Vibration Analysis for Predictive Maintenance: A Practical Guide

Key takeaways

Vibration analysis reads the tiny motions a running machine makes and turns them into an early warning of imbalance, misalignment, looseness and bearing or gear wear.
Velocity in mm/s RMS is the everyday severity measure; displacement suits slow shafts and acceleration or envelope suits early high-frequency bearing faults.
Judge readings against a standard like ISO 10816 and against each machine's own baseline trend, not a single universal number.
Characteristic fault frequencies (1x running speed, bearing defect frequencies, gear mesh) tell you not just that a machine is degrading but why.

Vibration analysis is the condition-monitoring technique that reads the small oscillations a running machine makes and uses them to detect faults long before they become breakdowns. Every rotating asset (a motor, a pump, a fan, a gearbox) vibrates in a pattern that is normal for its design and speed. When a bearing starts to spall, a coupling drifts out of alignment, or a rotor loses balance, that pattern changes in a way you can measure, trend and diagnose. Done well, it moves a plant from reacting to failures toward planning repairs on the plant's terms, which is the whole point of a predictive maintenance strategy.

What the sensor is actually measuring

A vibration reading can be expressed three ways, and each suits a different part of the frequency range. Displacement (microns) is the physical distance the part moves and is most useful on slow-turning shafts and proximity-probe monitoring of journal bearings. Velocity (mm/s) is the rate of that motion and correlates well with the fatigue-causing energy in the mid-frequency band where the most common faults live, which is why it is the default overall severity measure. Acceleration (g) emphasises high-frequency events and, combined with envelope or demodulation processing, is the sharpest tool for catching early bearing and gear defects that a plain velocity number would miss.

Most portable analysers and online sensors capture a time waveform and convert it with a Fast Fourier Transform (FFT) into a spectrum: amplitude against frequency. The spectrum is where diagnosis happens, because different faults deposit their energy at different, predictable frequencies.

Reading the spectrum: frequency tells you the fault

The single most useful idea in vibration analysis is that the frequency of a peak points to its cause. Working from the shaft running speed (often written 1x):

Imbalance shows a high peak at 1x running speed, mostly radial, that grows with the square of speed.
Misalignment typically raises 2x (and sometimes 3x) running speed, often with strong axial vibration across a coupling.
Mechanical looseness produces a run of harmonics (1x, 2x, 3x and up), and sometimes half-order peaks.
Rolling-element bearing defects appear at non-synchronous bearing defect frequencies and, early on, as raised high-frequency noise best seen with envelope analysis.
Gear problems show energy at the gear-mesh frequency (number of teeth times shaft speed) with sidebands spaced at shaft speed.

A worked example: severity and a bearing frequency

Two numbers put this on solid ground. First, severity. ISO 10816 (continued as ISO 20816) sets velocity zones A (good), B (acceptable), C (unsatisfactory) and D (unacceptable) by machine class. For a common medium machine of 15 to 75 kW on a rigid foundation, the A/B boundary sits near 2.8 mm/s RMS and the B/C boundary near 4.5 mm/s RMS. So a pump reading 3.1 mm/s is in Zone B, acceptable for long-term running but worth trending; the same pump climbing to 5.0 mm/s has crossed into Zone C and should be planned for intervention.

Second, the fault frequency. The outer-race defect frequency of a rolling bearing (BPFO) is:

BPFO = (n / 2) × f_r × (1 − (d / D) × cosφ)

where n is the number of rolling elements, f_r is the shaft speed in Hz, d is the ball diameter, D is the pitch diameter and φ is the contact angle. Take a motor at 1,480 rpm (f_r = 24.7 Hz) on a bearing with n = 8, d = 12 mm, D = 62 mm and φ = 0:

BPFO = (8 / 2) × 24.7 × (1 − 12/62) = 4 × 24.7 × 0.806 ≈ 79.6 Hz

Now the diagnosis is specific: a rising peak near 80 Hz (with harmonics and envelope energy) on that motor points to an outer-race defect, not imbalance or misalignment. That is the difference between "the motor is getting worse" and "change the drive-end bearing at the next window", which is exactly the kind of lead time that keeps a fault off your bearing troubleshooting and electric-motor troubleshooting firefight list.

