# How to find the cause of equipment vibration: a methodology, not a list of causes

> A fan starts humming after the impeller gets cleaned. You hold a sensor against it, see 7 mm/s, and stop there: you have a number, not a cause. A list of possible causes does not help much here, because almost every one of them can raise the same number. What helps is a working order that rules out theories one at a time.

**In short:** The cause is found not from a single spectrum (the breakdown of vibration by frequency) but from the whole set of signs together: context (what changed and when it started), how reliable the reading is, the ratio of vibration across all bearing supports in three directions, how the frequencies relate to running speed, the shape of the time waveform, repeatability, and phase — the angle showing at what point in the turn the vibration reaches its peak. You gather this data first, then formulate a single hypothesis, and test it with one controlled action. The diagnosis is not made by the instrument — it comes from the combination of the signs, the machine's behaviour, and a confirming check.

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

## Step 1. Gather context before you even open the instrument case

Half the diagnosis lives in the maintenance log and the operator's memory. The instrument answers the question "how much." Context answers the question "why."

Vibration that appeared the same day a bearing was replaced, and vibration that built up over three months, call for different theories. The first points to the installation. The second points to wear, build-up, or a settling foundation.

- [x] When it started. The exact date or shift. Suddenly or gradually.
- [x] What was done beforehand. A bearing replacement, cleaning the impeller, a foundation repair, relocating the machine, a new belt, welding on the frame.
- [x] How it depends on load. Rises with flow, falls, or does not react at all.
- [x] How it depends on speed. If there is a VFD, sweep the range and note where it gets worse.
- [x] How it depends on temperature. Measure on a cold machine and again after 40–60 minutes of running.
- [x] What you hear and what you see. A hum, a whine, periodic knocking, shaking pipework, contact marks on the housing.
- [x] What has already been tried. Balancing, alignment, torquing bolts. With what result, and in what order.
- [x] Whether a measurement history exists. Even two numbers a month apart are worth more than one taken today.

> After cleaning the impeller, a machine can hum for two reasons at once: the build-up was removed unevenly and created unbalance, and the housing mounting got shaken loose in the process. Context narrows the field of theories, but it does not replace the measurement.

## Step 2. Make sure the reading reflects the machine, not your mounting

Before you explain a number, check whether you can trust it. False readings account for nearly half of all "mysterious" cases, and people then spend weeks chasing these artefacts as if they were a real defect.

How the sensor is mounted sets the frequency band you can actually see anything in. A hand-held probe cuts off everything above roughly 500–1000 Hz. A magnet on a clean, flat pad covers a wider band. A stud gives you the widest band and the best repeatability. A magnet sitting on a layer of paint or on rust turns into a spring with a resonance of its own.

- Take the reading twice without touching anything, then reposition the sensor and repeat. A spread of more than 10–15% means you are measuring the mounting, not the machine. A peak that disappears after repositioning was yours.
- Check the laser phase sensor. A speed jumping around between 1450–1490 RPM usually means a dirty reflective marker, not the drive. With a clean marker, the 1x phase (the vibration component at the running speed) holds within ±2°.
- Check the time waveform (a recording of the oscillation over time, without breaking it down by frequency) for clipping. Flat tops and bottoms mean the sensor or the input has saturated. In the spectrum, this produces a false comb of harmonics, and you end up diagnosing looseness where none exists.
- Compare like with like. RMS (root mean square) with RMS, peak with peak. A setpoint (a trigger threshold) in mm/s RMS and a peak reading differ by a large factor.
- Keep the settings unchanged: the same point, the same direction, the same Fmax (the top of the spectrum's frequency range), the same number of lines. Do not compare spectra taken with different Fmax values directly.

> On the Balanset-1A, RMS vibration-velocity measurement runs in the 5–200 Hz band, while the ISO 20816 zones are written for the 10–1000 Hz band. Keep this in mind: this kind of reading will not show the high-frequency bearing content — that gets found separately, through acceleration and the envelope spectrum, a method that isolates weak bearing impacts.

