Machine learning-enhanced combustion modeling for predicting laminar burning velocity of ammonia-hydrogen mixtures using improved reaction mechanisms

• Published in: Energy (Oxford) Vol. 320; p. 135259
• Main Authors: Rao, Anas, Li, Wei, Abbasi, Muhammad Salman, Shahid, Muhammad Ihsan, Farhan, Muhammad, Zulfiqar, Sana, Chen, Tianhao, Ma, Fanhua, Li, Xin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2025
• Subjects: ammonia; Ammonia/hydrogen fuel; combustion; energy; fuels; hydrogen; Laminar burning speed; Machine learning algorithm; nitric oxide; prediction; PSO-SVM; Reaction kinetics; stoichiometry; support vector machines; Ammonia/hydrogen fuel; PSO-SVM; Laminar burning speed; Reaction kinetics; Machine learning algorithm
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2025.135259

The ammonia/hydrogen mixture is a promising zero-carbon fuel for internal combustion engines, with optimal efficiency in spark ignition engines at equivalence ratios of 0.6–1 and hydrogen fractions of 0.4–0.6. However, existing reaction mechanisms lose accuracy in this range, limiting combustion modeling. To address this, prediction accuracy is improved using refined reaction kinetics and machine learning algorithms. GRI Mech3.0 is refined by enhancing H/O, N2O, HNO, NH, and NH2 mechanisms, forming Model I for pure ammonia and Model II for lean burn ammonia-hydrogen mixtures. The simulated laminar burning speed of these models is compared with literature and seven other mechanisms. Combustion analysis includes sensitivity analysis of coefficients, nitric oxide emission rates, sub-mechanisms, and reaction sensitivities to assess their impact, with machine learning algorithms to improve accuracy. Model II achieves the highest accuracy under stoichiometric (RMSE = 1.4590 cm/s) and lean burn conditions (RMSE = 1.3701 cm/s). As different mechanisms suit various conditions, machine learning algorithms further enhance prediction accuracy. The support vector machine with particle swarm optimization improves computational accuracy by 2.9745 times over reaction kinetics, demonstrating its effectiveness. This study develops refined reaction mechanisms and machine learning models for practical applications in ammonia-hydrogen fueled SI engines. [Display omitted] •Ammonia/hydrogen kinetics has been improved corresponding to the best thermal conditions.•It has been upgraded with the addition of H/O, N2O, HNO, NH & NH2 species.•It has been compared with other mechanisms and machine learning algorithms.•PSO-SVM has the overall best computational accuracy.