David L. Wetzel Department of eain Science and Industry Kansas Slate University Manhattan. Kan. 66506
ReflectanceAnalysis Sleeper Among SpectroscopicTechniques Near-infrared reflectance analysis (NIRA), which relies effectively upon chemometrics, is a “sleeper” because it is unknown, illogical, or presumed a priori to he illegitimate by spectroscopists and analytical chemists. Examination of the facts concerning near-infrared reflectance m Table I will acquaint the reader with the potential of this technique. Traditionally applied quantitative infrared spectroscopy of mixtures in powder form could hardly be described as a user-friendly technique since extensive sample workup IS required. Thus, for such analytical proh, . 2 methods or inlems,chroi ’
direct separation are necessary for individual determinations of components. Since the 1978 work of Peter Griffiths ( I ), diffuse-reflectance infrared Fourier transform (DRIFT) spectroscopy has received considerable attention. This mid-infrared technique uses spectral subtraction to uncover hidden features within mixtures and has been successfully applied to analysis of pharmaceuticals, highly absorbing samples such as coal, and trace quantities such as catalysis studies of adsorbed chemicals. To date over 200 diffuse reflectance attachments have been sold for FT-IR instruments.
~
-
able l. Facts Concerning Ne include: molecular weight of propylene and ethylene gl moisture content 01 coal and hematite t@xtIleblends of cotton-polyester and of finishes on textile I l k s nt of cross linkage in chemically modified star genation of unsatwated law acid esters h y d W b o n mixture of *hexane. benzene. cyclobe volatiles (loss on drying) and moisture in cosmetics total alkaloids. nicotine. and reducing sugar in tobacco processin moisture in pharmaceutical excipients and in deter giculhral commodities in worldwide unnmerce ar I water, protein, lipid, etc. RBrmacButicals have been identitied by lriables resuming from mukiwavelenm measurem
dusblal applicatlons of NlRA
-
! ~
wid Bnd timely analysis for process control in various processi ion NlRA memodg.
uuu~2700/83/A351-1165$01.50/0 ~
~
0 1983American Chemical Society
~
~~~~~~~
What the Instrument Sees
11
ample preparatlon priw to spectroscopicmeasurem I powders.
Quantitative diffuse reflectance in the near-infrared region has probably not yet received the attention it deserves. In the case where components, present in the amount of 1%or more, of weakly absorbing samples require quantitation, near-infrared (diffuse) reflectance at multiple wavelengths coupled with multivariate statistics might he applied as the method of choice. The convenience of sample handling, computer assistance, and additivity of the near-infrared response makes this possible. Pioneering work in near-infrared transmission spectroscopy was done hv Kave (2.3) ~ ~ ~ . . , of Beckman Instniments”in the 1950s. Classical spectros~ o p in y this region prior to 1969 has been reviewed by Whetsel ( 4 ) . Although commercial single-purpose lual-wavelength (near-infrared) difuse-reflectance Droduction line moisture monitors had existed previously, it was Norris (5) of the Beltsville ‘JSDA laboratory who recognized the iotential of diffuse-reflectance meaxrement in the near-infrared for routine quantitative analysis of major components in agricultural commodities. This latter work caused the appearance of commercial multiwavelength filter instruments in the 1970s by Dickey-John, Neotec (Pacific Scientifie), and Technicon. European instruments have subsequently appeared in the 1980s.
..
~
.n*-.
Most filter instruments in routine use generate raw optical data for only a few wavelengths. However, the spec-
ANALYTICAL CHEMISTRY, V M . 55. NO. 12, OCTOBER 1983
*
ll65A
. . ~ _
__ ._,
Fig-.____: (b) r Data taken horn Reference 6
I
Table II.
