Modeling the characteristics and quantification of adulterants in gasoline using FTIR spectroscopy and chemometric calibrations

Figures & data

Table 1. Ratio of gasoline–adulterant admixture prepared in the laboratory

Sample no.	% v of gasoline–kerosene mixture (GK)	% v of gasoline–premix mixture (GP)	% v of gasoline–naptha mixture (GN)	% v of gasoline–diesel mixture (GD)
1	95:5	95:5	95:5	95:5
2	90:10	90:10	90:10	90:10
3	85:15	85:15	85:15	85:15
4	80:20	80:20	80:20	80:20
5	75:25	75:25	75:25	75:25
6	70:30	70:30	70:30	70:30
7	65:35	65:35	65:35	65:35
8	60:40	60:40	60:40	60:40
9	50:50	50:50	50:50	50:50

Figure 1. Spectrum of pure gasoline (a), diesel (b), kerosene (c), premix (d), naphtha (e).

Figure 2. Spectra of gasoline–kerosene mixtures at different concentrations.

Samples GK1-GK10 are gasoline–kerosene admixtures with concentrations of kerosene from 5–50% v/v, respectively.

Figure 3. Spectra of gasoline-diesel mixtures at different concentrations.

Samples GD1-GD10 are gasoline–diesel admixtures with concentrations of diesel from 5–50% v/v, respectively.

Figure 4. Spectra of gasoline–premix mixtures at different concentrations.

Samples GP1-GP10 are gasoline–diesel admixtures with concentrations of premix from 5–50% v/v, respectively

Figure 5. Spectra of gasoline–naphtha mixtures at different concentrations.

Samples GN1-GN10 are gasoline–diesel admixtures with concentrations of naphtha from 5–50% v/v, respectively.

Figure 6. Spectra of calibration set for SIMCA model.

Samples with labels GD are adulterated with diesel, GK with kerosene, GP with premix and GN with L-naphtha.

Table 2. Percent variance captured by PCA model–diesel class

Principal component number	Cov (X)	%Variance—this PCA	%Variance—total
1	1.07 × 10^−8	62.63	62.63
2	5.70 × 10^−9	33.30	95.93
3	5.81 × 10^−10	3.40	99.33

Table 3. Percent variance captured by PCA model–kerosene class

Principal component number	Cov (X)	%Variance—this PCA	%Variance—total
1	1.22 × 10^−5	94.76	94.76
2	4.09 × 10^−7	3.17	97.93
3	2.50 × 10^−7	1.94	99.86

Table 4. Percent variance captured by PCA model–naphtha class

Principal component number	Cov (X)	%Variance—this PCA	%Variance—total
1	2.47 × 10^−9	80.30	80.30
2	5.23 × 10^−10	17.01	97.32
3	5.95 × 10^−11	1.94	99.25

Table 5. Percent variance captured by PCA model–premix class

Principal component number	Cov (X)	%Variance—this PCA	%Variance—total
1	9.36 × 10^−9	74.81	74.81
2	2.14 × 10^−9	17.14	91.95
3	9.24 × 10^−10	7.39	99.33

Table 6. Confusion matrix (A) and confusion table (B) for the calibration set

A
	TPR	FPR	TNR	FNR	N	Err	P	F1
Diesel	1.00	0.00	1.00	0.00	6	0.00	1.00	1.00
Kerosene	1.00	0.00	1.00	0.00	6	0.00	1.00	1.00
Naphtha	1.00	0.00	1.00	0.00	6	0.00	1.00	1.00
Premix	1.00	0.00	1.00	0.00	6	0.00	1.00	1.00
B
	Actual class
	Diesel		Kerosene		Naphtha		Premix
Predicted as diesel	6		0		0		0
Predicted as kerosene	0		6		0		0
Predicted as naphtha	0		0		6		0
Predicted as premix	0		0		0		6

Note: TPR—true positive rates; FPR—false positive rates; TNR—true negative rates; N—number of samples; Err—error rate; P—precision; F1—F scores.

Figure 7. Spectra of validation (test) set for SIMCA.

Samples with labels GD are adulterated with diesel, GK with kerosene, GP1 with premix and GN with L-naphtha

Table 7. Confusion matrix (A) and confusion table (B) for the validation set

A
	TPR	FPR	TNR	FNR	N	Err	P	F1
Diesel	1.00	0.00	1.00	0.00	4	0.0000	1.00	1.00
Kerosene	0.75	0.00	1.00	0.25	4	0.0625	1.00	0.857
Naphtha	0.75	0.00	1.00	0.25	4	0.0625	1.00	0.857
Premix	0.50	0.00	1.00	0.50	4	0.1250	1.00	0.667
B
	Actual class
	Diesel		Kerosene		Naphtha		Premix
Predicted as diesel	4		0		0		0
Predicted as kerosene	0		3		0		0
Predicted as naphtha	0		0		3		0
Predicted as premix	0		0		0		2

Note: TPR—true positive rates; FPR—false positive rates; TNR—true negative rates; N—number of samples; Err—Error rate; P—precision; F1—F scores.

