J. Chil. Chem. Soc., 48, N 3 (2003) ISSN 0717-9324

ESTIMATING THE DENSITY AND PULPING YIELD OF E. globulus WOOD BY DRIFT-MIR SPECTROSCOPY AND PRINCIPAL COMPONENTS REGRESSION (PCR)

Juanita Freer,* José Ruiz,* Miguel A. Peredo , Jaime Rodríguez,* and Jaime Baeza,*

*Laboratorio de Recursos Renovables, Universidad de Concepción,
Casilla 160-C, Concepción Chile. E-mail: jfreer@udec.cl

(Received: March 20, 2003 ­ Accepted: April 3, 2003)

SUMMARY

The ability of mid-infrared (MIR) spectroscopy as a quick technique for determining wood properties of Eucalyptus globulus plantations, specifically basic density and pulping yield, has been examined. Twenty-seven samples were used as a calibration set and other three, for prediction making based on calibration (validation set). Calibrations and predictions through principal components regression (PCR) were obtained through the Quant+ chemometrics. Spectral data for a PCR model based on diffuse reflectance infrared Fourier transform (DRIFT), gave standard prediction error values of 21.44 kg m^-3 and 1.73 % for basic density and pulping yield, respectively. The method allows to predict density (R² = 0.84) and pulping yield (R²= 0.40) from a single spectral MIR measurement.

Keywords: DRIFT-MIR, PCR, wood, properties prediction

INTRODUCTION

Wood is a complex material chemically heterogeneous and its components can be divided in two groups: structural or major components (cellulose, hemicelluloses, and lignin) and non-structural components of low molecular mass (extractives and inorganic compounds). An arrangement of more or less standardized classical procedures^1-3methods (wet-lab) are available for primary quantitative analysis of wood, but they are time consuming and implies destructive procedures.^1-3 On the other hand, instrumental techniques such as Fourier transform infrared (FTIR) spectroscopy, have been used for wood analysis.^4-7 Two vibrational regions - the mid and near-infrared- have been used to obtain information about woods. Samples can be examined by using either traditional transmission or reflectance. Diffuse reflectance infrared Fourier transform (DRIFT) proved to be very convenient as secondary analysis of wood. DRIFT involves minimal sample preparation, short analysis time, high sensitivity and high linear range for quantitative analysis. In this technique, a set of samples analyzed by traditional methods are divided into two subsets. One subset is used to calibrate and to verify the calibration quality and the other one, is used for prediction.^8-9 Multivariate analysis allows to relate spectral data and any feature of a sample, which is reflected in its spectrum. Quant+ is a powerful chemometrics quantitative analysis software package, ideally suited to multicomponent spectroscopic analysis.¹⁰ This software provides a choice of multivariate calibration algorithms (PCR and two types of PLS). Relationships between spectral data and the property may be used for subsequent prediction in unknown samples.^7-9

From an economic perspective, one of the best ways to evaluate wood pulping is by means of the pulp yield for unit of wood, processed in a reactor. Traditional evaluation method of the pulping yield is cooking wood chips until a determined kappa number, in laboratory reactors. This is a slow and expensive process, limiting the amount of samples that could be processed. Therefore, becomes essential to implement techniques to predict yield, as well as other properties of the wood, by means of procedures that allow analyzing a high amount of samples in a short period.

This paper describes the application of PCR model to the prediction of basic density and pulp yield (relative to kraft pulping to kappa number 15), from measurements of absorbance of bands in DRIFT-MIR. A set of wood samples of Eucalyptus globulus was used.

EXPERIMENTAL

2.1. Sample preparation.

Wood chips (approximately 2.5X1.5X0.2 cm) of E. globulus obtained from plantations located in the VIII Region- Chile were used. For FTIR analysis, wood chips were oven-dried at 60°C and stored under dry conditions until used. Chips were milled in a knife mill (RetschMuhle) to pass through at 0.5 mm screen, and then, sieved to pass through a 150 mm screen. Sieved samples were dried at 60°C overnight and stored under silica gel.

2.2. Pulping and density determinations.

Wood samples pulping was carried out in a reactor M/K systems of 10 L, with liquor recirculation cooking cycling automatically controlled. The operation conditions were:

Kappa number was determined by TAPPI standard method 236 cm-85.¹
The basic density was determined by TAPPI standard method 258 om-94.¹

2.3 DRIFT spectra

Twenty-five mg of each one of the samples were homogenized with 225 mg of KBr for 1 min and placed in a macro-cup of the Spike Technologies attachment for DRIFT spectroscopy. FTIR were recorded between 4000 and 800 cm^-1 in a Perkin Elmer FT-2000 FTIR spectrometer using 64 scans, triangular apodization and resolution of 4 cm^-1. Data points collection was set at the same resolution value (each 4 cm^-1) giving 795 data points for each FTIR recorded between 4000 and 800 cm^-1. Reflectance spectra were transformed to Kubelka-Munk (KM) units to minimize scattering contributions to the measured absorption. The baseline was corrected to the regions near 800, 2000 and 3800 cm^-1 and the spectrum normalized to the band nearest to 1510 cm^-1 .^8-9

2.3 Principal component regression modeling

The data set was collected to evaluate the potential of MIR spectroscopy to predict density and pulping yield of Eucalyptus wood. Spectral data and properties values (density and yield) were analyzed with Quant+ software package, available with the Perkin Elmer spectrometer. Spectral data were analyzed in digitized form at regular intervals of 4 cm^-1. Normalized spectra were mean scaled (covariance about the mean) by subtracting each sample spectrum from the mean spectrum. PCR models were based on principal component analysis (PCA) performed on spectral information followed by multiple linear regression between each property and chosen principal components. This algorithm, calculates regression models for all properties (density and yield) simultaneously and it is suitable for samples with a high degree of correlation among properties. Cross-validation was used to check standard outlier in PCR models. For "n" samples in the calibration set, "n" separate calibrations were performed.⁶

