NIR spectroscopy: a rapid-response analytical tool

240

trends in analytical chemistry, vol. 21, no. 4, 2002

NIR spectroscopy: a rapid-response analytical tool M. Blanco*, I. Villarroya `noma de Departament de Quı´mica, Unitat de Quı´mica Analı´tica, Facultat de Cie`ncies, Universitat Auto Barcelona, E-08193 Bellaterra, Barcelona, Spain

In recent years, near-infrared (NIR) spectroscopy has gained wide acceptance in different fields by virtue of its advantages over other analytical techniques, the most salient of which is its ability to record spectra for solid and liquid samples with no prior manipulation. Also, developments in instrumentation have resulted in the manufacture of spectrophotometers capable of quickly providing spectra that are flexible enough for use in different situations; thus, portable equipment can record spectra on site or even at production lines. This article discusses the features of NIR spectroscopy that have driven forward its dramatic development in a wide range of analytical fields in the last few years. # 2002 Published by Elsevier Science B.V. All rights reserved. Keywords: NIR spectroscopy; Instrumentation; Chemometrics; Applications

1. Introduction Although Herschel discovered light in the near-infrared (NIR) region as early as 1800, even spectroscopists of the first half of the last century ignored it in the belief that it lacked analytical interest. The earliest applications of NIR spectroscopy were reported in the 1950s, but it was not until the 1970s that the group headed by Norris used it to analyze agricultural food samples. The development of equipment featuring improved electronic and optical components and the advent of computers capable of effectively processing the information contained in NIR spectra facilitated the expansion of this technique in an increasing number of fields. *Corresponding author. E-mail: [email protected]

0165-9936/02/$ - see front matter PII: S0165-9936(02)00404-1

The literature reflects the changes in the potential and the appreciation of NIR spectroscopy. Thus, Wetzel in 1983 deemed it a ‘‘sleeper among spectroscopic techniques’’ on account of its high potential, but scant use [1]. McLure in 1994 published a paper entitled ‘‘The giant is running strong’’ because of its increasing number of applications in different fields [2]. Davies in 1998, putting forward future possible advances and applications, aptly described its potential as taking it ‘‘from sleeping technique to the morning star of spectroscopy’’ [3]. The interest in NIR spectroscopy lies in its advantages over alternative instrumental techniques. Thus, it can record spectra for solid and liquid samples with no pretreatment, implement continuous methodologies, provide spectra quickly and predict physical and chemical parameters from a single spectrum. These attributes make it especially attractive for straightforward, speedy characterization of samples. This article emphasizes the fundamental aspects of NIR spectroscopy that account for its currently wide use as a fast analytical tool. It provides a brief description of its foundation, the equipment it uses and the mathematical tools needed to extract the whole information contained in NIR spectra. Finally, there are comments on selected applications in a number of fields.

2. Foundation The NIR region spans the wavelength range 780–2500 nm, in which absorption bands correspond mainly to overtones and combinations of fundamental vibrations. # 2002 Published by Elsevier Science B.V. All rights reserved.

241

trends in analytical chemistry, vol. 21, no. 4, 2002

The vibration of molecules can be described using the harmonic oscillator model, by which the energy of the different, equally spaced levels can be calculated from
 rffiffiffiffi 1 h k Evib ¼ u þ 2 2  where  is the vibrational quantum number, h the Planck constant, k the force constant and m the reduced mass of the bonding atoms. Only those transitions between consecutive energy levels (
= 1) that cause a change in dipole moment are possible, DEvib ¼ DErad ¼ hn where n is the fundamental vibrational frequency of the bond that yields an absorption band in the middle IR region. However, the harmonic oscillator model cannot explain the behavior of actual molecules, as it does not take account of Coulombic repulsion between atoms or dissociation of bonds. As a result, the behavior of molecules more closely resembles the model of an anharmonic oscillator, by which energy levels are not equally spaced. Thus, energy difference decreases with increasing :   DEvib ¼ hn 1  ð2u þ Du þ 1Þy where y is the anharmonicity factor. The anharmonicity can result in transitions between vibrational energy states where
= 2,  3. . .. These transitions between non-contiguous vibrational states yield absorption bands known as overtones (first and second overtone, respectively) at, approximately, multiples of the fundamental vibrational frequency. Also, they are much less likely than the fundamental transitions, so the bands are much weaker (the band for the first overtone is 10–100 times weaker than that for the fundamental frequency, depending on the particular bond). These bands appear between 780 nm and 2000 nm, depending on the overtone order and the bond nature and strength.

