Vibration Analysis in Predictive Maintenance, With Example

What is vibration analysis

Vibration analysis measures how a rotating machine vibrates and uses the spectrum of that signal to find the cause: unbalance, misalignment, looseness or a bearing defect. It is the most widely used technique in predictive maintenance, because each mechanical fault leaves a signature at its own frequency long before the machine stops.

The core idea is simple: overall vibration tells you there is a problem, and the spectrum tells you which one.

Vibration analysis visual map

The map shows how unbalance, misalignment and looseness appear in the spectrum, using the 1780 rpm motor example from this article.

From signal to spectrum

An accelerometer mounted on the bearing housing produces a time signal. The FFT (fast Fourier transform) converts it into a spectrum: frequency (Hz) on the horizontal axis, amplitude on the vertical. Three concepts are enough to start:

• Running speed (1X): rpm ÷ 60. At 1780 rpm, 1X = 29.67 Hz. Multiples (2X = 59.3 Hz; 3X = 89.0 Hz) are harmonics.
• Measured quantity: acceleration emphasizes high frequencies (bearings); velocity, in mm/s RMS (or in/s), is the best measure of overall severity; displacement suits low frequencies. Relationship: acceleration = velocity × 2π × f. A peak velocity of 6.0 mm/s at 29.67 Hz equals 1.12 m/s², about 0.114 g.
• RMS and peak: for a sine wave, RMS = 0.707 × peak. The 6.0 mm/s peak becomes 4.24 mm/s RMS.

Acquisition settings decide what you can see. With Fmax = 1000 Hz and 1600 lines, resolution is 1000 ÷ 1600 = 0.625 Hz, enough to separate 1X (29.67 Hz) from neighboring peaks. The sampling rate must be at least 2.56 × Fmax = 2560 Hz, and acquisition time is 1 ÷ 0.625 = 1.6 s.

Signatures of the most common faults

For bearings, defect frequencies depend on geometry. For the outer race, BPFO = (n ÷ 2) × f × (1 − d/D × cos φ), and for the inner race, BPFI = (n ÷ 2) × f × (1 + d/D × cos φ), where n is the number of rolling elements. With n = 9, d/D = 0.2, φ = 0 and f = 29.67 Hz: BPFO = 4.5 × 29.67 × 0.8 = 106.8 Hz and BPFI = 4.5 × 29.67 × 1.2 = 160.2 Hz. In practice, use the values from the bearing manufacturer's catalog. A quick check: BPFO + BPFI = n × f = 267.0 Hz.

Example: 1780 rpm motor and fan (fictional)

On the monthly route, the fan bearing showed overall velocity of 4.2 mm/s RMS against a baseline of 1.8 mm/s. The spectrum shows:

The ratio 2X ÷ 1X = 0.4 ÷ 4.2 ≈ 0.095, far below 0.5, and there are no harmonic series or BPFO peak. The diagnosis is rotor unbalance (for example, buildup on the blades). The action is to clean and balance in place, and confirmation comes at the next reading: 1X should return close to 1.8 mm/s.

Baseline, alarms and route

Two simple alarms work well to start, set from each point's baseline: alert at 2 × baseline (3.6 mm/s in the example) and critical at 3 × baseline (5.4 mm/s). The fan's 4.2 mm/s crosses the alert level and calls for diagnosis and a scheduled fix. These factors are an internal example criterion; for limits by machine class use ISO 20816, which defines severity zones (A to D).

• Fixed points: mark the sensor location, with the same direction and position every reading.
• Frequency: monthly for most assets; shorten it if the trend rises.
• Prioritization: start with critical assets, defined by reliability-centered maintenance (RCM).
• Records: log the reading, the diagnosis and the work order in the CMMS to follow the trend.

The result shows up in the indicators: fewer unplanned stops raise MTBF and cut MTTR, as in the article on MTBF and MTTR. Vibration analysis also helps detect the start of the wear-out phase of the bathtub curve.

Common mistakes

• Measuring without a fixed point: sensor in different positions produces a false trend.
• Looking only at the overall value without opening the spectrum.
• Insufficient resolution, which merges close peaks (for example, 1X and line frequency).
• Diagnosing without checking actual speed: with a drive, 1X moves.
• Mistaking a structural resonance for a machine fault.
• Not closing the loop: measure after the repair to confirm the peak dropped.

Frequently asked questions

What is vibration analysis used for?

To detect and diagnose mechanical faults in rotating equipment, such as unbalance, misalignment, looseness and bearing defects, before breakdown.

How do I calculate 1X?

Divide speed in rpm by 60. For 1780 rpm: 1780 ÷ 60 = 29.67 Hz.

What is the difference between unbalance and misalignment?

Unbalance is dominated by 1X in the radial direction. Misalignment shows 1X and 2X, with a significant 2X (rule of thumb: above 50% of 1X) and a high axial component.

Should I measure acceleration or velocity?

Velocity in mm/s RMS for overall severity; acceleration for bearing and gear faults, which show up at high frequencies.

How often should I measure?

Monthly in most cases. Shorten the interval when the trend is rising or the asset is critical.

Do I need ISO for alarm limits?

Not mandatory, but ISO 20816 gives severity zones by machine class. Many plants combine these zones with limits based on each point's baseline.

About the author

Vagner Soares

Lean Manufacturing & Behavioral Management Specialist

Over 20 years in the automotive and metalworking industries (GM and Dana), Lean Manufacturing practitioner since 2006. SENAI instructor and mentor in Brazil’s Brasil Mais Produtivo program, delivering consulting, training and audits for 50+ companies, combining quality, productivity and people development.

LinkedInInstagramFull profile →