## Step 3. Take a baseline map: every support, three directions

One point is not a diagnosis. Go over every bearing support in three directions: horizontal (H), vertical (V) and axial (A). Label them the standard way: 1H, 1V, 1A, 2H and so on, numbering the supports from the drive end. A two-channel instrument takes two points at once and saves you half the runs.

The diagnostic information is not in the absolute figure but in the ratio between directions. Here is how to read this map.

### H is roughly equal to V, axial is small

The classic picture of unbalance. The rotor drags the support around in a circle, so the radial directions come out comparable. Axial stays small, except on overhung rotors and wheels with face run-out.

### H is 3–5 times larger than V

A pronounced directionality. This is how a resonance of the support or frame in one direction, or loose mounting, tends to behave. Unbalance on its own does not usually produce a skew like this.

### V is larger than H

A rare and very telling picture. Look at the fixing to the foundation: a soft foot (a support that does not sit flush against the frame and twists the casing when the bolt is tightened), a cracked foot, soft grout, loose bolts. On a properly mounted machine, vertical stiffness is higher than horizontal, and if V has pulled ahead, that stiffness is gone.

### Axial is comparable to radial

Look for shaft misalignment, especially the angular kind, as well as a bent shaft, a tilted bearing, or a problem with the thrust assembly. Balancing will not help here: it only brings down radial 1x.

### One support is loud, the neighbouring one is quiet

The cause is local: that specific bearing assembly, its mounting, or a resonance belonging to that one support. Rotor unbalance is almost always visible at both supports.

> Take the whole map under one operating condition and log it in a table right away, along with the speed, the load and the support temperatures. Two hours later, you will not remember what flow rate you took point 2A at.

## Step 4. Convert hertz into multiples of running speed

The spectrum shows hertz. Diagnostics lives in multiples. Divide the peak's frequency by the running-speed frequency and you get an order: 1x, 2x, 0.5x, 3.7x. The order itself names the class of cause.

It is a one-line calculation. 1480 RPM is 1480/60 = 24.67 Hz. A peak at 49.3 Hz gives exactly 2.00x. A peak at 100 Hz at the same speed gives 4.05x, and that is not a harmonic at all — it is twice the mains frequency. The difference matters: one gets fixed by mechanics, the other by electricians.

- Whole orders (1x, 2x, 3x) mean phenomena tied to the shaft's rotation: unbalance, shaft misalignment, looseness, a bent shaft, pulley run-out.
- Half-orders (0.5x, 1.5x, 2.5x) mean mechanical looseness, rubbing, or oil-film instability.
- Fractional, non-integer orders mean something that is not rotating in sync with this shaft: rolling-element bearings, belts, a second shaft through a transmission.
- Blade-pass frequency equals the number of blades multiplied by 1x. On its own it is normal; what is diagnostic is a rise in its amplitude.
- A frequency that does not change when the speed changes is not an order at all. It is a natural frequency of the structure, or an external source.

> On a drive with a VFD and a floating speed, an ordinary spectrum smears the lines. There, the approach is order analysis: the axis is referenced to a once-per-turn mark, so the running-speed components stay put regardless of speed. How the spectrum itself works is covered in a separate article on reading a vibration spectrum.

## Step 5. Open the time waveform, check repeatability and phase

A spectrum averages, and so it hides one-off events. An early-stage bearing fault produces an impact once every few shaft turns. In the spectrum, that impact smears into the noise; in the time waveform, you see the individual pulses. Look at both, always.

Record for at least 4–6 shaft turns. At 1500 RPM that is 0.2 s, in practice people record 1–2 s. On a slow-turning roll at 60 RPM you need about 6 s, or a rare event simply will not land in the sample.

- [x] Periodic impacts. Work out the period and convert it into a multiple. Once per turn means rubbing or a defect on that shaft.
- [x] Modulation. The amplitude "breathes" at a steady rhythm. The modulation rate points to the source more precisely than the peak's height does.
- [x] Clipped tops. A sign the signal chain has saturated, not a sign of a machine defect.
- [x] Asymmetry up and down. Often looseness: the joint knocks more freely in one direction.
- [x] 1x phase between H and V at one support differs by around 90°. This is an argument for unbalance.
- [x] 1x phase in the axial direction on the two sides of the coupling differs by around 180°. This is an argument for shaft misalignment.
- [x] 1x phase at the two supports is in phase: the unbalance is more likely static. Out of phase: it is couple unbalance, and one correction plane will not remove it.
- [x] The phase jumps around from run to run. Looseness, rubbing, or resonance. There is no point balancing: the influence coefficient (the link between "fitted a weight" and "vibration changed," which the calculation is built on) will not hold.

