Printer Friendly

Predicting moisture content of yellow-poplar (Liriodendron tulipifera L.) veneer using near infrared spectroscopy.


On-line measurement of the moisture content (MC) of veneer sheets is paramount to process control in the veneer-based panel and engineered wood products industry. This study examined the feasibility of using near infrared (NIR) spectroscopy (800 to 2400 nm) combined with multivariate data analysis to predict MC of yellow-poplar veneer sheets. Multivariate data analysis employing principal component regression (PCR) and partial least squares regression (PLS1) analysis techniques indicated clustering of veneer samples of the same or close MC range with a clear distinction between samples of low and high MC. Both PCR and PLS1 veneer MC predictive models had correlations ([R.sup.2]) greater than 0.94. The spectra window, 1400 to 1900 nm, between the two moisture peaks (1450 and 1930 nm) gave correlation coefficients ([R.sup.2]) of 0.985 and 0.986 for PCR and PLS1, respectively. There is no clear distinction between the PCR and PLS1 models developed using the NIR spectra region of 1400 to 1940 nm. However, the PLS1 models with lower root mean square error of prediction (RMSEP), standard error of prediction (SEP) and Bias were better when compared to the PCR models developed using the whole NIR spectra region and restricted NIR spectra region not associated with the hydroxyl band.


Moisture content (MC) of wood, an important physical attribute, affects many wood properties such as strength, drying, glue curing, and bond performance. Methods currently employed to measure MC of veneer include electronic moisture meters, ovendry methods, and the Karl Fisher titration method. As an important part of process control during industrial veneer production, MC of veneer at various stages of manufacture needs to be determined. In the industrial setting, rapid methods of monitoring MC are mandatory in order to keep pace with the flow of materials. In North America, the veneer industry typically employs electromagnetic wave technology to measure MCs of veneer (Wagner Electronics, Rogue River, Oregon). Methods that are suitable for such applications need to be noncontact, nondestructive, requiring no sample preparation and can be employed online. Near infrared (NIR) technology meets these requirements.

The NIR portion of the electromagnetic spectrum, which covers the 800 to 2500 nm frequency range, represents overtone and combinations of notably C-H, NH, and O-H functional groups. NIR is increasingly becoming an important tool of Process Analytical Technology (PAT). PAT combines multivariate data analysis with process analytical chemistry and chemistry to monitor and better comprehend industrial processes (Baughman 2005). Norris and his coworkers first demonstrated that NIR could be used to characterize both physical and chemical properties of grains and forage (Hart et al. 1962, Ben-Gera and Norris 1968, Norris et al. 1976). The last two decades has seen increasing application of NIR to the characterization of physical properties and chemical properties of lignocellulosic materials. (McLellan et al. 1991a, 1991 b; Aber et al. 1994, Newmann et al. 1994, Hoffmeyer and Perdersen 1995, Michell 1995, Schimeleck et al. 1997, Schimeleck and Michell 1998, Svedas 2000, Thumm and Meder 2001, Kelley et al. 2004)

In West Virginia, yellow-poplar (Liriodendron tulipifera L.) is a major wood species used in the manufacture of veneer-based engineered wood products such as laminated veneer lumber (LVL) and parallel strand lumber (Parallam[TM]). NIR spectrum is known to contain information of more than one hydrogen-bonded subspecies at water bands at 1450 and 1930 nm (Walrafen 1972, Shenk et al. 2001). These bands consist of multiple peaks that can be used to develop predictive models for water using multivariate data analysis. Over the years, multivariate calibration techniques such as principal component regression (PCR) and partial least squares regression (PLS) have withstood the test of time and have become regarded as standards of multivariate calibration. In this paper, we report on the suitability of NIR technology for rapid determination and monitoring of MC during industrial veneer production. The objective of this study was to investigate the use of NIR diffuse reflectance spectrometry coupled with multivariate data analysis to develop models for predicting MCs of yellow-poplar veneer, and to compare the performance of the models developed using the two regression methods.

