Monitoring of frying process in canola oil blend using fourier transform infrared and chemometrics techniques

Abstract The production of trans‐fats and chemical changes during the process of frying are serious public health concerns and must be monitored efficiently. For this purpose, the canola oil was formulated with different ratio of extra virgin olive oil and palm olein using D‐optimal mixture design, and the best formulation (67:22:11) based on free fatty acid (FFA) content, peroxide value (PV), and iodine value (IV) as responses was selected for multiple frying process. The data on FFA, PV, and IV along with Fourier transform‐infrared (FT‐IR) spectra were taken after each frying up to ten frying. The spectral data were preprocessed with standard normal variate followed by principal component analysis which is clearly showing the differentiation for various frying. Similarly, partial least square regression was applied to predict the FFA (0.37%–1.63%), PV (4.47–13.85 meqO2/kg), and IV (111.51–51.39 I2/100 g) which demonstrated high coefficient of determination (R2) 0.84, 0.83, and 0.81, respectively. It can be summarized that FT‐IR can be used as a novel tool for fast and noninvasive quality determination of frying oils.


| INTRODUC TI ON
Fats and oils play a crucial role in our daily food consumption as they give us vital fatty acids and intense energy supply. Frying in fats and oils is one of very old techniques used in processing of food for quick preparation and provides some special effect on sensation of food (Bou et al., 2012). This deep fat frying (DFF) process is performed by fully immersing the foods in hot oil based on the continuous heat and the mass transfer at elevated temperatures ranging from 150 to 200°C (Moreira, 2014). DFF efficiency relies on either the process of frying itself, the nature of the food, and/or type of frying oil (Marikkar et al., 2011). During the process, thermooxidation and cleavage of triglycerides occur at higher temperatures in the presence of oxygen, particularly if the oil is polyunsaturated, resultantly, the formation of volatile and nonvolatile oxidative compounds. Repeated and persistent use of fried oils produces undesirable compounds that may affect nutritional quality and pose potential threats to the nutrition and human health (Andrikopoulos et al., 2002). Pure oils present low stability against oxidation during frying, alternatively, encouraging the use of mixture of oils for better stability (Alireza et al., 2010). Mixing two or more oils with distinct properties is among the techniques for developing new special oil products. In addition to improving the fatty acids' profile, combining multiple forms of vegetable oils often increases the amounts of natural antioxidants and bioactive phytonutrients in the blends and offers higher quality stable oils (Tiwari et al., 2014). Therefore, first a blend of oil having canola, extra virgin olive oil, and palm olein was optimized using D-optimal design against different quality indicators of frying oils.
Currently, the legislative authorities and food industries are looking for analytical solutions for rapid and noninvasive quality control tools. Different spectroscopic techniques such as Fourier Transform Infrared (FT-IR) have now been extensively studied to analyze various chemical changes and correspond such changes with instrumental data using statistical tools importantly chemometrics. FT-IR is an analytical instrument based on the scanning of samples using infrared light to identify polymeric, chemical, and inorganic materials (Velasco & Dobarganes, 2002). The sample detects some of the infrared radiation and produces its molecular fingerprint in the form of a spectrum. The difference in the characteristic pattern of the absorption bands indicates a modification in the substrate composition or existence of contaminants (Chippie et al., 2002). Furthermore, the sensitivity and precision of FT-IR detectors and a wide variety of automated algorithms have greatly increased the realistic use of infrared for quantitative research analysis. This apparatus is quick to establish and calibrate quantitative approaches and can be used in basic repetitive analytical procedures (Ma, 2018).
Different studies have been conducted on the application of FT-IR to analyze the quality of either adulterated oil, frying oil, or fried products, for instance, monitoring frying quality of palm and soybean oils used for preparing Falafel (Al-Degs et al., 2011), determination of peroxide value of red fruit oil (Andina et al., 2017), degree of degradation of frying rapeseed oil (Chen et al., 2015), monitoring of free fatty acids in commercially available Nigella sativa oil (Mahesar et al., 2014), tracking performance of virgin coconut oil during frying (Srivastava & Semwal, 2015), iodine value of blends with refined soybean oil, peanut oil, rapeseed oil and sunflower seed oil (Meng et al., 2017), correlating the acid values with FT-IR spectroscopy (Jiang et al., 2016), adulteration in wheat germ oil (Arslan & Çağlar, 2019), and others. All these monitoring studies based on spectroscopic techniques are generally combined with suitable chemometrics statistical models (Amit et al., 2020). Unfortunately, no data were found on the application of FT-IR on oil blends which are commercially available and are used for frying at consumer level.
Based on the literature cited above, the present study was planned to optimize the suitable blend for frying using D-optimal design statistical tool and then observe the chemical changes during frying by FT-IR. These chemical changes observed in the form FT-IR spectra were correlated with hectic and time-consuming conventional analytical methods with help of chemometrics.