3.1 Data analysis

DRIFT spectra measured directly from wood surfaces showed distortions at the intensities of the strongest absorption bands present in the 1150 - 950 cm^-1 region (Figure 1). A similar effect was observed in the DRIFT spectra reported by Pandey.¹¹ This anomaly in DRIFT spectra is attributed to the contribution from specular component. The use of finely divided samples and dilution in KBr can minimize the scattering effect.^8-9 IR spectra of wood showed a strong hydrogen bonded O-H stretching absorption around 3400 cm^-1 and a prominent C-H stretching absorption around 2900 cm^-1. In the fingerprint region between 1800-900 cm^-1, many sharp and discrete absorption bands due to several functional groups present in wood components, were observed. O-H (at around 3400 cm^-1), C-H (at around 2900 cm^-1), C=0 (at around 1740 cm^-1) and C=C aromatic (around 1510 cm^-1) bands are relatively pure, whereas other bands in the fingerprint regions below 1460 cm^-1 are complex having contributions from various vibration modes in carbohydrates and lignin.^4,6,11 The only "pure" band is related to the aromatic moieties present in lignin, which gives a distinct absorption near 1510 cm^-1.^8,11 In the E. globulus samples tested, this vibration mode appeared at 1505 cm^-1 (Figure 1).

Fig. 1 DRIFT-MIR spectra of E. globulus milled wood (showing at 1505 cm-1 the aromatic skeletal vibration).

3.2 Principal component regression (PCR)

Quant+ allows determining a number of factors (or principal components) used in each model, based on the cutoff point defined as the number of factors closest to the 10% significance level plus one. They express the maximum amount of spectral variance not accounted by previous factors and indicating the minimum SEP values.^6,10,11

The results for E. globulus properties on density and yield previously measured by traditional procedures are shown in Table 1. Samples had been chosen in order to cover a wide range of pulp density and pulping yield.

Table 1. Basic density and pulping yield of E. globulus wood of individual trees.

(*) Samples used for prediction

Data from 27 of the samples, were used to establish calibration and validation model and the samples 1, 15 and 30 were used for the prediction. The PCR models were developed to relate MIR spectral bands and properties for each component shown in Table 1. Characteristics of the model for three factors are summarized in Table 2. As expected, density and yield correlated using PCR models, with standard error prediction (SEP) 21.44 and 1.73, respectively. Plots of the property values obtained using the PCR model (predicted) versus the value measured directly are shown iN Fig. 2.

	Fig. 2 Plot of predicted PCR vs. determined values for basic density (a) and pulping yield (b).

Correlation for density was better than that for pulping yield, having correlation coefficients (R²) of 0.84 and 0.40, respectively. These results are indicating that yield property has a strong leverage on the result owing to the laboratory measured error itself. The prediction from samples 1, 15 and 30 using PCR, correlates with laboratory values with deviation (%) of 13.5 to -5.6 for density and 2.4 to 2.7 for yield (Table 2).

Table 2. Prediction: Basic density (kg m^-3) and pulping yield(%) predicted from PCR with 3 factors.

PCR models are being carrying out in the near-infrared region, due that NIR includes higher energy sources, lower SEP values and better correlation than those obtained in MIR spectral region would be expected.

CONCLUSION

PCR models based on MIR can be used to estimate the basic density and pulping yield of E. globulus wood. More precise results may be obtained based on NIR spectra were the PCR models, are probably more robust and better predictors not only for density and yield if well developed for other wood properties and components as glucan, total lignin, hemicellulose, carbohydrate, xylan etc.

ACKNOWLEDGEMENTS

Authors would like to thank FONDECYT (Grant. N° 1020161) and University of Concepción Research Council (Grant DIUC 201.023.027-10 and 201.023-028-1.4).

REFERENCES

1. TAPPI Test Methods, Technical Association of the Pulp and Paper Industry, Atlanta, Ga.

2. ASTM Standards, American Society for Testing and Materials, Philadelphia, Pa.

3. Appita Standards, Australian Pulp and Paper Industry Technical Association

4. A. Evans, Differentiating "hard" from "soft" woods using Fourier transform infrared and Fourier transform Raman spectroscopy, Spectrochim. Acta 47:A1441-1447(1991).

5. H. Anti, D. Alexandersson, M. Sjöström and L. Wallbacks, Detection of kappa number distributions in kraft pulps using NIR spectroscopy and multivariate calibration,Tappi J. 3:102-108(2000).

6. A. Michell, Vibrational spectrometry-a rapid means of estimating plantation pulp wood quality, Appita 47:29-32(1994).

7. O. Faix and J. Bottcher, Determination of phenolic hydroxyl group contents in milled wood lignins by FTIR spectroscopy applying partial-least squares (PLS) and principal component regression (PCR), Holzforschung 47:45-49(1993).

8. A. Ferraz, J. Baeza, J. Rodriguez and J. Freer, Estimating the chemical composition of biodegraded pine and eucalyptus wood by DRIFT spectroscopy and multivariate analysis, Biores. Technol. 74:201-212(2000).

9. K. Beebe and K. Kowalski, An introduction to multivariate calibration and analysis, Anal. Chem. 59:1007A-1017A(1987).

10. Quant+ guide book,Perkin Elmer, 1992

11. K. Pandey, A study of chemical structure of soft and hardwood and wood polymers by FTIR spectroscopy, J. Appl. Pol. Sci. 71:1969- 1975(1999).