2. Materials and methods2.1. Materi

2. Materials and methods
2.1. Materials and apparatus

Six kinds of edible oils, including corn oil (CO), sunflower oil (SFO), rapeseed oil (RO), peanut oil (PO), soybean oil (SBO) and sesame oil (SO), were bought from Trust-Mart supermarket in Chengdu, China. A UV–Vis spectrophotometer (UV-2100, Beijing Rayleigh Co., Ltd., China) was employed for spectral measurement. A thermostatic magnetic stirrer (DF-101S, Yuhua Instrument Co., Ltd., China) was used to heat oil samples. A micro vortex apparatus (WH-2, Shanghai Huxi Analytical Instrument Factory Co., Ltd., China) and a 0.0001 g precision electronic balance (AR2140, Sartorius Co. Ltd., Germany) were used. All the chemicals and reagents used were of analytical grade.

2.2. Sample preparation

Aliquots (50 mL) of edible oils (CO, SFO, RO, PO, SBO, SO) were placed in a round bottom flask and heated at 180 °C (Gonzaga & Pasquini, 2006) using the thermostatic magnetic stirrer for 20, 40 min, 1, 3, 5, 7 and 10 h, respectively. For each kind of oil, 64 samples were obtained, including 1 unheated sample, 7 heated samples and 56 mixed samples prepared by mixing accurate amounts (4 digit significant figures after the decimal point) of each kind of oil and its heated oils of different heating time in proportions of 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70% (w/w), respectively, and shaking vigorously using the vortex apparatus to ensure complete homogenization.

2.3. Determination of AV of unmixed samples

Determination of AV of unmixed samples was carried out according to the standard method described in Method for analysis of hygienic standard of edible oils of the People’s Republic of China (Ministry of Health of the People’s Republic of China and Standardization Administration of the People’s Republic of China, 2003). The AV of mixed samples was calculated based on the proportion of unheated oils to heated oils.

2.4. Spectral acquisition

The spectra of samples were measured in a 1 cm path length quartz cell by the UV–Vis spectrometer, against an empty quartz cell, in the region of 320–800 nm. Both the original spectral data and the first derivative spectral data of each sample were recorded. Each sample was measured three times and the average of them was used as the data of this sample in the next calculation process.

2.5. Model building and evaluation

Principle component regression (PCR) and partial least squares regression (PLS) are powerful multivariate statistical tools that have been successfully and widely applied to the quantitative analysis of spectral data (Hemmateenejad, Akhond, & Samari, 2007). PCR can compress high-dimensional data into a lower-dimensional space, thus, making data more comprehensible by extracting useful information with a minimal loss of information. The PCR may be outlined as follows: (1) spectral matrix X is decomposed into score matrix T (referring to the principal component) and load matrix P by PCA in order to eliminate useless noise information; (2) use the front f score vectors of T to form a new matrix Tf, which contains most information of the original variables; (3) perform multi linear regression (MLR) by using concentration matrix Y and Tf to obtain the PCR model. PLS is also a linear algorithm for modeling the relation between two datasets ( Zheng & Lu, 2011), in which not only the spectral matrix X but also the concentration matrix Y are decomposed to reduce useless noise information. That is the biggest difference between PLS and PCR. More details about PLS were described in the paper written by Mehmood, Liland, Snipen, and Sæbø (2012).

The 64 samples of each kind of oil were split into a calibration set (n = 48) and a prediction set (n = 16) with the principle of uniform distribution of AV. PLS and PCR were used to establish calibration models. Model optimization was carried out in three aspects: optimum wavelength selection, spectral-pretreatment and outlier exclusion. All the data processing was implemented via the Unscrambler ver 9.7 (CAMO PROCESS AS, Oslo, Norway) and MATLAB 7.0 (The Math Works, Natick, USA).

The performance of each model was evaluated by determination coefficient (R2) and root mean square error (RMSE) of calibration, cross-validation and prediction, respectively. Generally, a good model should have high values of R2 and low values of RMSE. R2 and RMSE were calculated as follows:


Turn MathJax on


Turn MathJax on

where and yi are the calculated and measured acid values of sample i, respectively. represents the mean of all the measured acid values and n is the number of samples in a calibration set or a prediction set.