| Raw material
Canola oil, extra virgin olive oil (EVOO), and frozen French fries were purchased from market of Faisalabad-Pakistan. Palm olein (degummed and neutralized) was procured from the local oil processing industry located at Faisalabad, Pakistan.

| D-optimal mixture design for oil blends
D-optimal mixture design was applied to develop formulations of canola oil with EVOO and palm olein in different ratios with the final volume of each formulation equal to 100 (table 1).
Generally, D-optimal mixture design was applied as an efficient statistical tool to determine the best combination among wide range of studied components (Mohamad Zen et al., 2015). Best suggested formulation of oil blend was interpreted from the chemical analyses as response variables and was further subjected to multiple frying with structural description using FT-IR spectroscopy. The quadratic regression model with three factors is as follows:

| The process of frying
The process of frying was performed in a "Cool Touch Electric Deep Fryer" (Anex, AG-2012). Fourteen formulations with different combinations of canola oil, EVOO, and palm olein blends were used in frying as per D-optimal design. For each trial, 50 g of frozen French fries was fried in 1 L of preheated oil blend at a temperature of 180°C for 4 min. After each frying process, oil samples were collected, cooled at ambient temperature, and subjected to chemical analysis. After screening out the best formulation using D-optimal mixture design, the safety and quality of oil blend were tested up to 10 frying using the above described procedure.

| Free fatty acid
Free fatty acid (FFA) was found out by adopting the AOCS Official Methods Ca 5a-40 (Firestone, 2009). For this purpose, a 10 g of canola oil blend was taken into a conical flask (250 ml) and heated with 100 ml ethanol with addition of 1 ml of phenolphthalein solution as indicator. The mixture was then boiled for about 4-5 min and titrated against 0.1 N KOH till the persistence of pink color for at least 30 s. The value of FFA was calculated using the following where Vs = Volume of sodium thiosulfate titrated, Vb = Volume of sodium thiosulfate used in a blank test, N = Normality of sodium thiosulfate solution, W = Weight of sample.

| Iodine value
The AOCS Official Methods Ca 5a-40 (Firestone, 2009) was followed to determine the iodine value. For this purpose, a 0.25 g of canola oil blend was taken into conical flask (500 ml) with a plastic stopper and combined with 25 ml of each CCl 4 and Wij's solution.
The mixture was placed in the dark for 30 min, and 50 ml of 10% potassium iodide solution was added. After diluting the whole mixture with 100 ml of distilled water, titration was executed against standardized sodium thiosulfate solution (0.1 N) using 0.5% starch as indicator. The titration process was stopped on the disappearance of blue color of starch solution. The same procedure was adopted to conduct blank test and value was calculated as: where B = Volume of standard sodium thiosulfate solution used for the blank, S = Volume of standard sodium thiosulfate solution used for the sample, N = Normality of standard sodium thiosulfate solution, W = Weight of sample.

| FT-IR spectroscopy
The fourier transform infrared (FT-IR) spectroscopy is a useful tool for investigating the chemical modifications in edible oils, both qualitatively and quantitatively, while using little amount of sample and minimizing the error (Xu et al., 2016). Spectroscopic data of oil samples after each frying were obtained with an infrared spectrometer (Alpha II FT-IR, Bruker) fitted with an attenuated total reflection (ATR) accessory. Each sample was preconditioned in a ( TA B L E 1 D-optimal design suggested formulations and chemical analyses of canola oil blends after frying

| Statistical and chemometric data analyses
The statistical software package Design Expert 7.0 (Stat-Ease Inc.) was used to develop the formulations of canola oil blends using Doptimal design and get optimized and best formulation based on the chemical analyses.

| Preprocessing of spectral data
The optimized canola oil blend was used for frying of French fries to observe the structural and chemical changes in frying oil. Optimized canola oil blend was taken in triplicate and used for potatoes frying up to ten times. Three spectral fingerprints were taken after each frying operation, and their average was calculated. Then, the mean spectral signatures were pretreated using various preprocessing techniques. Different types of preprocessing treatments such as Standard normal variate (SNV), multiplicative scatter correction (MSC), normalization, and base line corrections were applied to remove the abnormalities in the data datasets (Lee et al., 2018).

| Principal components analysis
Principal components analysis (PCA) is one of the most widely and unsupervised chemometric tool used for exploratory data analysis.
It reduces the dimensionality and provides the pattern recognition without loss of information. The SNV preprocessed data were utilized for application of PCA using single value decomposition algorithms and full cross-validation process (Ahmad et al., 2016).

