Pass or Fail in Seconds: How End-of-Line Noise and Vibration Testing Works

Every production line has that one inspector who can hear a bad unit. A faint click in a driveshaft, a whine in a motor, a rattle in a pump — they spin it up, listen for a few seconds, and wave it through or pull it off. The problem is not their skill. The problem is that their ear does not scale: it is not repeatable across shifts, it cannot be audited when a warranty claim comes back, and it cannot judge every unit when the line needs a verdict in seconds, unit after unit, all shift long.

End-of-line NVH testing writes that ear down as numbers. The unit mounts on a fixed test fixture, a sensor picks up its vibration signature, a DAQ digitises it, and software compares the spectrum against a stored limit — pass if it stays under, fail if it breaks through. In the TIERA stack, that chain is an accelerometer or microphone into a PhonoVibe DAQ, judged by the NDT RAM spectral-alarm module in TVIB. Same unit, same verdict, every time — and every verdict is logged.

The tempting shortcut is a single overall vibration level with one threshold: below the line ships, above the line fails. It rarely works. A narrowband defect — a bearing tone, a gear whine, a joint click — concentrates its energy in a thin slice of the spectrum. Riding on top of normal broadband energy, it can barely move the overall number while being obvious to any trained ear.

The spectrum is where faults live at known addresses. An unbalanced rotor raises the amplitude at 1x rotation speed. A bearing defect shows up at its characteristic BPFI, BPFO and BSF frequencies with sidebands. A driveshaft with an inner joint defect produces elevated amplitude around 6x to 8x rotation frequency, depending on ball count. Judge those addresses individually and the defect has nowhere to hide.

This is exactly what the NDT RAM module in TVIB does: it sets spectral alarm bands on a per-component basis. The spectrum is divided into frequency bands placed over the addresses that matter for that product — 1x for unbalance, the joint-defect region for a shaft, bearing frequencies for a motor. Each band carries its own upper threshold, and together they form a mask over the spectrum. The unit passes only if every band stays under its limit; a single band exceeded means fail. The comparison runs band by band and returns the verdict in seconds.

Watch the figure below. The known-good unit hugs the floor, comfortably under the mask everywhere. The failing unit matches it almost everywhere too — except in one watched band, where a defect tone climbs straight through the limit. That one band is the whole verdict, and the mask is the inspector's ear, written down as numbers anyone can audit.

Where do the threshold numbers come from? Not from a datasheet and not from intuition — there is no standards body that hands down a pass/fail number for a specific product's NVH signature. A common working method is to measure a sample of known-good units (often 20 to 30) on the same fixture, let the software compute the spectral envelope of that population, and then place each band's limit a deliberate margin above the population's mean level in that band — typically 3 to 6 dB, which is the reference both figures below use. Save that as the template, and every future unit is judged against what good actually looks like on this line.

The population matters because good units vary: tolerances, batches and assembly all spread the spectrum. A limit set from one golden sample will fail normal product; a limit guessed generously will pass defects. The template is only as trustworthy as the measurement chain behind it — which is why a fixed fixture and mounting, low-noise cabling, the PhonoVibe's 24-bit resolution and its factory calibration certificate matter: they keep the measurement consistent across shifts and across test stands, so the limit means the same thing on Monday night as it did when it was set.

A limit can be wrong in two directions, and both cost money. Set it too loose and defective units escape: warranty claims, field failures, and the slow damage of a customer learning to distrust the product. Set it too tight and the station fails good product: scrap and rework climb, the line slows, and operators who watch known-good units fail will start to override the station — at which point the test protects nothing.

So the margin is a business dial as much as an engineering one, and it should be tuned with evidence rather than argument. This is where logged data earns its keep: when a unit fails, the full waveform and spectrum are saved automatically — the rework team sees exactly which band tripped and why, disputed false-fails can be audited instead of debated, and the template can be re-derived as the good-unit population drifts over months of production.

A representative line, not a specific customer: a driveshaft station spins each shaft at a fixed test speed on a dedicated fixture, with an IEPE accelerometer on a stud mount feeding a 4-channel PhonoVibe. The NDT RAM template was built from a batch of known-good shafts, with alarm bands over 1x rotation for unbalance and over the 6x to 8x region where an inner joint defect raises the amplitude. A shaft with a joint defect trips that band before it can reach a vehicle, the verdict lands in seconds, and the saved spectrum goes to the rework bench with the shaft. The same pattern applies to electric motor QA and pump impeller inspection — only the fixture, speed and band placement change.

When a failure traces back to airborne noise rather than a mechanical defect, the fix often lives in the product's acoustic materials — and that work happens in the lab, where the T-Sonus impedance tube characterises absorption coefficient and transmission loss of liners, foams and enclosure panels from 50 Hz to 10 kHz under ISO 10534-2 and ASTM methods. If the ideas here are new, TIERA's free primers at 101.tieraonline.in — Acoustics & Noise 101 and Measurement Setup 101 — cover the theory at no cost (they are primers, not accredited certifications). For teams who need formal, assessed training, the TCAT programme on our services page runs structured courses with proctored exams at exams.tieraonline.in.

The station described above is a stock TIERA configuration, not a custom build. A PhonoVibe DAQ is the front end — 24-bit, simultaneous sampling, IEPE/ICP/CCLD sensor power, and a factory calibration certificate so the limit means the same thing on every shift and every test stand. TVIB's NDT RAM module does the judging: measure 20–30 known-good units on your fixture, let it compute the spectral envelope, place your band limits above it, save the template. From then on every unit gets a band-by-band verdict in seconds, with the waveform and spectrum logged on every fail.

If you'd rather not work out fixtures, band placement and margins alone, TIERA runs on-site engineering visits — we can measure your good-unit population with you and help commission the template on your line. And to be clear about one thing: T-Sonus, our impedance tube, is a lab instrument for characterising acoustic materials — absorption and transmission loss of the liners and panels you reach for when a failure traces to airborne noise. It belongs next to the line's problems, not on the line itself.