ผลลัพธ์ (อังกฤษ) 2:

2. Materials and methods
2.1. Materials and Apparatus Six kinds of edible oils, including corn Oil (CO), Sunflower Oil (SFO), rapeseed Oil (RO), Peanut Oil (PO), soybean Oil (SBO) and Sesame Oil (SO), were Bought from Trust. -Mart supermarket in Chengdu, China. A UV-Vis spectrophotometer (UV-2100, Beijing Rayleigh Co., Ltd., China) was employed for spectral measurement. A thermostatic magnetic stirrer (DF-101S, Yuhua Instrument Co., Ltd., China) was used to heat oil samples. A micro vortex apparatus (WH-2, Shanghai Huxi Analytical Instrument Factory Co., Ltd., China) and a 0.0001 g precision electronic balance (AR2140, Sartorius Co. Ltd., Germany) were used. All the chemicals and reagents used were of analytical grade. 2.2. Sample Preparation aliquots (50 mL) of edible oils (CO, SFO, RO, PO, SBO, SO) were Placed in a round bottom Flask and heated at 180 ° C (Gonzaga & Pasquini, two thousand and six) using the thermostatic Magnetic stirrer for 20. , 40 min, 1, 3, 5, 7 and 10 h, respectively. For each kind of oil, 64 samples were obtained, including 1 unheated sample, 7 heated samples and 56 mixed samples prepared by mixing accurate amounts (4 digit significant figures after the decimal point) of each kind of oil and its heated oils of different heating. time in proportions of 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70% (W / W), respectively, and using Vigorously shaking the Vortex Apparatus to ensure Complete homogenization. 2.3. Determination of AV of Unmixed samples Determination of AV of Unmixed samples was carried out according to the standard method described in Method for Analysis of hygienic standard of edible oils of the People's Republic of China (Ministry of Health of the People's Republic of China and Standardization Administration. of the People's Republic of China, 2003). The AV of mixed samples was calculated based on the proportion of unheated oils to heated oils. 2.4. Spectral Acquisition of Spectra The samples were measured in a 1 cm path UV-Vis Length Quartz Cell by the spectrometer, against an Empty Cell Quartz, in the Region of three hundred twenty to eight hundred NM. Both the original spectral data and the first derivative spectral data of each sample were recorded. Three times each sample was measured and the average of them was used as the Data of this sample in the next Calculation Process. 2.5. Model Building and evaluation Principle Component regression (PCR) and partial Least Squares regression (PLS) are powerful multivariate Statistical Tools that have been successfully and widely Applied to the Analysis of quantitative spectral Data (Hemmateenejad, Akhond, & Samari, in 2007). PCR can compress high-dimensional data into a lower-dimensional space, thus, making data more comprehensible by extracting useful information with a minimal loss of information. The PCR may be outlined as follows: (1) spectral matrix X is decomposed into score matrix T (referring to the principal component) and load matrix P by PCA in order to eliminate useless noise information; (2) use the front f score vectors of T to form a new matrix Tf, which contains most information of the original variables; (3) perform multi linear regression (MLR) by using concentration matrix Y and Tf to obtain the PCR model. PLS is also a linear algorithm for modeling the relation between two datasets (Zheng & Lu, 2011), in which not only the spectral matrix X but also the concentration matrix Y are decomposed to reduce useless noise information. That is the biggest difference between PLS and PCR. More Details About PLS were described in the Paper written by Mehmood, Liland, Snipen, and Sæbø (2 012). The 64 samples of each Kind of Oil were Split Into a calibration SET (n = 48) and a prediction SET (n = 16. ) with the principle of uniform distribution of AV. PLS and PCR were used to establish calibration models. Model optimization was carried out in three aspects: optimum wavelength selection, spectral-pretreatment and outlier exclusion. All the Data Processing was implemented via the Unscrambler Ver 9.7 (CAMO PROCESS AS, Oslo, Norway) and matlab 7.0 (The Math Works, Natick, USA). The Performance of each Model was evaluated by determination coefficient (R2) and root Mean square. error (RMSE) of calibration, cross-validation and prediction, respectively. Generally, a good model should have high values ​​of R2 and low values ​​of RMSE. R2 and RMSE were calculated as follows: Turn on MathJax MathJax Turn on where and Yi are the calculated and measured acid values ​​of sample I, respectively. represents the mean of all the measured acid values ​​and n is the number of samples in a calibration set or a prediction set.