(255b) Application of Machine Learning Algorithms for the Selection of Jet Fuel from Hydrocarbon Blends

Authors: 
Mukherjee, R., Gas and Fuels Research Center, Texas A&M Engineering Experiment Station
Mohamed, N., Texas A&M Qatar
El Wahsh, M., Texas A&M University at Qatar
Elbashir, N., Texas A&M University at Qatar
El-Halwagi, M. M., Texas A&M University
Application of Machine Learning Algorithms for the Selection of Jet Fuel from Hydrocarbon Blends

Rajib Mukherjee1, Noof Abdalla 2, Nasr Mohamed 2, Marwan El Wash2 , Nimir O. Elbashir2, Mahmoud El-Halwagi11. Gas and Fuels Research Center, Texas A&M Engineering Experiment Station, College Station, TX 77843, USA

2. Department of Chemical Engineering, Texas A&M University, Qatar, Education City, Doha
Abstract

Blending is an essential part in transportation fuel preparation. In this paper application of machine learning algorithms will be presented for accurate prediction of the properties synthetic fuel composed of blend of hydrocarbons obtained from gas to liquid (GTL) technology. Statistical measures have been used to ensure that the blended fuel meets American Standard for Testing Materials (ASTM) D1655 standard. The analysis is performed with GTL fuel building blocks: normal paraffins, iso paraffins as well as cyclo paraffins and aromatics. Four different properties of the blend are monitored including density, freezing point, flash point, and heat of combustion for the comparison of synthetic fuel with aviation grade fuel standard. Limited experimental data on estimated property as obtained from different composition across a broad range of different components of the synthetic blends is used for our analysis. Multivariate regression with partial least squares (PLS) regression method for linear as well as support vector machine (SVM) for nonlinear properties have been used. A multivariate statistical tool has also been developed using Principal Component Analysis (PCA) to investigate any new composition for its quality. This machine learning algorithms based blend selection can identify optimal composition for any number of components. This is followed by blend property characterization using already developed regression models to confirm the qualification of the selected blends as aviation grade fuel. Experimental verifications are also conducted on the selected blends for model validation.

Topics: