📊

Vibration Spectrum Comparator

Capture a baseline spectrum of a running machine, then later capture a current reading with the same microphone and gain. The tool overlays the two 1/3-octave spectra, color-highlights the frequency bands where the level rose, and lists them largest-first so you can match them against your calculated bearing, gear-mesh, imbalance or blade-pass frequencies. Save named baselines to your browser to trend a machine across sessions, watch specific fault frequencies over time, and export an annotated comparison report.

This microphone hears AIRBORNE SOUND, not accelerometer vibration. It reveals the frequency content of the sound your machine radiates — genuinely useful for spotting which tones grew — but it is not a calibrated vibration measurement: the absolute level is uncalibrated and is not in mm/s velocity or g acceleration, and this is not an ISO 10816 / ISO 2372 severity assessment (those need a calibrated accelerometer). The trustworthy part is the difference between two captures: with the same mic and gain the unknown calibration offset cancels, so “this band rose +9 dB” is reliable. For real vibration measurement use an accelerometer + analyzer. Auto-gain and noise suppression are requested off. Nothing is recorded or uploaded.

Idle — press Start, allow the microphone, then capture a Baseline and a Current reading.
Microphone idle.

Keep the microphone in the same spot, at the same distance, and don’t touch the input gain between captures — that’s what makes the comparison valid. The microphone is connected to the analyzer only; it is never played back.

Overlaid 1/3-octave spectrum (25 Hz–16 kHz) — Baseline vs Current; pink bands rose ≥ 6 dB
No captures yet. Press Start, then capture a Baseline and a Current reading.
Comparison of baseline and current capture metrics
Metric Baseline Current
Overall level
Peak sample
Dominant bands
Clips

Capture both a Baseline and a Current reading to see which frequency bands increased.

Saved baselines — trend a machine across sessions

Save your current Baseline capture under a name (e.g. a machine ID), then on a later day load it back and capture a fresh Current reading to see what changed. Baselines are stored only in this browser’s local storage — nothing is uploaded.

Watch frequencies — track specific fault tones

Add the exact frequencies you care about — a calculated bearing BPFO/BPFI, gear-mesh frequency, the 1× imbalance (shaft) frequency or a fan blade-pass frequency. The tool reads each one’s level from the nearest 1/3-octave band in your Baseline and Current captures so you can watch it rise or fall. Watch frequencies persist in this browser.

How It Works

The comparator takes two short snapshots of your microphone’s spectrum — a Baseline (the machine in a known-good state) and a Current reading (today’s condition). For each capture it averages a high-resolution FFT over the whole window and sums the energy into 1/3-octave bands from 25 Hz to 16 kHz. It then overlays the two band spectra, draws a pink stripe over every band whose Current level rose by 6 dB or more, and ranks those risen bands largest-first — because a band that grows between captures is exactly where a developing fault tends to show up.

Why the before/after difference is trustworthy even without calibration

A browser microphone has an unknown sensitivity, and it is positioned some distance from the machine, so every absolute number it reports is in dBFS (decibels relative to digital full scale) carrying a fixed but unknown offset. The key insight is the same one behind any good before/after test: when Baseline and Current are captured with the same microphone, in the same position, at the same gain, that unknown offset is identical in both, so it cancels exactly when you subtract Current − Baseline. A band that reads +9 dB really did get about 9 dB louder. That is why the difference and the change in spectral shape are meaningful even though neither absolute level is calibrated.

Matching risen tones to mechanical faults

Rotating machines radiate sound at characteristic frequencies: the shaft turning frequency (, RPM ÷ 60) and its harmonics for imbalance and misalignment; bearing defect frequencies (BPFO, BPFI, BSF, FTF) from the bearing geometry; the gear-mesh frequency (teeth × shaft Hz) and its sidebands; and the fan/pump blade-pass frequency (blades × RPM ÷ 60). Calculate the ones that apply to your machine using the related tools below, add them as watch frequencies, and see whether their bands grew between captures. A rise at a calculated fault frequency is a useful early warning — but the order-to-fault and band-to-defect mappings are diagnostic conventions, not a guaranteed diagnosis, and you must verify your own inputs.

What stays uncalibrated

The single-capture numbers (“Baseline overall is −46 dBFS”) are not a vibration amplitude. They are not in mm/s velocity or g acceleration, and they cannot be used for ISO 10816 / ISO 2372 severity zones — that requires a calibrated accelerometer mounted on the machine and a proper analyzer. Consumer microphones also roll off at the frequency extremes and generally cannot pick up true low-frequency vibration that radiates little airborne sound, so a quiet low-speed shaft tone may simply not be audible. Treat this tool as a fast, free screening and trending aid that points you at frequencies worth a closer look with the right instrument.

Frequently Asked Questions

Does this measure real vibration like an accelerometer?
No. It uses your device microphone, which captures the airborne sound a machine radiates, not the structural vibration an accelerometer would measure. It reveals the frequency content of that sound, which is genuinely useful for matching against calculated fault frequencies, but it is not a calibrated vibration measurement. It does not give amplitude in mm/s velocity or g acceleration, and it cannot be used for ISO 10816 or ISO 2372 severity. For a true vibration measurement, use an accelerometer and a vibration analyzer.
If the mic is uncalibrated, why is the comparison reliable?
Because both captures share the same unknown calibration offset. When you capture the Baseline and the Current reading with the same microphone, in the same position, at the same gain, that offset is identical in both, so it cancels exactly when you subtract Current minus Baseline. The result is that a band reading "+9 dB" really did rise by about 9 dB, and the change in spectral shape is real, even though neither absolute level is a calibrated figure. The whole tool is built around trusting the difference, not the absolute numbers.
How do I match a risen band to a specific fault?
Calculate the frequencies that apply to your machine — the shaft 1× frequency (RPM ÷ 60) for imbalance, bearing defect frequencies (BPFO, BPFI, BSF, FTF) from the bearing geometry, the gear-mesh frequency (teeth × shaft Hz) and its sidebands, or the blade-pass frequency (blades × RPM ÷ 60). Add those as watch frequencies, then see if their bands grew. A rise at a calculated fault frequency is a useful early warning. But order-to-fault and band-to-defect mappings are diagnostic conventions, not a guaranteed diagnosis, and you must verify your own inputs (correct RPM, bearing geometry from the datasheet, tooth counts).
Where are my saved baselines and watch frequencies stored?
Entirely in your own browser's local storage on this device — nothing is uploaded or sent anywhere. Saved baselines let you load a known-good spectrum on a later day and capture a fresh Current reading to trend a machine over time; watch frequencies persist so you can keep tracking the same fault tones. Clearing your browser data, or using a different browser or device, will remove them. The exported report is a plain-text file downloaded locally.
What makes a fair before/after capture?
Run the machine in the same operating condition (same load and speed) for both captures, keep the microphone in exactly the same spot and distance, and do not touch the input gain or move other noise sources in or out of the room between them. Use the same capture length. The entire validity of the comparison depends on everything except the machine's condition staying identical, so the difference reflects only what changed in the machine — not a different mic position or a passing truck.
Is any audio recorded or uploaded?
No. The microphone signal is analyzed in real time to compute spectrum magnitudes only — it is never recorded, saved, or transmitted, and the mic is connected solely to an analyzer node, never to the speakers. Auto-gain, noise suppression and echo cancellation are requested off so the readings are not altered. The microphone is released when you press Stop or close the tab.