Quantitative structure-property relationships (QSPRs) using machine learning tools to relate molecular structural features of pure hydrocarbons to a range of physical and chemical properties is of interest. Mathematical modelling of these relationships can be very complex, which makes the deep learning networks (DNNs) ideal candidates for the QSPR development. Here we show QSPR development for a chemical property related to the sooting tendency of a fuel, which is directly linked to high and undesirable particulate matter emissions to the atmosphere, and therefore is recognized as a key parameter to be controlled in combustion process. After developing the QSPR model, generative adversarial networks (GANs) are used to generate new chemical species for which the YSI is predicted. Finally we show how this approach can be used to predict molecular properties and design fuels.
Dr. Sarathy is currently Associate Director of the Clean Combustion Research Center at KAUST and Sr. Manager Technology & Innovation at NEOM Hydrogen and Green Fuels. He is also working with the Saudi Ministry of Energy to develop a Circular Carbon Economy National Program.
Dr. Sarathy employs engineering modeling tools to determine the net environmental, economic, and social impact of fuel engineering technologies and processes. Specific tools include engineering process modeling, life-cycle assessment, cost/benefits analysis, energy auditing, materials accounting, and environmental assessment to examine current engineering issues. A primary focus is using models to study greenhouse gas emissions from renewable energy systems, unconventional fuels, gas-to-liquid fuels, and biofuels.