Materials and methods


Inorganic salts--potassium acetate, lithium chloride, ammonium monophosphate, sodium bromide, sodium acetate and potassium nitrate and deririte--used for conditioning veneer specimens were purchased from Sigma-Aldrich (St Louis, Missouri). Green rotary-peeled yellow-poplar veneer of 3-ram thickness was provided by Trus-Joist, a Weyerhaeuser iLevel Veneer Technology Group (Buckhannon, West Virginia) and kept frozen until use. In order to account for possible variation in raw materials, 10 veneer sheets were selected from the series of veneers produced at different batches. The veneers were thawed at room temperature for 24 hours before use. Veneer moisture conditioning chambers consisted of plastic desiccators (height 311 mm, flange diameter 225 mm, Fischer Scientific Company, Pittsburgh, Pennsylvania) with a 12.5-mm square cooling fan (RadioShack, Fort Worth, Texas) suspended on the desiccators' perforated plastic platform (divider) on four stainless legs. Veneer specimens were placed above the desiccators' perforated platform (divider). Saturated solutions in 14.5-mm diameter shallow glass was placed at the bottom of each desiccator below the platform. Temperature and humidity inside each chamber were monitored with a traceable temperature and humidity monitor (Fischer Scientific, Pittsburgh, Pennsylvania).


Conditioning of veneer samples to different MCs

Ten veneer sheets selected from different batches of production were cut into small test specimens of 40 mm by 40 mm, and their green weights were recorded using an analytical electronic balance with a sensitivity of 0.1 mg. Veneer samples with MCs below the fiber saturation points (FSP) were obtained by conditioning green veneer samples in a desiccator (225-mm diameter) in which the relative humidity of its chamber is affected by use of saturated inorganic salts solutions as shown in Figure 1. A fan (12.5-mm square, RadioShack) was installed in each desiccator to facilitate circulation and expedite establishment of equilibrium MCs in the veneer samples. To achieve a relative humidity of 100 percent, deionized water was used. Sixteen veneer replicates were kept in each desiccator (treatment) containing the appropriate inorganic salt, and deionized water to give a total of 160 samples, and each veneer sample was weighed once a week until a constant weight was attained. At this point, a veneer sample was considered to have attained equilibrium MC (EMC). The inorganic salts used with their respective water activities and the various MC ranges attained in all the 10 conditioning chambers are given in Table 1. According to Table 1, all of the samples were equilibrated to MCs below fiber saturation point. Veneer samples with MC above FSP were obtained by soaking veneer samples in deionized water for 1 hour and 24 hours to achieve MC ranges of 60.1 to 69.5 percent and 70.5 to 81 percent. The EMC attained in each of the conditioning chambers of different relative humidity (%) correspond to the report on sorption isotherms of wood at three temperatures (Rasmussen 1961).


NIR measurements

When veneer samples attained the required equilibrium MCs, NIR spectra were collected with a Bruker Matrix-F FTNIR spectrometer (Billerica, Massachusetts) fitted with a fiber-optic sampling probe for solids and liquids (IN263E) and operating in a diffuse reflectance mode with a wavelength range of 833 nm to 2500 nm. Each sample was scanned 20 times to obtain an average spectrum and its green weight immediately taken in ambient laboratory condition (20 [degrees]C, 65% RH). Then each veneer sampled was ovendried at 103 [+ or -]2[degrees]C to a constant weight and data used with the original green weights to determine exact MC of samples at point of spectral scanning. The NIR spectra collected were converted to JCAMP file using Bruker OPUS[TM] software (version 5.0, Billerica, Massachusetts). Multivariate data analysis of converted spectral data was performed using Unscramble[R] software version 9.1 (Camo Smart, Woodbridge, New Jersey). All the spectral data were preprocessed using the second derivative transformation in order to improve spectral interpretation and the quality of the models due to the fact that in this form, band intensity and peak location are maintained with those of log (1/R) spectral pattern, and apparent band resolution enhancement (Shenk et al. 2001).

