Anal. Chem. 1990, 62, 1778-1785
1778
Determination of Microstructure and Composition in Butadiene and Styrene-Butadiene Polymers by Near- Infrared Spectroscopy Charles E. Miller,*,*B. E. Eichinger,' Thomas W. Gurley,2and James G . Hermiller2 Center for Process Analytical Chemistry, Department of Chemistry, BG-10, University of Washington, Seattle, Washington 98195, a n d Analytical Science Department, Goodyear Tire and Rubber Company, 142 Goodyear Boulevard, Akron, Ohio 44305
Transmission spectroscopy In the near-infrared reglon ( 1100-2500 nm) is used to determine the microstructure and the composition of poly(butadlene) (PBD) polymers and styrene-butadiene (SBR) copolymers in bulk and in carbon tetrachlorlde solutlon. The multlvarlate method of classical least squares (CLS) Is used to analyze near-infrared spectra of polymers wlth NMR-determlned microstructures and compositions. Although the near-Infrared spectra of the pure analytes (c/s-1,4-butadiene, trans-1,4-butadiene, 1,2-butadlene, and styrene) are highly overlapped, the CLS method provkles accurate predlctlons of analyte concentratlons, because all available spectral frequencies are used for quantitation. The results indicate that the near-infrared spectrum In the flrstovertone CH stretchlng regkm (1570-1850 nm) can be used to predict 1,2-butadlene within 0.947% (mass), c/s1,Gbutadlene withln 1.03 %, trans-1,4-butadiene within 1.15 % , and styrene within 1.38 % In PBD and SBR polymers. The second-overtone CH stretchlng region (1 100-1350 nm) and the CH comblnatlon band region (1950-2500 nm) are also very useful. The sensttMty of near-lnfrared spectroscopy to intermolecular interactlons and nelghboring-group effects in these polymers is demonstrated.
INTRODUCTION Poly(butadiene) (PBD) and styrene-butadiene copolymer (SBR) are used extensively in the tire and rubber industries ( I ) . These polymers contain sites of unsaturation that can react with cross-linking agents to form elastomers. Three different types of unsaturation can be present in PBD polymer chains (1,2) (see Figure 1): 1,2-butadiene, cis-1,4-butadiene, and trans-l+butadiene, which result from different stereospecific additions of monomers to growing polymer chain ends during polymerization. The relative amount of these three structures (called the microstructure) greatly affects the physical properties of PBD elastomers formed from the polymers (2). For SBR copolymers, the butadiene microstructure and the amount of styrene incorporated into the polymer affect the physical properties. As a result, microstructure and composition information about PBD and SBR polymers can be used to predict the physical properties of their elastomers. NMR (3-6) spectroscopy and infrared ( 7 , 8 )spectroscopy can accurately determine the microstructure and the composition of PBD and SBR. However, these methods usually require extensive sample preparation, typically dissolving the polymer in a solvent or pressing the polymer into a thin film.
* Present address: Max-Planck-Institute fur Polymerforschung, Postfach 3148, D-6500 Mainz, FRG. University of Washington. Goodyear Tire and Rubber Co.
Both of these preparations are time-consuming and might alter the spectral properties of the samples. As a result, these methods are usually not suitable for rapid process analysis. Therefore, a method for polymer analysis that involves minimal or no sample preparation is desired. Spectroscopy in the near-infrared (near-IR) region (1100-2500 nm) (9-11) has been used to analyze bulk agricultural products and can be similarly used to analyze bulk polymers. Near-IR spectroscopy can perform these simple analyses because the absorptivities of vibrational overtone and combination bands in the near-IR region are orders of magnitude lower than the absorptivities of vibrational fundamental bands in the infrared region. In this work, it will be shown that useful near-IR transmission spectra of relatively unprepared bulk polymer samples can be obtained. For the most part, near-IR spectra contain bands that arise from OH, CH, and NH groups in a sample. Near-IR bands of individual analytes are generally broader and more overlapped than bands in mid-infrared spectra. As a result, it is difficult to determine an analyte concentration from only one or a few near-IR absorbances. However, full-spectrum multivariate calibration methods, such as partial least squares (PLS) (12-15), classical least squares (CLS) (15-17), and principal components regression (PCR) (12),have been used to accurately determine analyte concentrations from near-IR spectra. Earlier near-IR analyses of PBD polymers (6) and unsaturated hydrocarbons (18) demonstrated that near-IR spectroscopy can be used to determine the microstructure. However, these analyses used calibrations with only one or two specific near-IR absorbances, which had difficulty discriminating between highly overlapped cis-1,Cbutadiene and trans-1,4-butadiene absorptions. In this work, the method of CLS (16,17)is used to determine the concentrations of all structural units in PBD and SBR polymers. The theory and the application of the CLS method (also called the K-matrix method) are reported in earlier papers (15-17, 19). In the calibration phase of CLS, the near-IR spectra of the calibration samples and the reference concentrations of all possible analytes in the calibration samples (in this case, determined by I3C NMR spectroscopy) are used to determine the least-squares estimate of the pure analyte spectra. The prediction of analyte concentrations in an unknown sample is then accomplished by fitting the estimated pure analyte spectra to the spectrum of the unknown sample. At this point, the relative contributions of the different pure analyte spectra to the sample's spectrum, hereafter called the prediction coefficients, are determined. In order to perform a CLS calibration, reference analyte concentrations for all analytes must be known, and the Beer-Lambert Law must hold for all analytes in the Calibration samples. It is also assumed that each of the four different structural units in the PBD and SBR polymers
0003-2700/90/0362-1778$02.50/0C 1990 American Chemical Society
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990
H \
/H
5'\ /c= CH
\H
1,2 Butadiene
Cis-l,4 Butadiene '2
Trans-I ,4 Butadiene
I-
C
+5C,H
Figure 1. Chemical structures of four different analytes in PBD and SBR polymers.
(Figure 1)contributes the only unique spectral features in any spectrum. Spectral effects that depend on unknown chemical properties, like molecular weight, sequencing of structural units, and crystallinity, can make this assumption inaccurate. Although a CLS calibration can approximately model unknown spectral effects (16), significant errors in prediction can occur because the unknown effects are not explicitly modeled in the calibration. I t should be mentioned that the near-IR spectra of polymers can be affected by the relative amounts of block and random segments in the polymers, As a result, separate calibrations for random and block polymers might be necessary. A final requirement for CLS calibrations is that each of the calibration spectra is normalized to a reference sample concentration and thickness. In this work, the normalization of a polymer solution spectrum with respect to the total concentration of the polymer in the solution is accomplished by dividing each absorbance value in the spectrum by the factor C/Co, where C is the concentration of the polymer in the sample and Co is an arbitrarily chosen reference polymer concentration. If the requirements for CLS calibration iwe satisfied, several advantages for prediction are obtained. Accurate CLS predictions from a spectrum with unknown base-line offset effects can be achieved by fitting a constant base-line spectrum, in addition to the estimated pure analyte spectra, to the spectrum during CLS prediction (16). Constant base-line shifts in the spectrum, caused by nonreproducible sample placement in the spectrometer and by varying physical states of the samples, only affect the fit of the constant base-line spectrum to the prediction spectrum and do not affect the fit of the estimated pure analyte spectra to the prediction spectrum. Furthermore, if sample absorbances are within the linear range of the spectrometer, the relative values of the prediction coefficients (which are equivalent to the predicted analyte concentrations in percentage) for the pure analytes do not depend on the sample thickness or the polymer concentration in solution. As a result, a CLS calibration can predict the analyte con-
1779
centrations of a relatively unprepared sample of unknown thickness. In this work, CLS calibrations for polymer solutions can be constructed directly from polymer solution spectra, because the polymer concentrations of the solutions are known and the sample path length is constant for all calibration samples. However, bulk polymer spectra, which contain unknown multiplicative and base-line effects, cannot be used for CLS calibration. Fortunately these effects can be estimated with CLS calibrations constructed from polymer solution spectra, provided that the differences between polymer spectra in solution and in bulk are not very large. The estimated multiplicative and base-line effects can then be removed from bulk polymer spectra to prepare them for CLS calibration. The CLS method not only offers advantages in quantitative analysis but also provides valuable qualitative results. The estimated pure analyte spectra, determined during a CLS calibration, are the closest approximations of the spectra of the pure analytes in the samples. In the case of polymer analysis, these spectra can be used to assign near-IR bands to specific functional groups in the polymers, to determine the presence of interactions between different functional groups in the polymer chains, or to detect differences in the structural states of specific groups in different states of the polymer.
EXPERIMENTAL SECTION Materials. Twenty-three PBD polymer samples and 17 SBR copolymer samples were obtained from the Goodyear Tire and Rubber Co. AU polymers were obtained in a bulk state. Reference microstructures and styrene contents of the polymers were obtained by carbon-13 NMR spectroscopy (at 300 MHz) of the polymers dissolved in CDCl, (20,21). The estimated error of the NMR reference method is 1%mass for all analytes. All polymers were analyzed by near-IR spectroscopy in bulk and in CCl, solution. Polymer solutions of approximately 1% (w/v) were prepared by dissolving a weighed amount of polymer (approximately 0.25 g) into 25 mL of CCl& A polystyrene solution was prepared by placing approximately0.25 g of polystyrene (M,, = 5.56 X lo6, Goodyear Tire and Rubber Co.) into 25 mL of CCL. Near-IR spectra were taken with a Pacific Scientific 6250 grating monochromator instrument with a lead sulfide detector. The nominal resolution was 10 nm, the wavelength accuracy was h l nm (NBS Standard Reference Material 1920 was used to align the wavelength scale), and the spectral region was 1100-2500 nm. All spectra were obtained in transmission mode. Each scan lasted about 30 s. Duplicate scans were obtained for most of the solution and bulk samples used in this analysis. These duplicate analyses, and the NMR reference analysis, were performed on different physical samples of the polymer. Spectra of solutions were obtained by placing approximately 10 mL of the solution in a 4 mm path length quartz cuvette with a Teflon cover. The cuvette was then placed in the spectrometer for near-IR sampling. A cuvette filled with CCl, was used as a reference. Each solution spectrum was corrected by the subtraction of the CCl, reference spectrum, the subtraction of the absorbance value at 1100 nm from all absorbance values, and the subsequent division of each absorbance in the spectrum by the total concentration of polymer in the solution in grams/25 mL (Le., Co = 4 % w/v or 1 g/25 mL). Bulk polymers were analyzed by placing a piece of polymer, approximately 0.5-1.0 mm thick and 10 mm wide, on one quartz plate of a two-plate quartz cell. A cover plate was then placed on the sample to form a "polymer sandwich" between quartz plates. For some samples, the thickness of the sample was reduced by compression of the sample between the two plates. The two-plate cell containing the sample was then placed in the spectrometer for near-IR analysis. A reference spectrum was obtained by scanning the empty two-plate cell. Each bulk polymer spectrum was corrected by the subtraction of the reference spectrum before multivariate analysis. The thicknesses of some bulk samples were determined with a caliper (Randall and Stuc kney).
