Vibration Analysis Explained: From Sensor to Decision

This explainer is part of our complete guide to predictive maintenance for SMEs. Vibration analysis is the highest-leverage diagnostic technique inside any condition-monitoring programme — everything below explains why.

Why Vibration Matters

Almost every fault in rotating equipment shows up in vibration before it shows up anywhere else. A worn bearing produces characteristic frequencies. A misaligned coupling generates twice-shaft-speed vibration. An unbalanced rotor produces synchronous vibration. A loose foundation produces sub-synchronous content.

Each fault has a signature. Once you know the signatures, you can diagnose problems from a measurement that takes seconds. That's why vibration analysis is the foundation of any serious condition monitoring programme — and why vibration is one of the six core engineering data types every team should be tracking. To understand how vibration analysis fits into a broader strategy, see our complete guide to predictive maintenance for SMEs.

The Three Things Vibration Sensors Measure

A standard accelerometer measures acceleration, but it can be integrated to derive velocity and displacement. Each domain reveals different fault characteristics:

• Acceleration (g or m/s²) — sensitive to high-frequency events. Best for early-stage bearing defects.
• Velocity (mm/s RMS) — the workhorse measurement. Sensitive across mid-frequency range. Used in international standards including ISO 10816 / ISO 20816 for machine vibration evaluation. [1]
• Displacement (microns or mils) — sensitive to low-frequency content. Used for shaft motion in fluid-film bearings.

Most condition monitoring on rotating equipment uses velocity as the primary measurement, because the standards and the tribal knowledge are built around it. A practical example: a pump running at 1,500 RPM with healthy bearings might measure 1.2 mm/s RMS velocity at the bearing housing; the same pump with developing bearing wear might rise to 3.5 mm/s RMS over the course of several weeks, well before the failure becomes audible or causes a process trip. That early warning is the whole point.

The three domains aren't interchangeable. Choosing the wrong one is one of the most common mistakes engineering teams make when they first start vibration monitoring — an acceleration reading on a slow-speed shaker (under 600 RPM) will hide problems that displacement would catch; a displacement reading on a high-speed turbine will miss bearing defects that acceleration would flag. The pragmatic rule: use velocity as the default, add acceleration for early bearing detection, add displacement when you need shaft-orbit data on fluid-film bearings.

The Standard: ISO 10816 / ISO 20816

The ISO 10816 series (now being superseded by ISO 20816) defines vibration severity zones for different machine types. The standard categorises machinery vibration into four zones:

Zone	Condition	Action
A	Newly commissioned	Normal — no action
B	Acceptable for unrestricted long-term operation	Monitor
C	Unsatisfactory for long-term operation	Plan corrective action
D	Sufficient severity to cause damage	Immediate action

The exact velocity thresholds for each zone depend on machine size and class. ISO 10816 / 20816 defines specific limits per machine category. [1] These standards provide a defensible, internationally recognised baseline — useful when you need to make a maintenance decision and back it up with documentation.

To put numbers on it: for a medium-sized industrial pump (Class II machine in ISO 10816-3), the velocity thresholds run roughly 2.3 mm/s RMS for Zone A/B boundary (acceptable for long-term operation), 4.5 mm/s RMS for Zone B/C boundary (plan corrective work), and 7.1 mm/s RMS for Zone C/D boundary (immediate action). A reading of 5.0 mm/s RMS sits firmly in Zone C: the pump can keep running for a short period but maintenance needs to be scheduled, ideally during the next planned outage rather than as a reactive callout. That's the kind of decision the standard is designed to support.

The Time Domain vs Frequency Domain

Raw vibration data is a time series — amplitude vs time. That's useful, but it doesn't reveal what's actually causing the vibration. The breakthrough is the Fast Fourier Transform (FFT), which decomposes the signal into its constituent frequencies.

