Getting Started with Vibration Analysis: A Plant Engineer's Guide

Why Vibration Is the First Thing to Measure

If you could only monitor one parameter on a rotating machine, vibration would be the right choice. Research from the NASA Glenn Research Center and decades of reliability engineering practice converge on the same finding: approximately 80% of rotating equipment failures produce detectable vibration changes before any other symptom appears — before temperature rises, before current draw changes, before the operator hears something.

A bearing developing an outer race defect generates a characteristic vibration signature 3-6 weeks before functional failure. A misaligned coupling shows up in vibration immediately. An unbalanced impeller, a loose mounting bolt, a cracked gear tooth — vibration catches them all, and it catches them early.

The challenge isn't whether vibration analysis works. It's that traditional vibration analysis requires a specialist with years of training, an expensive analyzer, and hours of interpretation time per machine. Most plants don't have that resource. This guide covers the fundamentals — enough to understand what your sensors are telling you, what the standards mean, and where AI changes the equation.

The Key Metrics: What Your Sensor Actually Measures

A vibration sensor (accelerometer) on a bearing housing produces a time-domain signal — a waveform showing acceleration over time. From that raw signal, several metrics are derived. Each tells you something different.

RMS (Root Mean Square) velocity — the overall severity indicator. RMS is the most widely used vibration metric. Measured in mm/s (or in/s), it represents the overall energy of the vibration signal. Think of it as the "average intensity" of the machine's vibration. ISO 10816 and most OEM specifications use RMS velocity as the primary assessment parameter. A healthy pump bearing might run at 1.5-2.5 mm/s RMS. The same bearing with a developing defect might read 4-6 mm/s. A bearing approaching failure could hit 10-15 mm/s or higher.

Peak velocity — the maximum excursion. Where RMS averages the signal, peak captures the single highest point. A machine with a mechanical impact (a chipped gear tooth, a cracked bearing element) might have a normal RMS but an elevated peak value — because the impact is brief but intense. The ratio between peak and RMS tells you about the nature of the vibration.

Crest factor — the spikiness detector. Crest factor is simply peak divided by RMS. A pure sine wave has a crest factor of 1.414. Normal machine vibration typically falls between 2.5 and 3.5. When crest factor climbs above 4 or 5, it means the vibration signal contains sharp impacts — the kind produced by a bearing with a localized defect (a pit, a spall, a crack). Crest factor is particularly useful in early fault detection because it can rise significantly while RMS is still within normal limits.

Kurtosis — the statistical anomaly flag. Kurtosis measures how "heavy-tailed" the vibration signal distribution is. A normal (Gaussian) distribution has a kurtosis of 3.0. Healthy machines typically show kurtosis between 2.5 and 4.0. Values above 5 strongly indicate impulsive events — bearing defects, gear tooth damage, or intermittent contact. Like crest factor, kurtosis can flag early-stage damage that RMS misses.

ISO 10816: The Universal Severity Chart

ISO 10816 (now updated as ISO 20816) provides vibration severity classifications based on machine type and mounting. It defines four zones:

| Zone | Classification | Action | |---|---|---| | A (Green) | Good | Newly commissioned or excellent condition. Typical: <2.3 mm/s for Class III machines. | | B (Yellow) | Acceptable | Normal long-term operation. No action required. Typical: 2.3-4.5 mm/s. | | C (Orange) | Alert | Conditionally acceptable. Investigation recommended, plan maintenance. Typical: 4.5-7.1 mm/s. | | D (Red) | Danger | Risk of damage. Immediate action required. Typical: >7.1 mm/s. |

Note: These values are for Class III machines (large rotating equipment on rigid foundations, 15-300 kW). Class I (small machines) and Class IV (turbomachinery) have different thresholds. Always check the specific standard for your equipment class.

ISO 10816 is useful as a starting point, but it has limitations. The zones are based on absolute RMS velocity and don't account for machine-specific baselines, operating conditions, or multi-parameter patterns. A pump that has always run at 3.8 mm/s is in a different situation than one that used to run at 1.5 mm/s and just jumped to 3.8. Both are "Zone B" by the standard, but only one is trending toward failure.

Bearing Fault Frequencies: The Fingerprints

Every rolling element bearing produces mathematically predictable vibration frequencies when a defect develops. These frequencies depend on the bearing geometry and shaft speed, and they act as fingerprints — each defect type has its own characteristic frequency.

BPFO (Ball Pass Frequency, Outer Race): The frequency at which rolling elements pass over a defect on the outer race. This is the most common bearing defect — outer race faults account for roughly 40% of bearing failures. BPFO typically falls between 3-5x shaft speed. On a motor running at 1,800 RPM (30 Hz), BPFO might be around 105 Hz for a common 6205 bearing.