Development of NIR calibration

Data from the entire MC range studied (0.3 to 80%) was used. The total sample size consisting of 160 samples (16 replicates per conditioning chamber) was divided into calibration and prediction sample sets. The calibration data set (80) consisted of all odd number spectra between 1 and 159, while the validation set (80) comprised all even numbers between 2 and 160. To obtain an overview of data, principal component analysis (PCA) was carried out on the samples to observe any clustering or separation in the sample set. The calibration spectra data were used in computing the principal component regression (PCR) and the partial least squares regression (PLS1). All the NIR spectra were combined into a single data matrix (X-matrix) while the ovendry MC was combined into a response matrix (Y-matrix). Separate calibration models were developed at five different wavelength regions in order to know which wavelength region provides useful quantitative information capable of estimating the amount of moisture in the veneers. The calibration models were constructed with an X-matrix of wavelength ranges of the whole NIR spectra region (800 to 2500 nm), the combination of second overtone and first overtone NIR spectra regions (1000 to 2500 nm, and 1300 to 2100 nm), and the first overtone NIR spectra region (1400 to 1900 urn, and 1500 to 2000 nm), and moisture ranges of 0.3 to 80 percent MC were combined into the Y-matrix. Both the X and Y matrices were mean centered variance normalized prior to the development of PCR and PLS1 models. The calibration models were constructed with 77 samples, after removing three outlier samples, using full cross-validation method. This fully cross-validated model was then used to predict the response of the validation set (test set) that comprises of 75 samples that were not included in the original calibration model.

Results and discussion

Table 1 shows the ovendry MC for all the samples conditioned with saturated salt solutions, dririte and water, and the corresponding relative humidity attained. The MC for all samples conditioned with the saturated salts solution ranged from 3 to 27.5 percent. The MC of samples conditioned with deionized water ranged between 28 to 32 percent, while MC of veneer sample conditioned by soaking in water for 1 and 24 hours ranged from 60 to 70 percent and 71 to 80 percent respectively. The ovendry MC for samples dried over drierite ranged between 0.31 and 0.79 percent.

Figure 2 shows the representative raw spectra of veneer samples of different MCs. The absorption peak at 1450 and 1930 nm increases with increase in MC. These peaks are as a result of water derived from the combination of O--H stretching and O--H deformation, while Figure 3 shows the second derivatives spectra with the absorption peaks pointing down rather than up (Walrafen 1972, Shenk et al. 2001).

In a data set of N observations and K variables, principal component is a line or hyperplane in K-dimensional space. Principal component analysis is a reductive process. The score plot represents coordinates of all observations projected down onto the reduced dimensional line or hyperplane (Newmann et al. 1994), and shows the relationship between the observations. Figure 4 shows clustering of veneer samples of the same or close moisture range with a clear distinction between samples of high moisture and low MC. Five spectra regions (windows, Table 2) were used to develop PCR and PLS1 predictive models the optimum number of PCs varied from one to four. The calibration and predictive models for the MC of yellow-poplar veneers contained 77 and 75 samples respectively. The respective calibration and predictive statistic for the PCR and PSL1 models and their optimum number of PCs are presented in Table 2. All the models performed very well with high correlation (R) with low prediction error. The performance of all the PCR and PLS models developed for this study were evaluated using three different parameters. First, the correlation between predicted and measured values is a good way to judge the performance of a model. A high correlation value is an indication that the model may be good. Secondly, the standard error of calibration (SEC) is also an expression of the amount of error to expect when using the calibration models, while the root mean square error of calibration/prediction (RMSEC/RMSEP) are direct estimates of the prediction error and the modeling error in the Y-matrix respectively. RMSEP expresses the average error to be expected associated with future predictions; it is also useful in comparing different models regardless of how the models were developed with regard to weighting, preprocessing of the X-variables or the number of components used. Thirdly, the Bias is used to measure the accuracy of a prediction model. It is a measure of the average difference between predicted and measured Y-values for all samples in the prediction set, and it is also used to check if there is a systematic difference between the average values of the calibration set and prediction set. The closer the bias to zero, the better the model (Esbensen 2002). Figures 5 and 6 are representative plots of predicted values against measured for models developed at NIR spectra range of 1400 nm to 1940 nm respectively for PCR and PLS1 models. Table 2 gives a detailed results and comparison of PCR and PLS models developed for estimating MC (%) in veneers at different NIR spectra region. The NIR spectra region of 1400 to 1940 nm (region between the two prominent peaks) which is known to be associated with the hydroxyl groups gives the best models in terms of its repeatability. It has the highest correlations ([R.sup.2]) of 0.985 and 0.986 respectively for the PCR and PLS1 models. It also has the lowest RMSEP values of 0.28 and 0.275 for both PCR and PLS1 models respectively. There is no clear distinction between the performances of the PCR and PLS1 models developed using the NIR spectra region of 1400 to 1940 nm. The two methods may be adjudged to be equal or similar, since both methods have the same number of optimum principal component (PC) of 4 to give almost the same [R.sup.2] and RMSEP values. Wentzell and Montoto (2003) compiled together a series of research work done using the two regression methods, and found that there was no significant differences in prediction error reported by PCR and PLS in all cases, except when artificial constraints were placed on the number of latent variables retained. They also found out that though PLS usually required fewer latent variables than PCR, but this did not appear to influence predictive ability. PLS1 models developed using the whole NIR spectra region of 800 to 2500 nm with 2 PC and [R.sup.2] value of 0.973 with low RMSEP value of 0.38; and PLS1 model developed with the restricted wavelength region of 1000 to 2500 nm with a [R.sup.2] value of 0.971 and RMSEP value of 0.39 with 2PC did appear to be slightly better than the PCR models with low [R.sup.2] value of 0.951 and RMSEP value of 0.514 and [R.sup.2] value of 0.938 and RMSEP value of 0.573 respectively for models developed using the whole NIR spectra region of 800 to 2500 nm and restricted wavelength region of 1000 to 2500 nm. Other PLS1 models developed at restricted wavelength of 1300 to 2100 nm and 1500 to 2000 nm not associated with the hydroxyl group also appear to be slightly better than the PCR models developed because of their lower PC factors, and RMSEP values.