,-,
Composition of Cereal Grains and Oilseeds St.rCh
ROdUCl
(%)
Mustard seed Millet Durum
60-65
Soybean
60
-
tral features of commodities scanned over the entire near-infrared region by use of a grating monochromator permit one to observe the available analytical information. Figure l shows the reflectance value expressed as log llreflectance in the near-infrared region for four agricultural commodities composed primarily of starch, protein, water, lipid, cellulose, and other fibrous polysaccharides. A look a t the region of 2100 nm indicates the presence of structures common to starch and cellulose. Data in Table I1 indicate high levels of starch in those commodities exhibiting this peak. On either side of the band a t 2100 nm are absorptions a t 2055 nm and 2180 nm corresponding to amide structures present in protein. In both of the oilseed (soy and mustard) samples the protein peaks are prominent because the baud of carbohydrate is essentially absent. Relative prominence of the central carbohydrate peak to the side amide peaks (i.e., shoulders) provides a pattern typical of the other higher starch commodities. Such preliminary qualitative information is useful as an aid in quantitative work, since high-starch samples may require that measurements be made a t the 2100-nm wavelength and that this term be included in the quantitation equation to com11S6A
a
(d) so
PmMeln (%)
Llpld (%)
FlbN
35.0 10.5 15.5 40.0
25-40 3.9 2.0 21.0
12.0 2.9 3.5 5.5
(%I
pensate for the effect of starch on the background of the analyte (protein) absorption a t 2180 nm. In the areas of 2310 nm and 2348 nm prominent peaks appear for the oilseed samples, indicating the presence of lipid. Data in Table I1
ANALYTICAL CHEMISTRY, VOL. 55, NO. 12. OCTOBER 1983
confirm the high lipid content of these samples. Also, a subtle difference due to CH bands present in ceUulose but not in starch allows one to explain the greater absorption in the 2336-nm and 2352-nm regions for low-fat samples containing high fiber. Figure 2 shows how the reflectance value is subject to differences of scattering and sample penetration due to differences in particle size. The data shown are for a coarse wheat-milling fraction (first mids) from the Kansas State University pilot flour mill. Note that the same material reground on a sample mill (UD-Cyclotec with 1-mm screen, Boulder, Colo.) and by ball mill for 28 h gave respectively higher reflectance (lower log l/R), and less contrast between the “absorption” maxima and baseline of the reflectance spectrum. It can also be noted from the absorption peak a t 1940 nm that water was partially evaporated during the regrinding. Thus, relative granulation and relative water content of the samples are made obvious by selective near-infrared reflectance measurement. It is readily apparent that a visual look a t the spectrum is inadequate to observe many of the spectral features discussed. For this reason, a computer is used to accumulate multiple digitized readings a t selected wavelengths. Comparison of time-integrated signals at anal* (indicator) wavelengths with reference wavelength readings dlows dual-wavelength spectrometry techniques to be applied. The resulting dual-wavelength measurements are then related to differences in chemical comnnsition. Differences
New tools for protein separations. New HPLC column technology from Bio-Rad gives researchers a number of worthwhile alternatives for protein separations. For example, here are three separations of the same sample, using three different approaches- high performance gel filtration, our new high performance ion exchangg column, and our new Hi-PoreTM reverse phase column. Looking at those separations, it's apparent that these HPLC columns provide fast separations with excellent resolution. Not only that, certain compounds appear in the ion exchange and reverse phase separations that don't reveal themselves in the gel filtration separation. HPLC lets you "read between the peaks" with both speed and precision. These new columns complement rather than compete with each other. Our HPLC ion exchange column for analysis of ionizable biochemical compounds is an excellent complement to our Bio-Si1 gel filtration column. By using all three columns, better characterization of proteins (including large ones) may be obtained. As always Bio-Rad is prepared to assist you in plotting a n effective strategy for your protein separations, using any or all of these techniques. Give our applications staff a call, or contact us at
Bio-Rad Laboratories- Chemical Division 2200 Wright Avenue Richmond, CA 94804 (415) 234-4130
Also in Rockville Centre, N.Y., Australia, Austria, Canada, England, Germany, Italy, Japan, The Netherlands, and Switzerland. CIRCLE 29 ON READER SERVICE CARD
ANALYTICAL CHEMISTRY, VOL. 55, NO. 12, OCTOBER 1983
1167A