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990
1780
Table 11. NMR-Determined Microstructures and Compositions of Styrene-Butadiene Copolymers
Table I. NMR-Determined Microstructures of Polytbutadiene) Samples sample
sample type'
1 2
C P P P P P P C C P P C P
3 4
5 6 7
8 9 10 11 12
13 14 15 16 17
C C C C
analyte concn, % mass 1,2-BDb cis-1,4-BD trans-1,4-BD 86.2 84.4 9.7 10.1 1.7 10.2 1.9 10.2 9.7 10.4 1.9 12.3 64.7 33.8 12.5 0.4 64.1 43.5 4.6 43.2 28.6 23.6 18.7
4.4 6.0 37.1 37.9 95.8 37.2 95.2 37.3 35.9 36.4 96.8 4.6 13.2 25.9 3.3 98.3 14.2 21.0 89.5 22.3 31.4 5.2 7.0
9.4 9.6 53.3 52.0 2.5 52.6 2.9 52.5 54.4 53.3 1.3 83.1 22.1
40.3 84.3 1.2
21.8 35.5 5.9 34.5 40.0 71.2 74.4
18 C C 19 20 P C 31 C 39 P 40 Squared Correlation Coefficients of Cross-Correlation of Analyte Concentrations
1,2-BD cis-1,4-BD @
cis-1,4-BD
trans-1,4-BD
0.373
0.0486 0.406
C: calibration sample, P: prediction sample. BD: butadiene.
Data Analysis. All CLS analyses were performed on an IBM-AT microcomputer. The near-IR spectra of the polymers were split into three spectral regions for separate analyses: region I (1100-1350 nm), region I1 (1570-1850 nm), and region I11 (1950-2500 nm). Region I11 contains combination bands from CH vibrations, which are stronger than the overtone bands in regions I and 11. Region I11 was useful for the analyses of polymer solutions, because the solvent used in this work (CC14)is transparent in this region. For each analysis, the ability of the near-IR/CLS method to determine the microstructure and the composition was determined by the construction of calibration from approximately half of the available samples and the subsequent use of the calibration to predict the microstructures and the compositions of the remaining samples (which are the prediction samples) from their spectra. For the PBD studies, the calibration set contained samples that were distributed over the range of the cis-l&butadiene, transl,4-butadiene, and 1,2-butadieneconcentrations of the available samples and contained the samples with maximum and minimum concentrations of these components. As a result, the concentrations of the remaining samples (the prediction samples) were distributed within the concentrations of the calibration samples. The choice of calibration and prediction samples for the SBR studies was done in a similar manner, where the distribution of only styrene and 1,2-butadiene concentrations was considered. The analyte concentrations and the sample selections of the PBD and SBR samples used in this work are listed in Tables I and 11, respectively. PBD and SBR analyses were performed separately. In each case, a polymer solution analysis was performed in each of the three spectral regions and a bulk polymer analysis was performed in both regions I and 11. In polymer solution analyses, solution spectra were used for calibration, and solution spectra were used for prediction. For each bulk polymer analysis, a CLS calibration constructed from solution spectra was used to determine base-line and multiplicative effects in the spectra of the bulk calibration samples. These spectra were then corrected for these effects and subsequently used to construct a CLS calibration. This calibration
sample 21 22
23 24
25 26 27 28 29 30 32 33 34 35 36 37 38
sample type@ P C P C C C C P P P C P P C C P C
analyte concn, % mass trans-l,4styrene 1,2-BD cis-l,l-BD BD 28.5 17.7 25.9 31.3 26.1 10.3 33.2 17.4 12.3 15.5 16.9 19.6 13.2 22.9 10.8 22.6 12.0
26.9 38.2 36.6 39.4 7.1 55.3 23.0 39.8 49.3 11.6 8.2 27.0 9.3 22.9 8.5 30.3 50.1
15.6 15.8 13.1 10.8
25.3 11.3 13.9 15.0 14.2
27.8 27.9 19.1 30.1 18.6 32.9 19.5 14.7
29.0 28.3 24.3 18.4 41.4 23.2 29.9 27.7 24.2 45.1 47.0 34.3 47.4 35.6 47.8 27.7 23.1
Squared Correlation coefficients of Cross-Correlation of Analyte Concentrations
styrene 1,2-BD cis-l,l-BD
1,2-BD
cis-1,4-BD
trans-l,.l-BD
0.180
0.128 0.749
0.0765 0.826 0.926
was then used to determine microstructure and composition from the spectra of bulk prediction samples. For some analyses, second-derivative spectra (Pacific Scientific Co.) were used for calibration and prediction. The statistic used to evaluate prediction error is the standard error of prediction (SEP):
where Ci,jAisthe known concentration of analyte i in prediction sample j , CL is the concentration of analyte i in prediction sample j that is predicted by using the CLS calibration, and NP is the number of prediction samples (including replicates). In addition, an evaluation of the repeatability of the predicted concentrations for replicate samples was made for each analysis. The root mean square deviation of replicate predictions (hereafter abbreviated RMSD,,,) is calculated according to
