Processing

AR INFRARED (NIR) ANALYSIS IN THE SUGARCANE FACTORY

M.A. Clarke1, E.R. Arias2 and C. McDonald-Lewis3 I a

Sugalt Processing Research Institute, Inc., 1100 Robert E. Lee Blvd. 'New Orleans, Louisiana 70124, USA

Sugar Cane Growers Cooperative of Florida, P.O. Box 666 I Belle Glade, Florida 33430, USA

NIRSystems, Inc., 12101 Tech Road, Silver Spring, Maryland 20904, USA

ABSTRACT

Near infrared (NIR) analysis has been finding widespread application in beet sugar manufacture, in 'sugarcane breeding programs, and in analysis of sugar products in Europe, the USA and Australia. Experience with the use of a scan- ning NIR spectrophotometer in a sugarcane factory laboratory is reported. Previ- ous work has described use of filter instruments; these scanning instruments offers wavelength selectivity, wavelength combinations and accuracy far greater than those previouslyb available. Juice samples (grower samples) were analyzed at different ppiods during the crop, and in different geographical locations. Results from this very rapid analysis compared favorably with the factory's pol, Brix and purity analysis1 Analysis of bagasse for moisture and sugar is also reported. Applications to analysis of raw sugars for pol, sucrose, invert, ash, colof, 'dextran and starch are also presented and discussed.

DUCTION

lysis is used for quality control and product evaluation in the grain, cereal, forage and fiber industries. It8 is finding increasing application in quality control of, chemicals, pharmaceuticals and food products, including sugar containing foods andrdry mixture (Osborne7) (Williams13, Williams14). The possibilities of near infrared analysis have been studied in several areas of the sugar industry; there are current applications in several areas in both the cane and

I beet industries. Early studies were made on whole sugarcane, shredded, by

I Sverzutg, who found'satisfactory correlation of NIR analyses. with standard labo- 1 I

ratory analyses for'pol, fiber, sugar and moisture. Wood and coworkers in South I Africa found NIR analysis for nitrogen in sugarcane leaves to be rapid and

accurate (Wood15). In studies on analysis of Brix, pol and sucrose in sugarcane juice,^ however, they concluded that, although NIR was rapid and required no sample clarification, it was not.accurate enough for payment purposes (Meye$). Samples were filtered but not otherwise treated. The same group tried direct NIR

829

I PROCESSING

analysis of shredded cane for sucrose, dry matter and fiber, but found results not as satisfactory as those for juice analysis. All samples were run in reflection mode, using multiple filter intruments.

A study at the Australian Bureau of Sugar Experiment Stations used a 19-filter instrument, in reflectance mode, on sugarcane juice and,wholei fibrated cane (Berding2). The purpose was to develop >rapid analyses for Brix, pol, purity, recoverable sugar fiber, moisture and' conductivity in cane breeding selection studies. Problems with sample homogeneity in fibrated cane, and temperature of flow-through cell for juice, were encountered, but data were sufficiently~promis- ing to expand the study and acquire scanning NIR equipment.

Another Australian study (Amesl) on raw sugar used a 20-filter instrument to measure pol, moisture, ash, reducing sugars in raw sugars. Calibrations were developed in an analytical laboratory and used on-site at a factory. Results were satisfactory for process control; but the~unresolved question of bias led the au- thors to conclude that the analyses could not yet be used for payment. Similar conclusion were drawn in a subsequent paper from the same group (Nguyen6) on further raw sugar studies using a slow scanning instrument or a filter instrument. Studies on molasses and fermentation broths for sugar showed promise. The authors pointed out that error in calibration of raw sugars~from a single mill was less than error in an overall laboratory calibration. ,

$

A successful application of the NIR technique to analyses for moisture, starch and invert sugar in powdered and agglomerated sugars was reported by Stevens8. The slight increase in error was more than compensated,by the time and labor savings over the tedious traditional methods.