in near-infrared optical response of samples with different compositions are very small compared to typical ultraviolet, or mid-infrared ana. lytical curves. However, they are reproducibly measurable and are the basis of the success of NIRA as a quantitative technique. Overcoming Apparent Shortcomings It is the nature of absorptions in the near-infrared to he weak since they consist of overtones or combinations of fundamentals. The subtle differences among samples previously referred to require a careful measurement of the signal. NIRA is concerned with observing differences between two samples in milliahsorhance units. For differences in this range to he meaningful, the noise levers must he kept to only a few microahsorhance units. Such instrumental requirements are more stringent than those for most other routine spectroscopic quantitation. The character of diffuse reflectance requires special consideration in the areas of optical design, operation sequence, sample handling, data accumulation, and statistical treatment. Near-Infrared. A few simple fundamental vibrational hands in the mid-infrared region of the spectrum of a particular compound will produce multiple overtones yielding many higher frequency hands in the nearinfrared region that overlap and that are difficult to interpret. Unlike the mid-infrared region, which is valuable as a tool for obtaining structural information, in the near-infrared region such structural information is ohscured. When working in the near-infrared region there is a lack of reference spectra to help one predict what one will see. Knowledge of the midinfrared absorption wavelengths of certain functional groups is of some help in knowing where to look in the near-infrared for overtones. Analytical information may also he obtained by observing a shift in the frequency of the bands due to influence of neighboring molecules. A salt that does not absorb in the near-infrared may he detected and quantitated hy its effect on the water absorption. Such shifts often explain the success of determining nonabsorhing materials because these materials do affect the ahsorption spectra of other molecules present. Weak absorptions in the near-infrared have been cited as a shortcoming. Instrumentally this is true, hut relative absorption strength also serves as a convenient self-limiting factor to restrict which vibrations are observed. For example, a fundamental vibration occurring at 15 pm, with over-
Light incident on a powder or granular material changes directions as it encwnte each individual boundary of powder granule. Reflection. refraction,and diffraction tal place. Diffuse reflection occurs when a portion of the light entering the sample is Scatter1 by boundaries within the sample and exits the body of the sample from the surface entry. In the absence of absaption and with sufficient (effectively infinite) thickness ai multiple scatteringinduced direction changes, a maximum of limt is scattered back a of the powder. Abswption may occur as the scattered light is transmitted between the scatteril boundaries within the sample befwe being scattered back. c he m t of tramrnittance back and farm between me scawing boafhK the opportunity for absorption w the effective thickness (cell path length) of 11 sample. The size and shapa of the sample particles. the voids between them, and the atmu 01 -acting affects the m t (concentration)of material t hw w h i i scam* light must be transmitted between individual scattering boundaries. Some of the preceding factors also affect the mean free path between SCatterii boundaries. The transmission path length (thickness) is shater fw strong scattering (opaque) materials and longer fw translucent materials. Reduction of particle size increaws the scattering of transparent materials. specular reflection. though geometrically well defined. comes ofi individual grains with surfaces of random slope in any direction and is a facta in the measurement of diffuse reflectance. Weakly absabing samples allow a simple specular reflectancecorrection (the mstant b a c k g r d pimariiy dependent on refractive index effectcan be Jubbacted out to move the specular component)handled by regression. A constant specular background results at any wavelength when the refractive ind does not change and the effect of absorption is neolioibl
I
1168 A
-
I ~
! '
.
.
tonesat7.5.5.0.3.75.and3.0umis not likely he observed in the nearinfrared because the fifth or sixth overtones lack intensity. Weak ahsorptions in the near-infrared provide selectivity. The commonly used region is from lo00 to 2500 nm where overtones of fundamental vibrations no higher than 5-8 pm, depending on their intensity, appear. The long wavelength end of the mid-infrared beyond 8 pm does not contrihute to the near-infrared. This means that the overlapping hands in the near-infrared produced by many comhinations and overtones, although spectroscopically complicated, are from only a few molecular groups. In the near-infrared we predominantly see the result of vibrations of light atoms that have strong molecular bonds. If the chemical bond is weak, or the atoms are heavy, the vibrational frequency is low and its overtone will not he detectable in the near-infrared. Therefore, we primarily see chemical bonds containing hydrogen attached to atoms such as nitrogen, oxygen, or carbon, thereby limiting the chemical structures that are observable to fairly simple ones that are common in many organic compounds. These weak overtone hands are more subject to their environment than is the fundamental of the same vibration. A slight perturbation in the bonding scheme causes small changes in the fundamental, hut drastic frequency shifts and amplitude
