Heat of vaporization (HoV) is one of the key physical properties of fuels that impact mixing, autoignition, and soot formation along with particulate matter index, octane number, and cetane number. Many studies have quantified the effect of HoV on combustion performance for various biofuel candidates and fuel surrogate blends, however, a discovery of new desirable fuels demands a rapid prediction of HoVs for efficient screening of fuel candidates. There are a few known methods for estimating HoVs, but they require other thermodynamic properties including vapor pressure, boiling point, critical temperature and pressure, and acentric factor. Group contribution methods also estimate HoVs as a summation of contribution values of each functional group to HoV, but a certain functional group always has a constant contribution value regardless of the neighboring atoms. Considering interactions between the adjacent atoms would lead to a more reliable prediction of HoVs. In this regard, here we will present a graph attention network (GAT) model for predicting HoVs of pure fuel compounds and fuel mixtures.<br/>For predicting HoV of pure fuel compounds, we trained our GAT model using the NIST Web Thermo Table database which contains ~100,000 HoVs at different temperatures (100–1000K) for ~8,000 compounds having C, H, O atoms. The accuracy of our model is compared with the other existing HoV estimation methods. As an application of the model, we discuss how HoV values affect the accuracy of our in-house cetane number prediction model. To develop a model for fuel mixtures, we used the vapor-liquid equilibrium (VLE) data of 746 mixtures stored in the Dortmund Data Bank (DDB) to build an HoV database for mixtures. Their HoVs were calculated by using the Clausius-Clapeyron equation and vapor pressure values of mixtures in DDB. We also used the VLE data calculated by using the COSMO-SAC model to gain more training data. Using the prediction model for the mixtures, we discuss how the non-linear increase or decrease of HoVs occurs for some mixtures, and how this non-linear behavior would affect their fuel properties.

Heat of vaporization prediction of pure fuel compounds and fuel mixtures using graph attention networks

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.