Using optical NIR sensor for on-line virgin olive oils characterization

Using optical NIR sensor for on-line virgin olive oils characterization

Sensors and Actuators B 107 (2005) 64–68 Using optical NIR sensor for on-line virgin olive oils characterization A. Jim´enez Marquez a, ∗ , A. Molina...

163KB Sizes 0 Downloads 21 Views

Sensors and Actuators B 107 (2005) 64–68

Using optical NIR sensor for on-line virgin olive oils characterization A. Jim´enez Marquez a, ∗ , A. Molina D´ıaz b , M.I. Pascual Reguera b a

Instituto Andaluz de Investigaci´on y Formaci´on Agraria, Pesquera, Alimentaria y de la Producci´on Ecol´ogica, Estaci´on de Olivicultura y Elaiotecnia CIFA ‘Venta del Llano’, PO Box 50, E-23620 Mengibar, Ja´en, Spain b Universidad de Ja´ en, Departamento de Qu´ımica, F´ısica y Anal´ıtica, E-23071 Ja´en, Spain Received 7 April 2004; received in revised form 24 October 2004; accepted 12 November 2004 Available online 19 February 2005

Abstract Near-infrared transmittance spectroscopy was applied to on-line control quality and characterization of virgin olive oils. The transmittance spectrum (log 1/T) was obtained through dispersal equipment, scanning the samples between 750 and 2500. A 1 mm optical path length flow cell and holding 120 ␮l was employed. Calibration models for: acidity value (AV), bitter taste (k225) and fatty acid composition (FAME), were previously developed in the laboratory using partial least squares (PLS) regression. A total of 190 virgin olive oil samples, gathered during the usual harvesting time of three olive crop seasons, were used. The validation set gave correlation coefficients and standard error of prediction of 0.999 and 0.35, 0.936 and 0.058, 0.998 and 0.604, 0.992 and 0.674% for AV, k225, fatty acid oleic and fatty acid linoleic, respectively. These PLS models, prior to a slope/bias correction, were used to monitor on-line values of these parameters during the processing of virgin olive oils in real olive mills. Samples of oils were obtained for chemical analysis. Application of NIR transmittance spectroscopy on-line to virgin olive oil production line, has allowed to detect, at real-time, the changes produced in the characteristics of the oils during their production. The results indicate similarity between information both NIR and reference laboratory methods. © 2005 Elsevier B.V. All rights reserved. Keywords: Near-infrared transmittance spectroscopy; On-line analysis; Virgin olive oil; Control quality

1. Introduction The virgin olive oil is obtained from olive fruits, Olea europaea L., by extraction using only physical methods: crushing of olives in a hammer crusher, mixing of olive paste in a mixer and separation of oil using presses or centrifuges. Before storage, the oil obtained is percolated or centrifuged for clarification. Depending on the state of the fruit and on the conditions of production, the obtained oil can be classified under different categories of quality [1]. In virgin olive oil extraction process, characterization and quality control of olive oil for its classification and ∗

Corresponding author. Fax: +34 953 374017. E-mail address: [email protected] (A.J. Marquez).

0925-4005/$ – see front matter © 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.snb.2004.11.103

separation, prior to its storage, have great importance. Acidity value (AV), bitterness (k225) and oleic–linoleic relation (MP), are analytical parameters important for this [2]. The acidity value measures free fatty acids according to its hydrolytic deterioration, oils with acidity >2% cannot be consumed directly and must be refined. k225 is a chemical index of oil bitterness. High bitterness is not well accepted by the consumer. MP is estimated on the basis of fatty acids profile and is calculated by: summa of monounsaturated fatty acids percentage/summa of polyunsaturated fatty acid percentage. Oleic and linoleum fatty acids are the most important ones. Resistance to authorisation is directly related. High values of MP indicate oils with high stability.