ANALYTICAL CHEMISTRY, VOL. 55, NO. 12, OCTOBER 1983
changes in the near-infrared. Since these weak hands are broad and overlapping, resolution is not a problem, but reproduction of the same wavelength is essential. The practical result of this is that there are numerous regions in the near-infrared where if wavelength reproducibility is high, signals maximized, and noise minimized, then the optical responses are sensitive to the environment of the absorhing molecules and the number of the molecules present. Thus quantitative measurement can he made and successfully correlated to chemical data obtained by other means. From these data and application of a suitable statistical relationship with appropriate constants, determination of analyte concentrations of unknowns can he made with surprising success. Diffuse Reflectance. Scattering is greater a t the shorter wavelengths of the near-infrared than in the mid-infrared region. In diffuse reflectance, scattering is an important factor. Diffuse reflectance characteristics are summarized in Tahle 111. A volley of photons fired into anonabsorbing, scattering sample will not interact identically. Not all will be scattered hack the same way since some will he transmitted through more of the sample and undergo many more scattering boundaries before exiting the sample. They will have different penetration depths, a different number of scattering events (encoun-
I
250p ESCA
200A Scanning Auger
Multiarray Detector
VG Instruments adds important new dimensions to the analysis of surfaces-the flexibility of ESCALAB will greatly expand the range and in. crease the efficiency of your surfaceanalysis in. vestigations. Now, without moving or changing the sample, you can utilize ESCA, Auger, SAM, SIMS. and other surface methods. How will ESCALAB enhance the results of your analyses? VG’s incorporation of a new high. transmission analyser, a multiarray detector, and the capability of selectivearea analysis offer you: faster analysis time; improved spatial resolution; lower limits of detection-improved sensitivity; and, cleaner, less cluttered spectra for more detailed surface studies. What are the new dimensions that will add these significant new perspectives to your surface analysis investigations? * Small Spot ESCA ( 2 5 0 ~ ) 200A Scanning Auger Multiarray Detector 2000A SIMS For information on how VG can add new dimen. sions to your analyses of surfaces, circle the inquire number or VG instruments.
= --
VG INSTRUMENTS VG Instruments Inc 300 Broad Street Stamford. CT 06901 (2031322-4546 CIRCLE
I
ter a different number of scattering boundaries), and a different length transmission path through the sample. Individual photons of a volley fired into an absorbing, scattering medium will have different probabilities of being absorbed since the effective thickness of the path within the sample will vary with multiple scattering events and penetration depth (Figure 3). Consideration of these facts makes it necessary to accumulate and average data to develop empirical constants and to treat the data statistically. From the detailed list of assumptions concerning the nature of diffuse reflectance as found in Table 111, we see that the following sample treatments are necessary. In practice, the variable of particle size difference from sample to sample and from sample to standards must be minimized. In the development of any NIRA method it is prudent to establish the limits within which one can deviate between mean particle size within an experiment and between different particle size distributions within a given type of sample preparation. Since the wavelength is on the order of 1-2 pm, the lower limit of particle sue should be no smaller than a few micrometers in diameter. The upper limit is based on the need to get a large number of particles in the beam to adequately represent the sample and the need to optimize scatter. This is particularly true in the case of heterogeneous materials. In general, the median diameter is on the order of 100 pm, and it is desirable to avoid tc broad a distribution. The overriding factor on particle size, however, is to reproduce a mean particle size and particle size distribution for each type of material and standards of that material. The nature of diffuse reflectance involves a change in the direction of the light which comes off randomly in all directions hut varies with the angle of observation from the normal. As a detector is moved away from the angle of incidence, the intensity will decrease. From this consideration, we observe that the instrument design is important. There is radiation coming off 360° around the sample and coming out in a solid angle around the incident beam. It is advantageous to collect a maximum amount of this radiation for three reasons: first, to maximize the signal; second, to average all directions; and third, to represent all desirable angles from the normal. For samples of interest in certain analyses, peculiar circumstances favor a simplification to correct for the specular effect referred to in Table 111. The specular contribution involves both refractive index and absorption.
