (root mean square) in the 10–1000 Hz band, against zones A/B/C/D. Residual unbalance against a G balance quality grade is a third, separate figure. Check which part and edition of the standard applies to your specific machine.

## Limits of the method, and where we can help

The vibration-velocity spectrum is strong in the roughly 10–1000 Hz band. That is where unbalance, shaft misalignment, looseness and rubbing live. Early bearing defects sit higher, in the 2–10 kHz zone, and only show up in acceleration and the envelope spectrum. Let's be direct: the Balanset-1A measures RMS (root mean square) vibration velocity in the 5–200 Hz band, and displays FFT, the time waveform and the 1x vector on two channels at once. That is enough to decide "balance or not," compare two bearings, find a resonance from a coast-down, and check the result. Hunting for pitting — small flaking of metal — on a bearing raceway needs an analyser with a high-frequency channel.

There are also things case-mounted sensors simply cannot give you. Machines on sleeve bearings need proximity probes and orbit analysis. A modal picture — the structure's natural frequencies and mode shapes — needs FRF and coherence measurements. If the spectrum points that way, the honest answer is to hand the job to specialists in that field, not to balance.

AXILINE's engineers design and manufacture the Balanset instruments and use them to balance on site themselves. We come to your site with our own instrument, fit sensors to your bearings, take the spectrum and the 1x vector, and tell you plainly whether it is unbalance or not. If it is unbalance, we balance the rotor in place, in its own bearings, without removing it. If it is looseness, shaft misalignment or a bearing, you will hear that before you pay for a balancing job that would not have helped.

You can also buy the instrument for your own use. The kit includes two accelerometers, a laser phase-and-speed sensor, a two-channel USB module with preamplifiers, integrators and an ADC, plus Windows software: one- and two-plane balancing, FFT, a polar diagram, weight-splitting across fixed positions, drilling calculation, stored influence coefficients, trim balancing, tolerance calculation against G grades, and an archive for reports. Consultation support is included.

Sources: [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) · [ISO 21940-11:2016](https://www.iso.org/standard/54074.html) · [ISO 13373-3:2015](https://www.iso.org/standard/40840.html) · [ISO 13373-5:2020](https://www.iso.org/standard/62202.html) · [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/) · [Balanset-1A manufacturer specification](https://vibromera.eu/product/balanset-1/)

## Frequently asked questions

**The spectrum shows a high 1x. Can I go ahead and balance right away?**

Almost always, yes, but run two quick checks first. First, compare overall vibration with 1x. If overall is several times higher, something other than unbalance is the main contributor, and balancing will only bring down a small part of it. Second, check that the 1x phase is stable. If the phase jumps by tens of degrees from one reading to the next, you are dealing with looseness or resonance, and the influence coefficient will come out unreliable.

**How large does 2x need to be before I should suspect shaft misalignment?**

There is no hard threshold; what matters is the ratio. A working rule of thumb: 2x above roughly half of 1x already deserves attention. 2x on its own is not a diagnosis. It gets confirmed by a rise in axial vibration and a phase difference of around 180° across the coupling. After that, you check shaft alignment mechanically, with dial indicators or a laser.

**Why have two peaks merged into one, and how do I fix that?**

Your frequency resolution is coarser than the spacing between the peaks. Work with the formula Δf = Fmax / N: bring Fmax down to what you actually need, increase the number of lines N, and record for longer, because T = 1 / Δf. For Δf = 0.2 Hz you need 5 seconds of recording at a steady running speed. If the speed drifts over that time, you will not get fine resolution, and you will need order analysis instead.

**Can you make a diagnosis from a single spectrum?**

No. A spectrum gives you a hypothesis. Confirmation comes from the time waveform, phase, repeatability of the reading, the machine's kinematics, trend history and a plain visual inspection. The same set of lines can come from different causes: harmonics from looseness are easy to mistake for a bearing, and resonance amplifies any line that happens to sit near a natural frequency.

**Will balancing bring down every peak in the spectrum?**

Only 1x, and only the part of it caused by rotor unbalance. The 2x peak, the comb of harmonics, bearing lines, the noise floor and the blade-pass frequency stay where they are. That is why overall vibration sometimes drops only a little even after a perfectly executed balancing job. That is a normal outcome, not a calculation error.

**Does a bearing defect show up in the vibration-velocity spectrum?**

At a late stage, yes: a group of lines higher up the frequency range, not multiples of running speed, blurred peaks and a raised noise floor. At an early stage, no. The first impacts from micro-spalling sit in the 2–10 kHz zone, where vibration velocity understates them. For early detection you take acceleration and an envelope spectrum after band-pass filtering, and you also watch temperature and trend.