A.J. Marquez et al. / Sensors and Actuators B 107 (2005) 64–68

Therefore, the determination of these parameters, through conventional analytical methods is carried out in a chemical laboratory. In order to carry out an effective control and classification of olive oils, at the time that they are obtained, it is necessary the instant measurement of all these parameters and, of course, without using chemical products. Near-infrared reflectance spectroscopy has been recognised as a powerful analytical technique for rapid determination of various constituents in food [3], it is non-destructive, low-cost and provides a safe working environment. It is also a technique suitable for on-line work. In the last year, application of NIR to oils and fats has become more popular for quality and composition studies. In olive oils, NIR is used for prediction–identification of adulteration [4], differentiation–classification of vegetable oils [5] and monitoring on-line carotenoid and chlorophyll pigments [6]. The aim of this work has been to test the feasibility of the optical NIR sensor, in transmittance method, for the monitoring of the changes in olive oils during their production in olive-mill, in order to evaluate its utilization as a potential on-line method for characterization and quality control.

2. Experimental 2.1. Samples A total of 190 virgin olive oil samples have been used for building the calibration–validation NIR models. The 64% of these samples are from ‘Picual’ cultivar, gathered during the usual harvesting time (November–March) of three olive crop seasons; the rest of the samples (36%) are from 23 different cultivars of olive oil in a unique harvesting state (December) and along the three olive crop seasons. This sampling provides a source for the incorporation in the NIR models of variability varietal, seasonal and technical. The oils were obtained by a continuous industrial system and taken directly in the last step of the production process, after cleaner vertical centrifuge and before being sent to the rinsing tanks. The samples, non-filtered, were stored at 15 ◦ C up to their chemical and spectroscopic analyses. 2.2. Reference methods Determination of acidity value and fatty acid composition (FAME), were carried out according to the analytical methods described in Regulation EEC/2568/91 of the European Union Commission [7]. Bitter taste was determined by extraction in solid phase with octadecyl (C18 ) packing for the isolation of hydrophobic species. Oil dissolved in hexane is put into column and the polyphenols fraction recovery with methanol:water (7:3) and measured in a spectrophotometer instrument at 225 nm [8].

65

Ten oil samples showing different values on these parameters were analysed twice in order to obtain the laboratory standard error values, S.E.L. [9]. 2.3. Optical NIR equipment The transmittance spectra (log 1/T) in the near-infrared have been obtained by a dispersal equipment, model V-570 by Jasco Ltd., scanning the samples between 750 and 2500 nm at a speed of 3000 nm/min and an interval of 2 nm. A 1 mm optical path length flow cell with an internal volume of 120 ␮l (Hellman 170.IQS) was used. The samples, non-filtered, were set in the flow cell using a peristaltic pump. This system allows for the analysing equipment to be adapted to the line process easily, since with the help of the pump a by-pass is created, which pumps the oil sample from the main oil current and gives it back once it has been scanned. 2.4. NIR calibration All the chemometrical analysis has been performed by means of the program Umscrambler 7.1 (Camo Casa). The prediction models have been achieved by application of the regression by PLS. In order to get the optimisation of the models, different levels of smoothing and mathematical treatments of first and second derivative over signal were tested. The best model was chosen on the basis of the greatest coefficient of determination, R2 and the lowest standard error of calibration, S.E.C. The wavelength region for the development of PLS models was selected by means of a graphical analysis of β-coefficients and the optimal number of PLS factors were estimated from cross-validation analysis. The lowest values of predicted residual error sum of square (PRESS) define the fit number of factors to be used in the PLS models [3,6]. The prediction PLS models were validated with the validation set. Linear correlation coefficient (r) and standard error of prediction (S.E.P.) into NIR predicted versus laboratory reference were examined. Univariate slope/bias correction of the NIR predicted values [10] were used to translate the laboratory PLS model to industrial PLS model. During olive oil production, 10 oil samples were collected and analysed by reference methods. Linear relationship NIR prediction versus reference methods were tested, NIR prediction = bias + slope (reference method) and slope/bias corrected if necessary to obtain similar values. 2.5. On-line measurement NIR equipment has been installed on virgin olive oil production line, in order to take samples to the exit of the clarifying centrifugal and from the hopper of the oil continuous weighed, as shown in Fig. 1. Through the system of constant production of the experimental olive-oil mill, a total of 6000 kg of olives were ground

