Most people encounter NVH as a term before they understand what it actually means. It shows up in automotive press releases, product spec sheets, engineering job descriptions. It sounds technical enough that people nod along without asking the obvious question.
So let's try to understand it together.
NVH stands for Noise, Vibration, and Harshness. Three things that are related but not the same, and the difference between them matters when you're trying to fix a problem rather than just describe one.
Noise
Noise is airborne sound. A machine runs, its surfaces vibrate, those vibrations disturb the air around them, and that disturbance travels outward until it reaches your ears. Simple enough in principle.
What makes it complicated in practice is frequency. Sound doesn't just have volume, it has pitch, and pitch is determined by frequency, measured in Hertz. A low frequency sound, say 60 to 80 Hz, is the deep, chest-felt rumble of a large diesel engine or a heavy compressor. A high frequency sound, say 3000 to 5000 Hz, is the sharp whine of a fast-spinning motor or a gear mesh. Both can measure identically loud on a decibel meter. They feel nothing alike, they travel through buildings differently, they're absorbed by different materials, and they need completely different solutions to control. This is why engineers don't just measure "how loud" something is. They plot the sound across the full frequency spectrum to see where exactly the energy is concentrated, because that's the map you need before you can do anything useful about it.
The human ear also doesn't treat all frequencies equally. It's most sensitive to sounds between roughly 1000 and 4000 Hz, which is roughly the range of human speech. This is why noise measurements are often expressed in dB(A), a weighted scale that adjusts the raw numbers to reflect how the ear actually responds, rather than what a completely neutral microphone would record.
Vibration
Where noise travels through air, vibration travels through solid material. It starts the same way, something moves, surfaces oscillate, but instead of disturbing the air, that energy moves through the structure itself.
Through a chassis. Through a mounting bracket. Through a floor panel. Which is exactly why you can sometimes feel a problem in your feet or through a steering wheel before you hear anything at all in the cabin.
Engineers measure vibration using accelerometers, small sensors bolted or glued directly onto surfaces, which capture the motion of that surface in precise detail. Amplitude tells you how much it's moving. Frequency tells you how fast. What engineers are really looking for is something called resonance.
Every structure, every panel, every housing, every bracket, has what are called natural frequencies. These are frequencies at which the structure "wants" to vibrate, determined by its shape, its mass, and how stiff it is. Under normal conditions, this is harmless. The problem comes when an external source, an engine running at a certain RPM, a motor spinning at a certain speed, happens to coincide with one of those natural frequencies. When that happens, the structure doesn't just vibrate in response. It amplifies the input, sometimes dramatically, turning a small, manageable excitation into a large, sustained shake that wasn't in anyone's original calculations.
Finding and understanding those natural frequencies before they cause problems in a real product is most of what structural NVH work is about. Engineers also look at mode shapes, essentially a visualisation of how a structure deforms at each of its natural frequencies, which tells them not just that a resonance exists but exactly where on the structure the motion is happening and which direction it's moving in
Harshness
This is where NVH stops being purely a measurement problem and becomes a human perception problem, and that shift changes everything about how you approach it.
Harshness describes how uncomfortable, crude, or unpleasant a noise or vibration feels to the person experiencing it. Not how loud it is. Not how much displacement the accelerometer recorded. How it feels.
Consider two sounds at the same measured volume. One is a smooth, steady hum at a consistent frequency, the kind a well-maintained electric motor makes. The other is an irregular rattle that varies slightly in pitch and rhythm, the kind a loose panel makes when it's not quite secured. A meter treats them as equivalent. A person sitting next to them does not. The hum fades into the background within minutes. The rattle becomes more noticeable with time, not less, because the brain keeps being surprised by its irregularity and can't fully tune it out.
This happens because human perception is sensitive to unpredictability. We are reasonably good at ignoring consistent, steady inputs after a while. We are not good at ignoring things that keep changing slightly, because change is what the nervous system is wired to pay attention to. NVH engineers have to account for this, which is why the field can't rely entirely on instruments.
Jury evaluations are the main tool for capturing harshness. A panel of trained listeners, sometimes also untrained ones, to get a range of responses, sits in a controlled environment and rates recorded or live sounds and vibrations against a set of descriptors. Solidity. Smoothness. Irritation. Refinement. Booming. Harsh. Each descriptor gets a score, and those scores get cross-referenced against the actual measured acoustic and vibration data to build a picture of which physical characteristics drive which perceptual responses.
Psychoacoustic modelling attempts to do this mathematically, predicting how a human listener would rate a sound based on its frequency content, its temporal pattern, its sharpness, its roughness. These models are genuinely useful, particularly in early development when you don't have a physical product to test yet. But they get validated against real listener panels before anyone puts serious weight on them, because the gap between a model's prediction and an actual person's reaction is still meaningful enough to matter.
Two of NVH's three components give you hard numbers to work with. The third gives you human beings, which are considerably harder to optimise against, and considerably more important to get right.
How testing actually works
Component level testing usually happens on shaker tables. A subassembly gets subjected to a controlled sweep of vibration frequencies, and engineers watch where it misbehaves, where it resonates, where it rattles, where it transmits energy in unexpected directions.
Full system testing happens in semi-anechoic chambers. These are rooms built to absorb reflected sound, so that measurements capture only what the vehicle or product itself produces, not what bounces back off the walls. Microphones map the airborne acoustic field. Accelerometers map the structural response. Laser vibrometers can scan entire surfaces without contact, building a high resolution picture of where vibration energy concentrates across the structure.
On-road or proving ground testing adds the layer that lab environments can't fully replicate: real world inputs, road surfaces, temperature variation, aerodynamic noise at speed. The combination of lab and real world data is what gives engineers enough confidence to call something refined.
The electric vehicle problem
For most of automotive history, engine noise masked a lot of smaller problems. Gear mesh noise, minor rattles, accessory motor whine.... all of it sat underneath the dominant sound of combustion and effectively disappeared from perception.
Electric powertrains removed that masking. The smaller sounds didn't go away. They just became audible, sometimes for the first time in a product that had been in production for years with a combustion variant.
Wind noise at the A-pillar, tyre roar transmitting through the suspension, the hum of a cooling pump, all of it moved from background detail to foreground problem the moment the engine stopped covering for it.
This is why NVH engineering on EV programmes is more demanding, not less, despite the powertrain being mechanically simpler. The quieter a product gets overall, the more precisely every remaining sound has to be managed. There is no longer anything to hide behind.
Why timing matters more than most people realise
An NVH problem caught at the design stage costs relatively little to fix. The same problem found after tooling is locked costs significantly more, sometimes by an order of magnitude.
And the reality is that you often cannot properly measure NVH until a physical prototype exists to test.
Simulation has narrowed this gap considerably. Finite element models can predict structural resonances before anything is built. Acoustic simulation can estimate how sound will propagate through a space.
But simulation results still get validated against physical prototypes, because the gap between a model and a real object, with real manufacturing variation and real material behaviour, remains meaningful.
The practical consequence is that NVH has to be considered earlier in development than it feels necessary, before the problems are visible, precisely because fixing them later is so expensive.
