DOI: 10.1016/j.fuel.2022.123941
Contribution
In this project, my role as an advisor was crucial in the development of the artificial neural networks (ANNs) for predicting the threshold sooting index (TSI) of fuels. My background in computer science and data science was vital in ensuring the accuracy and reliability of the ANN model. This task required a meticulous approach to testing and validating the software, which was essential for proper and efficient data processing, model training, and prediction of outcomes.
In addition to this, I also played a substantial role in the writing and review of the paper, particilarly in conveying the technical details of the research. I worked on refining the structure and flow of the paper, ensuring it effectively communicated the significant findings and methodologies to our readers.
Abstract
In this work, a machine learning based model using artificial neural networks (ANN) was developed for the prediction of threshold sooting index (TSI) of fuels containing oxygenated chemical classes like alcohols and ethers, along with hydrocarbon classes such as paraffins, olefins, naphthenes, aromatics, and their mixtures. Experimental TSI data of 342 fuels including 124 pure compounds, 212 fuel surrogate mixtures and 6 gasolines was used as a dataset for developing the model. Ten features (eight functional groups, molecular weight (MW) and branching index (BI)) have been used as inputs in this model. The eight functional groups and the two structural parameters (MW and BI) represent the composition and structure of the fuel. The ANN model was trained, validated, and finally tested on randomly split sets of 70%, 15%, and 15% of the data, respectively. The observed regression coefficient (R2) between the real and predicted TSI values was 0.97 as obtained for the test set. The absolute error of prediction obtained was 2.46, which is promising as this number is closed to the uncertainty observed in experimental measurements. The results indicate that a fuel’s TSI is dependent on the fuel functional groups, and thus can be used as modeling criteria. The model can be applied towards the prediction of TSI of pure compounds, fuel surrogate mixtures and petroleum fuels containing alcohols and ethers.