It should be pointed out that the laboratory experiment and results presented here are only meant for a feasibility study, and will only be useful in the real industrial production environment if calibration models are to include many other factors that are not included in this study. Such factors that can influence the performance of the calibration models in practice include raw materials variations, instrument and process variables, and environmental changes. Calibration models should include all possible variations from the various sources of raw materials, and production processes over a period of time. A possible source of raw materials variation is density. Wood density variation influences NIR absorbance. There is a shift upward in absorbance across all wavelengths as density increases; also according to the Beer-Lambert law, baseline shifts in absorbance with density occur when spectra are not pretreated (Via et al. 2003). In practical applications, accuracies of direct in-line measurements are subjected to many instrument factors that may include location of probes along the process line, distance and angle between the probes and the process stream, adjustable pathlengths and stability of the instruments. Sahni et al. (2004) reported that positioning the NIR measurement system after a heat source, in the production system, could introduce error into the prediction and make it difficult to detect actual changes in product performance. They suggest placing the NIR instrument after a cooling instrument in the production process, while other instrument problems could be corrected by appropriate design of the optics equipment or by incorporating them into the calibration models. The manufacturing environment is another factor that could influence the performance of the calibration models in practice. Thygesen and Lundqvist (2000 part 1) reported that there is an interaction between wavelength and temperature on the prediction of MC. They found that the two hydroxyl bands at approximately 1450 nm and 1930 nm peaks shifted toward shorter wavelengths as temperature increased from -20 to +25 [degrees]C. They also show that PLS models built at one temperature exhibit increased prediction errors as the absolute (actual temperature--temperature at calibration) temperature difference increased. They propose a solution to temperature fluctuations by including the full range of temperatures during calibration of NIR for MC prediction. This global calibration model worked better than the transformation model coupled with PLS (Thygesen and Lundqvist 2000 part II).



PCR and PLS1 models capable of predicting MC of yellowpoplar veneer within the accuracy of the calibration models were developed. The spectra window of wavelength 1400 to 1900 nm, a region between the two main hydroxyl absorption peaks, gave the best models with low RMSEP values when compared to other wavelengths used for model development. The two regression methods of PCR and PLS1 performed almost equally in prediction of new samples using calibration models developed at 1400 nm to 1900 nm. The PSL1 models with lower RMSEP did appear to work better than corresponding PCR models when the blind box approach is used (throw all wavelengths into the matrix) or when other restricted wavelengths not associated to the hydroxyl bands were used. However, determining potentially important wavelengths a priori not only narrows the gap between the two modeling methods, but lowers the overall error of prediction and thus is a recommended procedure as opposed to blindly modeling with all wavelengths in the matrix.



Literature cited

Aber, J., K. Bolster, S. Newman, M. Soulia, and M. Martin. 1994. Analyses of forest foliage If: Measurement of carbon fraction and nitrogen content by end-member analysis. J. Near Infrared Spec. 2:15-23.

Baughman, E. 2005. Process Analytical Tech. Spectroscopic Tools and Implementation Strategies for the Chemical and Pharmaceutical Industries. K.A. Bakeev, Ed. Blackwell Publishing Ltd., Oxford, United Kingdom. pp. 1-12.