where e k , and e k ,2 are the concentrations of a n a l e i predicted from the two replicate scans of the sample and NRP is the number of replicate sample pairs in the prediction set (which is not necessarily equal to NP/2, because some prediction samples did not have replicate scans). The ability of the CLS method to model the PBD and SBR spectra is investigated by a comparison of spectral residuals of prediction with spectral noise levels. The percent spectral residual of prediction (abbreviated % SR) is calculated according to
where a j p is the absorbance value of prediction sample j a t wavelength k , C j p is the value of the modeled spectrum (the part of the spectrum explained by the CLS calibration model) of prediction sample j a t wavelength k , NP is the number of prediction samples, and NW is the number of wavelengths in the spectrum. For the purposes of spectral residual analysis, base-line shifts in bulk polymer spectra were not considered to be meaningful spectral variations. As a result, only CLS analyses that
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990
1781
Table 111. Prediction Errors for Near-IR-Determined Microstructures of Poly(butadiene) Samples, Using the CLS Method spectral region used
spectral correction'
I I I1 I1 I11 I11
none second deriv none second deriv none second deriv
I none I second deriv I1 none I1 second deriv Second deriv: second-derivative spectral correction.
SEP (RMSD,,,,) for different analytes, % mass 1,2-BD cis-l,l-BD trans-l,a-BD
Solution Analyses 11.2 (10.6) 3.31 (5.61) 4.27 (2.56) 0.947 (0.10) 2.37 (1.15) 0.952 (0.16) Bulk Analyses 17.1 (12.5)
33.8 (24.7) 7.46 (5.45)
0.978 (1.67) 1.03 (1.35) 2.44 (2.27) 2.46 (2.78)
7.45 (6.43) 5.31 (4.94) 2.93 (4.15) 1.80 (2.86)
1.90 (1.71)
1.78 (0.85) 1.51 (1.43)
use second-derivative spectra were considered for spectral residual analyses. The spectrum of a SBR solution (about 1% w/v in CCl,) and the spectrum of a bulk SBR sample (about 1mm thick) were used to estimate the noise levels for near-IR spectra in all three regions. The percent root mean square of spectral noise (abbreviated % RMSJ was determined for both samples in each spectral region:
where ae&is the absorbance of a "noise spectrum" (obtained by subtracting two spectra of the same sample taken in rapid succession) at wavelength k, de& is the average absorbance value of the noise spectrum, and aN,k is the absorbance of the sample spectrum at wavelength k. Only second-derivativespectra were used for determination of %RMS, values.
RESULTS AND DISCUSSION PBD Analyses. The experimental design for PBD polymers was dictated by sample availability. Fortunately, the known analyte values of available PBD samples (see Table I) are well-distributed. The intercorrelations of analyte values for the available PBD samples (shown in Table I) are low in all cases. The prediction results for PBD solution analyses are shown in Table 111. In most cases, the S E P value for 1,a-butadiene content is the lowest of all three analytes. This result is caused by the unique near-IR spectral features of the 1,2-butadiene group, which will be discussed later. Predictions that use second-derivative spectra in region I1 have very low S E P values for all three analytes. The errors of predictions that use region I11 are comparable to or greater than the errors from predictions that use region 11, even though spectral resolution is better in region I11 than in region 11. Second-derivative spectral correction greatly improves prediction results in all spectral regions. The improvement of predictions with the use of second-derivative spectra is caused by two factors: (1) the decrease of spectral overlap and (2) the removal of base-line offset and slope variations in the spectra (11). The CLS-estimated spectra in region I1 for cis-174-butadiene, trans-1,4-butadiene, and 1,2-butadiene in PBD solution are shown in Figure 2. Note the sharp 1,2-butadieneband a t 1636 nm, which is not present in the spectra of trans-1,4butadiene and cis-1,Cbutadiene. This unique spectral feature of 1,2-butadiene enables the accurate determination of 1,2butadiene in PBD. Note also the high degree of overlap of the cis-lP4-butadieneand trans-1,4-butadienespectra. This high degree of overlap causes higher prediction errors for cis-1,4-butadiene and truns-1,4-butadiene by CLS analysis and essentially prevents prediction of these analytes by univariate
25.9 (20.4)
5.01 (4.56) 2.29 (2.30) 1.15 (1.27) 2.59 (1.39) 3.29 (2.09) 22.3 (14.2) 6.17 (5.68) 3.89 (4.72)
2.57 (3.40)
Estimated Pure Analyte Spectra
,
I . "
-
Owl2
own om1 00009
-
o m 8
-
f
o
c
o
m-
o
m
00007
m
--
O m 3 00002
-
00001
0
Figure 2. CLS-estimated spectra in region I1 of l,Bbutadiene, cis1,4-butadiene, and trans-1 ,Cbutadiene in PBD solution.
