Response of the gas sensor array to adulterated CEO
The e-nose response to the unadulterated CEO was initially evaluated for various successive cycles (Fig. 2). Each curve represents a different sensor response against time. The response of the array to air (baseline) can be seen first, followed by the volatiles from the CEO. The voltage increases sharply with the interactions of volatile compounds and then stabilizes after 30 s. The results revealed that the e-nose, composed of polymeric nanocomposite gas sensors, had a high response-ability to the volatile components produced by the CEO.
The gas sensor array was applied to CEO detection and adulteration with the addition of different percentages (0, 3, 6, 12, 25, 50, 75, and 100%) of vaseline (Fig. 3). The temperature and humidity were monitored during the experiments, showing values of 35 ± 1°C and 47 ± 3%, respectively.
Vaseline, when inserted at 100% in the sampling chamber, did not cause changes in the voltage response, and the values remained the same as at baseline (Fig. 2). Selectivity in the context of gas sensors refers to the ability of the sensor to specifically detect and respond to the target gas while not being affected by other gases or substances present in the environment. In this case, the presence of vaseline did not cause any changes in the sensor's response. This can be explained by the fact that vaseline is odorless and composed of a mixture of saturated hydrocarbons (alkanes, cycloalkanes) and aromatic hydrocarbons between C15 and C50 (Speight 2002; Pochivalov et al. 2018). The gas sensors used have measurable changes in voltage only when volatile compounds are released and interact with the sensitive layer. Thus, the vaseline acted as a diluting agent in the CEO´s volatile release.
In the adulteration from 75–0% (pure CEO), there was a progressive increase in voltage, indicating a high concentration of CEO present in the sample. It was also possible to verify that during all cycles; the sensors returned the signal to the baseline. The higher voltage response presented by the sensor array was for the CEO (0% vaseline, unadulterated CEO).
The selectivity of the polymeric gas sensor for CEO detection was studied by investigating other interfering compounds, including mineral oil (paraffinum) and vegetable oils (sunflower and soya). The result demonstrates that the voltage response caused by interfering compounds was similar to vaseline. Meanwhile, it indicates that PANI has no interaction with the sensitive layer of the polymeric gas sensor layers evaluated, and PANI has no recognition ability for mineral and vegetable oils because they are composed by saturated hydrocarbons. The presented results (Table 2) demonstrate the selectivity of the e-nose system, and the checking interferences do not disturb the gas sensor response.
Table 2
Effect of interfering substances (100%) on the determination of CEO.
Interference substance
|
Voltage (mV)
|
Vaseline
|
0.11
|
Mineral oil
|
0.20
|
Vegetable oil (sunflower)
|
0.32
|
Vegetable oil (soya)
|
0.34
|
The hierarchical CA method was utilized to categorize CEO samples based on the responses obtained from the six-sensor array. The similarity distance employed was the squared Euclidean, and Ward's clustering method was chosen as the amalgamation rule. Figure 4 shows the dendrogram produced by the CA method. The responses of the samples were separated into two clusters, comprising eight distinct groups. The first cluster encompassed two groups: one with 0% adulteration and another with 3% adulteration. The second cluster comprised adulterated CEO samples ranging from 6–100% adulteration. Therefore, the CA method provided an initial classification, although the group divisions varied when employing different distance measures.
The data obtained in the e-nose for adulterated CEO was analyzed by PCA (Fig. 5). PCA has been performed to describe the volatile compound changes in the adulteration. Figure 5 shows a grouping of the scores according to the different percentages of vaseline added to the CEO, showing discrimination among the various clusters representing the amount of adulteration. The different percentages of vaseline were distributed in the four quadrants of the plane. Pure vaseline (100%) was in the same quadrant as a percentage of 75% and was on the opposite side of pure CEO, near the percentage of 3%. The pure vaseline and CEO were in opposite quadrants, indicating that the sensor array was able to discriminate the different percentages of vaseline added to the CEO.
