Contribution
In our study, my primary contribution was developing and implementing machine learning models, particularly Artificial Neural Networks (ANNs), to predict the Derived Cetane Number (DCN) of oxygenated fuels. My role was instrumental in processing the data to accurately reflect the diverse chemical compositions of these fuels, a vital step for the ANNs to effectively learn and predict DCN values. Additionally, I introduced innovative techniques for model optimization, including a methodology for tuning hyperparameters and the use of a genetic algorithm, which significantly enhanced the robustness and accuracy of the final models.
Beyond the technical scope, I significantly contributed to data visualization and the communication of our findings in the final manuscript. My responsibility in creating clear and insightful visual representations greatly aided in the interpretation and presentation of our complex results. Moreover, my active role in writing the manuscript ensured that the intricate details of our models, encompassing the advanced methodologies for hyperparameter tuning and genetic algorithm optimization, were communicated with precision and depth. These efforts were integral to enhancing the clarity, comprehensiveness, and overall impact of our publication.
The developed code in addition to the results can be found here.
Abstract
Artificial intelligence-based computing systems like artificial neural networks (ANN) have recently found increasing applications in predicting complex chemical phenomena like combustion properties. The present work deals with the development of an ANN model that can predict the derived cetane number (DCN) of oxygenated fuels containing alcohol and ether functionalities. Experimental DCNs of 499 fuels comprised of 116 pure compounds, 222 pure compound blends, and 159 real fuel blends were used as the dataset for model development. DCN measurements of sixty new fuels were carried out in the present work, and the data for the rest were collected from the literature. Fuel chemical composition expressed in the form of eight functional groups, namely, paraffinic CH3 groups, paraffinic CH2 groups, paraffinic CH groups, olefinic -CH=CH2 groups, naphthenic CH-CH2 groups, aromatic C-CH groups, alcoholic OH groups, and ether O groups, along with two structural parameters, namely, molecular weight and branching index (BI), were used as the ten input features of the model. The qualitative and quantitative determination of functional groups present in real fuels was performed using 1H nuclear magnetic resonance (NMR) spectroscopy. A robust ANN methodology was then applied to prevent overfitting, using a multilevel grid search and genetic algorithm. The final developed model with two hidden layers was tested with 15% of randomly generated unseen points from the dataset, and a regression coefficient (R2) of 0.992 was observed between the experimental and predicted DCN values. An average absolute error of 0.91 obtained from the test set indicates that the developed ANN model is successful in predicting the DCN of oxygenated fuels and captures the dependence of the fuel’s ignition quality (i.e., DCN) on its constituent functional groups.