FELA · 2D FNO
Instant Thermal What If?
Setup
Returns the steady state temperature field of various battery packs.
Pros
~7 MB one-time download · 0 bytes per solve
More Pros
No meshing, no solver setup, no wait · Fast BMS iteration, no workstation.
Here, a compact 2D Fourier Neural Operator reads a battery pack layout, a simulated heat load factor,
and coolant temperatures. It then returns a visualization of the steady state temperature field.
It was trained on families of similar battery pack cross sections, so it
stands in for a "finite element thermal solver" on those designs... but way faster!
Who might like this?
Battery pack and EV thermal engineers. The process of designing great chemical storage technology
means building a mesh and waiting minutes for a finite element or CFD solve
(ANSYS, COMSOL, Fluent) every time you nudge a cell's spacing, or coolant temperature.
This model drops that slow solve loop like a hot potato and returns a good estimate of the field rapidly,
so you can sweep pack layouts and cooling choices interactively
and visually see the hot spots move. Explore -> then confirm the
final design in your full solver of choice.
... How accurate can that BE?
This is a compact browser build. The field pattern is
about 2% relative error against the full model, so the map of where
heat concentrates is appreciably visually very similar. The peak temperature readout is
approximate, about 15% on hot, extreme, cases here, versus about 3% for
our full model that runs server side. Use the browser build for fast
"what if" explorations and the server model when you need the exact peak. Probably still beats waiting on COMSOL!
Demo Limits
This is a thermal surrogate trained on a few families of battery pack
geometries. It handles a few different pack layouts as given but what is pictured
is not general across arbitrary geometries. Heat sinks, enclosures, and motors
can be similarly treated with the same technique and are on the roadmap, not a current capability.
Full card on Hugging Face ↗