The variances of the two main components (PC1 + PC2) yield 99.85% of the total information collected by the matrix. PC1 presented the highest amount of information (97.85%), and, therefore, the analysis of the detection of CEO adulteration must be based on this axis. Although small, the contribution of PC2 must be considered for analysis. In the loading plot, it was possible to observe that the vectors indicate the gas sensors used in the electronic nose array. The contribution of sensors is represented by their direction and magnitude. The sensors S1, S2, and S3 elaborated with PANI-MWCNT_COOH doped with HCl, CSA, and DBSA have discrimination in the PC1 positive quadrant, and S1 showed the largest positive coefficient. However, the sensors with PANI-GO fim doped with HCl, CSA, and DBSA (S4, S5, and S6) have discrimination in the PC1 negative quadrant, and S6 has a high negative coefficient. This result demonstrates that the gas sensor responses exhibited clustering based on their inherent patterns or similarities. These differences in the sensor signals can be associated with the interactions between the target compounds and the sensing materials. Based on these results, the sensors S1 and S6 have the largest contribution to the first principal component during the process of PCA.
The difference in their sensing mechanisms was associated with different types of composite materials and dopants used in the gas sensor layers. PANI-MWCNT composite sensors typically operate based on changes in electrical conductivity. PANI changes its conductivity when exposed to target analytes, and MWCNT is added to enhance the electrical properties and sensing performance of the composite. The interaction between the analyte and PANI-MWCNT leads to changes in the electrical conductance of the composite, allowing for detection and measurement. On the other side, PANI-GO refers to graphene sheets functionalized with oxygen-containing groups. The GO serves as a sensing element that interacts with the analytes. The interaction between the analyte and the PANI-OG composite causes changes in electrical conductivity as well as other properties such as charge transfer, surface area, and electron mobility, resulting in improved sensitivity in the detection of the analyte.
Rañola; Santiago; Sevilla (2016), used chemiresistors based on conducting polymers to detect different types of coconut oils (virgin coconut oil, refined, bleached, and deodorized coconut oil; flavored virgin coconut oil, homemade virgin coconut oil, and rancid virgin coconut oil), found responses with different signals for all polymeric sensors. Chemometric techniques such as PCA and cluster analysis showed different clusters for the coconut oil samples, which explains the influence of the dopants used in the polymers for the sensor array. The authors also report the importance of PCA analysis, an exploratory analysis that reduces the number of variables and generates an easy-to-view punctuation chart.
Figure 6 shows the LDA results, where around 98.3% of the total variance of the data was displayed by the linear discriminant analysis. The linear discriminants 1 (LD1) and 2 (LD2) contributed 78.1% and 20.2% of the total variance, respectively. A classification of the data into groups can be seen according to the amount vaseline used to adulterate the CEO. Thus, both linear dimension reduction methods were able to separate the pure and adulterated CEO samples correctly. The results of this study are in agreement with the work of Graboski et al. (Graboski et al. 2018b) who employed an e-nose to classify gummy candy samples with different artificial aromas (apple, strawberry, and grape). Their results were then analyzed using different pattern recognition techniques (PCA and LDA), observing 100% classification for both techniques. Xu et al. (Xu et al. 2016) applied the LDA method and succeeded in classifying the fresh and spoiled oils with a precision of 100%. Karami et al. (Karami et al. 2020) classified different percentages of mixed edible oils into oxidized and nonoxidized oils and obtained a 98% detection accuracy.
Tables 3 and 4 present the score diagram and performance parameters of the LDA method. Specifically, Table 4 provides performance metrics derived from the confusion matrix, such as precision, accuracy, specificity, sensitivity, and area under the curve (AUC), which assess the classifier's performance. In the confusion matrix, the classification accuracy of the LDA model can be evaluated. In the matrix, the classification accuracy of the LDA model can be evaluated. It provides an overview of how well the LDA model is correctly classifying samples from different classes, and it demonstrates 100% correct classification. The classification of pure oil (without adulteration) was 100%. In the accuracy results (Table 4), the samples with adulterations showed values < 1, whereas the pure oil was = 1. The sensors being referred to have high sensitivity and differentiation ability, enabling them to detect variations between volatile substances. These sensors are capable of discerning even subtle differences in the chemical composition or concentration of volatile compounds. Their heightened sensitivity allows for precise discrimination and identification of various substances based on their unique odor profiles or volatile emissions.