| Partial least square regression
Partial least square regression (PLSR) models are being used for determination of various parameters. The preprocess datasets (SNV, MSC, normalization, and baseline correction) were subjected to PLSR modeling using leave one out cross-validation. In this type of modeling, the calibrated models were developed using ten spectral signatures and the eleventh was left out for the prediction models. In this way, all the samples were left out once for prediction models for determination of FFA, PV, and IV using FT-IR spectral fingerprints after each frying process. The accuracy of the models was determined using root mean square error of cross-validation (RMSECV) and coefficient of determination (R 2 ). The RMSECV and R 2 can be calculated by using the Equations 4 and 5, respectively.
Here, n, mi, pi, and m stand for number of samples, measured, predicted, and mean value, respectively.

| Optimization of canola oil blend
D-optimal statistical design was used to formulate the 14 combinations randomly with each contributing the total amount equal to 100 (table 1)

| Chemical characterization in multiple frying
The optimized canola oil blend (  The chemical changes analyzed through FFA, PV, and IV have also been confirmed with FT-IR spectra during multiple frying (Figure 2).
The spectral bands of fried oils obtained were correlated with the available literature. The spectral peaks at 1110 cm −1 and 1710 cm −1 can be allied to C-O and carbonyl (C = O) bonds of aliphatic esters, respectively (Shen et al., 2014). Next to these peaks, two more strong peaks at 2820 cm −1 and 2990 cm −1 can be assigned to symmetrical and asymmetrical C-H bonds of CH 2 structures. With the increase in number of frying, the generic transmittance peaks were diminished and new transmittance peaks were appeared at 1595 cm −1 and 3320 cm −1 which indicate the formation of saturated bonds (C-O) and FFA formed after the hydrolysis of ester groups. The information obtained through the spectroscopic data is presenting the noticeable differences among various frying; however, it is essential to apply multivariate calibration to validate the information from such complex and overlapping spectra (Jamwal et al., 2021).

| Principal component analysis for FT-IR spectra
PCA was applied on the SNV preprocessed data which shows clear differentiation among different frying of optimized canola oil blend ( Figure 3a). The first principal component (PC1) explains 95% variance whereas the second principal component (PC2) provides its share of only 3%. Against PC1, maximum value was observed for canola oil blend with no frying whereas the lowest value was recorded for the oil blend after 10th frying. On the other hand, the maximum scores were recorded oil blend after eighth frying whereas lowest scores were found for oil blend after 10th frying. Hence, only first two PCs provide the maximum variance in the spectral datasets taken after each frying. It can also be observed that up to fifth frying the stability of the oil blend was not affected as one can see the spectral points overlap each other, but after sixth and seventh frying, the changes were observed. However, the most changes were taken place after eighth frying onward as they lie away from each other. These variations in the spectral fingerprints can be observed from the loading plots for PC1 and PC2 as presented in Figure 3b. The loading plot for PC1 shows peaks in various regions of the spectral fingerprints that may responsible for this clear differentiation due to multiple frying of canola oil blend. The peaks were observed in 3600-3200 cm −1 , 2900-2700 cm −1 , 1500-1700 cm −1 , 1200-1000 cm −1 , and 700-500 cm −1 against PC1 and PC2 loadings. These peaks can be correlated to different structural changes taken place due to the frying operation that ultimately results in the differentiation of oil after multiple frying. It can be inferred from above discussion that PCA can differentiate the oil spectra taken after multiple frying.
F I G U R E 2 Fourier transform-infrared spectra for canola oil, optimized blend, and oil samples after each frying

| CON CLUS ION
The best canola oil blend was formulated with olive oil and palm olein which is usually more stable in frying as compared to pure single oil. Moreover, FT-IR spectral data in combination with PCA show clear differentiation for the oil samples acquired after multiple frying. Similarly, PLSR model was used to predict FFA, PV, and IV. The changes in chemical analyses and on FT-IR fingerprints during the process of multiple frying are mainly attributed to the hydrolysis of lipids, breakdown of double bond, and oxidation of oil due to high temperatures and the presence of oxygen. All these changes are important to consider the stability of oil that is ultimately important for the production of good quality fried product having least health concerns. Previously, these changes were governed using chemical-based standard methods that are laborious and time taking. The prediction of these parameters helps in the development of fast and noninvasive tools for regulatory bodies and industrial applications which may smoothen the way toward sensor development.

CO N FLI C T S O F I NTE R E S T
The authors declare that there are no conflicts of interest regarding the publication of this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
All the data that support the findings of this study are included in the manuscript.
TA B L E 3 PLS models applied on FT-IR preprocessed data for prediction of oil stability parameters after multiple frying