Researchers Amirali Shateri, Zhiyin Yang, and Jianfei Xie from the University of New South Wales have published a study in the journal Combustion and Flame that explores how adding ethanol and methanol to ammonia-methane fuel mixtures can reduce harmful nitrogen oxide (NOx) emissions. Their work combines molecular dynamics simulations with machine learning to better understand and predict the chemical processes involved in combustion.
Ammonia-methane blends are being considered as a potential hydrogen-carrier energy source, but their use is hindered by high NOx emissions. To address this, the researchers designed ten fuel mixtures and evaluated the effects of adding ethanol and methanol at concentrations of 0%, 5%, and 10%. They conducted ReaxFF molecular dynamics simulations at temperatures of 2,000 K and 3,000 K to study the formation pathways of NOx.
The findings indicate that adding alcohol can significantly suppress NOx production, particularly at higher temperatures. At 3,000 K, a 10% addition of ethanol reduced NOx production by approximately 39.6%, while a 10% addition of methanol reduced it by about 30.1%, compared to the base fuel. This suppression is due to the redistribution of charges and the redirection of nitrogen intermediates through stabilizing pathways such as HNO, HNO2, and N2O.
To predict NOx emissions more efficiently, the researchers trained machine learning models using descriptors derived from the simulations. The models included Random Forest Regression (RFR), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Fully Connected Neural Networks (FCNN). RFR performed best, achieving an R2 score of 0.993 and a mean absolute error (MAE) of 0.661. These models successfully predicted NOx emissions for both simulated and non-simulated alcohol ratios, demonstrating the potential of hybrid physics-informed machine learning algorithms for extrapolating complex chemical behaviors.
The study highlights the promise of using ReaxFF-informed machine learning frameworks to design cleaner fuels and establish reliable predictive models for combustion chemistry. This research could aid the energy sector in developing more efficient and environmentally friendly combustion processes, particularly for ammonia-methane fuel blends.
The research was published in the journal Combustion and Flame.
This article is based on research available at arXiv.