Table 3
Confusion matrix obtained to identify CEO and adulterated CEO with vaseline LDA
Samples
|
100% adulteration (vaseline)
|
75%
|
50%
|
25%
|
12%
|
6%
|
3%
|
Pure oil (0% vaseline)
|
100% adulteration (vaseline)
|
6
|
0
|
0
|
0
|
0
|
0
|
0
|
0
|
75%
|
0
|
6
|
0
|
0
|
0
|
0
|
0
|
0
|
50%
|
0
|
0
|
6
|
0
|
0
|
0
|
0
|
0
|
25%
|
0
|
0
|
0
|
6
|
0
|
0
|
0
|
0
|
12%
|
0
|
0
|
0
|
0
|
6
|
0
|
0
|
0
|
6%
|
0
|
0
|
0
|
0
|
0
|
6
|
0
|
0
|
3%
|
0
|
0
|
0
|
0
|
0
|
0
|
6
|
0
|
Pure oil (0% vaseline)
|
0
|
0
|
0
|
0
|
0
|
0
|
0
|
6
|
Correct classification rate 100% |
Table 4
Performance parameters for LDA.
Class
|
Accuracy
|
Precision
|
Sensitivity
|
Specificity
|
AUC
|
Pure oil (0% vaseline)
|
1.000
|
1.000
|
1.000
|
1.000
|
1.000
|
3%
|
0.978
|
0.978
|
0.978
|
0.979
|
0.979
|
6%
|
0.970
|
0.971
|
0.974
|
0.974
|
0.974
|
12%
|
0.986
|
0.988
|
0.997
|
0.992
|
0.997
|
25%
|
0.953
|
0.959
|
0.967
|
0.961
|
0.966
|
50%
|
0.944
|
0.953
|
0.960
|
0.953
|
0.957
|
75%
|
0.985
|
0.995
|
0.998
|
0.987
|
0.992
|
100% adulteration (vaseline)
|
0.942
|
0.956
|
0.959
|
0.943
|
0.948
|
Average per class
|
0.970
|
0.975
|
0.979
|
0.974
|
0.977
|
The data obtained by the e-nose of CEO adulteration using vaseline was submitted to IDMAP information visualization techniques. Figure 7a shows the graphs with excellent discrimination for the non-linear IDMAP technique. Here, the silhouette coefficient calculated by the IDMAP was 0.93, and the total information represented by the PCA, Fig. 7b, was 99.81%. The silhouette is a metric used to evaluate the quality of a data cluster, and higher values indicate better cluster quality. When applied to groups of samples representing different concentrations, the silhouette value for each point measures its similarity to other points within its own cluster, compared to points in other clusters, using predefined similarity metrics. However, in this case, PC1 has the highest amount of information (99.25%), and, therefore, the analysis of the detection of CEO adulteration must be based on this axis. Although small, the contribution of PC2 (0.56%) must be considered for analysis. Through the PCA, it can be seen that the CEO adulteration is observed by a change in the quadrants of the second main component, that is, CEO samples adulterated with vaseline are found in the upper quadrant, while pure vaseline is located in the lower quadrant (opposite).
For both IDMAP, CA, LDA, and PCA, it is clear that the gas sensor array was able to discriminate between the different percentages of vaseline added to the CEO. Thus, the multidimensional projections of IDMAP, LDA, and PCA for analysis of pure CEO demonstrated discrimination against adulterated CEO´s, with the samples being quite distinct in the quadrants, indicating that they are valid methods for promoting a better visualization of the results. In this way, the e-nose showed a characteristic fingerprint of the release of the volatile compound by the CEO. This is a preliminary study that shows the potential of the gas sensor array for detecting adulteration with vaseline, and more studies are needed in order to actually validate the sensor array for the proposed application.