Estimated Pure Analyte Spectra 1 f 3mo
~
o mom2a -
OM)32
omw -
-
00024 om22
_I
om2 4
om2
O
m
0-
om2 o
/
,
1650 lee0
,
, 2070
, 2ilO
, , 2l50 2lw 2230 2270 2 3 0 2360 23w 2430 2470
wweiergth fnmi * cis-1.4 en +++ trans-l,4 BD Figure 3. CLS-estimated spectra in region I11 of 1,P-butadlene, cisl,Cbutadiene, and frans-l+butadiene in PBD solution.
-
12 80
analysis (6). Similar situations are encountered for pure analyte spectra in the other spectral regions. The pure analyte spectra in region 111(Figure 3) indicate the presence of unique 1,2-butadiene peaks (at 2118 and 2230 nm). Spectral features of cis-1,4-butadiene and trans-1,4-butadiene are more separable in region 111 (Figure 3) than in region I1 (Figure 2) as a result of increased spectral resolution. However, the shape of the trans-1,4-butadiene peak a t 2316 nm suggests the presence of nonlinear absorption a t that wavelength for the
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990
1782
Table IV. Prediction Errors of Near-IR-Determined Microstructures and Compositions of Styrene-Butadiene Copolymers spectral region used
O
M
1070
-
ldd0
10W
l720
l770
la0
Solution Analyses none 7.78 (13.4) second deriv 3.67 (5.48) none 2.75 (3.59) second deriv 1.40 (1.75) none 1.38 (0.98) second deriv 1.92 (0.59)
21.6 (33.6) 7.40 (9.29) 2.65 (3.01) 1.12 (0.96) 1.23 (1.54) 0.852 (0.25)
I I I1 I1
Bulk Analyses none 3.10 (4.84) second deriv 2.37 (3.41) none 1.83 (2.48) second deriv 2.56 (2.09)
4.39 (3.84) 3.30 (5.42) 0.861 (1.12) 1.47 (1.08)
1MO
WAVELENQTH (NHJ PenSOlUlon
++
Penbulk
Flgure 4. CLS-estlmated spectra in region I1 of cis-l,4-butadiene in PBD solution and in bulk PBD polymer.
solution samples. This effect probably causes the errors of cis-1,4-butadieneand trans-1,4-butadiene predictions that use region I11 to be greater than the errors of predictions that use region 11. The PBD bulk polymer prediction results are also shown in Table 111. It should be noted that the prediction errors for bulk analyses that use region I1 are significantly greater than prediction errors for solution analyses that use region 11. This result could be caused by the error associated with multiplicative and base-line correction of the bulk PBD spectra or by the inability of the CLS calibration model to describe all variations in the bulk PBD spectra. Nevertheless, the results of bulk PBD analyses are encouraging. They indicate that the microstructure of bulk PBD polymers of unknown thicknesses can be determined within 1.51% (mass) for 1,2butadiene, 1.80% for cis-1,4-butadiene, and 2.57% for trans-1,4-butadiene. The differences in the spectra of PBD polymer in bulk and in solution are indicated by a comparison of CLS-estimated spectra of the pure analytes in the two states. A comparison of CLS-estimated spectra of trans-l,.l-buhdiene and 1,2-butadiene in bulk PBD and in PBD solution revealed no significant differences. However, there are significant differences between the CLS-estimated spectra in region I1 of cis-1,4butadiene in PBD solution and in bulk PBD (Figure 4). The most notable difference is the relative height of the peaks at 1712 and 1772 nm for the two spectra. I t is possible that a nonlinear response of the strong 1712-nm band for the bulk samples causes this difference. The cis-1,Cbutadiene band a t 1712 nm is the strongest of all bands for all analytes in region I1 (see Figure 2). It is quite possible that the thicknesses of the bulk PBD samples are large enough to cause nonlinearity of this band for samples with high cis-l,.l-butadiene contents. SBR Analyses. The microstructures and the compositions of the copolymers used in all SBR analyses are shown in Table 11. The styrene contents of the copolymers are distributed between 10% and 33% mass. Unfortunately, the correlation between cis-1,4-butadiene and trans-1,4-butadiene contents in the available SBR copolymers is very high, as indicated by the cross-correlation of concentration values shown in Table 11. As a result, the discussion of the results of SBR analyses will focus on the 1,2-butadiene and styrene components. The prediction results for SBR solution analyses are shown in Table IV. In most cases, 1,2-butadiene content is more accurately predicted than styrene content. In fact, the SEP value for the 1,Zbutadiene prediction that uses second-derivative spectra in region I11 is below the estimated error of the NMR reference method. S E P values for styrene predictions are below 2% mass for analyses that use second-
correction
I I I1 I1 I11 I11
0 1670
SEP (RMSD,,,) for different analytes, % mass styrene 1,2-BD
spectral
O m 7
00x4
€ 4
5
OaoW
o
m
OM03
00002
Olxxx 0
-om 1570
lM0
lOl0
-
wr-
loa0
V30
l770
la0
fddo
WAVELENQTH (NMJ 12bm.dlms
+
Figure 5. CLS-estimated spectra in region I1 of styrene and 1,2-butadiene in SBR solution.
