Impact of blending alcohols, ethers and ketones with TPRF mixtures on smoke point: an experimental and neural network analysis

The combustion of conventional hydrocarbon fuels adversely affects air quality through soot emission. However, quantitative smoke point (SP) data for oxygenated surrogate fuels remain limited despite their importance for soot mitigation. This study presents 89 new laminar SP measurements for toluene primary reference fuel (TPRF 80–20) blends, covering eleven oxygenates including alcohols, ethers, and ketones tested at concentrations of 2,4,6,8,10,12 and 14 vol%, along with one measurement for pure ethanol. All oxygenates increased SP, with 14 % dibutyl ether achieving the highest SP value of 35.5 mm. Ether blends produced the most significant SP enhancements, while branched alcohols exhibited more complex trends influenced by oxygen content and volatility. To generalize these trends, four machine learning models were developed using molecular descriptors: support vector regression (SVR) achieved the highest accuracy with an R² of 0.99 and RMSE of 0.54 mm; Nu-SVR closely followed with an R² of 0.98 and RMSE of 0.55 mm; extremely randomized trees (ETR) attained an R² of 0.97 and RMSE of 0.62 mm; and artificial neural networks (ANN) yielded an R² of 0.96 and RMSE of 2.04 mm using a simple train/test split, improving to an R² of 0.95 and RMSE of 1.17 mm with 7-fold cross-validation. The superior performance of the 7-fold ANN over the simple split and the overall best accuracy of SVR highlights the influence of dataset size on model generalization. This combined experimental and computational framework provides a robust tool for predicting soot formation and guiding the design of cleaner combustion fuels.