In the frequency domain, faults become recognisable as peaks at specific frequencies:

• Imbalance: Peak at 1× running speed (1×).
• Misalignment: Peaks at 1× and especially 2× running speed.
• Looseness: Multiple harmonics (1×, 2×, 3×, etc.) and sometimes sub-synchronous content.
• Bearing defects: Specific frequencies determined by bearing geometry — BPFI, BPFO, BSF, FTF.
• Gear mesh problems: Peaks at gear mesh frequency and its sidebands.

SKF's bearing failure literature, which is the industry reference for bearing diagnostics, describes the four-stage progression of bearing damage from early defect through audible/tactile failure — with each stage producing recognisable frequency signatures. [2]

The Four Bearing Fault Frequencies (BPFI, BPFO, BSF, FTF)

Rolling-element bearings generate four characteristic defect frequencies that any vibration analyst (or modern AI platform) uses to localise the fault to a specific component. Each frequency depends on the bearing's geometry — number of rolling elements, pitch diameter, contact angle — and the shaft's rotational speed. [3]

• BPFI — Ball Pass Frequency, Inner race. The frequency at which rolling elements pass a defect on the inner race. Typically around 5× to 7× shaft speed for common deep-groove bearings. Modulates strongly with shaft speed (sidebands at ±1× running speed) because the defect rotates with the shaft.
• BPFO — Ball Pass Frequency, Outer race. Rolling elements passing a defect on the outer race. Typically 3× to 5× shaft speed. Sidebands are less prominent because the outer race is usually stationary. Often the first defect frequency to appear because outer-race spalling is the most common bearing failure mode in industrial pumps and motors.
• BSF — Ball Spin Frequency. Rolling element rotation. Typically 2× to 3× shaft speed. Appears with rolling-element damage (spall on the ball/roller itself). Modulates with the cage frequency.
• FTF — Fundamental Train Frequency. Cage rotation. Typically less than 1× shaft speed (sub-synchronous). Appears with cage damage — relatively rare on its own but a late-stage indicator when present.

For a typical pump bearing (Conrad-type, 9 rolling elements, running at 1,500 RPM = 25 Hz shaft speed), BPFI might appear around 138 Hz, BPFO around 87 Hz, and BSF around 53 Hz. An analyst seeing a clean peak at 87 Hz with 1× (25 Hz) sidebands knows immediately: outer-race defect, monitor closely, plan replacement. The same data shown to a maintenance engineer with no vibration training would just be "a peak somewhere". That's the expertise gap that AI now closes.

Why It's Hard to Do Manually

Reading vibration spectra is a learned skill. A trained vibration analyst can look at a spectrum and identify problems in seconds. But the skill takes years to develop, and qualified analysts are scarce. The Vibration Institute and similar bodies certify analysts through Cat I–IV qualifications, with most production reliability work requiring Cat II or higher.

For SMEs, this creates a problem: vibration analysis is the most valuable condition monitoring technique, but it requires expertise most teams don't have in-house. The traditional answer was to bring in a contractor for periodic walk-arounds — effective but expensive and infrequent. Typical UK contract rates for periodic vibration surveys run £800–£1,500 per visit covering 30–60 assets, repeated quarterly — useful, but a snapshot rather than a continuous signal. To make the business case for moving beyond periodic walk-arounds to continuous monitoring, see our guide to building a business case for predictive maintenance. For a UK-specific view of how this fits into a wider analytics stack, see our manufacturing analytics software for UK SMEs overview.

How AI Changes the Equation

AI doesn't replace vibration analysts, but it does change who can benefit from vibration analysis. Modern AI-powered platforms can:

• Learn each asset's normal vibration signature automatically and flag deviations — no manual threshold setting required.
• Recognise common fault patterns (imbalance, misalignment, looseness) without requiring an expert to interpret the spectrum.
• Correlate vibration with other signals like temperature, current, and pressure to build a richer diagnostic picture.
• Maintain continuous monitoring instead of relying on quarterly walk-arounds — catching trends earlier.
• Translate findings into plain language so a maintenance engineer can act on them without years of vibration training.

