Kurtosis in Vibration Analysis for Fault Detection

Devices

Portable balancer & Vibration analyzer Balanset-1A

€1,975.00 + VAT (if applicable)

Parts

Vibration sensor

€90.00 + VAT (if applicable)

Parts

Optical Sensor (Laser Tachometer)

€124.00 + VAT (if applicable)

Devices

Balanset-4

€6,803.00 + VAT (if applicable)

Parts

Magnetic Stand Insize-60-kgf

€46.00 + VAT (if applicable)

Parts

Reflective tape

€10.00 + VAT (if applicable)

Devices

Dynamic balancer “Balanset-1A” OEM

€1,735.00 + VAT (if applicable)

Kurtosis is a statistical parameter that describes the shape of a probability distribution. In vibration analysis it is applied to the time waveform to measure its “peakedness” or impulsiveness. A signal with high kurtosis is characterised by sharp, distinct peaks or impacts, while a signal with low kurtosis is flatter and more rounded. The great practical value of kurtosis is that it condenses this character into a single number that can flag a change in the nature of a vibration signal even when the overall energy — the RMS value — has not yet moved significantly.

1. The Statistic Behind the Number

Kurtosis is the normalised fourth statistical moment of the signal’s amplitude distribution. Because each deviation from the mean is raised to the fourth power before averaging, occasional large excursions — exactly the kind a brief impact produces — dominate the result far more than they would in an RMS calculation, which uses only the second power. That mathematical emphasis on extremes is what makes kurtosis so alert to short, sharp transients buried in an otherwise modest signal. In effect, it answers a different question from RMS: not “how much energy is present?” but “how spiky is it?”

2. The Diagnostic Value of Kurtosis

The primary value of kurtosis in condition monitoring is its high sensitivity to early-stage, impulsive faults. Many mechanical defects begin as microscopic cracks or spalls. As these developing faults make contact, they generate short-duration, high-frequency impacts. Those impacts cause sharp peaks in the vibration time waveform that drive the kurtosis value up dramatically — long before the fault has grown large enough to affect the machine’s overall RMS level.

Kurtosis is therefore an excellent tool for:

• Early bearing fault detection: it is one of the most effective methods for catching the very first signs of spalling on a bearing race or rolling element, and complements envelope analysis in bearing diagnostics.
• Gear-tooth fault detection: a cracked or broken tooth produces a distinct impact once per revolution, which a rising kurtosis value detects readily — a useful cross-check on gear defects.
• Detecting intermittent rubs or impacts: any non-uniform, impacting event within a machine, such as rubbing or mechanical looseness, is highlighted by this measurement.

3. Interpreting Kurtosis Values

Kurtosis is a normalised value. For a perfect Gaussian (normal) distribution — typical of the random background vibration in a healthy machine — the kurtosis value is 3.0. Deviations from this figure are diagnostically significant:

• Kurtosis ≈ 3.0: the vibration is random and Gaussian, suggesting normal, healthy operation.
• Kurtosis > 3.0: the signal is becoming more peaky or impulsive than normal. A rising value is a clear warning of developing impacts; values of 5, 10, or higher are common when significant bearing or gear defects are present.
• Kurtosis < 3.0: the signal is flatter than a normal distribution. This can occur with certain types of rubbing, or when the signal is dominated by a very clean, sinusoidal vibration such as pure unbalance.

It is worth noting that some instruments report excess kurtosis, which subtracts 3.0 so that a healthy Gaussian signal reads 0 rather than 3.0. The interpretation is identical; only the reference point shifts. Kurtosis is closely related to the crest factor, which compares peak to RMS and responds to the same impulsive events from a slightly different angle.

4. The Kurtosis Lifecycle of a Bearing Fault

When tracking a bearing fault from inception to failure, the kurtosis value often follows a predictable — and at first counter-intuitive — pattern:

1. Healthy stage: kurtosis is stable and close to 3.0.
2. Early fault stage: a microscopic defect forms. Sharp, distinct impacts are generated, causing the kurtosis to rise significantly (for example to 5.0 or higher). The overall RMS vibration may still be low. This is the ideal time to detect the fault.
3. Developed fault stage: as the defect grows and spreads, the impacting becomes more frequent and less distinct. The signal starts to resemble random noise again, though at a much higher energy level. Consequently the kurtosis value may decrease back towards 3.0, even as the RMS level begins to climb dramatically.
4. Late / failure stage: the bearing is heavily damaged and the vibration is high and largely random. Kurtosis sits near 3.0, but the RMS value is now in alarm.

This lifecycle is exactly why kurtosis is so valuable. The “sweet spot” for detection is the early stage when kurtosis rises; relying on RMS alone would leave the fault undetected until it had already become significant. The non-monotonic behaviour is also a caution: a kurtosis reading that has fallen back to 3.0 does not, on its own, prove the bearing is healthy — it must be read alongside the RMS trend.

5. Measuring Kurtosis in the Field

Kurtosis is computed directly from a clean, well-sampled time waveform, which makes it a natural companion to the work an engineer is already doing on site. A portable analyser such as the Balanset-1A captures the vibration time waveform and spectrum from a machine running in its own bearings, so when a confirmed unbalance is corrected by field balancing the same dataset can be screened for the impulsive bearing or gear signatures that kurtosis brings to the surface. Trending that value across periodic measurements turns a one-off reading into an early-warning indicator.

6. Limitations

Powerful as it is, kurtosis should be used alongside other techniques such as spectrum and waveform analysis. It can be sensitive to random, non-machine-related shocks, so it is best treated as a trending parameter rather than an absolute threshold. A consistent rise in kurtosis over time is a far more reliable indicator than a single high reading, and confirming the source in the spectrum prevents a stray bump from being mistaken for a developing fault.

← Back to Main Index