Laminar Flame Speed modeling for Low Carbon Fuels using methods of Machine Learning

Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods are designed to accurately predict Laminar Flame Speed (LFS) over the entire engine operating range for Ammonia (NH3), Hydrogen (H2), and Methanol (CH3OH). These are promising zero-carbon or low-carbon alternative fuels for the transportation sector but require combustion models to optimize and control the engine performance. These developed Machine Learning (ML) methods provide an LFS prediction that requires several orders of magnitude less computation time than the original thermo-kinetic combustion mechanisms but has similar accuracy. Then an SVM and an ANN LFS model for blends of the three fuels was developed by combining LFS datasets of different fuels. Results show that for single fuels, ANN shows better performance than SVM and can predict the LFS with a correlation coefficient R2test higher than 0.999. For fuel blends, SVM has better performance with R2test close to 0.999. These predictive ML LFS models can be integrated into 0D and 1D engine models and their low computation time makes them useful for engine development and for future model-based combustion control applications.

Automated summary

Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods are used to predict the laminar flame speed (LFS) of ammonia, hydrogen, and methanol. These machine learning methods have significantly reduced computation time compared to traditional methods while maintaining similar accuracy. ANN outperforms SVM for single fuels, while SVM performs better for fuel blends.