BPFI (Ball Pass Frequency, Inner Race): The frequency produced by inner race defects. Usually higher than BPFO (4-6x shaft speed) and often modulated by shaft rotation — appearing as sidebands around the BPFI frequency. Inner race faults account for about 30% of bearing failures.

BSF (Ball Spin Frequency): The rotational frequency of the rolling elements themselves. Ball defects show up at 2x BSF (because the defect contacts both races per revolution). Less common than race defects, but distinctive in the spectrum.

FTF (Fundamental Train Frequency): The rotational frequency of the bearing cage. Cage faults are relatively rare but dangerous because they can cause sudden catastrophic failure with less warning than race defects. FTF is typically 0.35-0.45x shaft speed.

You don't need to calculate these by hand. Bearing manufacturers publish defect frequency tables, and any vibration analysis software will calculate them from the bearing model number and shaft speed. The point is knowing they exist and what they mean: if you see energy at 105 Hz on a 30 Hz shaft, and the bearing catalog says BPFO for that bearing at that speed is 105 Hz — you have an outer race defect.

Sensor Placement: Getting Useful Data

Where you mount the accelerometer determines what you can detect. Three principles:

Mount on the bearing housing, as close to the load zone as possible. Vibration signals attenuate rapidly through structural interfaces. A sensor mounted on the machine frame 30 cm from the bearing sees a fraction of the signal compared to one mounted directly on the bearing housing. For horizontal machines, the load zone is typically at the bottom of the bearing (gravity loads) or in the direction of belt/chain pull.

Measure in three axes. Radial (horizontal and vertical) and axial vibration carry different information. Imbalance shows up primarily in radial vibration. Misalignment often appears in axial vibration. Bearing defects can appear in all three axes but may be strongest in one. Triaxial sensors (or three single-axis sensors mounted orthogonally) give you the full picture.

Ensure rigid mounting. A loosely mounted sensor adds its own resonance to the measurement. Stud-mounted or adhesive-bonded sensors are best for permanent installations. Magnetic mounts work for route-based portable measurements but introduce a resonance around 2-4 kHz that can mask high-frequency bearing defect signatures. For permanent online monitoring, always use stud or adhesive mounting.

Common Fault Patterns: What to Look For

Four fault types account for the majority of rotating equipment vibration problems:

Imbalance: Vibration at 1x shaft speed, predominantly in the radial direction. The most common vibration problem. Causes: residual manufacturing imbalance, fouling buildup (fans, impellers), broken/eroded impeller vanes, thermal bow. Signature is clean and sinusoidal — high RMS but normal crest factor.

Misalignment: Vibration at 1x and 2x shaft speed, with a significant axial component. Angular misalignment emphasizes 1x axial; parallel (offset) misalignment emphasizes 2x radial. Often the first symptom after a motor or pump has been reinstalled after maintenance.

Mechanical looseness: Vibration at multiple harmonics of shaft speed (1x, 2x, 3x, 4x and higher), sometimes with half-harmonics (0.5x, 1.5x). The spectrum looks "noisy" with many peaks. Causes: loose mounting bolts, cracked frame, loose bearing in housing, excessive bearing clearance.

Bearing defect: Vibration at bearing defect frequencies (BPFO, BPFI, BSF, FTF) and their harmonics, often with sidebands. Early-stage defects show up first in the high-frequency range (acceleration envelope) before appearing in the velocity spectrum. This is why kurtosis and crest factor are valuable early indicators — they detect the impulsive hits before the overall vibration level rises.

Where AI Changes the Game

Traditional vibration analysis requires a trained analyst to collect data, review spectra, identify fault frequencies, compare to baseline, and write a recommendation. A skilled analyst can assess 15-20 machines per day. A plant with 500 rotating assets needs weeks of analyst time per survey cycle.

AI-based vibration analysis doesn't replace the analyst's knowledge. It encodes it and scales it. A trained model monitors all 500 machines continuously, compares current signatures against learned per-machine baselines, detects multi-parameter patterns (vibration + temperature + current), and flags anomalies with explainable attribution — telling the engineer which features are driving the detection and what fault pattern they match.

The analyst's role shifts from screening hundreds of spectra (mostly normal) to investigating the 5-10 machines that the AI flagged as genuinely changing. That's a better use of scarce expert time.

For plants without a dedicated vibration analyst — which is most small and mid-size facilities — AI makes vibration-based predictive maintenance accessible for the first time. You don't need to read spectra. You need sensors that collect the data and a platform that interprets it.

Start Listening to Your Machines

Prevly combines continuous vibration monitoring with AI-powered fault detection and diagnostics. Connect standard industrial accelerometers, and the platform handles the rest — learning each machine's baseline, detecting anomaly patterns, estimating remaining useful life, and explaining every alert in terms your maintenance team can act on. No vibration analysis certification required.

Start your free trial at prevly.org and find out what your rotating equipment has been trying to tell you.