In polyatomic molecules, two or more vibrational modes can interact in such a way as to cause simultaneous energy changes and give rise to absorption bands called combination bands, the frequencies of which are the sums of multiples of each interacting frequency. NIR combination bands appear between 1900 nm and 2500 nm. The intensity of NIR bands depends on the change in dipole moment and the anhar monicity of the bond. Because the hydrogen atom is the lightest, and therefore exhibits the largest vibrations and the greatest deviations from harmonic behavior, the main bands typically observed in the NIR region correspond to bonds containing this and other light atoms (namely C–H, N–H, O–H and S–H); by contrast, the bands for bonds such as C=O, C–C and C–Cl are much weaker or even absent. Interactions between atoms in different molecules (for example hydrogen bonding and dipole interactions) alter vibrational energy states, thereby shifting existing absorption bands and giving rise to new ones, through differences in crystal structure. This allows crystal forms to be distinguished and physical properties (such as density, viscosity, and particle size in pulverulent solids) determined. In other words, the NIR spectrum contains not only chemical information of use to determine compositions, but also physical information that can be employed to determine physical properties of samples.

3. Instrumentation NIR spectroscopy instrumentation has evolved dramatically in response to the need for speed in analyses and flexibility in adapting to different sample states. Spectrophotometers used to record NIR spectra are essentially identical with those employed in other regions of the electromagnetic spectrum. But NIR equipment can incorporate a variety of devices (Fig. 1), depending on the characteristics of the sample and the particular analytical conditions and needs (such as speed, sample complexity and environmental conditions), so the technique is very flexible.

242

trends in analytical chemistry, vol. 21, no. 4, 2002

Fig. 1. Principal features of NIR spectroscopy equipment.

NIR spectrophotometers can be of two types with respect to wavelength selection, namely discrete wavelength and whole spectrum. The former are simpler, as they irradiate samples with only a few wavelengths. As a result, they can be used in only applications with analytes absorbing in specific spectral zones. In discrete-wavelength spectrophotometers, wavelengths can be selected by using as light sources filters that allow the passage of variably broad wavelength bands or light-emitting diodes (LEDs) that emit narrow bands. The absence of moving parts makes LED-based spectro-

photometers straightforward and robust, and hence amenable for use in portable equipment. Whole-spectrum NIR instruments usually include a diffraction grating, although they may be of the Fourier transform (FT)-NIR type. They are much more flexible than discretewavelength instruments, so they can be used in a wider variety of situations. Other wavelength-selection devices incorporated into NIR spectrophotometers in recent years include Acousto-Optic Tunable Filters (AOTFs) [4]. These choose wavelengths by using radio-frequency signals to alter the refractive

243

trends in analytical chemistry, vol. 21, no. 4, 2002

index of a birefringent crystal (usually TeO2) so that it transmits light of a given wavelength or performs a wavelength scan much more rapidly than with the previous designs. The absence of moving parts in AOTFs ensures more reliable, reproducible wavelength scans than those provided by other devices. This makes AOTFs especially suitable for equipment subject to aggressive conditions, as in production plants. Detection in NIR spectroscopy uses devices comprising semiconductors (PbS or InGaAs). In multi-channel detectors [5], several detection elements are arranged in rows (diode arrays) or planes [charged coupled devices (CCDs)] in order to record many wavelengths at once, so as to increase the speed at which spectral information can be acquired. This type of detector has given rise to NIR-imaging spectroscopy, in which spectra are recorded by using cameras that can determine composition at different points in space and record the shape and size of the object. Making measurements at different wavelengths provides a three-dimensional image that is a function of the spatial composition of the sample and the irradiation wavelength used. Analysis is fast not only because the NIR technique records spectra quickly but also because there is virtually no need to pretreat samples. This has helped the development of quartz-fiber-optic-based devices that allow the spectra of samples with different characteristics to be recorded simply by selecting the most suitable mode for each (usually reflectance for solids, transmittance for liquids, and transflectance for emulsions and turbid liquids). Another factor influencing speed of analysis is the ability to perform field measurements instead of having to collect samples for subsequent analysis in the laboratory. Some NIR spectrophotometers can make measurements on-site. The miniaturization of optical components has boosted development of portable NIR spectrophotometers. Currently available models [6,7] using such optical devices include hand-held instruments and equipment that can be carried in a backpack or mounted on a vehicle, such as a tractor [8].

