WeChat account 弧光星云 — First published on the WeChat quality-engineering column 弧光星云; republished here with attribution. Reviewed and formatted by 8D.wiki before publication.
Reviewed & edited by: Alvin Jin — IATF 16949 expert · Certified Product Engineering (PE) coach · DRBFM expert
Published 2026-11-19 · Last reviewed 2026-10-25
The customer said "remove the acoustic package". What he actually wanted was 3 dB less cabin noise, and what the engineers eventually had to change was a rad...
The customer asked for one thing. The engineering team heard a second. What finally shipped was a third — and everyone involved worked extremely hard to deliver it. This is the most expensive failure mode in product development, and it never shows up in a test report.
A supplier of electric drive units was asked by its customer to delete the acoustic package from the bill of materials. The reasoning was simple and, on the surface, beyond argument: the acoustic package is real cost — material, weight, assembly time, tooling. Remove it, and the BOM goes down by a number purchasing can put on a slide.
That is the customer's direct voice. Every programme manager has heard a sentence like it, and the fastest way to look responsive is to agree on the spot and open the engineering change.
The engineering team did not agree on the spot.
Acoustic package. Absorptive and barrier materials bonded to the body shell — think of it as a soundproof jacket for the car. It genuinely makes the cabin quieter, and it is genuinely heavy and expensive. It sits on the vehicle because of one function: suppressing cabin noise.
Decibel. The unit of sound level, on a logarithmic scale. A 3 dB reduction is half the sound power — one clear step that a passenger notices without being told to listen for it. That is why "2 to 3 decibels" is not a rounding error in a specification. It is the whole deliverable.
E-axle. The drive unit of an electric vehicle: the traction motor, the reduction gearbox that brings motor speed down to wheel speed, and the inverter that controls the motor. The whine a passenger hears in an EV mostly comes out of those three assemblies.
With that in place, the story reads properly.
The team started from the function rather than the action. The acoustic package exists to suppress cabin noise. Take it out, and cabin noise has to fall by roughly 5 dB just to keep today's comfort level — and cross-checking the industry roadmaps, no competitor was in a position to deliver 5 dB inside the foreseeable two years.
So "remove the package" is not a requirement. It is a prescription the customer wrote for himself. The disease is "cheaper and quieter at the same time".
Read the same measurement the other way round and the requirement appears: a 2 dB reduction puts the vehicle at the good end of the industry benchmark, and 3 dB or more puts it ahead of every competitor. The thing the customer is actually buying is a number, not an action.
Customer voice: "cancel the acoustic package." Translated requirement: "3 dB less cabin noise at unchanged comfort, with the package gone."
This is the first translation — from a prescription to a performance target. Teams that skip it execute the action, cancel the package, fail to reach the noise level, and are then told the result is not what the customer wanted. Of course it is not. Nobody was buying the action.
"3 dB at the vehicle" is still not something anyone can work on. The e-axle has three major assemblies, so the target has to be split along the noise paths that actually dominate.
Contribution analysis comes first: the inverter contributes very little to what the passenger hears. The dominant paths are the electromagnetic noise of the motor and the gear mesh noise of the reducer. So the 3 dB is decomposed along those two paths.
| Path | Contribution to cabin noise | Share of the 3 dB target |
|---|---|---|
| Inverter | small | — (it shapes the excitation, not the level) |
| Motor, electromagnetic | dominant | 2 dB |
| Reducer, gear mesh | dominant | 2 dB |
Two independent 2 dB improvements is how a 3 dB vehicle-level improvement is actually reached: the paths do not add linearly in decibels, and the split deliberately keeps margin so that the vehicle target survives the case where neither path lands perfectly at the same operating point.
Notice what this step produces: a target per subsystem, with an owner and a test bench. Before it, the whole programme was arguing about one number nobody could influence directly.
Still not designable. Motor noise and gear noise are outputs. No engineer can put "noise" on a drawing.
Motor electromagnetic noise is driven mainly by radial force — the one-sided electromagnetic pull between stator and rotor. The stronger the pull, the more the stator deforms, and the more noise it radiates. Reducer noise is driven mainly by transmission error — the deviation between the actual and the ideal rotation angle as the gears mesh, which is where the "zzz" whine comes from.
Both of those are parameters a design engineer can set. The chain therefore ends in two numbers:
The request that began as "take the acoustic package out" now fits into a CAE model and onto a bench test. Whether −10 % is reachable is a separate feasibility question — and gear macro-geometry may well have to change to get there — but at least the question is now well posed, which it was not at any of the earlier layers.
| Layer | What it says | Who owns it |
|---|---|---|
| Customer voice | "cancel the acoustic package" | customer |
| Performance target | −3 dB cabin noise, package removed | vehicle attribute owner |
| System decomposition | motor −2 dB, reducer −2 dB | system engineering |
| Design parameters | radial force −10 %, transmission error −10 % | component design |
Three ways this chain breaks, and all three are common:
Customer satisfaction rarely fails at the last step. It fails at the first: the sentence the customer actually said was never translated into something the organisation can design.
Translation is not a document. It is three questions asked in order — what result does this action buy, which part of the system produces that result, and which parameter on a drawing controls that part. Answer all three and delivery becomes ordinary engineering. Skip one, and the harder everybody works, the further the result drifts from what the customer wanted.
First published on the WeChat account 弧光星云.
Turn this method into a submittable 8D report
Generate a D0–D8 report skeleton with a 5-Why chain and containment actions.
8D.wiki — Quality Engineering Knowledge Base