> Phase on its own does not show you where the heavy point sits. What it shows is a difference: between directions, between supports, between runs. It is the difference that does the work.

## Step 6. Bring the signs together into a hypothesis using the table

| Dominant frequency | Direction and behaviour | Hypothesis | Check |
| --- | --- | --- | --- |
| 1x, clean spectrum | H roughly equal to V, axial small, phase stable, amplitude grows roughly as the square of speed | Rotor unbalance | A trial weight of known mass at a known radius. A valid trial run changes 1x by 20–30% or phase by 20–30° |
| 1x, clean spectrum | A sharp peak over a narrow speed range, vibration drops past the peak, phase shifts by around 180° | Resonance of the support, frame or rotor | A controlled start-up and coast-down with 1x amplitude and phase recorded. A bump test on the stopped machine |
| 1x and 2x are comparable | Axial has risen noticeably, 1x phase across the coupling differs by around 180° | Shaft misalignment | Check for a soft foot first, then align the shafts with dial indicators or a laser system on the warmed-up machine |
| A comb of 1x, 2x, 3x plus 0.5x and 1.5x | Pronounced directionality, phase does not repeat from run to run | Loose mounting or rubbing | Torque to sequence with a check measurement after each foot. Looseness will give itself away as a jump in the number at a specific bolt |
| Twice the mains frequency, 100 Hz on a 50 Hz supply | Radial, weak dependence on mechanical load | An electromagnetic cause: air-gap eccentricity, a loose stator | Cut the power during coast-down. The 100 Hz line cuts off instantly; the mechanical one decays smoothly along with the speed |
| High frequencies, not multiples of running speed, blurred, noise floor rising | Localised at one support, that support's temperature is higher than its neighbour's | Rolling-element bearing defects | An acceleration envelope spectrum in the 2–10 kHz band, a repeat measurement in a week, a look at the lubrication |
| A subsynchronous line at 0.4–0.5x | A machine on sleeve bearings, the line is unstable in frequency | Oil-film instability, oil whirl | Case-mounted sensors will not work here. You need proximity probes (non-contact shaft-position sensors) and the orbits built from them. Check the oil's viscosity, temperature and pressure |
| Blade-pass frequency and broadband noise at 1–10 kHz | Grows when moving away from the operating point, sounds like "gravel in a pipe" | Cavitation or flow separation | Bring the operating condition back to the design point, raise the inlet head. The cause is process-related, not the rotor |
| Two close lines, the vibration pulses audibly | Pulsing with a period of around a second, both lines stable | Beating between two independent sources, for example two fans on a shared duct | Stop the second machine. The pulsing will disappear; the machine is fine |

> The table gives you a hypothesis, not a conclusion. The same picture can come from two different causes, and the reverse is also true: looseness can raise 1x, and shaft misalignment can add harmonics. It is the right-hand column that turns a hypothesis into a diagnosis.

Sources: [ISO 13373-3:2015](https://www.iso.org/standard/40840.html) · [ISO 13373-5:2020](https://www.iso.org/standard/62202.html)

## Step 7. Change one parameter at a time and watch the response

This is where you stop observing and start experimenting. There is one rule: change exactly one thing per run. Otherwise the number changes, and you will never know what the change actually belongs to.