Ben-Gera, I. and K.H. Norris. 1968. Direct spectrophotometric determination of fat and moisture in meat products. J. Food Sci. 33(1):64-67. Esbensen, K.H. 2002. Multivariate Data Analysis in practice: An Introduction to Multivariate Data analysis and Experimental Design, 5th ed. CAMO Process AS, Oslo, Norway. 598 pp.

Hart, J.R., K.H. Norris, and C. Golumbic. 1962. Determination of the moisture content of seeds by near-infrared spectrophotometry of their methanol extract. Cereal Chem. 39:94-99.

Hoffmeyer, P. and J. Perdersen. 1995. Evaluation of density and strength of Norway spruce wood by near infrared reflectance spectroscopy. Holz als Roh-und Werkstoff 53:165-170.

Kelley, S., T. Rials, R. Snell, L. Groom, and A. Sluiter. 2004. Use of near infrared spectroscopy to measure the chemical and mechanical properties of solid wood. Wood Sci. and Tech. 38:257-276.

McLellan, T., M. Martin, J. Melillo, K. Nadelhoffer, and B. Dewey. 1991 b. Determination of nitrogen, lignin, and cellulose content of decomposing leaf material by near infrared reflectance spectroscopy. Can. J. Forest Res. 21:1684-1688. --,--,--,--,and--. 1991a. Comparison of wet chemistry and near infrared reflectance measurements of carbon-fraction chemistry and nitrogen concentration of forest foliage. Can. J. Forest Res. 21 : 1689-1693.

Michell, A.J. 1995. Pulpwood quality estimation by near infrared spectroscopic measurements on Eucalypt wood. Appita J. 48(6):425-428.

Newmann, S., M. Soulia, J. Aber, B. Dewey, andA. Ricca. 1994. Analyses of forest foliage I: Laboratory procedures for proximate carbon fractionation and nitrogen determination. J. Near Infrared Spec. 2:5-14.

Norris, K.H., R.F. Barnes, J.E. Moore, and J.S. Shenk. 1976. Predicting forage quality by infrared relectance spectroscopy. J. Anita. Sci. 43: 889-897.

Rasmussen, E.F. 1961. Dry Kiln Operators Handbook. US DA Handbook 188, USDA, Washington D.C. 197 pp.

Sahni, N., T. Isaksson, and T. Naes. 2004. In-line near infrared spectroscopy for use in product and process monitoring in the food industry. J. Near lnfrared Spec. 12(2):77-83.

Schimeleck, L.R., P.J. Wright, A.J. Michell, and F.A. Wallis. 1997. Near infrared spectra and chemical compositions of E. globules and E. nitens plantation woods. Appita J. 50:40-46. --and A. Michell. 1998. Determination of within-tree variation of Kraft pulp yield using near infrared spectroscopy. Tappi J. 81(5):229-236.

Shenk, J.S., J.J. Workman, and M.O. Westerhaus. 2001. Application of NIR Spectroscopy to Agri. Products. Handbook of Near-Infrared Analysis, 2nd Ed. Burns, D.A., and E.W. Ciurczak, Eds. Marcel Dekker, Inc. New York. 814 pp.

Svedas, V. 2000. Transformation of the near infrared bands of cellulose surface hydroxyls under the influence of adsorbed water molecules. Appl. Spec. 54(3):420-425.

Thumm, A. and R. Meder. 2001. Stiffness Prediction of radiata pine clearwood test pieces using near infrared spectroscopy. J. Near Infrared Spec. 9:117-122. Thygesen, L. and S. Lundqvist. 2000. NIR measurement of moisture content in wood under unstable temperature conditions. Part 1. Thermal effects in near infrared spectra of wood. J. Near Infrared Spec. 8(3):183-189. --and--.2000. NIR measurement of moisture content in wood under unstable temperature conditions. Part II. Handling temperature fluctuations. J. Near Infrared Spec. 8(3): 191-199.

Via, B., T. Shupe, L. Groom, M. Stine, and C. So. 2003. Multivariate modeling of density, strength and stiffness from near infrared spectra for mature, juvenile and pith wood of longleaf pine (Pinuspalustris). J. Near Infrared Spec. 11(5):365-378.

Walrafen, G.E. 1972. Raman and Infrared Spectral Investigation of Water Structure. Franks, F., Ed. Plenum Press, New York. Vol. 1, Chap. 5, pp. 195.

