Surrogate-Based Optimization of Nozzle Geometry to Prevent Flame Flashback in Premixed Hydrogen Burners

Hydrogen combustion presents key challenges, particularly the risk of flame flashback at low flow rates in premixed systems. This study focuses on optimizing the nozzle geometry of a pilot-scale hydrogen burner using a combined CFD and machine learning approach. A large set of randomized axisymmetric nozzle internal geometries was generated to introduce ample shape variability. The minimum diameter at the throat of each geometry was kept constant to keep the controlling section of the nozzle fixed among the different designs. Steady-state CFD simulations were performed to identify the critical flow rate for flashback onset. Flashback detection was based on upstream temperature monitoring in the CFD simulations.

The database of CFD results was used to identify correlations between geometric features and flame stability. The surrogate model was then employed to explore a reduced-dimensional nozzle design space and guide an optimization algorithm toward flashback-resistant nozzle profiles. Future work will focus on experimental validation of the optimized nozzle profiles and further refinement of the machine learning model to incorporate a wider range of operating conditions.