Figure 3. Various pathways of diffusereflectance from weakly absorbing scattering particles
Fortunately, in the near-infrared the refractive index effect is essentially constant. Also, the absorptions are usually weak, and thus constancy of the specular background is common for many organic samples. The right combination of high scatter, weak absorption, and nearly constant index of refraction allows scattered back transmission within the sample to be measured in the presence of a steady-state specular reflectance background. Under optimum conditions the response is sufficiently linear over the measurement range. Much of the past success with plant materials, for example, may be due to the constancy found in nature. Consideration of the nature of nearinfrared absorption and diffuse reflectance measurement allows one to realize both the quantitation possibilities and practical limitations and the necessity for empirical statistical input. InstrumentalConsiderations. The primary instrumental considerations are those that provide a high
1170 A * ANALYTICAL CHEMISTRY, VOL. 55, NO. 12, OCTOBER 1983
signal-to-noise ratio. A high-intensity stahle source and a low-noise highly sensitive detector are required. ALSO the ability to capture the maximum amount of the diffuse component of the reflected radiation a t multiple angles and to reject the specular component is needed. The mechanism and frequency of optical referencing may contribute to favorable noise and drift characteristics. Intensity of the light scattered back varies with angle from the normal, diminishing a t larger angles. In practice, the angle of observation is fixed by the design of the instrument. Since a slight vertical difference in positioning of the sample would result in a difference in the angle observed by the fixed optical components, a mechanism for reproducible positioning of the sample is necessary. Packing a granular sample against a window regiments the orientation of the sample so that the orientation is no longer random. The degree of regimentation affects the intensity mea-
THERE ARE STILL A FEW THINGS AS PORIABLEAS AN OHAUS BALANCE. It's getting harder every day to stay one hop ahead of everyone else. Now Ohaus makes it easier with our Port-0-Gram electronic balances. They not only go anywhere to do almost any weighing job. They do it for as little as $395. They're small enough to fit in the palm of your hand (or pouch). Inside, microprocessor electronics compute weights instantly and pop up bright, bold rock-steady readings. Just touch the pushbutton controls and you're ready to weigh in grams, ounces or pennyweight. Some models even allow select parts counting or your own customized weighing units. Long battery life lets you take it with you on the line, in the labs, indoors or out. When you're staying put, an AC adapter's included. For over 75 years, Ohaus has been making dependable balances of rare quality The Port-0-Gram balances are no exception. Now why would you go
with any other balance?Send for our new catalog today. Ohaus Scale Corporation 29 Hanover Road, Florharn Park, NJ 07932
0 1983 Ohms Scale C o r p ~ r a t ~ oOhaui n and Porl~O~Gram n e regsitered tradernarks of Ohaur Scale Cormation. Prices and rpeallcalanr are rublecf Io change withoul nollce. CIRCLE 156 ON READER SERVICE CARD
sured a t any angle. The amount of regimentation varies with the compacting pressure and with particle size and shape. The direction of observation can have an effect on the intensity measured a t a given angle depending on whether the observation is against the grain or with the grain of a textured sample. Such is the case with naturally textured samples as in meat. Surfaces may be oriented preferentially in one direction by polishing, slicing a section of sample, or by striking off as in the case of an uncovered powder or granular surface. Sample observation variables addressed by the instrument design, including the median angle of observation, angular boundaries of observance, and the direction of observation, are summed up in the designation of solid angle. The observation may be simultaneous for different angles and directions, or sampling of a particular angular segment may be in one direction continuously or in different directions intermittently. The signal intensity depends on the fraction of reflected light hitting the detector (solid angle of observation) and on the total flux from the source that strikes the sample. Wavelength selection by interference filters or interferometry does not limit optical throughput; however, in using a grating monochromator this is a major consideration and optics with f/2 or less are desirable. The instrument operator is responsible for sample preparation and must ensure that procedures for unknowns and calibration standards are kept the same. This includes grinding (homogenization of liquid suspensions), storing of the sample, and packing and positioning of the sample cell. Any change requires experimental verification that analytical results have not been affected. Instrumentation The typical operating sequence of a routine instrument for NIRA involves measurement of reflected intensity off the sample surface at a number of wavelengths and off a standard reference reflecting surface at those same wavelengths. The reflectance measurement in practice is thus a relative measurement to a standard reflector. The term Io (incident intensity) does not appear in the practical relationship. Reflectance = I(sample)/I(reference). Such a relationship exists at every wavelength measured. The traditional Kubelka-Munk function with appropriate scattering and absorption constants has generally given way among NIRA practitioners to logarithmic reflectance terms used with empirical coefficients that accommodate the scattering effects as part of the 1172A