66

A.J. Marquez et al. / Sensors and Actuators B 107 (2005) 64–68

3. Result and discussion Table 1 shows the chemical characterization of oil samples used for calibration of the PLS models and their validation. The number of samples, the mean, standard deviation and range for both calibration and validation set are indicated for each parameter. The oils used in this work allowed us to obtain an adequate degree of variability for acidity value (0.12–15.14% oleic acid), bitter taste (0.06–0.66 k225), oleic fatty acid (55.04–83.58%) and linoleic fatty acid (2.26–20.25%). Shenk and Westerhaus [12] indicate that the statistical tests to determine NIR spectroscopy equation acceptance are the residual standard deviation, named as standard error of calibration (S.E.C.) and the coefficient of multiple determination (R2 ). Application of a PLS analysis to the calibration set gives calibration equations models for these parameters with a low error standard of calibration (S.E.C.) and high coefficients of determination (R2 > 0.85). A smoothing, with a nine points signal, previous and posterior to a three points first derivative on spectral ranges 1100–2500 and 1000–2500 nm for acidity value, bitterness and oleic–linoleic fatty acid, respectively, was the best math treatment applied. On validation of these selected PLS models a high lineal correlation, r > 0.90, is obtained for each parameter. In Table 2 results of calibration and validation from PLS models are shown. The standard deviation for the residuals due to differences between reference values and the NIR predicted or standard error of performance (S.E.P.) obtained indicates that prediction PLS models developed for acidity value, oleic and linoleic fatty acid are excellent and acceptable for k225. The standard error of prediction (S.E.P.) is similar to standard error obtained with the reference methods on laboratory

Fig. 1. Schematic diagram of localization of the sensor NIR in the last pass in the olive oil extraction process: (1) oil from horizontal centrifuge decanter; (2) vertical centrifuge for oil clarification; (3) tank for oil sedimentation; (4) continuous oil-weigher; (5) oil to storage; (6) NIR equipment.

for 6 h. This included one batch of 2000 kg of Arbequina variety, followed by 4000 kg of Picual variety from two different batches, olive-tree and olive-soil, with 2000 kg each one. During the process, oil samples were taken for analysis in the laboratory, every 20 min approximately. Each sample taken was made up, at the same time, of three partial oil samples carried out for ten minutes. Simultaneously a spectrum NIR has been carried out, which was made to coincide with every partial sample, so each sample is represented by three NIR spectra, using the average spectrum for NIR prediction. The samples were immediately analysed in the laboratory, previous paper filtration, using the reference methods. A paired t-test [11] for comparison of means NIR prediction versus reference values was applied at each batch. Every batch is constituted by three olive oil samples taken consecutively every 20 min, avoiding the moments of transition between types of oils. A texp less than ttheo , at the P = 0.05 level, indicates that the averages obtained by both methods are not significantly different.

Table 1 Chemical characterization of calibration and validation set of the oil samples used for building PLS models Calibration set

Acidity value (% of oleic acid) Bitterness (k225) Oleic fatty acid (% of normalized area) Linoleic fatty acid (% of normalized area)

Validation set

Number of samples

Mean

S.D.

Minimum

Maximum

Number of samples

Mean

S.D.

Minimum

Maximum

131 149 160 160

1.25 0.208 75.65 8.46

2.275 0.132 7.05 5.19

0.12 0.06 55.04 2.26

15.14 0.66 83.58 20.25

45 30 35 35

2.38 0.276 72.20 5.31

2.739 0.155 8.63 5.310

0.16 0.07 55.83 2.50

12.22 0.61 83.60 20.00

S.D.: standard deviation. Table 2 Results of calibration and validation of the PLS models selected Calibration

Acidity value (% of oleic acid) Bitterness (k225) Oleic fatty acid (% of normalized area) Linoleic fatty acid (% of normalized area)

Validation

Number of factors

Spectral range (nm)

R2

15 13 8 9

1100–2500 1100–2500 1000–2500 1000–2500

0.998 0.870 0.990 0.991

S.E.C.

r

S.E.P.

S.E.L.

0.12 0.048 0.69 0.42

0.999 0.936 0.998 0.992

0.16 0.058 0.66 0.68

0.10 0.026 0.21 0.15

Number of PLS factors and NIR spectral range selected. R2 , coefficient of determination; S.E.C., standard error of calibration; r, coefficient of correlation lineal; S.E.P., standard error of prediction; S.E.L., standard error of laboratory methods.

