High frequency and kurtosis: why one number isn't enough
Why RMS velocity amplitude misses early bearing faults
The most common way to evaluate machine vibration is RMS velocity amplitude compared against ISO 10816-style zones. It's an excellent measure of overall low-frequency severity: imbalance, misalignment and looseness live there and that figure catches them well. But it has a blind spot: a bearing that starts to fail generates high-frequency impacts (thousands of Hz), and at those frequencies velocity amplitude is tiny by physics —velocity comes from dividing acceleration by frequency, and a large frequency "kills" the value. The result: a bearing can be entering failure while the velocity number stays green. That's why a dashboard watching only RMS velocity warns you late, when the machine already vibrates, already runs hot and is nearly audible. To anticipate, you must look at the high-frequency acceleration band and the shape of the signal.
What kurtosis is (and why it matters)
Kurtosis measures how "peaky" a signal is versus a normal bell curve. A healthy bearing vibrates randomly and its kurtosis sits around 3. When a defect appears, the signal fills with sharp, repetitive impacts and kurtosis rises (typically 5, 10 or more in the high-frequency band). The valuable part is that kurtosis is independent of amplitude and RPM: it detects that the shape of the signal changed, even while overall vibration level stays low. That's why it complements RMS velocity instead of repeating it: velocity tells you how much it vibrates, kurtosis tells you how it vibrates. In practice, a sustained rise in high-band kurtosis is one of the earliest signs of damage in bearings, gears or cavitation.
Crest factor and why you never watch a single indicator
Crest factor (peak divided by RMS) is another useful shape indicator, but it has a known trap: it rises early when the first defect appears and then falls again as the damage spreads and the signal becomes more random. Someone watching only crest factor may believe the problem "resolved itself" precisely when the bearing is worse. That's why serious diagnostics never rely on a single number: you read together the RMS velocity (low frequency), the high-frequency acceleration, the kurtosis and the crest factor. Each covers the other's blind spot, and their combined reading —plus their trend over time on the same asset— is what reveals the equipment's real state. Wertek AI collects and trends these indicators per measurement point, not just the velocity headline.
The three axes: what fails, not just that something fails
A triaxial vibration sensor measures in three directions: two radial and one axial. That direction is diagnostic. Dominant radial vibration points to imbalance or looseness; high axial vibration points to misalignment or a bent shaft. A single-axis sensor can't distinguish those cases because it lacks the axial channel. By combining direction (which axis dominates) with frequency band (low for running speed, high for bearings), you answer two questions at once: what kind of fault and how advanced it is. Wertek AI lands the three directions and every indicator the sensor produces —from any compatible brand— in one place, tied to the asset, to interpret them together.
Honest scope: detection and trend vs. fine spectral diagnosis
It's important to be clear about what each tool delivers. A fixed vibration/temperature sensor delivers band-processed values (RMS velocity, high-frequency acceleration, kurtosis, crest factor). With that you achieve early detection and trending: "this bearing is degrading, with weeks of lead time", which is most of the value for planning maintenance. Identifying the exact defect frequency (BPFO, BPFI, BSF, FTF for outer race, inner race, rolling element or cage) requires capturing the full waveform with a route instrument or the app's formal capture flow. The recommended strategy: let the fixed sensor detect and anticipate continuously, and confirm the specific defect with a formal measurement when screening shows high severity or before a costly decision. See How diagnostics works for the capture flow.