The Role of Chemical Functional Groups in Fuel Properties Prediction and Design of Surrogates

Hydrocarbon-based fuels have served as dependable energy sources for decades, boasting qualities like high energy density, safe storage capabilities, cost-effectiveness, and reliability. This study conducts a comprehensive examination of Chemical Functional Groups (CFGs), both as predictive descriptors and for formulating fuel surrogates. The research systematically explores the potential of machine learning models to predict Derived Cetane Number (DCN) based on CFG compositions. Multiple machine learning models are developed and trained using a dataset that includes UNIFAC CFG compositions. The results underscore the promise of CFGs in DCN prediction, with the Neural Network model achieving remarkable accuracy. Furthermore, the study explores an advanced semi-supervised learning approach aimed at improving DCN prediction accuracy. By integrating synthetic mixtures with optimized CFG compositions, the model exhibits an enhancements in performance. These outcomes highlight the method's robustness and its ability to generalize to novel data by covering regions not represented in the real dataset. The research underscores the significance of specific CFGs in influencing DCN values, emphasizing their role in analyzing fuel ignition behavior. In the subsequent section of the study, an innovative surrogate formulation methodology, known as the Chemical Functional Groups Optimizer (CFGO), is introduced. This data-driven technique aligns the CFG compositions of surrogates with those of their parent fuels. CFGO's efficacy is demonstrated through the formulation of surrogates for various fuels, aligning with the CFGs present in the parent fuels, thereby showcasing its potential for modeling complex fuel compositions. A comparison between CFGO-generated surrogates and existing literature surrogates illustrates the pivotal role of mirroring CFGs in accurately representing fuel behavior.