Akademik Veri Yönetim Sistemi
11th Thermal and Fluids Engineering Conference, TFEC 2026, Arizona, Amerika Birleşik Devletleri, 9 - 12 Mart 2026, ss.823-826, (Tam Metin Bildiri)
Combustion simulations with detailed chemical kinetics are computationally expensive because chemical substeps dominate the computational time. To address this challenge, we develop a neural network model for hydrogen/air combustion that predicts species mass fraction changes and is integrated into a custom solver. This allows the model to replace conventional chemistry computations in reactive flow simulations while maintaining accuracy. The training dataset is derived from one dimensional premixed flame calculations spanning a range of fuel–air ratios and temperatures. To improve the generalization of the model, the dataset is populated with perturbed thermochemical states. The neural network model takes temperature and species mass fractions as inputs and predicts mass fraction changes over a small timestep. A priori validations confirm accurate predictions for species mass fractions, while a posteriori validations reproduce the flame structure with good agreement. In addition, the model significantly reduces the computational cost, demonstrating the suitability of integrating data driven combustion chemistry models into computational fluid dynamics solvers.