Data-Driven Modeling of Laminar Flame Speed in Ammonia–Oxygenated Fuel Blends Using Machine Learning

Ammonia (NH3) and hydrogen (H2) are emerging as pivotal carbon-neutral energy carriers, offering promising pathways for reducing carbon emissions in power generation and transportation. Despite its advantages, ammonia exhibits unfavorable combustion characteristics, which can be significantly improved through blending with oxygenated fuels and hydrocarbons to enhance combustion kinetics. This study focuses on characterizing the laminar flame speed (LFS) of ammonia-based fuel blends using advanced data-driven modeling techniques. A feed-forward artificial neural network (FFANN) with back-propagation was developed to predict the LFS of premixed ammonia–air mixtures containing various oxygenated additives. The model was trained using an extensive data set comprising 8580 chemically simulated data points generated via chemical kinetic simulations, covering a wide range of operating conditions. In addition to ANN, several machine learning (ML) models including generalized linear regression (GLR), support vector machines (SVM), decision trees (DT), random forests (RF), extreme gradient boosting (XGBoost), stacking regressor (SR), and voting regressor (VR) were implemented and systematically evaluated. Hyperparameters were optimized using grid-search cross-validation, and all models were developed in Python using the Keras API. Among the evaluated models, the ANN exhibited superior predictive performance, achieving near-perfect accuracy on the independent test set (15% of the total simulation data set), with an R2 of 1.0, RMSE of 0.27, and MAE of 0.20. This performance demonstrates the ANN’s capability as a highly accurate surrogate for computationally expensive chemical kinetic simulations. Ensemble-based approaches, particularly XGBoost, stacking, and voting regressors, also yielded competitive accuracy following optimization. Furthermore, a generalized laminar flame speed correlation was developed by extending Gülder’s formulation to encompass ammonia–air mixtures blended with oxygenated fuels. The proposed correlation incorporates key governing parameters, including pressure, temperature, equivalence ratio, and molar blend fraction, achieving an R2 exceeding 99% against simulation data set. This confirms the robustness and generalizability of the proposed formulation. Overall, the results highlight the effectiveness of data-driven methodologies for rapid and reliable combustion characterization, supporting the development of ammonia-fueled sustainable energy systems.