derivative spectra in regions I1 and 111. The S E P value for styrene obtained by using second-derivative spectra in region I is also encouraging, because region I can be used to analyze solution samples with path lengths up to several centimeters. The estimated pure analyte spectra in region I1 for styrene and 1,2-butadienein SBR solution are shown in Figure 5. The distinguishing peak for 1,2-butadiene at 1636 nm once again enables the accurate determination of 1,Zbutadiene content in the samples. The major styrene peak a t 1679 nm, which also has very little overlap with the spectra of other analytes, enables accurate predictions of styrene contents in the SBR copolymers. The results of bulk SBR analyses are also shown in Table IV. The lowest prediction errors for bulk analyses are 1.83% mass for styrene and 0.861% mass for 1,2-butadiene, which were obtained by using uncorrected spectra in region 11. The differences between the estimated spectra of pure analytes in SBR solution and in bulk SBR will be discussed in a later section. Sources of Prediction Errors. There are two important sources of error in this analysis: (1)sampling error and (2) CLS modeling error. Sampling error originates from nonreproducible sample placement in the spectrometer, nonrepresentative sampling of nonhomogeneous samples, and instrumental inconsistencies. The RMSD,,l, which indicates the difference between predicted concentrations obtained from replicate sample scans, is used to estimate the sampling error for each analysis. A comparison of the RMSD,, values (the values in parentheses) and the SEP values obtained from the PBD analyses (Table 111) and the SBR analyses (Table IV)
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990 0.00ll
Table V. Comparison of Spectral Residuals for CLS Predictions and Estimated Spectral Noise Levels
1679
n
% SR
sample type spectral region" solution bulk
I I1 I11 I I1
PBD
SBR
%RMS,
11.7
22.0 2.51 2.46 6.28 2.69
15.5 0.760
2.28 4.59 6.62 4.60
1.41
1.50 0.194
Only second-derivative spectra are used for residual analyses.
o m 0
indicates that, in most cases, errors from sampling are comparable or slightly less than the overall prediction error. This result suggests that sampling error is the major source of prediction error in these analyses. As mentioned earlier, sampling error can originate from several sources, including sample placement nonreproducibilities. The most prominent result of sample placement deviations is a shift in the spectral base line. Therefore, calibrations that use second-derivative spectra are expected to be less affected by sample placement deviations. Because second-derivative correction results in improved prediction in most cases, it can be concluded that significant sample placement errors exist for analyses that did not use secondderivative spectra. However, sampling errors from other sources must also be present, because RMSD,,,] values are still comparable to overall prediction errors for analyses that used second-derivative spectra. Although the results indicate that sampling errors are largely responsible for prediction errors in these analyses, errors from inaccuracy of the CLS model can also be present. For the CLS method, it is assumed that the maximum number of independent variations in the spectra is equal to the number of known analytes in the samples. If unknown spectral variations, such as nonlinear responses or interaction effects, are present, the CLS method might be inaccurate. This inaccuracy can be determined by the observation of the spectral residuals of prediction, which are the part of the prediction spectra that are not explained by the calibration model. In Table V, spectral residual of prediction values (determined by eq 3) are compared to estimated spectral noise levels (determined by eq 4) for both bulk and solution spectra in each spectral region. In each case, the spectral residual exceeds the estimated noise level. This result indicates the presence of CLS modeling errors in all analyses. For the solution analyses, in which the residuals are not much greater than the corresponding noise level, the modeling errors are very small. In fact, the model error might be the result of sampling errors mentioned previously. However, the amount that the residual exceeds the noise level is much greater for the bulk polymer analyses. This result suggests that significant CLS modeling errors are present for the bulk polymer analyses. In a previous section, it was suggested that the strong 1712-nm cis-1,4-butadiene peak exhibits a nonlinear response with concentration. This phenomenon might explain the significant modeling errors for the bulk PBD analyses. In the following section, the presence of interaction effects in bulk SBR, as indicated by the observation of CLS-estimated pure analyte spectra, is discussed. These effects would explain the substantial CLS modeling errors for bulk SBR analyses. Interaction Effects in Bulk SBR Copolymers. The estimated spectra in region I1 of styrene in bulk SBR and in SBR solution (Figure 6A) show significant differences. The major peak in these spectra, a t about 1680 nm, is most probably an aromatic CH stretching first-overtone peak, previously observed in the near-IR spectrum of a polystyrene wavelength standard (11). The difference in the wavelength
-om
-
1638
!Au 111111111.11111..1../111111111.111111111
1783
1
I
1784
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990 n
Styrene Spectra
10
-
0 -
om20
.5:
-
OM2-.
c 0.0019
-.
