Sensors 2026 , 26 , 2049
13of 16
3.3.2. Optimization of the E-Nose Methodology The simplex design methodology was used to optimize the analytical methodology. The objective was to maximize the analytical signal (sensor response). Three variables were expected to have a major effect on the e-nose response, affecting sampling and analysis: (i) the nitrogen flow rate, (ii) the number of paper circles (quantity of paper used for sampling) and (iii) the exposition time of the SPME fiber to volatiles emitted from the paper. The experiments proposed by the Simplex method stopped after the 46th run, upon convergence, indicating that the optimal solution has been found. The optimum objective function was reached with 22 paper circles, 30 min of exposure of the fiber, and a desorption/transporting nitrogen flow rate of 105 mL min − 1 . 3.3.3. Results with the SPME/E-Nose The headspace extraction of volatiles by SPME from books has been used previously with success, followed by the analysis with a homemade e-nose, which led to book discrim- ination by paper degradation state [9]. In addition, it helped identify acidic papers made from wood, which were distinguished from those made from cotton or neutral/alkaline wood papers. The e-nose system was changed for the present work, as new sensors needed to be incorporated. Sampling needed to be adapted and experimental conditions optimized. All paper samples were analyzed according to the procedure described in Section 2.6.2. Hierarchical Clustering Analysis (HCA) HCA is an unsupervised chemometric technique that reveals natural groupings among samples characterized by the values of a set of measured variables. The results were described using a dendrogram, which has a tree-like structure. HCA was performed on 21 samples. For testing purposes, the results from three replicates of acacia P10 (paper from three different sheets from the same ream of paper) and two replicates of recycled P21 were included in the analysis. For all the other samples, only the result of the analysis of 22 circles of the same sheet was included to keep the dendrogram as simple as possible. Cluster analysis was conducted using hierarchical agglomerative clustering with average linkage (within groups). Euclidean distance was used as the dissimilarity metric to quantify similarities/differences among samples, and the resulting clustering can be visualized in Figure 5. At a distance level of 18, all the samples can be clustered into two groups (see Figure 5). The first cluster, I (blue line), consists of acacia samples (P10 and P11), and the second cluster, II (yellow line), consists of all the remaining samples. This second cluster aggregates all E. Globulus, birch, mix, and recycled papers, and includes all the remaining samples without information about the tree of origin. Replicates are together, as expected. The findings are quite promising and suggest a significant potential for future research on identifying the origin of office paper sheets. Figure 6 shows a plot of the frequency decreases observed for sensor 1 vs. sensor 3 for samples with known wood type. It can be observed that acacia paper elicits the strongest responses from both sensors 1 and 3, whereas recycled paper elicits the highest responses from sensor 3 but lower responses from sensor 1. Birch paper samples exhibit low responses from both sensors 1 and 3, while E. Globulus and mix papers are also characterized by low responses from sensor 3 but higher responses from sensor 1. These findings suggest that the two sensors can differentiate among various wood types used in paper production.
https://doi.org/10.3390/s26072049
Made with FlippingBook interactive PDF creator