4. Chemometrics Obviously, the analytical information contained in the typically broad, extensively overlapped bands of NIR spectra is hardly selective and is influenced by a number of physical, chemical and structural variables. In addition, differences between samples may cause very slight spectral differences that are difficult to distinguish with the naked eye. However, the powerful NIR instruments currently available quickly provide vast amounts of data that require speedy, efficient processing if it is to yield useful analytical information. For these reasons, NIR spectroscopy requires chemometrics to extract as much relevant information as possible from the analytical data [9]. The two techniques are closely related, as NIR spectroscopy would never have reached its present stage of development without chemometrics and NIR-spectroscopy results are frequently used to illustrate the power of new chemometric algorithms. A comprehensive description of chemometric techniques used in NIR spectroscopy would require a chemometrics treatise, so only the more frequently used chemometric methods are discussed in this article. The analytical information contained in NIR spectra can be extracted by using various multivariate analysis techniques that relate several analytical variables (as in a NIR spectrum) to properties (such as concentration) of the analyte(s). The multivariate techniques most frequently used allow samples with similar characteristics to be grouped, in order to establish classification methods for unknown samples (qualitative analysis) or to perform methods determining some property of unknown samples (quantitative analysis).

4.1. Pretreatment of spectra The spectra of solid samples are influenced by the physical properties of the solid samples. This poses some problems in evaluating aspects of samples for which physical appearance is not important (such as identification of raw materials and determination of composition). In these

244

trends in analytical chemistry, vol. 21, no. 4, 2002

situations, spectral pretreatment should be used to minimize those contributions incorporating irrelevant information into spectra in order to be able to develop more simple and robust models. Some of the more frequent pretreatments for NIR spectra include: normalization [10]; derivatives (usually first or second) [11,12]; the multiplicative scatter correction (MSC) [13]; the standard normal variate (SNV) [14]; de-trending (DT) [14]; or, a combination thereof.

4.2. Reduction of variables There is a need for variable-reduction methods because of the vast amount of spectral information provided by NIR spectrophotometers, the substantial number of samples required to construct classification and calibration models, and the high correlation in spectra. As a result, a number of multivariate-analysis methods rely on variable-reduction techniques that allow the dimensions of the original data to be reduced to a few uncorrelated variables containing only relevant information from the samples. The best known and most widely used is principal component analysis (PCA) [15]. PCA searches for directions of maximum variability in sample groupings and uses them as new axes called ‘‘principal components’’. In this way, the relevant information for the system is contained in a reduced number of variables. The PCA data thus obtained can be used as new variables, instead of the original data, in subsequent calculations.

4.3. Multivariate analysis methods The purpose of multivariate-analysis methods is to construct models capable of accurately predicting the characteristics and properties of unknown samples. The process involves the steps described in Table 1. A number of multivariate-analysis methods can be classified according to their purpose and the algorithms or computational procedures that they use. Fig. 2, although not exhaustive, shows the most frequently used. The method of choice will depend on the purpose of the analysis, the characteristics of the samples and the complexity of the system concerned (for example its non-linearity). Once models are constructed, their predictive capacity must be checked on samples subjected to the same treatment (spectrum recording conditions and spectral pretreatments) as those used for calibration but not employed to construct the model. Qualitative multivariate analytical methods are known collectively as ‘‘pattern-recognition methods’’ [9], which are labeled ‘‘supervised’’ or ‘‘unsupervised’’, depending on whether or not the class to which the samples belong is known. These methods establish mathematical criteria that allow similarity between two samples, or a sample and a class, to be expressed quantitatively. Usually, similarity is expressed as the coefficient of correlation between samples or as a distance (Mahalanobis or Euclidean) measurement; both types of parameter can be calculated using spectra or PCA results. The different types of classification method available

Table 1 Steps in the multivariate model-construction process Step 1. 2.