It is interesting to observe that in whole shredded sugarcane studies, only the work of Sverzutg found satisfactory data. Sverzut's study was performed at the Southern Regional Research Center, (site of the senior author's laboratory) using

, i

established NIR analysis for on-line process control on raw, thin and thick juices in the beet factory. Vaccari efi al.1° found, encouragingly, that calibrations set up for one factory could be used for others. )A filter instrument (20 filters) in reflect- ance mode was used. Although Vacoari*et a1, l2 used a smaller numb,er of sam- ples than other workers to set up calibration curves; the standard deviations,were comparable to those of laboratory measurements) , -

830

M.A. CLARKE ET AL.

In France, Burzawa et aL3 using a 9-filter near infra-red instrument in reflectance mode, have developed a satisfactory on-line control for Brix, pol and purity in beet factories (~urzawa~p~).

MATERIALS AND METHODS I

Bonds within the sucrose molecule were examined using a saturated sucrose I

solution and found to display unique absorptions at 2,088 nm and 2,272 nm, with HOH group (water) absorbing at 1,435 nm and 1,940 nm.

t

An ,NIRSystems Modek 6500 !scanning spectrophotometer (NIRSystems, Inc., Silver Spring, MD) was psed for the NIR work. This instrument scans from 400 to 2,500,nm7 using a silicon detector for wavelengths between 400 and 1,000 nm and 2 lead sulfide detector between 1,100 and 2,500 nm. All samples were scanned over both spectral ranges. The samples were scanned at an average of 32 qcans per minute, and the aveiage analysis time was 30 seconds. NIRSystems 'fNSAS" NIR software was used for all calibration and prediction work. Calibra- tions were developed using "conventional" NIR technology procedures, which ipclude selection of optimum wavelengths for log 1/R and first and second derivatives thereof by forward stkpwise regression.

Ninety-eight samples of unfiltered sugarcane juice were collected from the Sugar Cqne Growers Cooperative of Florida Factory in Belle Glade, Florida. All sam- ples were used to develop calibration equations for Brix, purity values and po- larization. An independent set of juice samples were obtained from the USDA Sugarcane Juice Laboratory, Ardoyne Farms, Louisiana, and used to test the validity of the developed equations. This paper will review the concept for devel- oping universal calibrations to predict Brix, pol, and purity in unfiltered cane juice derived from the two different geographical1 locations in Florida and Loui- siana. In addition, the sample presentation method, including calibration devel- opment,~ validation, and statistical interpretation will also be discussed. No sam- .pleatreatment, filtration, or chemical clarification was required. Unfiltered cane juice, -fromf,the normal laboratory growers samples, was pipetted (micropipet with rubber bulb) into a 1 mm quartz cuvette cell and the NIR spectrum recorded ~fromi1,100 to 2,500 nm using NIRSystems Model 6,500 Spectrophotometer. All standard laboratory analyses (for pol, Brix and purity) were run by factory or USDA personnel in their normal manner. Pols were run at the Belle Glade Faotory using a Membrex filtration and no chemical treatment on the juice sam- ples. Pols were run at the USDA lab using a calcium hydroxide-aluminum choride chemical clarification of cane juice. , I ' , i'

x r ;

831

', , PROCESSING

RESULTS AND DISCUSSION A

I * ) I * , * , ( 8 - t

The absorbance (log 1/T) spectra for three varying concentrations and a,saturated sucrose spectrum are shown in Figure 1.

21 12 10 56

SATURAIED SUCROSE S O C U ~ ION '

04 J t

1100 1300 , 1500 1700 1900 : O ? 2309 2500 WAVELENGTH (NANOHE TERI , > [

FIGURE 1. Cane juice and saturated sucrose. .. , \ , , , , s U > ~ * I ,

t , $ 1 ' ( 1 1 , ; I / I

L l I I

i

' I I I i , I (

The absorption bands in the NIR region are quite broadland overlapping, making it difficult to determine the optimum spectral bands for weak absorbing analytes, such as sucrose. By converting the spectra to the second derivative, shown in Figure 2, the overlapping weaker absorbers are separated and the peak intensities of these absorbers are enhanced, providing spectral bands which ma~ ,~be ' as- signed to each absorbing analyte. Note that the peak maxima are cofiverted to minima. I 9 A i s ! )

