8290614309

7. Opublikowane badania własne

| Orni et ol : Oemomema and Inteligent Labęrotory Systeml 110 (2012) 89-96    95

TaMr 2

Rooc mean squarr mor of calibration (RMSC). root nwn squarr mor o 1 predicnoa (RMSEP) and mon npmird as a percenuge of the response vorUble rangę for pamal kast squares regression (PIS) crrated for tolor and ash rontem on thr basb of riccUa-tion-emissłoo fluorescence landscape* and selected emission spectra for 230 and 290 nm e*dułton wavekngths for color and ash ronieni. rrspectively.

Modrlrd pro perty	Color		Ash ronrrnt
Type of diii	Landscape	230 nm	Landscape	290 nm
r	6	5	6	4
RMSE	1.671	1.696	1506	2012
	5.08*	5.14*	559*	8JM*
RMSEP	1-8S8	2.194	1535	1879
	5.638	6.65*	5.48*	7.51*

dislance-discance plot. which prescnts thc standardized leverage and residual distances computed using parameters of a robost model. In this study. the oitoff values for both distances were set to three as was explained in the Theory secton.

The PRM model with five latent factors built for color was used to detect outlying objects in the calibration set. In Fig. 4* and b, the dis-tance-distance plots for the model and test set samples are pre-sented. respectively. As is indicated in the figurę one bad leverage object is revealed in the model set (object no. 131). Seven objects nos. 46.67.92.95. 96.132 and 180 are high residual objects for the model set There is also one good leverage object bccause its residual distance is smali; however, its leverage distance is high. It Is object no. 71. It potentially extends the calibration rangę but has no influence on the consmiction of dassic models such as PIS and N-PLS. As can be seen in Fig. 4b. the test set also contained some objects that exceeded the cutoff values.

The distance-distance plots obcained from the PRM model for ash content (with nine latent factors) revtal one bad leverage objea in thc model set (object no. 71). This has a significant influence on dassic calibration models (see Fig. 4c). Moreover. one can spot a few good leverage objects in the model set The distance-distance plot constructed for the test set exhibits three outlying objects nos. 129, 162 and 171 (see Fig 4d).

In order to identify the characteristic exdtation wavelength that can provide a suitable spectrum for calibration purposes, PIS models for each emission spectrum observed at a given exatation wave-length for the color and ash content were constructed. Characteristic exotarion wavelengths were selected according to the optima! RMSE and RMSEP values obtained from a model for a calibrated propcrty. A satisfactory PIS model for color was built using the emission spectrum corresponding to the exdtation wavelength at 230 nm. whereas for the ash content it was the emission spectrum observed at the exdtation wavelength 290 nm. As can be observed by the results presented in Table 2. both PLS models (with two and four latent variables, respectively). are comparable to the PIS models constructed for the unfolded fluorescence bndscapes.

6. Condusions

The aim of this paper was to highlight the advantages of the robust calibration technique (PRM) used for monitoring sugar production. Toachieve this goal. difTerent calibration models - classic and robust -were built and compared for exdtation-emission fluorescence land-scapes. The predlctlon errors of the constructed models. given in per-centages relative to the total rangę of response variable. were equal to 3.24% and 4.37% for the color and ash content. respectively. It was confirmed that the sugar production process can be monitored using multivariate calibration models and fluorescence signals and that robust modeling has great potential in the conrrol of real pro-cesses. The results obtained from PIS. N-PIS and PRM indkate that robust PIS models are stable against new sources of variation and have better prediction properties than PLS and N-PLS. The PRM models allow for the direct Processing of data containing outlying objects. Therefbre. identificarion of outliers prior to the consmiction of a PRM model is unnecessary. In addition. it is possible to select characteristic excitation wavelengths and use the conesponding emission spectra for modeling for the purpose of monitoring

Adcnowtedgment

MD wishes to express his gratitude to the Minister of Science and Higher Education of the Polish Republic for funding the scholarship.

References

111 T. Woodcock. C 0'DonnelL C. Downey. Berrrr qiultty food and brmogrs: che rok of near Infrared spectroscopy. Journal of Near fnfrared Spectroscopy 16 (2008) 1-29.

[2| J.C Menem P. FeKzardo. MJ. Nctva-Conela. The use of process analyucal (ech-nology (PAT) (oots in biofuels production. Spectroscopy 23 (2008).

131 J. Munson. CF. Stanheld. 8. Cujral A review of process anatyucal technołogy (PAT) m rhe US płunnocrutkal mdusoy. Curm* Pharmocrutkol AiuJystj 2 (2006)405-414.

|4| AS. Rathorc. A Ihambure. V. Charę, frocni analytlcal lechnology (PAT) for bk>-phormaceutkal product. Analytkoi and BioanółyUcal Chemłjtry 398 ( 2010) 1J7-1S4.