1. **Speed** — On a VFD, sweep the range in steps of 5–10% and record 1x and phase at each step. Unbalance builds up smoothly, roughly as the square of speed. Resonance produces a narrow peak and a phase shift. A natural frequency of the structure stays put in hertz while the running-speed line moves past it.
2. **Load** — Hold the speed steady, change the flow or the feed rate. Hydraulics and aerodynamics respond immediately. Unbalance is almost indifferent to load. Shaft misalignment often grows with load because of how the coupling reacts.
3. **Temperature** — Take a reading on a cold machine and again after 40–60 minutes of running. A drift in vibration during warm-up points either to thermal growth and induced misalignment, or to thermal bow in the rotor. A machine like this gets aligned with pre-set cold offsets, not aligned to zero.
4. **Power supply** — Disconnect the motor from the supply and watch the suspect line at the moment of coast-down. The electromagnetic component cuts off instantly; the mechanical one decays along with the speed. The test costs nothing and gives an unambiguous answer.
5. **Mounting** — Torque the bolts to a sequence and measure after each one. This finds you the specific foot, rather than "looseness in general." The order matters: fasteners and soft foot first, then shaft alignment, and only after that, balancing.

> Log everything in a single table: speed, load, support temperature, overall vibration (the total level across all frequencies), 1x, 2x, phase. Half a day of work, and this table becomes your finding.

## Three alarm levels, and why the trend matters more than one figure

A single measurement tells you the condition. Two measurements taken apart in time tell you the trend, and it is the trend that drives the decision. A machine that has sat at 4.0 mm/s for three months is less of a worry than one that climbed from 1.5 to 3.0 in a month, even though the second one is formally "better."

That is why setpoints are not set as a single threshold but as three levels. That way you get time to prepare, instead of an emergency shutdown on the night shift.

### Warning

Note it and watch more closely. You stop nothing, just add a point to the chart and shorten the interval between readings. A working guideline is the boundary between zones B and C in the applicable part of ISO 20816, or a statistical threshold built from that machine's own history.

### Alarm

Investigate and plan an intervention. You look for the cause using the sequence described above, and get parts and a maintenance window ready. The ISO 20816 guideline is the boundary between zones C and D, but for a specific machine it is more honest to work from its own baseline.

### Shutdown

Shut down per the site's own procedure. This level gets justified separately and agreed with operations, because a false shutdown costs money too.

- Take the baseline on a sound, warmed-up machine at its working operating point, and save it. Without a baseline, there is no trend.
- A statistical threshold from a machine's own history is often more honest than a standard one: the mean plus a few standard deviations gives you fewer false alarms and fewer missed ones.
- Besides the overall level, set narrowband setpoints too: separately for 1x, for 2x, and for the high-frequency band. A localised defect raises its own line long before the overall figure moves.
- Set the measurement interval by how fast the defect develops. A practical rule: the interval between readings should be no more than half the interval from the first sign to failure. For unbalance that is weeks to months; for rolling-element bearings it is usually one to six months.

> Treat the zone A, B, C, D figures as working guidance, and check the applicable part and edition of the standard for your specific machine: power, speed, support type and exceptions by machine type are all set out separately there. We cover the three different "tolerances" in a separate article.

Sources: [ISO 20816-1:2016](https://www.iso.org/standard/63180.html) · [ISO 13373-3:2015](https://www.iso.org/standard/40840.html)

## One hypothesis, one check, and the honest limits of the method

State a single hypothesis in the machine's own terms, not the instrument's. Not "high 1x," but "impeller unbalance after uneven cleaning." Then pick the one action that tells this theory apart from the closest competing one, and do only that.

You do not diagnose from a single spectrum, and that is a property of the method, not excess caution. A spectrum shows the average frequency content and does not tell cause from effect. Loose mounting raises 1x and pretends to be unbalance. A soft foot twists the casing and creates a genuine misalignment, which you will honestly find and honestly correct — and it will come back within a week. Resonance amplifies any excitation, so under resonance almost everything looks like unbalance.

There are also things case-mounted sensors simply cannot see. The shaft's position in the clearance and oil-film instability need proximity probes and orbits. Early bearing defects need acceleration and an envelope spectrum up to 10 kHz. Broken rotor bars in a motor need fine frequency resolution plus stator current analysis. If your problem falls into one of these, you need a different instrument and a different specialist — not one more trial weight.

- [x] "Unbalance" theory: test it with a trial weight. No change of 20–30% in amplitude or 20–30° in phase means the theory is wrong — and that is a result too.
- [x] "Shaft misalignment" theory: soft foot first, then shaft alignment on the warmed-up machine, and only in that order.
- [x] "Resonance" theory: change the system, not the rotor mass. The foundation's stiffness, the mass, the operating speed.
- [x] "Looseness" theory: torque to sequence with a measurement after each foot, then a repeat map across all supports.
- [x] The signs contradict each other, or point toward sleeve bearings, electrics, or hydraulics: call in a vibration specialist, and do not waste a shift on weights.