1 a , * 2 $0 >

To isolate and identify the sucrose bands, 'a saturated sucrose solution,was scanned and overlaid with cane juice samples of varyihg concentrations. Three distinct absorption bands were displayed in the combination region of the NIR spectrum, 2090,2272, and 2380; with 2272 displaying the strongest absorption band, where the long chain C-H atoms absorb. Figure 3 isolates the 2272 band. This band was selected to monitor the sucrose content in the unfiltered cane juce.

- a - - p - p p lr- M.A. CLARKE ET AL.

/ I

55 . w , > I , , ! ( 3 - I >,, I I I -

Z 4

m a 0 8RlX 9OL ARl lA l ION wl 16 76 13 61 m <

67 18 92 ! 5 86 21 12 :0 56 I

I SATURATED SUCROSE SOLUTION

i .,' >(,I<

I

FIGURE 2. Cane juice and saturated sucrose.

' ( I I . , , I , L

FIGURE 3. Cane Juice and saturated sucrose.

PROCESSING

SATISTICAL RESULTS

Polarization

Linear least squares regressions were performed using the s spectra and the reported wet chemical pol values (sucrose) for regression results are shown in Table 1. The y-intercept K(0) and slope K(l) are shown along with Multiple R(MR), otherwise known :as the correlation coeffi- cient; this parameter explains the amount of variation in a data set. Example: A good correlation may equal -'I or +1 multiplied by 100 = 100% variation. Using

data.

TABLE 1. Calibration of pol. '

File name: FLORIDA Regression Results Instrument: 6500 Standard error .18$ ' Number of spectra: 98 I Multiple R = - .9909 Constituent: 2, POL Math: 2nd derivative Constants :Wavelength Simple

Segment: 20 K(0) = 11.465 Gap: 0 K(l) = - 259.333 2274 - .991

The standard error of the calibration (SEC) is the primary paramher fo$ evaluat- ing the accuracy of the NIR method. : , '

This parameter describes the total residual (Lab-NIR) error in \he ineasurement, and is comparable to the standard deviation of the reference meihod, (standard wet lab results). Emphasis is first placed on evaluating the SEC and se=ondly, on the multiple R, which is a range dependent variable and does not indicate total performance of a given calibration. \ '

, 1:

Comparison between the laboratory results vs. NIR calculated percents showed five samples with large residuals greater than two standard deviations of differ- ence; these samples were not deleted but kept in the calibration. These differ- ences may often be attributed to errors in the wet chemical method or in sample presentation to the instrument. Overall calibration results were ac and compared favorably with lab results.

A plot of the NIR vs. the la shown in Figure 4.

M.A. CLARKE ET AL.

LA8 REPORTEO

FIGURE 4. Calibration for pol in unfiltered cane juice.

Brix

Similar studies were performed using refractometric Brix values. Table 2 dis- plays the regression reylts. A 1inear.summation of two wavelengths, 2276 and 1428, were selected: 2276 corresponding to C-H atoms (sucrose) and 1428 HOH groups for water and used to monitor solids (Brix). A standard error of .212 and a multiple correlation coefficient of .99 was achieved.

TABLE 2. Calibration of Brix.

I / I / ' I

~ i l e name: : I , FLORIDA Regression Results , ,

Instrument: , , 6500 Standard error .212 , NUmhc=r nf m e ~ t r o . 97 J V l v l p y u u r l u .

1 i , : j I Multiple R = .9863

---.-Lituent: 1, ' . BRIX Math: 2id derivative Constants Wavelength Simple

Segment: 20 K(0) = 15.684

% i 1 PROCESSING

A tabular comparison between the lab and NIR data showed seven samples ranging from 18.66 to 20.72 with large residuals. These samples were not deleted. A 45" plot of the NIR calculated versus the lab reported is displayed in Figure 5. AH samples were very close to the 45" line, indicating a bood fit of the data set.