calibration. The general analytical expression is of the type: % = z a log 1/R1+ b log 1/Ra c log 1/Rs . . . A commonly used term involves a coefficient multiplied by log 1/R as shown above. The equation has been modified in some instances to allow for a transformation of R that may combine data from a number of wavelengths into one term. The result of this is that there are fewer terms in the equation but each of these terms becomes considerably more complicated. Modern Hardware Features. Commercial instruments commonly use tungsten-quartz-halogen sources and lead sulfide detectors, but the mechanisms for referencing to a stnndard reflector and collecting the diffusely reflected radiation differ considerably (8). Intermittent referencing of reflected radiation intensity from sample to maximum reflected intensity from a standard reflector allows each measurement to be expressed as a ratio. In the most elaborate referencing system, this is done a t each wavelength reading before indexing to the next sample. In other instances, sample measurements are made a t all selected wavelengths followed by measurement of the standard at those same wavelengths. Multidirectional (360") reflectance sampling allows simultaneous collection of radiation in all directions with an integrating sphere, time-averaging with rotating sample, or use of multiple detectors. A maximum solid-angle (approximately 2 T steradians) simultaneous reflectance collection is obtained with an integrating sphere. Available commercial NIRA instruments include interference filter and grating monochromator systems. Commercial grating monochromator instruments designed specifically for NIRA are specialized systems with appropriate computer software and high-throughput monochromators. The Technicon InfraAlyzer 500 performs incremental step scanning. The Pacific Scientific 6350 accumulates rapid scan data. Diffuse reflectance accessories are commercially available for use with existing FT-IR instruments. In filter instruments two types of mountings are the continuously moving type and the turret wavelength selection of discrete narrow band interference filters. In the second type (Figure 4),radiation intensity is measured at one wavelength and stored in the computer and, after rotation of the filter mount to the next preselected filter, the process is repeated. In the continuously moving filter mount,
+
+
ANALYTICAL CHEMISTRY, VOL. 55, NO. 12, OCTOBER 1983
+
multiple wavelengths are obtained from the same filter depending on the angle of tilt a t the instant of transmission through this filter. There are three levels of filter instruments (minimal, basic, enhanced) from which to choose for analysis by multichannel discrete wavelength measurement and computer data treatment. Enhanced systems have auxiliary computer hardware and software to allow rapid in-house custom calibration and method development. Other channels or wavelengths may be added to, or substituted for, those commonly used on the basic system. The system may allow direct raw data transfer to a computer equipped with statistical software suitable for identifying the channels best correlated to the quantity of analyte of interest by the "t7'test or other statistical functions. Statistical terms as well as calibration coefficients typically result from multiple linear regression of data from a suitable training set. Such a set is selected to include a suitable range of analyte concentration in a matrix that contains other constituents in a typical variation cross section for that product. Good reference classical laboratory analytical data are a prerequisite for success. Developing a Method In NIRA the principle of multiplewavelength spectroscopy is used. At least one wavelength is responding to the amount of a particular analyte in the sample matrix. The measurement of at least one other wavelength is used as a reference of the overall reflectivity of that particular sample matrix. In some instances a third, fourth, fifth, etc., wavelength may be used. These may also perform a reference function or may give either a negative or positive correlation to the analyte. To obtain the most transferable (robust) calibration, one should minimize the use of data at wavelengths that are not essential to the calibration. When simultaneous analyses are performed, data are collected at the wavelengths required for each of the analyses, and a different equation is solved for each different analyte. In each equation the corresponding correct wavelengths appear, and their appropriate coefficients are used. We have thus the capability for simultaneous multicomponent analysis. For example, one set of reflectance measurements for milk could be used to determine protein, fat, total solids, and lactose by the solution of four different equations using the reflectance data at the appropriate wavelength in each equation, and the proper regression coefficients for each wavelength. Experimental. To perform a calibration, it is necessary to assemble a
Microbore .HPLC. Compromise Is l’t n
T
i
1
Higher sensitivity and more resolution at very low flow rates -that’s the promise of micmtoE HPLC. Now that promise ismetmfidl by L W M i l t o n b y ’ s new components, offering you the first system truly designed for microtore HPLC-without compmmise! No adaptation of existing components. LDCNilton by’s new microMetric” Metering Pump offers true pulse-free flow from 1-999NUmin with better than 1%of thumbwheel setting accuracies. A 5 mlsyringe capacity permiamultiple chromat-~,grams.Refill isaccomplished~in oneminute. High pressurecapability to 10.000psiforboth
today’s packed columns and tomorrow’s capillary columns. New spectroMonitot. D with microCell” offers crossflow” optics to minimize thermal and flow noise without a bulky heat exchanger. A 1mm tubular quartzfused-silicacellpermits direct coupling to microbore columns for maximum efficiency. Uniquefiberopticsand high performancemirror collect and deflect light to the integral preamplifier circuit board. Learn more about the marvels of microbore HPLC, and all about how LDC/Milton b y has packaged all the benefits in one new. uncumpnnising set of components. Write or call for further information.