@on/
0
w6velergfh (om1
Flgure 7. CLS-estimated spectrum of styrene in SBR solution (A) and a spectrum of 1% ( w h ) polystyrene in CCI, (B). Spectrum A is offset for clarity; spectrum B is scaled and offset for clarity.
of the solvents are identical. With this in mind, the thermodynamic data support the suggestion that exothermic vinyl-phenyl interactions are present in bulk SBR copolymers. Neighboring-Group Effects. In polymer analysis, it is often necessary to determine the intramolecular sequencing, as well as the relative amounts, of structural units in the polymer chains. The sequencing of structural units is an important property, because it dictates the ability of the polymer chains to phase separate or crystallize. The ability of a method to determine sequencing requires the ability to distinguish between structural units that have different neighboring units. In order to demonstrate the sensitivity of near-IR spectroscopy to neighboring-group effects in polymer chains, the estimated spectrum of styrene in SBR solution is compared to a spectrum of a polystyrene homopolymer solution in region 111 (Figure 7). In this region, absorbances from aromatic CH groups (in the regions 2110-2210 nm and 2430-2500 nm) and absorbances from aliphatic CH groups (in the region 2270-2400 nm) are well-separated. Weak negative peaks at 2113 and 2228 nm in the spectrum of styrene in SBR solution (spectrum A), which are a t the same wavelengths as sharp 1,2-butadiene absorbances (Figure 3), are probably a result of weak nonlinearities of strong 1,2-butadiene peaks for the SBR solutions. The aromatic CH combination bands from 2110 to 2210 nm are very similar in the two spectra. However, a major difference is observed in the region 2230-2390 nm, where bands from CH vibrations of methylene and methyne groups in the polymer backbone are found. This result is expected, because the backbone groups are affected most by the identities of neighboring groups on the polymer chains. Maximum Allowable Thickness f o r Bulk Measurements. It was mentioned earlier that the relative magnitudes of the CLS prediction coefficients do not depend on sample thickness. However, the sum of the prediction coefficients is linearly related to the sample thickness. For bulk polymer analyses, this linear relationship will hold if the sample absorbance, which is linearly related to the sample thickness, is not large enough to cause nonlinear instrumental responses with concentration. The maximum allowable thicknesses for near-IR sampling of PBD and SBR polymers in regions I and I1 are estimated by performing CLS predictions of bulk PBD sample 4 using the spectra obtained from samples with five different thicknesses. Uncorrected spectra of the bulk samples are used for prediction. Two values are used to estimate the maximum allowable thickness of the polymer sample: the sum of the
Sm@s T H S k n N S fnn) @M/!
+
Sum of CLS prediction coefficients vs bulk sample thickness for microstructure predictions of PBD sample 4 that used uncorrected spectra in regions I and 11. Figure 8.
a4 3.2326
-
28 2.4 -
5
22-
I2 0
*/
20
snpk TNckmJs tnn) + @an/I
Flgure 9. Sum of squared prediction errors for the three analytes vs bulk sample thickness for microstructure predictions of PBD sample 4 that used uncorrected spectra in regions I and 11.
prediction coefficients and the sum of squared prediction errors for all three analytes (abbreviated SSPE). Figure 8 shows the relationship between the sum of the prediction coefficients and the thickness of the bulk polymer sample for analyses using regions I and 11. The plot for region I shows a linear relationship for all thicknesses considered, which indicates that region I has linear absorptions for bulk samples up to 22 mm thick. The plot for region 11, however, shows a linear relationship for low thicknesses and a change in slope a t a point corresponding to a thickness of approximately 2-3 mm. This trend is observed because at sample thicknesses greater than 2-3 mm the absorbances in region I1 are large enough to cause nonlinear instrumental response with concentration. As a result, it can be tentatively concluded that region I1 is useful for the sampling of bulk polymers up to approximately 2-3 mm thick. Although this maximum allowable thickness for bulk sampling depends on sample placement, sample shape, instrumental conditions, and sample composition, it should be a good approximation. The effect of sample thickness on prediction errors are indicated by plots of SSPE vs thickness for regions I and I1 (in Figure 9). For region I, the SSPE decreases sharply with increasing thickness for thin samples and then gradually decreases as the sample thickness increases. The results described in the previous paragraph indicate that linear absorption in region I was observed for all thicknesses studied. Therefore, the gradual decrease in SSPE with thickness for region I indicates that increased thickness of a bulk sample
ANALYTICAL CHEMISTRY, VOL. 62, NO. 17, SEPTEMBER 1, 1990
causes an increase in spectral signal vs a slightly changing spectral noise level, which should cause prediction errors to decrease. For region 11, this trend is also observed but only for thin samples. After the prediction error reaches a minimum value at approximately 2-3-mm thickness, it continually increases with sample thickness. The thickness a t the minimum of this plot roughly corresponds to the thickness at the discontinuity in the region I1 curve in the previous plot (Figure 8). This coincidence indicates that as the instrumental responses become nonlinear with concentration, the CLS prediction errors increase. In summary, these results indicate that bulk samples up to 2-3 mm thick can be analyzed by the near-IR/CLS method in spectral region I1 and that samples a t least 20 mm thick can be analyzed in spectral region I.
