Fuel property modeling by high-speed gas chromatography coupled with partial least squares data analysis

Area of Science:

• Chemometrics
• Analytical Chemistry
• Petroleum Science

Background:

• Partial least squares (PLS) regression is a key chemometric technique for property prediction.
• Gas chromatography (GC) is crucial for separating complex fuel mixtures.
• Effective PLS modeling depends on GC separation time and stationary phase selection.

Purpose of the Study:

• To investigate the impact of GC run time and stationary phase selection on PLS property prediction for aerospace fuels.
• To evaluate the robustness of PLS models for viscosity, density, and hydrogen content.
• To explore data smoothing techniques for enhancing fast GC separations.

Main Methods:

• Utilized partial least squares (PLS) regression coupled with gas chromatography (GC).
• Analyzed 50 aerospace fuel samples using both polar and non-polar stationary phase columns.
• Varied GC separation time windows (1-min to 10-min) and applied data smoothing.
• Assessed model performance using normalized root mean square error of cross-validation (NRMSECV).

Main Results:

• Fast GC separations (1-min TW) yielded robust PLS models for viscosity prediction with both column types.
• Polar stationary phase columns provided superior PLS models for density and hydrogen content prediction compared to non-polar columns.
• Data smoothing allowed shorter GC run times to achieve comparable model performance to longer, unsmoothed separations.

Conclusions:

• Fast GC separations are viable for robust PLS property prediction, especially for viscosity.
• Polar stationary phase columns are recommended for accurate density and hydrogen content prediction in aerospace fuels.
• Optimizing GC parameters, including stationary phase and run time, is essential for effective chemometric property modeling.