LDC/Milton Roy US. HEAOOUARTERS. P.O. Box 10235. Riviera Beach. F L 33404 Tollfree: 800-3276162. In Florida: 13051 844-5241 lnlernsfional Telex: 441098. DOrnslVE Telex: 513479 SUNNYVALE OPERATION, 560 Oakmaad Parkway, Sunnyvale. CA 94086 Toll-frw: 600-538-1595, I n California: 14081 736-0300 UNITED KINGDOM. LOC UK. Milton Roy HOYYI. 52 High Street. Stone. StaffordshireST15 8AR EnQlandPbane: 78581.3542 Telex: 36623 WEST GERMANY. Millon ROY10)cutsrhlandl GMBH JehnSrBYe 22.24 6467 Hal%?elrOth 2. Well Germeny Phone: 6055-2031 133651 Telex: 4184357 JAPAN. Mil,nn R n u K.K..Tstanawa Oai-lchi Ruildino 11.41. Takanawa ?-Chome. MinatwKu. Tokyo 108. Japan Phone: 1031473-3181 Telex: 24242%
-
CIRCLE 127
ON READER SERVICE CARD
Position 2-
-
A
Position I
-
0
Lens
Lens rret-mounted Filters
1,
Reference
I 'I/
4 \
/
Detector
Detector Sample
Flgure 4. Typical turret-mounted dis-
crete filter instrument (Technicon InfraAlyzer 400) Minm position 1, reR8ctBncB dl the sample: mirposition 2, refieCtance on the s t a n m re*ence rellecting &ace. Bath detectors collect radiation in the sphere wlth the m i r r a in either pwi-
.:L l Y l
collection of preanalyzed samples of the product of interest. This collection should contain the full range of the analyte. Fur example, if you are anal y i n g protein contenr in wheat you should have a range of perhaps 9 1990 protein. The learning set should also be representative in terms of the matrix material. If you plan to analyze wheat of a number of varieties. different varieties should then be used to assemhle the learning set. With this set of samples (minimum of 30), and classical laboratory analysis data such as the Kjeldahl protein value, optical data are collected for each ofthe Samples. While this occurs the Kieldahl value of each sample is entered into the computer. Correlation transformation, typically multiple linear regression, is then performed. The result of this statistical procedure is the best set of coefficients by which each respective logarithmic reflectance re. sponse is multiplied to solve the analytical equation w d spectroscopically determine the percent protein. Once this is done, tne coefficientx for each wavelength are programmed into the built-in computer of the NIR analyzer. Another set of samples, also prean. alyzed, is then analyzed by NIR. T o ohtain calculated (NIRA) values, the equation is solved using the appropriate coefficients and the specwosn)pic response. When these are compared t the corresponding classical laboratory values a standard error of difference
between the classical values and those determined spectroscopically gives the operator an idea of how good his calibration is. Standard errors of difference for routinely performed commodity analyses are expected to be on the order of 1or !2% relative. Computation. The analyte concentration is correlated to the spectroscopic response by individual wavelengths (log liR),by difference between pain (log Rz-log R1), and by trios (2 log Rz-log R1-log RB)where the reflectance of the central wavelength is mathematically weighted in comparison to the contribution of reflectance a t wavelengths incrementally spaced on either side. Plots or tables of correlation by wavelength result from single, pair, and trio reflectance measurement subjected to simole reeressiou. Correlation is also determined by direct application of multivariate statistical techniques. Multiple linear regression, for example, is commonly applied to groups of wavelengths preselected by filter choice. For example, reflectance readings taken at 12 disCrete wavelengths, not necessarily re. lated M each other. would appear as 12 separate terms in the multivariate treatment (Table IV, Equation a). Reverse stepwise multiple linear regression beginning with 12 individual reflectance terms is commonly used to select a simplified expression containing two to five of the original 12 terms which in use will be more robust. Similarlv the best combination of three or of two, found by successive multiple
linear regression of all different combinations, is used as a starting point. Wavelength terms may then be added by forward stepwise multiple linear regression. Ultimately, by either process, a usable expression is produced that contains the optimum number of wavelength terms, the optimim wavelength choice, and good regression coefficients for each of the terms in the expression. An alternative approach to generate an expression for analytical use based upon the measurement at 12 wavelengths is to composite them by pairs or trios. (Table IV, Equations b and d). In this way a restriction is imposed on the data prior to statistical treatment. Twelve measurements composited by trios produce an expression containing four complex reflectance terms. Remession or other multivariate treatmenta are subsequently used to generate coefficients or eliminate unneeded reflectance terms. Even the composite reflectance terms may be cornposited (Table IV, Quations c and e). Such is the case when 12 reflectance measurements are composited by trios and subsequently divided. Both the individual reflectance term and the composited reflectance term calculation methods have been used successfully with the instruments that they complement. Software needs are dictated by instrumental hard. ware, and the final test is performance of any analysis system as a whole. Developing a calibration for an analvte in a oarticular matrix involves 1) choosingthe wavelengths for incorpo-
4
Table IV.
