Industrial vibration analysis: when to use it and what it actually tells you

A bearing vibrating at 4 mm/s on a 1500 RPM motor has roughly 60 days of service life left. At 7 mm/s, 30. At 11 mm/s, it's already failing. The question isn't whether the bearing will break — it's when. And the gap between those three points on the curve is three months of margin to schedule the intervention into a planned shutdown instead of into a Friday 3 a.m. emergency. That's what vibration analysis delivers: turning unexpected breakdown into planned maintenance.

This article covers what failures it detects, on which assets, with what early-warning window, what's required to implement it well, and when it simply isn't worth it.

What vibration analysis actually detects

The standards that frame the discipline are ISO 10816/20816 (overall vibration severity) and ISO 13373 (vibration condition monitoring procedure). For Class II machines (15-300 kW general use), severity classes are: A — good (<2.8 mm/s RMS), B — acceptable (<4.5), C — degrading (<7.1), D — critical (>7.1). These are the values shown by continuous monitoring alarms or by a portable analyst's report.

The failure modes a trained analyst reads in the spectrum:

Bearing defects (4 stages). Stage 1: detectable only by ultrasound (>30 kHz). Stage 2: detectable by envelope analysis, 1-3 months pre-failure. Stage 3: visible in the standard vibration spectrum as harmonics of characteristic defect frequencies (BPFI inner race, BPFO outer race, BSF ball spin, FTF cage), 1-4 weeks pre-failure. Stage 4: high broadband vibration, imminent failure.

Imbalance. Dominant peak at 1× RPM, radial direction. Cause: uneven mass distribution (caked dirt, lost blade, deformation). Instant detection.

Misalignment. Peaks at 1× and 2× RPM, axial+radial. Instant detection with phase analysis between coupling halves.

Mechanical looseness. Integer multiples of running frequency (1×, 2×, 3×, 4×...) and subharmonics (0.5×). Loose foundations, missing bolts, degraded fits.

Resonance. Sharp peak at a structural natural frequency. Any excitation force matching that frequency is amplified 5× to 50×. Key diagnostic if mass or stiffness has changed recently.

Gear defects. Peak at gear mesh frequency (number of teeth × RPM) with sidebands at running frequency. The spacing and amplitude of sidebands tells you whether it's even wear, broken tooth, or eccentricity.

Pump cavitation. Random broadband vibration with no clean peaks. Linked to insufficient suction pressure.

Electrical motor defects. Peak at 2× line frequency (100 Hz in Europe) — broken rotor bars, air-gap eccentricity.

When vibration analysis is the right tool

It works exceptionally well on rotating machinery: motors, pumps, fans, compressors, gearboxes, turbines. In a typical industrial plant this represents over 50% of critical assets. The technique is mature, the standards are clear, the tools are available from €800 per measurement point.

It works less well on: hydraulic systems (better with oil and pressure analysis), electronic control (better with electrical diagnostics and thermography), batch chemical processes, low-speed equipment (<30 RPM, where you need very-low-frequency accelerometers and specialist analysis).

Vibration analysis is economically justified when at least one of these holds:

• The machine is critical to production (downtime = €1,000+/hour in losses)
• Observed MTBF is shorter than spec
• Access is difficult and visual inspection isn't viable
• Cost of catastrophic failure dwarfs programme cost

A medium-sized Belgian plant with 50 critical motors typically pays back the full investment in 8 to 12 months, counting avoided failures and reduced spare-parts inventory.

Permanent monitoring vs portable route

Permanent (wired or wireless IoT sensors). Cost per point: €600 to €2,000 installed plus analytics platform. Justified for the top 10-20 critical assets where one hour of early warning is worth thousands. Runs 24/7, integrates with the CMMS, triggers automatic alarms.

Portable monthly route. Cost per point: €150 to €300 per year. A technician with handheld equipment (Fluke 805, SKF Microlog, Adash) collects readings on a defined route. Suitable for the 50-200 secondary assets.

The combination is the norm in serious plants: continuous on critical, route on the rest. Data flows into a unified platform (Bently Nevada System 1, SKF Observer, Schaeffler ConditionAnalyzer) integrated with the CMMS.

How to read a vibration spectrum (the basics)

The spectrum is the machine's vibration signature. Three steps to read it:

Step 1: identify the running frequency. If the motor runs at 1,500 RPM, that's 25 Hz. Mark that line as your 1× RPM reference.

Step 2: look for peaks at characteristic frequencies. 1× RPM → imbalance/runout. 2× RPM → misalignment or looseness. Integer multiples → advanced looseness. Gear mesh frequency (Z teeth × RPM) → gear health. Bearing defect frequencies (calculated from bearing geometry, available in manufacturer catalogues or SKF/Schaeffler calculators) → race or rolling-element defect.

Step 3: compare to baseline and to a symmetric machine. A naturally quiet machine and a naturally noisy one can show the same absolute spectrum yet be in very different states. Trend matters more than the single value.

Envelope demodulation is a fundamental complementary technique: it reveals bearing defects that lie below the noise floor of the standard spectrum. Any professional analysis system includes it.

The 5 most common implementation mistakes

Wrong sensor placement. Must be on the bearing housing, in the load zone, in rigid contact (bolted, not glued). On the casing or frame, bearing vibration arrives damped and filtered.

Setting alarms before establishing a baseline. Factory thresholds are statistical averages. A machine that's noisy by design triggers constant alarms; a quiet one that's degrading slips through. You need 4 to 12 weeks of stable measurement before fixing your own thresholds.

Comparing across bearings without normalising. A small motor bearing and a mill bearing have very different baseline levels. Comparing absolute mm/s across different machines is meaningless; comparing each one to its own trend is.

Reacting to a single value. A point reading out of range can be electrical noise, momentary resonance, or an operating event. A sustained rising trend over 4-6 weeks is what counts.

No ISO 18436-2 Level II certified analyst. Software detects peaks; humans interpret what they mean. Without an analyst who understands machine dynamics, the system generates noise and false alarms until the team stops checking it.

When NOT to use vibration analysis

Low-speed equipment (<30 RPM): standard accelerometers lose signal below 2-5 Hz. Specialist sensors exist (low-frequency accelerometers, proximity probe displacement) but complexity and cost rise.

Reciprocating combustion engines: the vibration signature is dominated by combustion, not mechanical defects. Better with combustion and oil analysis.

Electronic equipment: failures don't manifest as vibration. Better with thermography.

Equipment with very short expected MTBF and very low part cost: if a €600 motor bearing is replaced preventively every 12 months by procedure, monitoring it is overkill.

Closing

Vibration analysis is the most mature and best-developed condition-monitoring technique for rotating machinery. Implemented well — sensors in the right place, baseline established, CMMS integration, certified analyst — it delivers 1-6 months early warning on most critical failure modes.

Implemented badly, it's expensive noise and dashboards no one reads.

If you want a free audit to identify which assets in your plant justify vibration monitoring (continuous or route-based) and a realistic implementation plan, reach out. You can also see our predictive maintenance offer, which combines vibration, thermography, and oil analysis based on the asset.