A.J. Marquez et al. / Sensors and Actuators B 107 (2005) 64–68

(S.E.L.) with a Gaussian distribution of errors, so that 80% of the residuals are within of ±S.E.P. for acidity value, oleic and linoleic fatty acid and 70% for k225. RPD (relative prediction deviation) expresses the relation between the standard deviation of the references values on validation set (S.D.) and the standard error of the prediction (S.E.P.). For acidity value, oleic and linoleic fatty acid, values of RPD are higher than 3; for k225 the RPD value is near to this value, suggested by Williams and Sobewring [13]. This indicates that these PLS models are sufficiently accurate in the ranges checked.

67

The evaluation of these PLS models in on-line analysis is carried out by comparing the NIR prediction from these models versus the values obtained in the laboratory, from the monitoring samples during olive oil production. Previously to NIR data acquisition, an adjustment slope/bias was carried out. Only offset correction (agreement bias) was necessary for acidity value, oleic and linoleic fatty acids. Correction of the bias and slope was necessary to adjust the NIR prediction to the reference values for bitter taste. In Fig. 2 the results obtained in an on-line monitoring value NIR versus the laboratory dates, on 6 h of work, are showed,

Fig. 2. Evolution of bitterness, in specific extinction at 225 nm (a), acidity value, in percentage of oleic acid (b), linoleic fatty acid, in percentage of normalized area (c) and oleic fatty acid, in percentage of normalized area (d), in an on-line monitoring values NIR () vs. laboratory values (). The shaded area shows the transition among types of oils (mixing of oils in tank 3 in Fig. 1. Every point of the curve is a sample taken every 20 min. Evolution of olive oil process in min.

68

A.J. Marquez et al. / Sensors and Actuators B 107 (2005) 64–68

every point of the curve is a sample taken every 20 min from procedure indicated in material and methods. They can be appreciated so that evolution of these four parameters during the olive oil production process presents similar curves using either method, NIR prediction and laboratory values. Changes in acidity values, when transition olive-tree to olive-soil was accomplished, are reflected clearly. The average results obtained in each batch with the laboratory methods and with NIR prediction are similar. During ‘Arbequina’ olives processing the acidity average was 0.49 ± 0.01 and 0.54 ± 0.15 for laboratory and NIR prediction, respectively. In the ‘Picual’ olives processing the change of the batch-tree, with acidity average of 0.33 ± 0.00, to the batch-soil, with acidity average of 2.77 ± 0.06, from laboratory values, was monitored by NIR equipment that provides the average values of 0.37 ± 0.16 and 3.03 ± 0.12 for tree and soil, respectively. Also, changes in the fatty acids profile can be monitored clearly. Transition on ‘Arbequina’ oils to ‘Picual’ oils can be appreciated through oleic and linoleic fatty acids. Laboratory reference proportionate average values of 72.49 ± 0.12 and 10.36 ± 0.12 versus 73.40 ± 0.63 and 9.91 ± 0.37 from NIR, for oleic and linoleic fatty acid, respectively. Monitorization of bitter taste can be carried out satisfactorily. Although the oils come from ripe olives, and their k225 values do not produce a high yield, we can clearly appreciate the changes. Average values of 0.10 ± 0.01, 0.20 ± 0.01 and 0.10 ± 0.00 from laboratory reference and 0.14 ± 0.00, 0.17 ± 0.03 and 0.10 ± 0.02 from NIR, were obtained for ‘Arbequina’ olive-tree, ‘Picual’ olive-tree and ‘Picual’ olive-soil, respectively. A paired t-test for these means was applied. The tabulated value of tther , at P = 0.05 and three degrees of freedom is 2.353. The calculated values of tcal are less than 5% ttheor , which indicates not significant differences between the reference values and NIR prediction on-line. In conclusion, the results indicate that by using an optical NIR sensor, through near-infrared transmittance PLS models, we can provide a rapid and non-pollutant means to on-line simultaneous monitoring of acidity, oleic, linoleic and bitterness levels during virgin olive oil production.