The result isn't that AI replaces the certified vibration analyst — it's that vibration analysis becomes accessible to teams that couldn't justify hiring one. AI anomaly detection excels here because vibration is exactly the kind of signal where subtle pattern changes precede major failures by weeks.

The goal isn't to make every maintenance engineer into a vibration specialist. It's to give them the same insights a specialist would deliver — without needing to be one.

Where to Start with Vibration Monitoring

If you're starting from no vibration monitoring at all, the route is roughly the same for every SME. Five steps, in order — skipping any of them is the most reliable way to waste budget.

1. Identify Critical Rotating Equipment

Pumps, motors, fans, gearboxes, compressors. Start with assets where failure stops production, creates safety risk, or has long lead-time replacement parts. Understanding the true cost of unplanned downtime on these assets is what justifies the monitoring investment to finance. Use our downtime cost calculator to quantify the value of early fault detection on your critical assets.

2. Begin with Periodic Measurements

Handheld instruments are cheap (a basic accelerometer-based meter is £500–£1,500). Use them monthly to build a vibration history for each critical asset. Without a baseline, you cannot tell whether a reading is normal or alarming. Most SME programmes spend the first three months purely building baseline data — that's not wasted time, it's the foundation everything else depends on.

3. Add Continuous Monitoring on Highest-Criticality Assets

Permanently mounted sensors with cloud connectivity. The cost has dropped dramatically — wireless vibration sensors now run £200–£500 per point compared with £3,000+ a decade ago. Start with the three to five assets that drive the largest share of unplanned downtime cost. Continuous monitoring catches the trends that monthly walk-arounds miss.

4. Use a Platform That Interprets the Data

Use a platform that interprets the data. Raw vibration data is useless without a way to make decisions from it — spectra, trend charts, and threshold alarms are necessary but not sufficient. AI-powered platforms close the gap between raw signal and actionable insight, translating fault signatures into plain-English diagnoses an in-house engineer can act on.

5. Build the Workflow

Vibration alerts that go to email get ignored. Vibration alerts that integrate with the maintenance work order system get acted on. The technology is only as effective as the operational routine wrapped around it. Define who owns the alert, how long they have to investigate, and what escalation looks like if they don't — before the first sensor goes on the wall.

Common Faults at a Glance

Symptom	Likely Cause	Confirmation
High 1× vibration	Imbalance	Phase analysis — 90° between H/V
High 2× vibration	Misalignment	Phase — 180° across coupling
Multiple harmonics	Looseness	Amplitude varies with load
High-frequency content	Bearing defect (early)	Envelope demodulation
Gear mesh sidebands	Gear damage	Sideband spacing = 1× or modulating frequency
Sub-synchronous	Oil whirl, rub, looseness	Frequency relative to running speed

Key Takeaways

• Vibration analysis is the most powerful condition monitoring technique for rotating equipment. Almost every fault appears in vibration first.
• Three measurement domains: acceleration, velocity, displacement. Velocity (mm/s RMS) is the workhorse.
• ISO 10816 / 20816 defines severity zones from "newly commissioned" (A) to "sufficient to cause damage" (D).
• FFT analysis reveals fault signatures: imbalance at 1×, misalignment at 2×, looseness as harmonics, bearing defects at characteristic frequencies.
• Manual analysis requires specialist training — historically a barrier for SMEs.
• AI-powered platforms make vibration analysis accessible by learning normal signatures, recognising fault patterns, and flagging anomalies before they become failures, all translated into plain language.
• Start with critical rotating equipment, build vibration history, add continuous monitoring on highest-criticality assets, and use a platform that turns data into decisions.

Vibration analysis was once gatekept by expensive specialists. Today, with the right platform, it's a competitive advantage any maintenance team can access — and the gap between sites that use it and sites that don't is widening.