The P-F curve: why lead time is the prize

Vibration analysis earns its keep because of where it sits on the P-F curve, the interval between the point a failure becomes detectable (P) and the point of functional failure (F). Vibration usually detects rotating-equipment faults early in that interval, giving weeks or months of warning, far more than heat or noise a technician would feel by hand. That lead time is what lets you order the part, schedule the labour and take the machine down on a planned stop instead of losing it mid-shift. It is also why measurement interval matters: readings must be frequent enough to catch the fault while usable warning remains.

How to start a route without boiling the ocean

You do not need a full online system on day one. A workable start looks like this: rank assets by criticality and pick the important, hard-to-spare rotating machines; establish a baseline reading for each in a known-good state; set a route and interval by criticality (monthly is a common starting cadence); take readings at the same points and speeds each time so trends are comparable; and act on trend changes, not just threshold breaches. Feed every finding and repair back into your work-order history so you build a picture of which assets actually fail and why, which is the raw material for reliability-centered maintenance and for a meaningful MTBF and MTTR calculation.

As coverage grows, most plants hit the same ceiling: monthly manual routes miss faults that develop between visits, and the data lives in a handheld or a spreadsheet, disconnected from the downtime it is meant to prevent. That is where continuous online monitoring and machine-health software earn their place, streaming vibration and other condition data against the same production and stop data your operators already generate.

Comparing condition-monitoring platforms?

See how machine-monitoring and predictive-maintenance tools stack up before you buy.

Read the machine-monitoring guide

When the goal is not just a bearing warning but linking machine health to real production loss and closing the loop to a work order, the platform we recommend is Fabrico: it reads OEE and stops directly from the machines and shows the true cause on video, so condition data and downtime data live in one place. Fabrico is EU-built, so your production data stays in EU jurisdiction (ISO 27001 / 20000-1 / 9001, supports audit-readiness). If that fits, you can book a Fabrico demo. The calculators and guides here are free regardless.

FAQ

Is vibration measured in displacement, velocity or acceleration?

All three, but for most rotating machinery overall velocity in mm/s RMS is the workhorse because it tracks fatigue-causing energy well across the mid-frequency band where imbalance, misalignment and looseness live. Displacement suits low-speed shafts and proximity-probe work; acceleration and envelope suit high-frequency faults such as early bearing and gear defects. A good analyst reads the one that best shows the fault, not just a single overall number.

What is a good vibration level for a motor or pump?

Judge it against a standard and against the machine's own baseline, not a single universal number. ISO 10816 (and its successor ISO 20816) sets velocity severity zones A to D by machine class. For a common medium machine of 15 to 75 kW on a rigid foundation, the good-to-acceptable boundary sits near 2.8 mm/s RMS and the acceptable-to-unsatisfactory boundary near 4.5 mm/s RMS. A rising trend from a machine's own baseline is often a stronger signal than the absolute value.

How often should we take vibration readings?

Set the interval by criticality and by how fast the failure mode develops. Critical unspared machines may warrant continuous online monitoring; important machines are often walked on a monthly or fortnightly route; lower-criticality assets quarterly. The rule is that the interval must be short enough to catch the fault on the P-F curve while there is still lead time to plan the repair.

Can vibration analysis replace all other maintenance?

No. It is one condition-monitoring technique and it is strongest on rotating equipment. It works best alongside oil analysis, thermography, motor-current analysis and basic inspection, all feeding a wider reliability strategy. Vibration tells you a machine is degrading and often why; you still need the planning, spares and work-order discipline to act on the warning in time.