Wentzell, P. and L. Montoto. 2003. Comparison of principal components regression and partial least squares regression through generic simulations of complex mixtures. Chemom. Intell. Lab. Syst. 65(2): 257-279.

The authors are, respectively, Graduate Student and Associate Professor, Division of Forestry and Natural Resources, Davis College of Agriculture, Forestry, and Consumer Sciences, West Virginia Univ., Morgantown, West Virginia (, The authors express their appreciation to Trus-Joist, a Weyerhaeuser iLevel Veneer Technology Group, Buckhannon, West Virginia, for supplying the veneer sheets for this study; West Virginia Univ. Wood Utilization Research Program funded by USDA/CSREES (Improving Quality and Utilization of Upland Hardwoods in the Appalachian Region); and to the Social Justice Dept. (Minority Doctoral Program), West Virginia Univ., for the support of the graduate work of Mr. Oluwatosin Emmanuel Adedipe. This paper was received for publication in April 2007. Article No. 10340.
Table 1.--Average MC of yellow-poplar veneer obtained by conditioning
to different relative humidity (%) in chambers containing saturated
inorganic solutions.

 values Temp. Relative Average
Salt/conditioner ([a.sub.w]) (degrees]C) humidity MC (a)


Lithium chloride 0.11 25 20 3.3
Potassium acetate 0.23 25 28 5.2
Sodium bromide 0.57 25 56 10.1
Sodium acetate 0.76 25 73 14.5
Potassium nitrate 0.94 25 85 24.5
Ammonium monophoshate 0.82 25 95 27.5
Drierite 0.11 20 0.31
Dcionized water 25 99 29.5

(a) Average ovendry MC (percent) of 16 samples at EMC.
MC ranges of 60.1 to 69.5 percent and 70.5 to 81 percent were obtained
by soaking yellow-poplar veneers in deionized water for 1 and 24 hours,

Table 2.--Comparison of PCR and PLS models developed for estimating
MC (%) in veneers at different NIR region.

 Principal component regression (PCR) models

 R-sq. R-sq.
Wavelength PC cal. RMSEC SEC value RMSEP SEP
 (nm) (a) (b) (c) (d) (c) (f) (g)

800 to 2500 2 0.961 0.471 0.474 0.951 0.514 0.516
1000 to 2500 1 0.952 0.527 0.530 0.938 0.573 0.575
1300 to 2100 4 0.990 0.250 0.251 0.982 0.306 0.307
1400 to 1940 4 0.993 0.204 0.206 0.985 0.280 0.281
1500 to 2000 4 0.991 0.223 0.224 0.984 0.290 0.290

 Principal Projection to latent structures
 component (PLS 1) models
 (PCR) models
 R-sq. R-sq.
Wavelength PC cal. RMSEC SEC value
 (nm) Bias (a) (b) (c) (d) (c)

800 to 2500 0.029 2 0.980 0.338 0.340 0.973
1000 to 2500 0.050 2 0.978 0.356 0.358 0.971
1300 to 2100 0.027 3 0.990 0.239 0.240 0.982
1400 to 1940 0.031 4 0.993 0.197 0.198 0.986
1500 to 2000 0.037 3 0.992 0.215 0.216 0.984

 Projection to latent structures (PLS 1) models

Wavelength RMSEP SEP
 (nm) (f) (g) Bias

800 to 2500 0.380 0.381 0.027
1000 to 2500 0.390 0.392 0.021
1300 to 2100 0.313 0.313 0.030
1400 to 1940 0.275 0.275 0.028
1500 to 2000 0.288 0.288 0.035

(a) No of principal components.

(b) Calibration R-square.

(c) Root mean square error of calibration.

(d) Standard error of calibration.

(e) Validation R-square.

(f) Root mean square error of prediction.

(g) Standard error of prediction.
COPYRIGHT 2008 Forest Products Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2008 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Adedipe, Oluwatosin Emmanuel; Dawson-Andoh, Ben
Publication:Forest Products Journal
Article Type:Report
Geographic Code:1USA
Date:Apr 1, 2008
Previous Article:Fingerprint traceability of logs using the outer shape and the tracheid effect.
Next Article:The effect of compressive load on the moisture content of oak blocks during radio-frequency/vacuum drying.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |