1. Introduction
Food-grade paraffin wax is a petroleum-derived product (PDP) used in a variety of food products as both a preservation material and an additive. For example, it is employed in the production of chewing gum and as a food preservative in cheeses, fruits, and cakes [
1,
2,
3,
4,
5]. Accordingly, to comply with the U.S. Food and Drug Administration (FDA) standards under 21 CFR 172.886 for its use in food [
6] and, 21 CFR 178.3710 for direct food contact [
7], paraffin wax for food applications is subjected to a high temperature and pressure hydrogenation process that allows for the removal of both sulfur compounds and aromatic compounds [
1,
8,
9]. Therefore, to ensure food safety, companies responsible for manufacturing and distributing food packaging and additives products must implement quality control programs to ensure product safety and sanitary compliance.
In this context, the petrochemical industry had to adapt both good manufacturing practice principles and legal requirements to implement food safety policies at paraffin wax production facilities. To do this, it was necessary to implement a quality management system, which was based on the American Society for Testing and Materials’ (ASTM International) standardized methods [
10]. Among food quality and food-package control parameters, odor intensity is considered one of the most important factors to be evaluated. In this sense, the lack of an undesirable odor is an indicator of final product quality [
11]. Therefore, its assessment is very general. For food-grade paraffin wax, the most common method for determining the odor level is governed by the ASTM D1833—Standard Test Method for Odor of Petroleum Waxes [
12]—which consists of assessing the intensity of the current odor in paraffin wax through a tasting panel of five experts. Then, the paraffin wax odor is graded on a 5-point scale from 0 to 4, with 0 being odorless (“None”) and 4 being the maximum odor level (“Very strong”). Disadvantages that this method can pose include the need for human resources, problems during the comparisons of the panel testing results among different laboratories or companies, lack of rapid response from sensory tests, and possible health risks to the evaluator due to long-term exposure to undesirable volatile organic compounds (VOCs) and semi-volatile organic compounds (SVOCs) [
13]. In this sense, the proper automated and instrumental-based characterization of the global aromatic profile responsible for the organoleptic properties of paraffin wax can facilitate production operations and redirect them toward more effective processes.
Today, there are many analytical techniques used for odor quality control in the food industry and other fields. Among those employed for the characterization of VOCs and SVOCs, methods based on gas chromatography (GC), often combined with mass spectrometry (MS), have been of excellent use due to their high sensitivity and reproducibility [
14]. Regarding sample preparation, many options for extracting VOCs and SVOCs can be coupled with the GC/MS system. Some of these options include thermal desorption (TD), solid-phase microextraction (SPME), and headspace generation (HS) techniques, with the last two being the most used [
15,
16]. In general, the information obtained by GC/MS, the total ion chromatogram (TIC), has been employed to find target compounds that allow for sample characterization [
17]. However, this task is time-consuming and requires skilled operators. While this task demands considerable time and expertise, advancements in the GC–MS field have enabled its utilization as a screening or direct method through total ion spectra (TIS). The TIS method involves summing the intensities for each mass-to-charge (
m/
z) ratio across the full chromatographic range to generate a time-averaged spectrum [
18,
19]. This approach serves as an alternative that addresses the challenges linked with retention times and the application of GC/MS as a chemical sensor, functioning similarly to an MS-based electronic nose [
20,
21,
22]. Focusing on this last point, when used as a chemical sensor, the GC/MS operates so that each fragment ion (represented by the
m/
z ratio) functions as an individual sensor. The intensity of each ion corresponds to the sensor’s signal. This approach enables the determination of the complete aromatic profile of each sample, which is as distinctive as a fingerprint. Additionally, as TIS is time-independent, it is more suitable for inter-laboratory comparison [
22]. The development of GC/MS methods involving the use of TIS results in the acquisition of a large amount of information, and it has become imperative to use advanced machine learning (ML) techniques to transform and manage these data into useful information [
22,
23,
24,
25,
26]. Unsupervised and supervised machine learning (ML) techniques frequently utilize data from global profiling or screening methods, such as GC/MS with total ion spectra (TIS) and MS-based electronic noses. For instance, unsupervised methods like cluster analysis (CA) and principal component analysis (PCA) are commonly used for pattern recognition [
19,
24,
26,
27]. Conversely, supervised techniques such as linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), k-nearest neighbors (kNNs), support vector machine (SVM), random forest (RF), and partial least squares regression (PLSR) are employed for tasks including discrimination, classification, and regression [
24,
27,
28,
29,
30,
31,
32]. Moreover, the creation of supervised regression and classification models, which are predictive in nature, can streamline data processing for future samples, thereby simplifying these tasks in the petrochemical and agri-food sectors.
Regarding the application of analytical techniques to the characterization of paraffin wax, Durret (1966) applied GC together with a flame ionization detector (FID), as well as MS, to identify target compounds related to the paraffin smell, indicating that toluene is mainly responsible for the undesirable odor [
33]. In turn, Yuan et al. (2013) investigated the presence of toluene and aromatic compounds in the volatile composition of paraffin wax by applying HS-SPME-GC/MS [
34]. The results of the study indicated that toluene was the component with the greatest influence on the odor, followed by other minor components such as xylene and other benzene derivatives. Palu et al. (2014) applied GC/MS to characterize de-oiled industrial paraffin wax [
35]. On the other hand, in a recent study conducted by our research group [
36], we employed TD-GC/MS to identify and quantify VOCs in paraffin waxes used in food processing. In this particular case, the focus was on identifying the VOCs involved in the aroma of the paraffins, and thus, the TIS strategy was not used. On the other hand, Wang et al. (2015) [
37] pioneered the use of a gas-sensor electronic nose for discriminating paraffin waxes based on their volatile profiles. They combined this technique with principal component analysis (PCA) for dimensionality reduction and data visualization. Additionally, kNNs, SVMs, and multilayer perceptrons were employed to classify the different qualities of petroleum waxes. For their part, Men et al. (2018) [
13] extended the use of an electronic nose integrating a gas sensor network with ML techniques such as SVM, RF, and extreme learning machine (ELM) to enhance odor analysis systems for paraffin wax, yielding commendable outcomes. Their work introduced an innovative screening method for the automatic assessment of odor levels in paraffin products. Despite these excellent results, the reliance on gas sensor-based systems introduces inherent limitations, including issues with sensor stability and sensitivity. To overcome these challenges, there is an increasing interest in investigating alternative technologies, such as HS–MS electronic noses or HS-GC/MS using the TIS strategy, that offer greater sensitivity and robustness. These methods provide a more detailed and comprehensive aroma profile of samples, making them particularly well-suited for critical quality assessments, such as odor evaluation in paraffin waxes.
Therefore, after all the above, in this study the integration of TIS with HS-GC/MS was explored to assess its efficacy as an electronic nose for capturing the global aroma profile of food-grade paraffin waxes. This innovative approach aims to capture the comprehensive aroma profile of food-grade paraffin waxes, addressing the limitations of traditional gas-sensor electronic noses and the targeted analyte approach typically employed in GC/MS with TIC. This approach has been successfully applied previously to other petroleum-derived products [
22]. By leveraging TIS, this research seeks to enhance the objectivity and reproducibility of sensory analysis methodologies, offering a viable alternative to the current ASTM standards used in the petrochemical industry. For this purpose, a Box–Behnken design (BBD) using response surface methodology (RSM) was applied to variables related to headspace generation. In the present investigation, GC/MS is used as a multi-sensor device with all the advantages that this brings. Therefore, TIS was proposed as a new robust approach to determine the optimal conditions required to best distinguish and quantify odor levels in paraffin waxes. Furthermore, the final goal of this study is to apply the developed method to paraffin wax samples with different odor levels, to detect and distinguish the degree of odor intelligently and simply. To do so, TIS was combined with several ML tools to establish the optimal algorithms to obtain better efficiency from the results and information contained in the data. The best-performing models generated were employed to be integrated into an interactive web application for sharing purposes.
4. Conclusions
The present research has successfully developed an automatic, effective, and objective methodology based on HS-GC/MS to characterize and quantify the odor grade in paraffin waxes. Specifically, both RSM–BBD and a univariate study have been applied for the optimization of the HS working conditions, obtaining accurate and repeatable results. The final method developed presents certain advantages based on the non-use of solvents and the analysis time, making it eco-friendly and applicable to routine analysis in petrochemical or agri-food laboratories using this material. On the other hand, the study of the relevant literature revealed that, to the best of our knowledge, this is the first time that TIS information has been employed as a strategy to characterize this parameter in paraffin wax. These chemical data, combined with the appropriate ML tools, allow for reproducible and fast processing of the results. Concretely, the unsupervised ML techniques, HCA and PCA, revealed the existence of a certain grouping tendency of the studied samples according to their odor level. Of the classification models generated, the one based on Gaussian SVM obtained excellent results (test set: 100% accuracy and kappa 1) for the discrimination of the odor level in the paraffin wax samples. On the other hand, of the regression algorithms explored for odor-grade quantification, the Gaussian SVR-based model obtained a higher performance (test set: 6.789 RMSE and 0.9667 R2). Furthermore, an interactive online application has been developed to share these two models with industry researchers and manufacturers as well as to make it simpler for this product to be controlled. These models can be retrained as more samples are examined. In this way, by creating a constantly updated database, the reality may be more precisely tuned to satisfy the quality control criteria of this PDP.
Future studies should focus on applying and evaluating this methodology across a broader range of real sample matrices to guarantee reliable prediction results in other samples. Moreover, expanding the database with real-time data from various sources will enhance the models’ predictive capabilities.