A prototype on a durability loop comes back with a complaint logged by the driver. A tonal whine appears on coast down between 55 and 65 km/h. It is faint, high pitched and clearly audible in a quiet EV cabin.

The NVH engineer now needs 3 answers: the exact vehicle build, the calibration running on the inverter and whether anything similar has shown up before.

In most programmes those answers live in 3 different places. The build sheet sits with the prototype shop. Calibration versions belong to the controls team. Previous findings are scattered across slide decks and email threads. Before any analysis begins, someone spends a day reconstructing context.

The fix starts here, at the last stop. Every subjective complaint and every objective finding should open as an issue tied to a named test article, its build and the test that found it. Logged that way, an issue record becomes the thread you can pull all the way back to the rig.

The car goes onto a chassis dynamometer inside a hemi anechoic chamber. Engineers place microphones at driver and passenger ear positions and accelerometers on the motor mounts, subframe and steering column. They run the coast down again and capture the order content.

The whine lines up with a gear mesh order of the e-axle. That narrows the source. It does not yet explain why the full vehicle result misses the NVH target that the e-axle passed on its own rig months earlier.

Full vehicle NVH tests are expensive. Chamber time is scarce, prototype vehicles are shared with crash, thermal and durability teams, and each prototype carries a build state that changes week to week. A chamber booking that ignores the vehicle's current build is how teams end up measuring a car that has had its mounts swapped since Tuesday.

Transfer path analysis separates the problem into source, path and receiver. Here the source is the e-axle, the paths run through the mounts, subframe and body structure and the receiver is the driver's ear.

TPA needs data from both ends of the chain. It needs blocked forces or operating loads from the subsystem test and frequency response functions from the vehicle body. If those two datasets came from different hardware revisions, the analysis will look tidy and still be wrong.

This is the stop where campaign management and NVH analysis meet. The engineer has to prove that the subsystem data and the vehicle data describe hardware that can actually be compared. That proof is a configuration record, and most labs don't have one in a form the analyst can trust.

Months earlier, the e-axle ran on a dynamometer inside a semi-anechoic chamber. It passed. Radiated noise at the target orders came in under the cascaded subsystem target with margin to spare.

Pull the test record and the first problem appears. The record says "e-axle prototype B2." The prototype shop says B2 existed in 2 variants, one with revised gear micro geometry introduced after a durability finding. Nobody wrote down which variant went on the dyno.

At the bottom of the chain sit the component tests. Modal testing of the motor housing. Mount stiffness characterization. Gear transmission error measured on a single stage rig. For interior parts, BSR testing runs on a shaker table under temperature cycling, chasing the buzz, squeak and rattle that EV cabins now expose.

Here is where the whine began. The gear set measured on the transmission error rig carried the revised micro geometry. The set fitted to the first prototype vehicles did not. A component result was used to sign off a subsystem, and the subsystem result was used to predict the vehicle, on hardware that never shared a configuration. A specialist UAV validation lab lost 6 weeks to this exact failure when an undocumented modification forced a propulsion retest. Once every airframe variant, firmware version and payload specification was recorded against its test article in TITAN, configuration discrepancies at the lab dropped to zero.

NVH is more sensitive to configuration than almost any other test discipline. A 5 Shore change in mount durometer, a firmware update to the inverter switching frequency or a revised gear tooth profile can shift a tonal peak enough to fail a target. Without configuration identity on every test, correlation from rig to vehicle is guesswork.

Walking the noise backwards shows exactly where an NVH test campaign needs structure. Run in the right direction, from rig to road, a well-managed campaign looks like this.

Cascade the targets before booking a single rig. A vehicle level target for interior noise at a given speed breaks down into subsystem targets for radiated noise and mount transmissibility, which break down into component targets for transmission error and housing modes. Tie each target to the test that verifies it inside your design verification plan, so a failure at any level points straight to the requirement it threatens.

Give every test article an identity. Part number, revision, variant, software version and build date go on the record before the test starts. A test without them should not be bookable.

Share prototypes across teams in one calendar. NVH competes with crash, thermal and durability for the same handful of vehicles. A global OEM with 3 engineering teams now books samples, equipment and users in a single action while each team keeps its own workspace, and its programme managers can commit capacity months ahead.

Keep one issue thread from complaint to root cause. The whine logged at Stop 1 should link to the chamber test, the TPA run, the dyno record and the rig result without anyone rebuilding the trail by hand.

The whine in our example came from gear physics, and its root cause traced back to a missing line on a test record.

NVH teams will keep getting better at hearing small sounds, because EV cabins demand it. The teams that ship quiet vehicles on time keep rig, dyno and chamber results inside one vehicle testing record, where a configuration change shows up long before the microphones catch it.