[2] M. Uceda, M. Hermoso, in: D. Barranco, R. Fern´andez, L. Rallo (Eds.), El cultivo del olivo, 2nd ed., Mundiprensa, Madrid, Spain, 1997, pp. 547–572 (Chapter 18). [3] B.G. Osborne, T. Fearn, T. Hindle, Practical NIR Spectroscopy with Applications in Food and Beverage Analysis, Longman Scientific and Technical, Harlow, UK, 1993. [4] I.J. Wesley, R.J. Barnes, A. McGill, Measurement of adulteration of olive oils by near infrared spectroscopy, J. Am. Oil Chem. Soc. 3 (1995) 289–298. [5] E. Bertran, M. Blanco, J. Coello, H. Iturriaga, S. Maspoch, I. Monteliu, Near infrared spectrometry and pattern recognition as screening methods for the authentication of virgin olive oils of very close geographical origins, J. Near Infrared Spectrosc. 8 (2000) 45–52. [6] A. Jim´enez, Monitoring carotenoid and chlorophyll pigment in virgin olive oil by visible–near infrared transmittance spectroscopy. Application on-line, J. Near Infrared Spectrosc. 11 (2003) 219–226. [7] European Union Commission, Regulations ECC/2568/91, On the characteristics of olive oil and olive pomace oils and on their analytical methods, J. Eur. Commun. L248 (1991). [8] F. Gutierrez Rosales, S. Perdiguero, R. Gutierrez, J.M. Ol´ıas, Evaluation of the bitter taste in virgin olive oil, J. Am. Oil Chem. Soc. 69 (4) (1992) 394–395. [9] J.J. Workman, in: D.A. Burns, E.W. Ciurczak (Eds.), Handbook of Near Infrared Analysis, Marcel Dekker, New York, 1992, pp. 247–270. [10] E. Bouveresse, C. Hartmann, D.L. Massart, I.R. Last, K.A. Prebble, Standardization of near infrared spectrometric instruments, Anal. Chem. 68 (1996) 982–990. [11] E. Morgan, in: ACOL (Eds.), Chemometrics Experimental Design, Wiley, Chichester, 1991. [12] J.S. Shenk, M.O. Westerhaus, in: A.M.C. Davies, P.C. Williams (Eds.), Near Infrared Spectroscopy. The Future Waves, NIR Publication, Chichester, UK, 1996, pp. 198–202, Calibration the ISI way. [13] P.C. Williams, D.C. Sobewring, in: A.M.C. Davies, P.C. Williams (Eds.), Near Infrared Spectroscopy. The Future Waves, NIR Publication, Chichester, UK, 1996, pp. 185–188, How do we do it: a brief summary of the methods we use in developing near infrared calibration.

Acknowledgements

A. Molina D´ıaz is Analytical Chemistry professor in the Departament of Physical and Analytical Chemistry from the University of Ja´en (Andalusia, Spain). He received his PhD from the University of Granada in 1982. He is the head of the research group “Analytical Chemistry from the University of Ja´en” and has published over 100 papers and reports concerning different topics, especially in the field of the flow-through spectroscopic sensors, which is his main research topic.

This paper has been supported by FAGA-FEOGA: program for the improvement in the olive oil production quality (project CAO 00-L5). This is a work of Ph.D. thesis by A. Jim´enez.

References [1] D. Boskou, Olive Oil, Chemistry and Technology, AOCS Press, Champaign, IL, USA, 1996, p.12.

Biographies A. Jim´enez Marquez received his PhD in Chemistry from the University of Ja´en in 2003. His PhD thesis focused in application of Near Infrared Spectroscopy for on-line virgin olive oil characterization. Currently, he is researcher of from I.F.A.P.A. (Andalusia, Spain) and he works in extraction process and application of non-destructives techniques for real time control of production and quality of virgin olive oil.

M.I. Pascual Reguera is PhD in Chemistry from the University of Granada in 1987. Currently she is a Lecturer in Chemistry in the Department of Physical and analytical Chemistry from the University of Ja´en (Andalusia, Spain). Main research interest: pesticides, olive oil, sensor flow-through optosensors applied to pharmaceutical analysis, GC-MS, HPLC-MS.