SPECTROPHOTOMETRIC METHOD TO MEASURE QUALITY AND STRENGTH PARA¬ METERS IN TREES, LUMBER, TIMBER, CHIPS, SAW DUST, PULP AND PAPER.
The background of the invention
The development of the quality concept in forestgrowing is more and more noticed. Optimal growth, fast or slow is correlated to desired quality for differ¬ ent uses. Various wood species yield products with different properties most useful for products ranging from window frames to fibre boards to paper. Place of growth and growing conditions results in different wood properties de¬ scribed, for instance, density. There are also large differences between core wood and splint which can be noticed in various processes. The water content in wood, lumber, chips and saw dust is strongly depending on season and is an important value parameter, as is for instance the ageing of wood, lumber, chips and saw dust. Frozen lumber, chips and saw dust may cause process distur¬ bances.
This can be analysed by various methods. These are today however both time consuming and costly.
It is important to obtain information from the whole process line from wood, lumber, chips, saw dust, pulping and bleaching to the final product. The line can also be customers, sellers and their different product demands. Intermedi¬ ate process-steps in the line or parts thereof also have to be determined in order to obtain control feedback to the manufacturing process.
Pulp is characterised by, for instance, the measuring of tensile and tear index, density and light scattering coefficient. It is thus evident that not only chemical parameters such as kappa number - a measure of remaining lignin in the pulp - cellulose and hemicellulose are crucial for the product characterisation.
Chemical as well as physical parameters related to the properties of the final product - what the customer wants and pays for - are asked for. Tearindex and tensiieindex are examples of this. During pulping and bleaching the pulp is changed as well in properties as in appearance. The same holds true for paper after addition of filler materials and treatment with different chemicals. Addi¬ tion of recycled fibre demands knowledge about the quality of virgin fibre as well as recycled. In more complex systems, for instance coated or in other ways treated or laminated paper, different parameters and quality properties are cru¬ cial for the final result.
The result of various process steps are not only related to the result of the previous process step but also to the quality of the initially charged raw mate¬ rial. The characterisation of the raw material and /or intermediate products en- ables classifying /sorting as well as monitoring of different process parameters.
Today we lack methods of determination, which are sufficiently fast and wide to be applied in handling of raw materials, during ongoing processes and for control of the product to obtain optimal results.
A future competitive manufacture will need improved measuring methods in order to be able to
• sort and select optimal raw materials
• change the process parameters according to raw material as well as de¬ sired customer quality
• monitor and adjust the process steps according to the present status of the process
• monitor and control an even and desired intermediate and end-product quality.
A review of desired basis for classification and sorting of wood raw materi¬ als "cost-efficiency-competition" STFI bulletin A1001 December 1993. In this review today's methods for measuring wood raw materials are described. De¬ crease of variation of various parameters at the production of thermo mechanic pulp by classifying and sorting of the wood raw material is described by Braaten IMPC, Proceeding from Int. Mec. Pulp Conference, Oslo 1993. Braaten shows that the variation decreases by 20-50% after classing and sorting. Access to a fast method for classifying by simultaneous determination of moisture and the quality of the dry part of wood should have a surprisingly large potential. As further examples variations in moisture in wood raw material may result in a difference of yield of pulp per ton of charged raw material from 160 to 330 kg of produced pulp. Certainly the production parameters during pulping are affected by moisture and quality of the charged raw material. Today, low and high yield wood have the same price. The value of wood raw material for pulping in Sweden only is 15 billion Skr.
Similar demand of classifying and determination of quality is present in the board industry. Manufacturers of fibre board today determine for the process manufacturing optimal ageing of saw dust by a sensorial determination based on human experience. Here, also, there is a big demand of instrumental analy¬ sis to determine the ageing as well as monitoring of the manufacturing process in order to optimise choice of glue brand and added amount of glue. The reject percent today is 5-7% of manufactured board. The above determinations demands that one can measure single parameters as well as several parameters at the same time on a short timescale i.e. minutes to seconds, and also correlating effects between different parameters. Not until then the producer can obtain a total description of customer relevant quality.
DESCRIPTION OF THE INVENTION
The invention is a method to determine properties, qualitative and quantita¬ tive, of cellulose fibre products such as trees, lumber, timber, chips, saw dust, pulp and paper.
The method can be performed before, during and after processing or parts thereof to a from the charge to the final product.
