Congrats to group members Mohammed and Yeonjoon for their work on the prediction of bio-oil output composition being published in ACS Sustainable Chemistry and Engineering!
In collaboration with NREL, we developed a preliminary simple ML model (random forest) capable of translating a mass spectra (MS) to a set of chemically-interpretable compositional descriptors, the Paraffins, Isoparaffins, Olefins, Naphthenes, and Aromatics (PIONA) fractions. The key spectral features were extracted from the MS, and these features are used as the input of the random forest model which outputs the predicted PIONA values. These studies are expected to provide generally applicable strategies to predict bio-oil output compositions.
For more information:
Our group has introduced a “Fuel property first” design approach to reduce emissions and increase performance. Traditional approaches for developing these mechanistic models require many years for each new molecule, a pace that is poorly suited to the large-scale search for new bioderived blendstocks. We have developed a quantitative structure-property relationship (QSPR) model for sooting tendency based on the experimental yield sooting index (YSI), developed by collaborators at Yale (Prof. L. Pfefferle and Dr. C. McEnally). This is the first fuel property predictive tool using ML (Machine Learning) approaches in combustion research. We have started to build kinetic mechanisms of soot precursor formation during combustion using DFT and flow reactor experiments (collaboration with Dr. R. McCormick, NREL) to show how the fundamental chemistry affects this practical engineering problem..