Purpose

5. 6.

Choosing the calibration samples Determining the target parameter by using the reference method Recording the NIR spectra Subjecting spectra to appropriate treatments Constructing the model Validating the model

7.

Predicting unknown samples

3. 4.

To select a set of samples representative of the whole population To determine the value of the measured property in an accurate, precise manner. The quality of the value dictates that of the calibration model To obtain physico–chemical information in a reproducible manner To reduce unwanted contributions (such as shifts and scatter) to the spectra To establish the spectrum–property relationship using multivariate methods To ensure that the model accurately predicts the property of interest in samples not subjected to the calibration process To predict rapidly the property of interest in new, unknown samples

trends in analytical chemistry, vol. 21, no. 4, 2002

245

Fig. 2. Classification of the major qualitative and quantitative multivariate-analysis methods used in NIR spectroscopy. ANN: Artificial Neural Networks; PCA: Principal Component Analysis; PLS: Partial Least Squares; PCR: Principal Component Regression; MLR: Multiple Linear Regression; SIMCA: Soft Independent Modeling Class Analogy; KNN: K-Nearest Neighbor; LDA: Linear Discriminant Analysis.

establish boundaries between the different classes, or they model the space occupied by a class and determine whether a sample belongs to it on the basis of distance measurements or the residual variance. Soft Independent Modeling of Class Analogy (SIMCA) is the best known and most widely used of the latter type [16]. Qualitative analyses by NIR spectroscopy usually rely on the use of spectral libraries constructed by using one of the above methods [17]. Such libraries are normally constructed using qualitative analytical tools included in the software bundled with commercially available equipment. Appropriate use of NIR libraries allows one not only to identify products chemically, simply by comparing the coefficient of correlation between the spectrum for an unknown sample and those contained in the library, but also to ascertain whether they possess the desired physical properties (such as grain size or moisture content). The simplest quantitative multivariate-analysis method is multiple linear regression (MLR) [18], which usually uses fewer than five spectral wavelengths. MLR assumes concentration to be

a function of absorbance, which entails the knowledge of the concentrations of not only the target analytes, but also all other components contributing to the overall signal. The multivariate-regression methods most frequently used in NIR spectroscopy are principal component regression (PCR) [19], particularly partial least-squares (PLS) regression [19]. Both can be used in specific spectral regions or the whole spectrum, and they allow more information to be included in the calibration model. PCR uses the principal components provided by PCA to perform regression on the sample property to be predicted, while PLS finds the directions of greatest variability by considering both spectral and target-property information, with the new axes called ‘‘PLS components’’ or ‘‘PLS factors’’. In some cases, however, the spectral data and the target property are not linearly related as a result of instrumental factors or the physico– chemical nature of the sample. These cases can be addressed using non-linear calibration methods, particularly prominent among which are artificial neural networks (ANNs) [20], which

246

trends in analytical chemistry, vol. 21, no. 4, 2002

are iterative computational algorithms that allow experimental variables to be fitted to a specific response. ANNs contain layers that serve different purposes, namely: to receive experimental information; to process it by using non-linear functions; and, to provide the response. In some cases, the original variables can be transformed by using a non-linear (logarithmic, exponential or quadratic) function and doing the regression with one of the above-described linear methods.

5. Applications The above advantages of the NIR technique, resulting from its technical developments and advances in chemometrics, have aroused inter-

est in it, as a powerful choice for routine control analyses in many industrial sectors. Fig. 3 shows the major fields of application and selected examples of the parameters determined. As can be seen, NIR spectroscopy allows the characterization of natural and synthetic products; for the latter, it allows the monitoring of production processes with a view to the early detection of flaws and preventing them from reaching the end product, thereby saving time and money. Rather than give an exhaustive review of the vast number of applications of NIR spectroscopy in different fields, this section illustrates its use as a rapid control tool. Readers interested in more comprehensive coverage are referred to specific literature reviews on the topic [42,45,46].