The method can be used for product classification, sorting and for monitor¬ ing of processes and additions during different process steps.
The method is characterised by the mathematical treating data from spec- troscopy of named fibre products by multivariate methods so-called chemometrics, with the spectra preferably obtained in the wavenumber region of 200-15400 cm"1.
The invention is not limited to the cellulose fibre products given above.
The method also can predict several parameters determined at the same time, chemical as well as physical as latent and collective parameters such as processability. Further, the status and changes in a process such as ageing or chemical conversion can be determined. Parameters and properties today tak¬ ing hours and days to determine now can be made in minutes, seconds or parts thereof.
Mentioned manufacturing processes generally consists of several steps where the product of one step is the raw material for the next. The advantage of being able to use comparative methods of determination during and in- between several process steps is obvious. The figure below show determination of dependent variables in processes with several steps.
P — P P — P p_„p P— P
11 li 21 2j 31 3k 41 41
Where P is relevant quality parameter for the processing step.
To judge the properties of a complex product, methods have been developed who try to simulate the use of certain customer category. These methods often comprise special apparatus and result in, with few exceptions, approximations of what can be called a customer relevant quality property. The present method makes it possible to obtain a more real determination of quality as the method is calibrated with the customers real vision of quality as objective.
The link between the desired quality properties and the properties of the raw material and different process parameters open entirely new possibilities to relevant optimisations. With the present method determinations can be made within seconds or fractions thereof in comparison with today's conventional methods, taking hours and days. One goal of the invention is to describe and estimate properties and quality parameters in a substantially more cost effec¬ tive way.
It is not always desirable or necessary to relate a sample or a single process step to the raw material. The method is also useful in the guiding and control of ongoing processes and determinations without the mentioned linking to the raw material.
A further possibility of characterisation by the method is determination of so-called latent information. By latent information we mean results of several parameters together. Single parameters and correlation may however be meas- urable, but the demanded total result can also be grouped in a way where the effects of underlying chemical and physical parameters appear as one entity.
Thus we can obtain a simple determination of quality. An example is deter¬ mination of the influence of saw dust composition on accepted quality of the final product - fibreboard. It is generally not possible to link the influence of one or more of today's measured parameters to the accepted quality mentioned above. The judgement today is usually based on human experience. This depends on the correlation between several parameters which is difficult to determine. Such knowledge based on experience may be used as a basis of calibration for the proposed method.
Another example is the ability to predict quality as for instance flaws in me¬ chanical properties of lumber used for glued wood. We can furthermore also determine single qualitative /quantitative parameters, season of growth and season of logging as well as how the tree has and been stored after logging. Further examples on qualitative parameters is the possibility to sort lumber after growth place. The total information can be used for decision of which end product the wood is best suited for. By our method we can determine such la¬ tent information before and during the processes to final product.
It has earlier been possible to determine certain chemical parameters in pulp by Near Infra Red spectroscopy, NIR, such as in Birkett, M.D. and M.J.T. Gam- bino, Estimation of pulp kappa number with near-infrared spectroscopy. Tappi Jour¬ nal, 1989. (September): p. 193-197. Easty, D.B., et al., Near-Infrared Spectroscopy or the Analysis of Wood Pulp: Quantifying Hardwood-Softwood Mixtures and Esti¬ mating Lignin Content. Tappi J., 1990. 73(10): p. 257-261. Friese, M.I. and S. Banerjee, Lignin determination by FT-IR. Appl. Spect., 1992. 46(2): p. 246-248. Wright, J.A., M.D. Birkett, and M.J.T. Gambino, Prediction of pulp yield and cellu-
lose content from wood samples using near infrared reflectance spectroscopy. Tappi J., 1990. 73(8): p. 164-166.
The works mentioned above require in their mathematical models that no disturbances in the form of non linear light scattering and matrix effects are present - a requirement that can be strongly questioned and also appears in the quality of their models, which are not as good as in the present invention. The models used are based on linear relations with the wanted concentrations at one or several spectral wavelengths. A simple survey of the spectra which are used in these models we note that absorption maximum, reported as charac- teristic for the required parameter show a drift in the wavelength plane hereby rendering the models non-valid.
