Physical kernel
- carbon diffusion in austenite
- temperature and steel dependence
- profile and case-depth calculation
Core model
∂C/∂t = ∂/∂x [D(C,T,steel) · ∂C/∂x]
-D · ∂C/∂x = JC(T, p, gas pulse, surface state, load history)
-D · ∂C/∂x = JC(T, p, gas pulse, surface state, load history)
What physics keeps / what AI assists
| Physics keeps | AI assists |
|---|---|
| diffusion equation and carbon transport | effective boundary carbon flux during boost stages |
| temperature dependence and steel response | furnace/load correction from calibration data |
| carbon profile and case-depth calculation | recipe calibration and transfer between load cases |
| validation against measured profiles | uncertainty estimate within the trained domain |
AI closure
- effective boundary carbon flux
- boost-stage behaviour
- self-calibration from measured profiles
Calibration data
- carbon profile
- surface carbon
- mass gain
- hardness / effective case depth
Use
- recipe comparison
- supplier review
- troubleshooting
- process transfer