Fig. 3. Selected examples of the use of NIR spectroscopy. [21–44]

247

trends in analytical chemistry, vol. 21, no. 4, 2002

5.1. Agricultural food sector The agricultural food sector was the first to adopt NIR spectroscopy as an analytical technique. In 1974, the Canadian Grain Commission replaced the traditional Kjeldahl method for determining protein in wheat with a new, NIR-based method. In 1995, the new method was estimated to save $2.5m and avoid the generation of 47 tons of caustic waste each year [47]. At present, advances focus on the speed and the non-destructive nature of NIR spectroscopy. Vegetables and fruits can be directly analyzed, after harvest or even during growth [48–51] by using optic-optic probes or special devices to record reflectance spectra. von Rosenberg Jr. et al., who developed a prototype NIR spectrophotometer mounted on a truck for determining the protein, moisture and oil content of crops at harvest, provides confirmation of the great potential of NIR spectroscopy as a rapid control tool [8]. The equipment avoided the need for farmers to wait in long queues for their produce to be analyzed and provided them with useful information to improve classification, drying, storage and marketing of their produce.

5.2. Petrochemical sector The properties typically examined in characterizing petrochemical products have traditionally been determined using specific tests for chemical composition and physical quantities that are time-consuming and delay decision-making in production. For these reasons, because of the many advantages of NIR spectroscopy, oil refineries and the petrochemical industry increasingly use it as a control technique. There have been a number of papers published on the NIR spectroscopic analysis of hydrocarbons, fuels, petroleum fractions, polymers and other petroleum derivatives. The parameters that characterize these products can be determined more quickly, occasionally from a single spectrum, with NIR spectroscopy. However, the greatest interest has been aroused by the benefits of implementing NIR

spectroscopy on the production line to characterize products in real time. Bu¨ ttner reported the advantages of using NIR spectroscopy in refineries [52]. The main advantages are the low costs of equipment maintenance, the fast response times and the great quantity of process data that can be collected and used for process control. All these advantages increase product yield, improve up-times, minimize waste production and yield substantial improvements in other cost-effectiveness factors.

5.3. Pharmaceutical sector Because of their intended end use, pharmaceuticals must be thoroughly controlled prior to release. This entails conducting a large number of analyses in the various steps of the manufacturing process. Pharmaceutical companies have gradually adopted NIR spectroscopy as their technique of choice for this purpose [45]. By using commercially available devices, the spectra for raw materials can be recorded from the products received in the warehouse, without the need to transfer samples to the laboratory. NIR spectral libraries allow the materials to be analyzed qualitatively. In some cases, spectra can even be recorded through the protective film used as packaging [53]. Another factor that speeds up the analysis of pharmaceuticals by NIR spectroscopy is its ability to determine composition at intermediate stages in production, thereby avoiding the production of out-of-specification formulations as soon as possible. Thus, some NIR spectroscopic methods allow the composition of pharmaceutical preparations to be controlled reliably even if the materials are in different physical forms (such as granules, tablets, or lacquered tablets); a single calibration allows the content of the active principle in each form to be accurately determined [32]. However, the ability of NIR spectroscopy to analyze intact dosage forms has dramatically helped to characterize products, at both the intermediate stages and the end of the process; thus, NIR spectroscopy has been used to determine the active principle content in intact tablets [54] and even to analyze products

248

in the packaging used to deliver them (such as tablets in blisters, injectables in capped vials) [34]. Although pharmacopoeias have adopted some NIR spectroscopic methods for the identification of pharmaceuticals, the technique has not yet been officially endorsed for quantitative analyses.

5.4. Clinical and biomedical sectors Although NIR spectroscopy is not as widely used in these sectors as it is in the previous ones, it has already proved useful in medical diagnostics [55]. The greatest developments of this technique in the clinical and biomedical fields came during the 1990s because it is noninvasive, so it avoids biopsy or even surgery in some cases, and because NIR radiation, unlike X-rays, is harmless to biological tissues. NIR spectroscopy has proved effective in medical applications, such as the determination in blood of biochemicals (such as glucose and proteins) used as indicators of specific diseases and the monitoring of the oxygen level in tissues during surgical processes. In the latter case, the portability and modest cost of NIR equipment compared with other instruments, such as NMR spectrometers, makes it especially attractive in those situations requiring speedy intervention (as in implanting skin grafts or replacing vital organs).