CONCLUSIONS Near-infrared spectroscopy can be used to accurately determine the microstructure and the composition in poly(butadiene) and styrene-butadiene polymers. The prediction ability of the near-IR/CLS method presented in this work depends on the sample state, the spectral pretreatment used, and the near-IR spectral region used. In most cases, the prediction error is only slightly greater than the estimated error of the NMR reference method. The results of replicate sample analyses indicate that sampling errors contribute the greatest amount to prediction errors of the near-IR method. These results suggest that the near-IR method can be improved with more reproducible sampling procedures, more representative sampling of nonhomogeneous samples, and reduced instrumental inconsistencies. In addition, it is advised that a t least two replicate spectra be obtained from different parts of each sample, in order to reduce systematic errors from nonrepresentative sampling. Errors from inaccuracies of the CLS model are also present, especially for the analyses of bulk polymer samples. Therefore, the use of a more appropriate model, such as a PLS model, might further improve prediction ability. It should be noted that the optimal spectral region depends on the sample state and the path length. Region I (1100-1350 cm-') is useful for thick (10-20-mm-thick) bulk samples, region I1 (1570-1850 cm-') is useful for thinner (up to 2-mm-thick) bulk samples and for solutions (in which the solvent is near-IR-transparent), and region I11 is useful for dilute solutions (ca. l % w / ~ ) . Studies of CLS-estimated pure analyte spectra indicate that near-IR spectroscopy is sensitive to the identity of neighboring units in polymer chains and can detect interactions between polymer chains. As a result, near-IR spectroscopy can be used to monitor the relative amounts of random and block copolymer and determine intermolecular properties, such as
1785
crystallinity and phase separation.
ACKNOWLEDGMENT We thank the Goodyear Tire and Rubber Co. for organizing a 1988 summer internship, during which the data were collected. Specific acknowledgments go to Lori Cianchetti at the NMR lab a t Goodyear for essential lab analyses and Dr. Adel Halasa, Neil Maly, and Dennis Romain for the acquisition of PBD and SBR polymers. Comments from Dr. James B. Callis and Dr. James Visintainer were quite valuable.
LITERATURE CITED ( 1 ) Morton, M. Rubber Technology; Rober E. Kreiger Co: Malabar, FL, 1981. (2) Odian, G. Principles of Polymerization, 2nd ed.; John Wiley 8 Sons: New York, 1970. (3) Koenig, J. L. Chemical Microstructures of Polymer Chslns; John Wiley & Sons: New York. 1980. (4) Alaki, Y.; Yoshimoto, T.; Dnanari, M.; Takenchi, M. Rubber Chem. Technol. 1973. 46. 350. (5) Claque, A. D. H.; vanBroekhoven, J. A. M.; Biaavin, R. P. Macromolecules 1974, 7,348. (6) Durbetaki, A. J.; Miles, C. M. Anal. Chem. 1985,37. 1231. (7) Zbinden, R. Infrared Spectroscopy of Hlgh Polymem; Academic: New York, 1964. (8) Binder, J. L. Anal. Chem. 1954,2 6 , 1877. (9) Stark, E.; Luchter, K.; Margoshes, M. Appl. Spectrosc. Rev. 1988, 22,335. (10) Weyer, L. G. Appi. Spectrosc. Rev. 1985,2 1 , 1. (11) Wliiiams, P.; Norris, K. Near-infrared Technology in the Agricultural and Food Industries; American Association of Cereal Chemists: St. Paul, MN, 1987. (12) Sharaf, M. A.; Iilman, D. L.; Kowalski, B. R. Chemometrics; John Wiley & Sons: New York, 1986. (13) Geladi, P.; Kowalski, B. R. Anal. Chim. Acta lS88, 185, 1. (14) Haaland, D. M.; Thomas, E. V. Anal. Chem. 1988,60, 1202. (15) Beebe, K.; Kowalski, 8. R. Anal. Chem. 1987,59, 1007A. (16) Haaland, D. M.; Easterling, R. G.; Vopicka, D. A. Appl. Spectrosc. 1988, 3 9 , 73. (17) Brown, C. W.; Obremski, R. J. Appl. Spectrosc. Rev. 1984,20, 373. (18) Goddu, R. F. Anal. Chem. 1957,2 9 , 1790. (19) Miller, C. E. Appl. Spectrosc., in press. (20) Duch, M. W.; Grant, D. M. Macromolecules 1970,3 , 165. (21) Bovey, F. A. Hlgh Resolution NMR of Macromolecules; Academic: New York, 1972; pp 226-229. (22) Renyuan, Q.; Xigao, J. Sci. Sin. Ser. B (Engl. Ed.) 1982, 2 5 , 137. (23) Thompson, R. E.; Butler, G. B. J . Polym. Sci., Polym. Chem Ed. 1978, 16, 1367. (24) Int. DATA Ser. Sel. Data Mixtures, Ser. A 1973, 100. (25) Int. DATA Ser., Sel. Data Mixtures, Ser. A 1975, 186. (26) Znt. DATA Ser., Sel. Data Mixtures, Ser. A 1978,55. (27) Znt. DATA Ser., Sel. Data Mixtures, Ser. A 1975, 188. (28) ASTM Committee D-2 and API Research Project 44. Physical Constants of Hydrocarbons C1 to C10; ASTM: Philadelphia, PA, 1971; pp 2, 14. (29) Buchat, R. K.; Richard, A. J. J . Chem. Thermodyn. 1975, 7 , 271. (30) Eichinger, B. E.; Flory, P. J. Trans. Faraday SOC. 1988, 64, 2035.
RECEIVED for review November 6, 1989. Accepted May 7, 1990. This work was done with support from the Goodyear Tire and Rubber Co. and the Center for Process Analytical Chemistry.
