What is acoustic simulation?

Acoustic simulation computes sound: pressure fluctuations propagating through a fluid, and the structures and flows that create them. The output is sound pressure level at points or over surfaces, resonant frequencies, and how much noise a treatment removes.

The organising question is where does the noise come from, because the answer decides which of three quite different analyses you need. Sound can be generated by a vibrating structure, by the flow itself, or it can simply be resonating in a cavity that was excited by something else.

Most acoustic analysis is done in the frequency domain. Rather than marching through time, you assume everything oscillates at a single frequency and solve for amplitude and phase — which reduces the wave equation to the Helmholtz equation, a much cheaper problem. You then repeat across the frequencies of interest.

The mesh requirement follows directly and is the thing that catches people out: you need enough elements per wavelength, conventionally six to ten. Wavelength shrinks as frequency rises, so doubling your top frequency of interest roughly doubles the mesh resolution needed in every direction. Acoustic models are cheap at 100 Hz and punishing at 10 kHz, and that scaling — not solver speed — is what limits how high you can go.

• Cavity and duct modes. An enclosed volume of air has natural frequencies just as a structure does. A modal acoustic analysis finds them with no excitation at all, and it explains why a particular boom appears at one engine speed or why a room has a dead spot. Cheap, and usually the first thing to run.
• Vibro-acoustics. A vibrating surface pushes on the air and radiates sound. This needs a structural answer first — modal or harmonic FEA — and then an acoustic solve driven by that surface motion. This is gearbox whine, panel drumming, motor noise and most machinery sound.
• Aeroacoustics. Flow generates sound on its own: turbulence, vortex shedding, jets, and flow over an edge or cavity. Getting at it needs a CFD solution first, because the acoustic sources are computed from the unsteady flow field. This is wind noise, fan noise and HVAC duct noise.

Non-reflecting boundaries. Sound radiating outdoors should leave and never come back. A plain boundary reflects it, producing resonances that are entirely numerical. Absorbing or infinite-element treatments exist for exactly this, and forgetting one is a common way to get a confidently wrong spectrum.

Impedance and absorption. Real surfaces absorb, and how much they absorb varies strongly with frequency. Acoustic foam is nearly transparent at low frequency and effective at high frequency, which is why an absorptive treatment can do nothing at all for the boom somebody actually complained about.

Damping. At resonance, the response is set entirely by damping, and damping is the least well-known property in most models. Undamped or lightly damped assumptions give peaks far sharper and taller than reality.

Sound pressure level is quoted in decibels, which is logarithmic — a 6 dB reduction is halving the pressure amplitude, and a 3 dB change is roughly the smallest most people notice. That scale is worth keeping in mind before celebrating a result: a change that looks dramatic on a linear pressure plot may be inaudible, and vice versa.

Perceived loudness is also frequency-dependent, which is what A-weighting corrects for. A model can be right about physics and useless about annoyance if it reports unweighted levels at frequencies nobody can hear well.

As always, comparison is where simulation is strongest: this silencer against that one, this rib pattern against a flat panel. Absolute sound pressure prediction is a much harder claim.

Elmer covers Helmholtz problems for duct and cavity modes and vibro-acoustics, with its own elasticity solver on the structural side. code_aster handles structural acoustics where the structure is the hard part. OpenFOAM provides the unsteady flow solution that aeroacoustic source terms are computed from.