Speaker
Description
Hydrogen micro-mixing burners require homogeneous fuel-oxidizer mixing because the local hydrogen distribution affects flame stability, flashback susceptibility,and emission behavior. Additive manufacturing enables compact burner geometries with small, inclined fuel passages and tailored injection arrangements. However, this geometric freedom creates a design space that cannot be efficiently explored using computational fluid dynamics alone. This work therefore presents a design-to-flame-plane surrogate workflow for an additively enabled hydrogen burner parameterized by fuel-inlet height, inlet diameter, number of injection arms, and injection angle.
A database of 190 geometries was generated by Latin hypercube sampling and evaluated using converged, non reacting, compressible Reynolds-averaged Navier-Stokes simulations in OpenFOAM. A design-conditioned DoMINO neural operator predicts the steady three-dimensional flow and hydrogen fields directly from geometric information. Because the full-field model smooths the low-amplitude downstream hydrogen fluctuations governing mixing quality, a second geometry-aware transformer based on GeoTransolver, denoted GALE, operates on eleven axial slabs. Using slab-normalized relative hydrogen fluctuations and the interface state predicted by DoMINO, GALE reconstructs the hydrogen distribution within the flame-plane band and predicts its spatial coefficient of variation.
For 18 unseen geometries, DoMINO achieved a mean relative root-mean-square error of 2.1% for the hydrogen field. The complete surrogate chain predicted the flame-plane coefficient of variation with a mean absolute percentage error of 11.3%. Inference required approximately 10 s on a single GPU, compared with approximately 4 h on 48 CPU cores for one CFD simulation. The proposed workflow therefore enables rapid, field-resolved screening and provides a foundation for future optimization of additively manufactured hydrogen burners.