5.5. Environmental sector A major contribution of NIR spectroscopy to environmental analysis is in recycling of plastic packaging. Effective reprocessing of these materials entails their prior separation according to the type of constituent polymer (PVC, PP, PET, or PS). This has been helped by the development of mechanical sorting devices [39,40] that discriminate among plastic types; the samples are transported to a NIR spectrophotometer on a conveyor belt. By using chemometric sorting techniques and appropriate hardware, the plastics are separated in real time and collected in different containers according to their major constituent polymer.

trends in analytical chemistry, vol. 21, no. 4, 2002

Also, the development of portable NIR equipment has enabled on-site measurements to determine, for example, the degree to which soil is contaminated with motor oils [41].

5.6. Process control One reason for the increasing industrial acceptance of NIR spectroscopy is the possibility of using it directly on the production line. This has meant adapting existing equipment to operate away from the carefully controlled conditions of the laboratory. The switch from offline to non-invasive, on-line spectrophotometric methods is driven by the desire to take the light to the sample rather than to take the sample to the light [56]. In spectrophotometric methods for process control, light is taken to the sample by using optic-optic probes, thereby separating the instrument and its operator from the aggressive environment of the production plant. The probes can be directly inserted into the process line or connected to a flow-cell, through which the sample can be diverted from the production line [57]. Reflectance measurements in the production line or at reactors can be made through a window. Transmission measurements can be made by inserting two opticoptic probes facing each other. Transflectance measurements can be made by sending the light first through the probe, then through the sample, and subsequently to the detector back through the probe following reflection [58]. Both the probe ends and the flow-cell are constructed from corrosion-resistant materials, into which are inserted sapphire windows (high resistance even under aggressive conditions). Finally, some instruments include multiplexers that can simultaneously receive signals from different points along the production line and subsequently decode the information from each measuring point.

6. Conclusions In recent years, NIR spectroscopy has increasingly been adopted as an analytical tool in

249

trends in analytical chemistry, vol. 21, no. 4, 2002

various fields and has superseded traditional methods. The main reasons for its current widespread acceptance are the speed with which samples can be characterized without manipulating them and the flexibility of NIR equipment. However, some characteristics of the technique restrict broader application or preclude particular uses. The main advantages and disadvantages of NIR spectroscopy as an analytical tool are set out below.

6.1. Advantages (a) It is a non-invasive, non-destructive technique. (b) It requires minimal or no sample preparation. Solid samples can be directly measured with little pretreatment, or no pretreatment, if an appropriate device is used. (c) Measurement and result delivery are quite fast. Dramatic developments in NIR equipment and chemometrics, used in conjunction with computers, have enabled the real-time extraction of analytical information from samples. (d ) There is no need for reagents or materials to prepare samples and the automation of the technique results in increased throughput, which, in turn, reduces analytical costs and decreases amortization time. (e) A single spectrum allows several analytes to be determined simultaneously. ( f ) The technique allows determination of non-chemical (physical) parameters. In fact, the influence of some parameters on the NIR spectrum allows the ready determination of properties such as density, viscosity or particle size. ( g ) Because of the great strength of optical materials and the robustness of NIR equipment, which in some cases has no moving parts, NIR instrumentation is most suitable for use in process control at production plants. (h) Fiber optics provides robust, strong sensors for at-line, on-line and in-line analyses to control processes.

(i) NIR spectroscopic results are comparable in accuracy to those of other analytical techniques; also, their precision is usually higher, because there is no need for sample treatment.