16 0 . , (>. ( , r ~ ; ,\!% ,, c ' , I

16 00 1 7 SO "' 19 00 ,'' q 2 0 5 0 ' ' " 2 2 0 0 , . * I < ;

LAB\ REPORTED I , , s > , i j i t , , t

' , % * I . t ,: \;: 4 8 7 i

FIGURE 5. Calibration for Brix intunfiltered cane juice. I \ I b

Purity

In addition to pol and Brix, a calibration was developqd to determine the feasibi; 1 , 1 ' 9

lity of using NIR to predict purity in unfiltered cane.'A ratio of two wavelengths were used to develop the linear least squares results; 230411782, botli' yeaks absorbing the C-H region to the NIR spectrum., The stakdard 'eribr displayed results of .786 and a multiple R of -.9199. ~esu l t s are displayid in ~ i b l e 3.

' ? ( I

Overall statistical rcsults were acceptable. I. '

q , 2 1 , 4&',

M.A. CLARKE ET AL.

TABLE 3. Calibration of purity.

File'name: , FLORIDA) Regression Results ,. Instrument: I I 6500 Standard error .786

I 'Number !of spectra: 981 4 Multiple R = - ,9199 % Constituent: 3, t PURITY

i,i. Math: 2nd derivative Constants Wavelength Simple Segment: 1 20 K(0) = 8.856

, Gap:' i 0 K(l) = -47.175 230411782 -- -.920

VALIDATION (PREDICTION)

To evaluate the overall performance of each calibration equation, an independent sCt bf la6 analyzed simples were uked as a validatioh set; these samples were the $kt of Lbuisiana cane juice sahples. Table 4 displays a summary of the results.

a / ( ' 3 1

T ~ L E 4. Stitistical Bhmmary of calibration and prediction. i '.

, , 5 d a l y t e " , " I ' . , 1

~alibration'l Validation I

C4 " MR3 " Bias1 ' SEP2 Simple Correlation5

~oldr imet r~ 185 1.99 , .304 .337 .97 &rix "' 212 -.99 ' .991' .243 .975

, ' > I ~ h : ~ v e r & e differencb bctween'lab k d AIR; 2~tiAdard ermr of performance;

j' ~ u l t i ~ l e correlation coe?ficient; staddard error of calibration; Simple cor-relation.

The standard error of prediction (SEP) is the primary parameter evaluated. This parameter describes the overall perfdrmance of the developed calibration, which describes the standard deviation of the residuals (differences between the pri-

A compal'ision )between the calibration results, the SEC and MR were compared to the results derived from the SEP and simple C, as shown in Table 4. An increase from :I85 SEC to .337 was observed with the validation set. The corkelatioris were comparable. Such an increase is expected since differences in geographical locations of samples, lab personnel, and error from the lab methods will'increase 'the prediction error. A slight bias of .304 was displayed. It should be noted that the lab data *ere! not refined; all data were included, even possible outliers. The different methods for laboratory pol readings in Louisiana and Florida no doubt account for the bias.

837 I

PROCESSING

Brix

Similar studies were performed for Brix values, comparable values for the SEC and SEP were displayed, with the SEC original value of .212 and a SEP of .243; the correlations were comparable. Deviations were observed with the, bias value (Average difference between the lab-NIR). This deviation was 1.00 and once corrected, yielded acceptable results. The purity of the calibration set. (Florida samples) is higher than purity of the validation (Louisiana) set; this difference may have caused the bias. Different filtration treatments in Brix analysis in Louisiana and Florida may also have contributed.