## What AXILINE does when the picture does not add up

We come to your site and go through this whole route: a reading at every bearing support in three directions, overall vibration and 1x with phase, an FFT spectrum and the time waveform on two channels at once, and a check for resonance from a start-up and coast-down. If the picture points to unbalance, we balance in one or two correction planes right in the machine's own bearings, with no disassembly and no removing the rotor. If it does not, you get a clear finding on what to do instead of balancing, and you do not pay for weights that would not have helped.

You can also buy the instrument and work with it yourself. The Balanset-1A kit is two accelerometers, a laser phase sensor, a two-channel USB module with preamplifiers, integrators and an ADC, and Windows software. The 1x vibration-velocity measurement range is 0.02–80 mm/s, speed from 100 to 100,000 RPM, phase measurement error ±1°. The software displays overall vibration and 1x, the spectrum and the time waveform, keeps an archive, and calculates the tolerance against the G grades (balance quality grades). You build the trend from your own saved data, not from memory.

Consultation support comes from engineers who design and manufacture the Balanset instruments and use them for on-site balancing themselves. Send us your readings, spectra and a description of the machine, and we will work through them with you.

Sources: [Balanset-1A manufacturer specification](https://vibromera.eu/product/balanset-1/) · [Balanset-1A operation manual](https://vibromera.eu/balanset-1a-operation-manual/)

## Frequently asked questions

**Where do I start if a fan starts humming right after the impeller is cleaned?**

Not with weights. Start with context: exactly what was done to the wheel and the housing. Then take a reading at both bearing supports in three directions and compare overall vibration with 1x. If 1x accounts for most of the overall figure, the spectrum is clean, H and V are comparable, and axial is small, that is unbalance from build-up removed unevenly. If a comb of harmonics and axial vibration have risen, check the housing and foot mounting you touched during cleaning first.

**Can you find the cause of vibration from a single spectrum?**

No. A spectrum shows the average frequency content and does not tell cause from effect. Loose mounting raises 1x and looks like unbalance, resonance amplifies any excitation, and a soft foot creates a genuine misalignment that will come back after you fix it. The minimum set for a diagnosis is the spectrum plus the time waveform plus phase plus the ratio between directions at every support plus the response to changing one parameter.

**Why measure three directions if horizontal is usually enough?**

Because the diagnostic value is in the ratio, not the figure. Axial vibration comparable to radial turns you away from balancing and toward shaft alignment and checking for a bent shaft. Vertical higher than horizontal means the fixing to the foundation has lost stiffness. Horizontal 3–5 times higher than vertical means resonance or looseness in one direction. From a single horizontal reading, these three cases all look the same.

**How do I convert a peak from hertz into a multiple of running speed?**

Divide the peak's frequency by the running-speed frequency in hertz. Running-speed frequency equals RPM divided by 60. At 1480 RPM that is 24.67 Hz, so a peak at 49.3 Hz gives 2.00x, while a peak at 100 Hz gives 4.05x and turns out to be twice the mains frequency, not a harmonic. Whole orders point to running-speed phenomena, half-orders to looseness and rubbing, fractional orders to bearings and belts.

**What setpoints should I use if there is no measurement history for the machine?**

Start with the guideline from the applicable part of ISO 20816: take the B/C zone boundary as a warning, the C/D boundary as an alarm, and agree the shutdown level separately with operations. In parallel, take a baseline on a sound, warmed-up machine, and after 3–5 rounds recalculate the thresholds from it statistically. After that, the standard's figures stay in the background, and the decisions get made by that specific machine's own trend.

**What do I do if the signs point to two causes at once?**

That is a normal situation, not a dead end. Fix things in order of dependency: whatever distorts everything else comes first. Fasteners and soft foot, then shaft alignment, then balancing. Take a check measurement after every step and record it. If the second cause shrinks on its own once you fix the first, it was a consequence, not a separate defect.