6.2. Disadvantages (a) NIR measurements are scarcely selective, so chemometric techniques have to be used to model data from which to extract relevant information. (b) There are no accurate models to take account of the interaction between NIR light and matter. As a result, calibration is purely empirical in many cases. (c) Accurate, robust calibration models are dicult to obtain as their construction entails using a large enough number of samples to encompass all variations in physical and/or chemical properties. (d ) The need to incorporate the physical and chemical variability of samples in calibration entails using as many dierent calibration models as there are sample types, and hence more than one model per analyte. (e) As NIR spectroscopy is a relative methodology, to construct models using it requires prior knowledge of the value for the target parameter, which must be previously determined using a reference method. ( f ) The technique is not very sensitive, so usually it can be applied only to major components. ( g ) The construction of NIR models requires substantial investment, which can, however, be amortized by transferring calibrations from the master equipment to several slaves. Despite significant advances in recent years, no specific methodology for this has yet gained widespread acceptance. Acknowledgements This work was funded by Spain’s DirectorateGeneral of Scientific and Technical Research

250

trends in analytical chemistry, vol. 21, no. 4, 2002

(DGICyT) within the framework of Project BQU2000–0234.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30]

D.L. Wetzel, Anal. Chem. 55 (1983) 1165A. W.F. McLure, Anal. Chem. 66 (1994) 43A. T. Davies, Analusis 26 (1998) M17. C.D. Tran, Anal. Chem. 64 (1992) 971A. Q.S. Hanley, C.W. Earle, F.M. Pennebaker, S.P. Madden, M.B. Denton, Anal. Chem. 68 (1996) 661A. J. Malinen, M. Ka¨nsa¨ koski, R. Rikola, C.G. Eddison, Sensors and Actuators B 51 (1998) 220. http://www.asdi.com. C.W. von Rosenberg Jr., A. Abbate, J. Drake, D.M. Mayes, Spectroscopy 15 (2000) 34. D.L. Massart, B.G.M. Vandenginste, S.N. Deming, Y. Michotte, L. Kaufmann, Chemometrics: A Textbook, Elsevier, Amsterdam, 1988. P.J. Griffiths, J. Near Infrared Spectrosc. 3 (1995) 60. W.F. McLure, NIR News 4 (6) (1993) 12. W.F. McLure, NIR News 5 (1) (1994) 12. P. Geladi, D. McDougall, H. Martens, Appl. Spectrosc. 39 (1985) 491. R.J. Barnes, M.S. Dhanoa, S.J. Lister, Appl. Spectrosc. 43 (1989) 772. S. Wold, K. Esbensen, P. Geladi, Chemom. Int. Lab. Syst. 2 (1987) 37. S. Wold, Pattern Recognition 8 (1976) 127. C.I. Gerha¨user, K.A. Kovar, Appl. Spectrosc. 51 (1997) 1504. E.V. Thomas, D.M. Haaland, Anal. Chem. 62 (1990) 1091. H. Martens, T. Naes, Multivariate Calibration, Wiley, New York, 1991. J. Zupan, J. Gasteiger, Neural Networks for Chemists: An Introduction, VCH, Weinheim, Germany, 1993. R. Frankhuizen, Handbook of Near-Infrared Analysis, Marcel Dekker Inc, New York, 1992. S.E. Kays, F.E. Barton II, W.R. Windham, J. Near Infrared Spectrosc. 8 (2000) 35. H.B. Ding, R.J. Xu, J. Agric. Food Chem. 48 (2000) 2193. O. Fumie`re, G. Sinnaeve, P. Dardenne, J. Near Infrared Spectrosc. 8 (2000) 27. C. Ridgway, J. Chambers, I.A. Cowe, J. Near Infrared Spectrosc. 7 (1999) 213. E.D. Yalvac, M.B. Seasholtz, M.A. Beach, S.R. Crouch, Appl. Spectrosc. 51 (1997) 1565. M. Blanco, S. Maspoch, I. Villarroya, X. Peralta, J.M. Gonza´ lez, J. Torres, Anal. Chim. Acta 434 (2001) 133. A.F. Parisi, L. Nogueiras, H. Prieto, Anal. Chim. Acta 238 (1990) 95. S.J. Foulk, B.E. DeSimas, Process Control and Quality 2 (1992) 69. R. Guchardi, Augusto da, Costa, P. Filho, R.J. Poppi, C. Pasquini, J. Near Infrared Spectrosc. 6 (1998) 333.