Table 8. Actual and predicted adulterants present in validation set

Sample ID	Actual class	Predicted class
GD7	1	1
GD9	1	1
GD3	1	1
GD6	1	1
GN2	3	0
GN5	3	3
GN7	3	3
GN9	3	3
Sample ID	Actual class	Predicted class
GK2	2	0
GK9	2	2
GK3	2	2
GK6	2	2
GP3	4	4
GP5	4	4
GP9	4	0
GP1	4	0

Note: 0—unassigned; 1—diesel; 2—kerosene; 3—naphtha; 4—premix.

Figure 8. Spectra of calibration data set for PLS-K.

Table 9. Percent variance captured by regression model

	X Block		Y Block
Comp	This	Total	This	Total
1	99.30	99.30	74.98	74.98
2	0.63	99.95	20.70	95.69

Figure 9. Measured vs. predicted concentration of kerosene in the calibration set.

Table 10. Actual and predicted concentration of kerosene in calibration set

Sample ID	Actual:Conc. Kerosene (% v/v)	Predicted:Conc. Kerosene (% v/v)	p-Value (at 0.05 significance level)
GK1	5	5	0.884
GK4	20	25
GK5	25	27
GK7	35	27
GK8	40	42
GK10	50	51

Figure 10. Actual vs. predicted concentration for the validation set (PLS-K).

Table 11. Actual and predicted concentration of kerosene in validation set

Sample ID	Actual:Conc. Kerosene (% v/v)	Predicted:Conc. Kerosene (% v/v)	p-Value (at 0.05 significance level)
GK2	10	14	0.622
GK9	45	42
GK3	15	16
GK6	30	29

Figure 11. Spectra of the calibration data set for PLS-D.

Samples GD1, GD2, GD4, GD5, GD8 and GD10 are adulterated gasoline with 5%, 10%, 20%, 25%, 40% and 50% v/v of diesel, respectively.

Figure 12. Actual vs. predicted concentration of diesel in calibration set (PLS-D).

Figure 13. Actual vs. predicted concentration of diesel in the validation set.

Table 12. Comparison of actual and predicted diesel concentration in calibration set

Sample ID	Actual concentration of diesel (% v/v)	Predicted concentration of diesel (% v/v)	p-Value (at 0.05 significance level)
GD1	5	8	0.390
GD2	10	9
GD4	20	22
GD5	25	27
GD8	40	39
GD10	50	48

Table 13. Comparison of actual and predicted diesel concentration in validation set

Sample ID	Actual:Conc. Diesel (% v/v)	Predicted:Conc. Diesel (% v/v)	p-Value (at 0.05 significance level)
GD9	45	44	0.812
GD3	15	17
GD6	30	34
GD7	35	38

Figure 14. Spectra of (L) naphtha-adulterated gasoline samples (calibration set).

Table 14. Comparison of actual and predicted naphtha concentration in calibration set

Sample ID	Actual:Conc. Naphtha in gasoline (% v/v)	Predicted:Conc. Naphtha in gasoline (% v/v)	p-Value (at 0.05 significance level)
GN1	5	6	0.993
GN3	15	13
GN4	20	21
GN6	30	30
GN8	40	39
GN10	50	50

Table 15. Comparison of actual and predicted naphtha concentration in validation set

Sample ID	Actual:Conc. L-naphtha Sample (% v/v)	Predicted:Conc. L-naphtha Sample (% v/v)	p-Value (at 0.05 significance level)
GN2	10	14	0.886
GN5	25	27
GN7	35	34
GN9	45	47

Figure 15. Actual vs. predicted concentrations of naphtha in the calibration set.

Figure 16. Actual vs. predicted concentration of naphtha (validation set).

Table 16. Percent variance captured by regression model

	X Block		Y Block
Comp		Total		Total
1	62.25	62.25	90.55	90.55
2	20.38	85.63	90.09	99.62
3	10.70	96.32	0.22	99.85
4	3.23	99.55	0.14	99.99

Figure 17. Spectra of gasoline samples adulterated with premix (calibration set).

Figure 18. Actual vs. predicted concentration of premix (calibration set).

Figure 19. Actual vs. predicted concentration of premix (calibration set).

Table 17. Comparison of actual and predicted concentration of premix in calibration set

Sample ID	Actual concentration—premix in gasoline (% v/v)	Predicted concentrations—premix (% v/v)	p-Value (at 0.05 significance level)
GP1	5	4	0.934
GP3	15	17
GP4	20	19
GP7	35	32
GP8	40	41
GP10	50	53

Table 18. Comparison of actual and predicted concentrations of premix in the validation set

Sample ID	Actual content of premix (% v/v)	Predicted content of premix (% v/v)	p-Value (at 0.05 significance level)
GP2	10	9	0.934
GP5	25	28
GP6	30	32
GP9	45	42

Table 19. SIMCA classification of commercial samples based on adulterants present

	Adulterant predicted to be present					Total
	Diesel	Kerosene	Naphtha	Premix	Unassigned
Number of samples	1	4	0	2	93	100

Table 20. Quantity of adulterants in adulterated commercial gasoline samples

Sample ID	Adulterant detected	Concentration of adulterant present (% v/v)
SG008	Diesel	39
SG015	Kerosene	29
SG030	Kerosene	32
SG033	Kerosene	17
SG043	Kerosene	21
SG039	Premix	34
SG082	Premix	25