151 T. Kmirti. Proces* analyóca) techno loty bryond real-ttme anałyzers: the role of multńrarijie analym. CRC Cndcal Reyirws in Analytkal Chemisiry 36 (2006) 257-278.

(6) 0. Kim. S.A Kim. DJ. Day. MJ. Oh. H.Y. Kim. A dipstkk method lo meomirt dnt-rron in raw sugar proceu sneam. Imernaiional Sugar Journal III (2009) 119-124.

|7| S.N. WolfardL KJ SchJnkr. P.C. Morel Du Bod. ChromatogTaphk Solutions for sugar procrsung. Internjuonal Sugar Journal 107 (2005) 90-1(0.

181 0.L Masurt. 8.CA4. Vandeginue. LM. Budgrns. S. de Jnng. PJ. lewi. J. Smeyrrs-Vertoeke. Handbook of Chemometna and Qualm>e<ncs Pan A. Eisevter. London. 1992.

|9| LM. Becker. J. Chratensen. C.5. Fredertksen, V.K. Haugaard. Front-face fluoces-cenre speciroscopy and chrmometrla In arułysis of yogurr. rapid arułysis of rt-boAamn. Journal of Dauy Saenc* 86 (2003) 2506-2515.

1101 R. Cheikhoutman. M. Zude. D. Bouvrmse. Ci. l/get. DJL Rot ledge. L Blrtouez-Aragon. Fluorescence specrroscopy for monitoring detertoraOon of extra vlrgtn ohve oil durtng heating. Analytkal and Bioaiulytical Chemtstry 382 (2005) 1438-1443. 1111 D. Saunsgaard. L Norgaard. MA. CodshalL Ruoresrence of raw cane sugars rval-uated by chemometrk*. Journal of Agrkohural aod Food Chemlstry 48 (2000) 4955-4962.

[ 121 L Norgaard. Classiflcation and prediction of quaKty and process parameteo of thick Jukę and beet sugar by fluorescence specrroscopy and chemometrk*. Zuckerind 120(1995) 970-981.

113) R. Bra. Eiploratory nudy of sugar production using fluorrscrnce specrroscopy and muhi-way arulysn. Chemomrtrks and Inielllgent Laboraiory Systems 46 (1999)133-147.

1141 I. Stanimirova. S. Semeels. PJ. Van Ej pen. B Walczak. How to construa a multtpłe regression model for data mih miss Ing elements and outlying objects. Analytka Chlmica Ana 581 (2007 ) 324-332

115| MC. Ortu. LA Saraba. A Herrero. Rabusi regression trchmques. A useful ahema-tlve for the detection of outlier data In Chemical aiułysis. Talanca 70 (2006) 499-512

|16| A Bro. Multiway caJtbrabon. mul t Hi near PLS. Journal ofChemornemcs 10(1996) 47-62.

I17( S. Semeeti. C. Crou*. P. Rlzmosrr. PJ. Van Epsen. Partia! robust M-regression.

Chemometncs and Intelbgem Laboraiory Systems 79 (200S) 55-64.

1181 M. Frltz. P. Fllzmoser. C. Crou*. A rompartson of algonthms for mulrtvartate Ll-median. Comput Siat (m press) 1-18. doi; 10 1007**00180-011 -4262-4.

| IS | O. Hbssjer. C. Crou*. General maj on rv a rutę signed rank staikstics for lesmg and estlmaiing a muHivariatr localion paromeier. Journal of Nonparametric Statisiki 4(1995)293-308.

|20| DJ. Cummins. C. Andrews. !terauveły reweighted pamal Ieast squares: a performance onaiysH by Monte Carto slmulatlon. Journal of Chemomrtrks 9 (1995) 489-507.

(211 M. Hubert. K Vanden Branden. Robust methods for partia! Ieast tquares regression. Journal of Chemometrtcs 17 (2003) 537-549.

|22] M. Hubert Robust Calibration in Comprehenstvf Chemometncs. vol 3. E3sevier. Amsterdam. 2010. pp. 315-343.

(23| 5. Frosch Molier, J. von Frese. A Bro. Rabusi methods for moKKanaie data anaty-sis. Journal of Chemomentcs 19 (2005) 549-563.

]24] M. DaszykowskL 8. Walczak DJ- Massart Representanwe subsei seiecnon. Analytka Clumlca Acta 468 (2002)91-103.

[251 AD. Snee. Validatno of regression models: methods amd esamples. Techno-methes 19(1977)415-428.

|36| i .W. K en nart. LA Sione. Computer aided design of enperunenu. Techno meirio 11 (1969)I37-14A

|27| T. Maes. T. Isaksson. T. Fearn. 7. Davies. A User-fnendly Guide to Mukivanate Calibration and Classlflratloo. NlR Publtcations. Chkhestrr, 2002.

Strona 58