[31] G. Lachenal, Analusis 26 (1998) M20. [32] M. Blanco, J. Coello, A. Eustaquio, H. Iturriaga, S. Maspoch, Anal. Chim. Acta 392 (1999) 237. [33] A.D. Trafford, R.D. Jee, A.C. Moffat, P. Graham, Analyst 124 (1999) 163. [34] B.F. McDonald, K.A.J. Prebble, Pharm. Biomed. Anal. 11 (1993) 1077. [35] K. Murayama, K. Yamada, R. Tsenkova, Y. Wang, Y. Ozaki, Fresenius J. Anal. Chem. 362 (1998) 155. [36] S.F. Malin, T.L. Ruchti, T.B. Blank, S.N. Thennadil, S.L. Monfre, Clinical Chemistry 45 (1999) 1651. [37] J. Wallon, S.H. Yan, J. Tong, M. Meurens, J. Haot, Appl. Spectrosc. 48 (1994) 190. [38] K.A. Shenkman, D.R. Marble, E.O. Feigl, D.H. Burns, Appl. Spectrosc. 53 (1997) 325. [39] W.H.A.M. van den Broek, D. Wienke, W.J. Melssen, C.W.A. deCrom, L. Buydens, Anal. Chem. 67 (1995) 3753. [40] M.K. Alam, S.L. Stanton, G.A. Hebner, NIR News 6 (1995) 10. [41] B.R. Stallard, M.J. Garcia, S. Kaushik, Appl. Spectrosc. 50 (1996) 334. [42] J. Workman Jr., D.J. Veltkamp, S. Doherty, B.B. Anderson, K.E. Creasy, M. Koch, J.F. Tatera, A.L. Robinson, L. Bond, L.W. Burgess, G.N. Bokerman, A.H. Ullman, G.P. Darsey, F. Mozayeni, J.A. Bamberger, M.S. Greenwood, Anal. Chem. 71 (1999) 121R. [43] E. Cleve, E. Bach, E. Schollmeyer, Anal. Chim. Acta 420 (2000) 163. [44] M. Blanco, T. Canals, J. Coello, R. Gasca, J. Gene´, H. Iturriaga, S. Maspoch, J. Soc. Leather Technol. Chem. Analyst 83 (1999) 204. [45] M. Blanco, J. Coello, J. Iturriaga, S. Maspoch, C. de la Pezuela, Analyst 123 (1998) 135R. [46] C.T. Mansfield, B.N. Barman, J.V. Thomas, A.K. Thomas, A.K. Mehrotra, J.M. McCann, Anal. Chem. 71 (1999) 81R. [47] E. Stark, in A.M.C. Davies, P. Williams (Editors), Near Infrared Spectroscopy: The Future Waves, NIR Publications, Chichester, UK, 1996, 704. [48] J. Guthrie, B. Wedding, K. Walsh, J. Near, Infrared Spectrosc. 6 (1998) 259. [49] K. Miyamoto, M. Kawauchi, T. Fukuda, J. Near, Infrared Spectrosc. 6 (1998) 267. [50] D.C. Slaughter, C.G. Cavaletto, L.D. Gautz, R.E. Paul, J. Near, Infrared Spectrosc. 7 (1999) 223. [51] V. Bellon, J.L. Vigneau, M. Leclercq, Appl. Spectrosc. 47 (1993) (1079). [52] G. Bu¨ ttner, Process Control and Qual. 9 (1997) 197. [53] M. Ulmschneider, E. Penigault, Analusis 28 (2000) 136. [54] A. Eustaquio, M. Blanco, R.D. Jee, A.C. Moffat, Anal. Chim. Acta 383 (1999) 283. [55] S.P. Rempel, H.H. Mantsch, Canadian J. of Anal. Sci. and Spectrosc. 44 (1999) 171. [56] B.D. Mindel, Process Control and Quality 9 (1997) 173. [57] D. Coombs, G. Poulter, A.I. Thomson, Spectroscopy Europe 10 (1998) 10. [58] C. Hassell, E.M. Bowman, Appl. Spectrosc. 52 (1998) 18A.