Text Size Default Text SizeDefault Text Size Large Text SizeLarge Text Size Largest Text SizeLargest Text Size Print Print this Page

Fall 2010

Comparison of Fuels Using GC-FID and Principal Component Analysis

Stephen M. Adams, Jr., Faculty Advisor Dr. Sarah E.G. Porter

Steven Adams 200px height


The gasoline used in cars and airplanes is assigned an octane rating by a knock test.  A test engine is run and the number of knocks is counted relative to hexane and iso-octane.  The octane rating is the ratio of iso-octane and hexane that has similar knocking characteristics as the fuel being tested.  This method is relatively costly and time consuming.  A simpler method is the use of principal component analysis (PCA) with data obtained from a gas chromatograph with a flame ionization detector (GC-FID).  PCA was used to characterize gasolines by octane rating.  It was also useful for distinguishing gasoline from other fuels, including diesel fuel and lighter fluid.  Target testing was used to classify an unknown fuel chromatogram.  This method might also be useful to members of the forensics community, specifically those who deal with arson investigation, as they often need to distinguish between different types of accelerants used at a fire scene.  The results presented here are from the early phases of an ongoing project.



A Shimadzu GC-2010 gas chromatograph was used, equipped with a flame ionization detector and an autosampler.  The carrier gas was helium at a flow rate of 1 mL/min.  The injector was heated to 250ºC with an injection volume of 1 mL and a split ratio of 1:10.  The column used was a Restek Rtx-5 capillary column, 15 m x 0.25 mm.  The oven was programmed to hold at 50 ºC for 2.50 min, increased to 300ºC at 15.0ºC/min, and held at 300ºC for 5.83 min.  The total run time was 25.00 min.  Gasoline and diesel fuel samples were obtained from several local gas stations.

Chromatograms were exported as ASCII files and analyzed in the Matlab® programming environment.  Chromatographic data was arranged into an n x p matrix, X, with n samples (different gasolines or fuel samples) and p properties (chromatographic time points).  R-mode PCA was used to rotate the data set and project it into two dimensional space.  This process is described in Figure 1.  Samples in a data set were compared by plotting the first two principal components in the scores matrix.  The R-mode PCA algorithm was written in house.

Porter Project Figure 1 Figure 1. R-mode PCA

Results and Discussion

The utility of the PCA method to distinguish obviously different fuels (gasoline and lighter fluid from diesel) is shown in Figure 2.  Gasoline and lighter fluid are comprised of low molecular weight, mainly aromatic hydrocarbons while diesel fuel is comprised of high molecular weight, mostly aliphatic hydrocarbons.  The comparison between the chromatograms, shown in Figure 3 shows that these substance are easily distinguished by their GC profiles. 

Porter Project Figure 2 Figure 2. PCA comparison of different fuels
Porter Project Figure 3 Figure 3. Chromatographic profiles of different fuels
Porter Project Figure 4 Figure 4. GC profiles of different octane gasoline
Porter Project Figure 5 Figure 5. PCA comparison of different octane gasoline

Figures 4 and 5 illustrate the comparison of different octane gasolines from the same gas station.  It is difficult, if not impossible, to make a comparison based on the GC profiles alone.  However, the PCA algorithm readily separates the different gasolines. 

The PCA algorithm was then used to compare 87 octane gasoline from different gas stations.  Figure 6 shows that one of the samples was distinguishable from all of the others.  We attribute this to the fact that this sample had been stored in a gas can for quite some time and the difference that we see is due to weathering.  This result raises the possibility of using PCA to study weathering patterns for gasoline.  Weathering in diesel fuel is quite well documented; however, fewer studies have been carried out on gasoline weathering.

Porter Project Figure 6 Figure 6. Comparison of older 87 octane to new.

Finally, the newer gasoline samples were compared, omitting the weathered sample.  Here, no distinction was found between the different gas stations until the third principle component was observed.  Using this multidimensional approach, one of the gas stations in particular (Sunoco) was distinguishable from the others.  We attribute this grouping to the fact that the Sunoco station where this sample was obtained does not use ethanol in their gasoline, while the others do.

Porter Project Figure 7 Figure 7. Comparison of 87 octane from different gas stations, omitting weathered sample

Future Work

We intend to increase our sample sizes and the number of different gas stations to further evaluate the potential of the PCA algorithm to distinguish different octane gasoline and gasoline from different sources based on GC profiles.  This study should include reformulated and oxygenated gasolines (commonly sold during summer months in cities).  Weathering of gasoline will also be studied, in light of the results seen here.  Multidimensional PCA will be further investigated, and target testing will be applied to determine the predictive power of the method.