You can put two exhausts on the same bike and they'll sound completely different. Both legal, both running fine, totally different acoustic signature. Most exhaust design is still craft and intuition, sketch a chamber, weld it up, listen, adjust, repeat. A custom muffler can take twenty iterations before it sounds right, and "right" is whatever the rider says it is.
I think there's a better way, so I'm building one.
The project
I'm building an open-source motorcycle exhaust simulation framework. It takes an engine spec and an exhaust geometry and predicts what comes out of the tailpipe, pressure waveform, frequency spectrum, perceived loudness, the lot. Eventually it will let you design an exhaust before you build it, hear how it'll sound, and quote real engineering numbers for what changes when you stretch a chamber by a centimetre.
Three things matter to me:
It has to be honest. Every number the model produces gets compared against either another model, a published reference, or eventually a recording. If the model can't reproduce something, I say so. No magic, no hidden tuning knobs.
It has to be open. The framework, the data, the validation cases, the equations, all public. The thing you pay for is the engineering that uses it, not the framework itself.
It has to actually work. The end state isn't a paper. It's a tool that an exhaust shop, a small OEM, or a rider with a Panigale and too much time on their hands can pick up and use.
Where it is today
The basic forward model runs end-to-end on real bike geometries. Engine firing pattern in, pipe acoustics applied, outlet pressure out, peak SPL, RMS, A-weighted dBA, the spectrum, the lot. Two independent implementations of the same physics cross-validate to within 3 % RMS, which is the strongest L1 validation you can do without recordings on a bench.
Today the model can tell you why two cylinders firing 90° apart sound different from two firing 180° apart, why a chamber three centimetres longer plugs a different acoustic null, and roughly how loud each design lands on a dBA meter. Not "vaguely louder" numbers.
Where it's going
The rough plan is three years. Fidelity climbs steadily as the work compounds: better boundary conditions, mean flow, perforated mufflers, hot-exhaust thermal effects, eventually a real 1D combustion-driven source. Calibration against real bike recordings lands about a year in. Sound synthesis from pure physics, playable audio with no recording behind it, lands about two years in.
I'll be posting Field Notes as each piece lands. Some will be technical (here's the math, here's the validation), some lighter (here's why an Audi R8 V10 sounds like an Audi R8 V10). All of them stay honest about what works, what doesn't, and what's still on the to-do list.
If you want to follow along, the next post walks through what an exhaust forward model actually does, source pulse, pipe filter, outlet pressure, on the simplest possible case. After that we start adding chambers and mufflers and things get interesting.
→ Next: The fidelity ladder, where we are and where we're going