Understanding Differentiation in Vibration Analysis
Differentiation in vibration analysis is the mathematical operation that converts a vibration signal from one measurement parameter to another by taking its time derivative — or, equivalently, by multiplying by frequency in the frequency domain. It turns displacement into velocity, and velocity into acceleration. Differentiation is the exact inverse of integration; it is performed far less often, because most field sensors are accelerometers and the usual need is to integrate down to velocity or displacement, not differentiate up. The case where it earns its keep is when displacement measured by a proximity probe must be compared against a velocity-based standard, or examined for high-frequency content.
The key behaviour to internalise is that differentiation is a frequency-weighting operation: it emphasises high-frequency components and suppresses low-frequency ones — precisely the opposite of integration. That makes it useful for pulling faint high-frequency diagnostic detail out of a displacement record, but it is a double-edged tool, because it amplifies high-frequency noise just as enthusiastically as signal. Used without care, it can bury the very information you were trying to reveal.
1. The Mathematical Relationships
The same physics can be expressed in two equivalent ways, and the choice between them has real practical consequences.
Time-domain differentiation
- Velocity from displacement: v(t) = d/dt [x(t)]
- Acceleration from velocity: a(t) = d/dt [v(t)]
- Acceleration from displacement: a(t) = d²/dt² [x(t)] — the second derivative, applied in one step
Frequency-domain differentiation
In the frequency domain the operation collapses into a simple multiplication, which is why modern instruments work here:
- Velocity from displacement: V(f) = D(f) × 2πf
- Acceleration from velocity: A(f) = V(f) × 2πf
- Net effect: every spectral line is scaled by its own frequency, so high frequencies are lifted and low frequencies pushed down — and double differentiation scales by (2πf)², an even steeper tilt.
This frequency dependence is the whole story of differentiation. Because each conversion multiplies one power of frequency, it links the family of parameters an engineer routinely switches between; converters such as a vibration acceleration calculator or a vibration displacement calculator apply exactly this single-frequency relationship for a pure tone.
2. Why Differentiation Is Used
Despite being the less common operation, differentiation has several legitimate uses:
- Proximity-probe applications: proximity probes measure shaft displacement directly, yet many vibration standards specify velocity limits. Differentiating displacement to velocity lets a displacement sensor be judged against those limits.
- Emphasising high frequencies: because differentiation lifts the top end, it can expose high-frequency defect signatures hidden in displacement data, and convert sluggish low-speed displacement into a more analysis-friendly acceleration record.
- Cross-comparing sensor types: to compare a displacement sensor with an accelerometer, both are converted to a common parameter — usually velocity — so their measurements can be checked for consistency.
3. The Challenges: Noise Amplification
The defining difficulty of differentiation is noise, and it follows directly from the multiply-by-frequency rule.
Why noise dominates
Because the operation multiplies by frequency, broadband noise — which lives across the whole spectrum — is amplified more at the top than the signal of interest is. A vivid illustration: 1 % noise at 10 kHz is amplified roughly 100× relative to a signal at 100 Hz, so a clean-looking input can emerge swamped. The defence is to apply a low-pass filter before differentiating, removing high-frequency content that would otherwise be blown up.
Sensor noise and double differentiation
Every displacement sensor carries its own electrical and quantisation noise. Single differentiation to velocity amplifies it; double differentiation all the way to acceleration compounds the effect dramatically and should generally be avoided. If you genuinely need acceleration, the right answer is almost always to measure it directly with an accelerometer rather than differentiate displacement twice.
Numerical errors
Time-domain differentiation also amplifies digitisation errors and is sensitive to sampling artefacts, which is the practical reason the frequency-domain method is preferred wherever accuracy matters.
4. Doing It Properly
A disciplined procedure keeps differentiation honest. Note the contrast with integration, which instead needs a high-pass filter to remove low-frequency drift — the two operations require opposite filtering strategies.
Single differentiation (displacement → velocity)
- Low-pass filter first: remove high-frequency noise, with a cut-off roughly 2–5× the highest frequency of interest.
- Check signal quality: confirm the input is free of obvious noise and artefacts.
- Differentiate: multiply by 2πf in the frequency domain.
- Sanity-check the result: compare against expected magnitudes for reasonableness.
Double differentiation (displacement → acceleration)
- Generally avoid it — it rarely produces good results.
- If unavoidable, apply aggressive low-pass filtering with the cut-off set right at the highest frequency of interest, and accept that the high-frequency band will be noise-limited.
- Better alternative: use an accelerometer and measure acceleration directly.
Frequency-domain implementation
The modern, robust recipe is to compute the FFT of the displacement or velocity signal, multiply each bin by 2πf (or (2πf)² for double differentiation), apply any low-pass filtering in the frequency domain, and read off the spectrum in the new parameter — taking an inverse FFT if a time waveform is wanted. This approach avoids cumulative errors, makes filtering trivial, is computationally efficient, and is the standard method built into today’s analysers.
5. When to Use It — and When Not To
Reach for differentiation when converting proximity-probe displacement to velocity for ISO comparison, when enhancing high-frequency content in low-speed displacement data, when comparing different sensor types on a common basis, and generally whenever proper filtering can be applied. Avoid it on noisy displacement signals, avoid double differentiation unless it is truly unavoidable, and — the recurring theme — avoid it entirely whenever an accelerometer is available, since measuring the desired parameter directly always beats deriving it.
6. Differentiation vs Integration, and Modern Instruments
The two operations are mirror images, and seeing them side by side clarifies both.
| Aspect | Integration | Differentiation |
|---|---|---|
| Frequency effect | Amplifies low frequencies | Amplifies high frequencies |
| Common use | Acceleration → velocity, velocity → displacement | Displacement → velocity |
| Main problem | Low-frequency drift | High-frequency noise amplification |
| Required filter | High-pass before integration | Low-pass before differentiation |
| How often used | Very common | Less common |
In practice the engineer rarely performs these conversions by hand. Modern analysers convert automatically between displacement, velocity and acceleration: the user selects the desired parameter and the instrument applies the correct filtering and scaling, which greatly reduces the chance of error. Many can display all three parameters at once — each emphasising a different part of the frequency range — to give a comprehensive view of the vibration. A portable two-channel instrument such as the Балансет-1А handles this conversion internally, presenting velocity for routine assessment against severity bands like those in ISO 20816-1 while retaining the underlying acceleration data, so the analyst never has to differentiate a raw record manually in the field.
Differentiation, then, is the less-used but genuinely valuable counterpart to integration: indispensable for converting displacement measurements to velocity or acceleration and for cross-checking sensor types, provided its noise-amplifying character is respected and the right low-pass filtering is applied. Understand that single trait — it lifts the high frequencies — and accurate parameter conversion follows.