SUMMARY , b

In summary, acceptable and reproducible results were achiived for polarization and Brix with favorable calibration results for purity. By testing each calibration with an independent set of samples derived from a different geographical loca-

I tion, the performance for each constituents as pol and Brix were determined. Pol displayed comparable results, whereas Brix sliowed a bias. The difference in chemical reference methods in the two locations contributed to bias. Such devia- tions pointed to changes in crop location and lab, etc.; however, with minor adjustments to the y-intercept, these deviations were eliminated. NIR has demon- strated that pol and Brix may accurately be nlonitored in unfi/tered cane juice. Calibration results for purity were favorable. The analysis requires no sample preparation and is a non-destructive technique which can be placed into a process or lab environment. The elimination of ,harsh chemicals, such as lead subacetate, is achievable with this technique. s his technique is adaptable to flow through cells for routine analysis.

' REFERENCES

1. Ames, R., Norton, S.W. and Nguyen, H.H. (1987). Control of sugar polari- zation and moisture using near infra-red reflectance analysis. Proc. Austr. Soc. Sugar Cane Technol. p. 246-252. (Also Sugar J. Sept. 1989. p. 7-10).

2. Berding, N., Brotherton, G.A., LeBrocq, D.G. and Skinner, J.C. (1989). Application of near infra-red reflectance (NIR) spectroscopy to the analysis of sugar cane in clonal evaluation trial. Proc. Austr. Soc. Sugar Cane Technol. p. 8-15. (Also Proc. Int. Soc. Sugar Cane Technol. 1989 p. 3-20). ,

3. Burzawa, E. and Melle, M. (1988). Caracterisation des produits de sucrbrie par reflectance dans la proche infra-rouge. Ind..Alim. Agric. (in press).

4. Burzawa, E., Melle, M. and Groutt, M. (1991). Application de I7analyse dans la proche infrarouge an contrble en ligne de la puretC. Pres. at C,.I.,T.S., Cambridge, UK. , .

1

M.A.'CLARKE ET hL.

5. Meyer; J.H. and Wood, R.A. (1988). Rapid analysis of cane juice by near infrared reflectance. ProC South Mr. Sugar Technol. Assoc. p. 203-207.

6. Nguyen, H.H. and Player, M.R. (1990). Near infra-red analysis of cane sugar products. Proc. Conf. Sugar Proc. Res. p. 57-77.

7. Osborne, B.G., Fearn, T., Miller, A.R. and Douglas, S. (1984). Application of near infra-red reflectance spectroscopy to the rcompositional analysis of biscuits and biscuit doughs. J. Sci. Food Agric. 49: 995-999.

8. Stevens, S.H. (1988). NIR analysis of sugar products. Proc. Conf. Sugar Proc. Res. p. 205-226.

9. Sverzut, C.B.., Verma, L.R. and French, A.D. (1987). Sugarcane analysis using near infrared spectroscopy. Trans. Amer. Soc. Agric. Eng. 39: 255- 258.

10. Vaccari, G., Mantovani, G., Sgualdino, G. and Goberti, P. (1987). Near infrared spectroscopy utilization for sugar products analytical control. I Zuckerind. 112: 800-807.

11. Xaccari, G. and Mantovani, G., (1988). Near infrared spectroscopy in sugar analysis:. Beet sugar manufacture. pro^. Conf. Sugar Proc. Res. p. 192-204.

i i

12. Vaccari, G., Mantovani, G. and Sgualdino, G. (1990), The development of : near infrared (NIR) technique on line in the sugar factory. Proc. Conf. Sugar I

Conf. Res. p. 40-56. 13. Williams, P.C.$!and Norris, K. (eds.) (1987). Near infrared technology in the

agricultural and food industries. The Amer. Assoc. Cereal Chemists. St: Paul, MN. 330 pp. ,

14. Williams, P.C., Cordeiro, H,M. and Harndeer, M.F.T. (1991). Analysis of ' oat bran products by NIR reflectance spectroscopy. Cereal Foods World 36:

571-574. 15. Wood, R.A., Meyer, J.H. and Govender, M, (1985). A rapid.system of cane

leaf analysis using X-ray spectrometry and infra-red reflectance. Proc. South Mr. Sugar Technol. Assoc. p. 195-201.