Typical ComputationalAlogorithms
+ b log i/Rz + c log i l 5 + . . + a ( M R z - log R,) + b(log R, - log 5) + .. .
(a) % = z + a log l l R , (indindual)
(b) % = 1 (pair: first dilference)
(c) % = z + a (logRz-108Rq)
loa R s -
+
logR3
(log&-
log Rs)] + . . . log Re- log RI
(pak: ratio f h t dillweme)
+
-
+
'
-
+ . ..
(d) % = 1 a(2 log Rz - log R t log R d b (2 MRs log R, - log Re) (trio:second differ-) 2 log Re - log RI - log Re (e) % = z + a 2 lag RZ - log R1 logR3 .'IogR,, - logRlo - log R, (2 log Rs log & - log Rs) (bio: ratio of semnd dllfw-s) Unt Rt. R2. Rs, R, . . R. represent reflmamss m ader by wavelmglh.
-
".-
1174A * ANALYTICAL CHEMISTRY, VOL. 55, NO. 12, OCTOBER 1983
+
3+
I
I The new LDC/Milton Roy spectroMonitorBDvariable wavelength detector. On the outside, the newest variable wavelength HIGHEST EFFICIENCY ANDSENSITIVITY detector from LDC/Milton Roy looks very much like its predecessors. But on the inside, where ir -"t** *e sWctroMonitor""D VariaMa wavelength detector sets new industry standards for WliN and performance.
STANDARD Our new max NTM series high efficiency, high pressure (1000 psi) fluid cell IS a standard feature in the spectroMonitor D. The competition makes you pay extra for high efficiency. In addition, the max N series fluid cell has optimized flow dynamics and a LOWEST NOISE AND DRIFT trOe 1Omm pathlength to maximize SensitivltY acThe new SpectroMonitor D variable wavelength decording to Beefs Law. Others have tried to beat our teetor deliversthe lowest (2 10-5 AU peak to peak) and drift (1 10-4 AIJ/~~.) avail. noise and drift specifications by cutting cell pathlength, but in the process, they decreased sensitivity. able today. Highly efficient dual beam optical system electronics make these specifi. With the spectroMonitor D you get it all - low noise, and low drift and high sensitivity. cations possible. UVNIS MEASUREMENTS The spectroMonitor D also features absorbance measurements over the entire UV/VIS spectrum (190700nrn) wirhwr a lamp change. One 35-Watt deuterium lamp is all you need to buy or replace. DIGITAL DIAGNOSTICS New LED diagnostic circuitry. introduced in the spectroMonitor D, provides you with a switch selectable digital display of absorbance units, reference and $ample energy in mV. lamp current in mA, and zero offset.
'
1 *e
ATRADITIONOF QUALITY ANDPERFORMANCE LDCIMilton Roy pioneered HPLC technology over a decade ago. We have put all of our technoloaical exparience and knowledae into makino an initrumant that is worth much m&e than i t s $5895.00 price. Let us sh SpectroMonitor D can work for you.
For further information call our Technical fiepresenrativesat
CIRCLE 128 ON READER SERVICE
800-327-6182.
CARD
ANALYTICAL CHEMISTRY, VOL. 55, NO. 12. OCTOBER 1983
-
1175,.
Electrochemical and Spectrochemical Studies of Biological 3edox Components
4dvances in Chemistry Series
uo. 201