1 I

' PROCESSING

ANALYSES DANS LE PROCHE INFRA ROUGE (NIR) t EN ,SUCRERIE DECCANNES L , r I

, / I 2 8

$ 6 , '

M.A. Clarke1, E.R. Arias2 et C. McDonald-Lewis3 . < ' , 1 ' 1 / * ) , I \

I /

Sugar Processing Research Institute, Inc.; 1100 Robert E. Lee Blvd. New Orleah~, Lousiana 70124,iUSA ' I

Sugar Cane Growers Cooperative of Fl~rida,~P.O. Box 666i 2 '

Belle Glade, Florida 33430, USA ' i i 1

NIRSystems, Inc., 12101 Tech Road, Silver Spring,xMaryland 20904, USA I \ t i < ' l ' 0 , 1 1 , !I r r / /

< i >

RESUME I ; ! , , i , , ' ' V

i I ' * , 8 ,

L'analyse dans le proche infra rouge (NIR) est utilisee en bettbrave: pour la selection des varietes de cannes et dans d'autres, domaines 'sucriers en Europe, aux Etats Unis et eh Australie. L'emploie du NIR dans le 'laboratoire d'une sucrerie de canne a sucre est decrit. Des:travaux dejai publies ont 'decrit l'utilisa- tion td'instruments a filttes; ces instruments 'a balayage* adtomatique offrent la selection de longeur d' onde, de combinaisons de longeur d'onde e t une~preci- sion bien plus grande que' precedement! Des echantillons de jus, (echantillonnage pour les planteurs) furent analyses pendant la campagne et les resuItbts.de cette analyse beaucoup plus rapide ont ete tres proche des resultats (pol, Baixj pdrete) du laboratoire de la sucrerie. On a' aussi dnalyse la )bagasse et le sukre: La calibration de l'intrument, la preparation d'echantillod'et la fiabilite de la inethode sont discutes. Finalement, on presente des applications en sucrerie, par exemple l'analyse duSpol, des reducteurs, des ckndres, de la couleur, dn dextranase et de I' amidon dans le sucre. I , i 7 a; , < > r e I ., .,, i, . I

I , 1 , , ( t ; > s :

M8A. CLARKE ET AL.

ANALISIS 'CERCA A INFRAROJO (NIR) EN FABRICAS DE AZUCAR

M.A. Clarke1, E.R. Arias2 y C. McDonald-Lewis3

Sugar Processing Research Institute, Inc., 1100 Robert E. Lee Blvd. New Orleqns, Louisiana 70124, USA

Sugar Cane Growers Cooperative of Florida, P.O. Box 666 Belle Glade, Florida 33430, USA ,

NIRSystems, Inc., 12101 Tech Road, Silver Spring, Maryland 20904, USA

.% 1 , , I , 2 i f 8 ,

, 1 I ' , RESUMEN

El analisis cerca a infrarojo (NIR) ha estado encontrando amplias aplicaciones en la fabricacidn de, az6car de remolaoha, en programas de propagacidn de la caiia defaziicar, p en el analisis de productos azucarados in Europa, Estados Unidos y Australia., Experiencias con el uso de un espectofotometro examinador cerca a infrarojo (NIR) en una f6brica de az6car son reportadas. Trabajos previos han descrito el uso de, instrumentos a base de filtracidn; estos instrumentos examinadores .ofrecen sekctividad de hondas, combinaciones de hondas y major exactitud que los::previa~fiente disponibles. Muestras de jugos: de caiia fueron 'analizadas en diferentes ocasiones durante la zafra. Resultados de estos rhpidos aliIlisis comparan favorablomonto con.10~ analisis de polarizaci6n, Brix y pureza de la fhbrica. Analisis de,humedad y sacarosa en bagazo tambiCn son reportados. La ,calibraoidn deli instrumento, preparaci6n de la muestra, confiabilidad y reproductibilidad del an6lisis,son,discutidos. TambiBn son presentadas y discutidos aplicaciones en analisis de, az6car para pol, sacarosa, invertidos, cenizas, color, dextrana y almidon.