Backa, S. and Brolin, A. in Determination of pulp characteristics by diffuse reflectance FT-IR. Tappi J., 1991. 74(5): p.218-226, use PLS to predict lignin and sugar contents by FT-IR in combination with PLS. However the suggested method demands that the samples are dried and homogenised by grinding. This limits the applicability of their method considerably.
Wallbacks, L., U. Edlund, and B. Norden, Multivariate data analysis of in situ pulp kinetics using 13C CP/MAS NMR. J. of Wood Chemistry and Technology, 1989. 9(2): p. 235-249. Wallbacks, L., Pulp Characterization Using Spectroscopy and Multivariate Data Analysis. 1991, University of Umea, Dept. of Organic Chemis¬ try, S-901 87, Umea, Sweden: Wallbacks, L., et al, Multivariate characterization of pulp using solid-state C13 NMR, FT-IR and NIR. Tappi J., 1991. 74(10): p. 201-206, show that NIR, compared to NMR and FT-IR is the spectroscopic method gen¬ erating most information from a system of birch pulp. He shows that xylose, glucose, Klason lignin and galactose can be predicted by NIR, FT-IR, NMR in combination with multivariate methods. The best result is obtained by adding spectra from the three methods to a combination spectrum. Wallbacks tech¬ nique does not generate as good models as those in the present invention. This is caused by non linearities mentioned above that can not acceptably be mod- elled. Possibilities are also shown by transforming classic physical descriptors (18#) originating from 7 commercially available pulp samples to a latent vari¬ able plane (with PCA), to follow the influence of certain unit operations such as grinding and with some doubt predict those. Wallback points out that it is "extremely important to be able to characterise pulp (physically) in a fast and efficient manner in the future. One way would be using spectroscopic measures such as NIR or FT-IR". We show exactly this in the present invention.
Wallback also show in: Characterisation of Chemical Pulps using Spectros¬ copy, Fibre Dimensions and Multivariate Data Analysis, in the 7th ISWPC. 1993. Beijing, PR of China. The possibility to predict physical descriptors by spectroscopy. It is however necessary that to independent data (spectra) add
information about grinding which generally is not available in the real case. This makes Wallback's method to differ from the present invention.
In the present invention the absorbance data obtained are analysed by multi¬ variate statistical analysis e.g. chemometrics. To obtain optimal analysis results the spectral data first have been linearised.
For NIRR (Near Infra Red Reflectance) the measured absorbance is not line¬ arly correlated to the desired properties of the samples. This depends mostly on light scattering effects, in turn depending on the physical nature and form of the samples. Several methods of transforming for linearising have been devel- oped to overcome with this effect. Four of the most common are:
• Kubelka-Munk transformation (K-M), which by a physical scatter model tries to correct for scatter-wavelength dependency and also the dependency of the aborbance-samplematrix effect (Kubelka, P. and F. Munk, Z. Tech. Physik, 1931. 12: p. 593) • "Multiplicative Scatter Correction" (MSC), where each spectrum is linearised against some overall feature of the calibration spectra, usually the grand mean of the wavelengths (Geladi, P., D. MacDougall, and H. Martens, Linearization and Scatter-Correction for Near-Infrared Reflectance Spectra of Meat. Appl. Spect, 1985. 39(3): p. 491-500. Isaksson, T. and T. N_es, The Effect ofMulti- plicative Scatter Correction (MSC) and Linearity Improvement in NIR Spectroscopy. Appl. Spect., 1988. 42(7): p. 1273-1284).
• The derivative of the spectrum. The method calculates the difference be¬ tween adjacent wavelengths (O'Haver, T.C. and T. Begley, Anal. Chem., 1981. 53: p. 1876. Savitzky, A. and M.J.E. Golay, Smoothing and Differentiation of Data by Simplified Least Squares Procedures. Anal. Chem., 1964.36(8): p. 1627-1639).
• "Standard Normal Variate Transformation" (SNV), where the grand mean of the model spectrum wavelengths is subtracted and the discretised wave¬ lengths are nor ed with their standard deviation (Barnes, R.J., M.S. Dahnoa, and S.J. Lister, Standard Normal Variate Transformation and De-trending ofNear- Infrared Diffuse Reflectance Spectra. Appl. Spect., 1989. 43(5): p. 772-777).
In this invention we can with advantage use 2" factorial designs on discrete levels (Box, G.E.P., W.G. Hunter, and J.S. Hunter, Statistics for Experimenters. 1978, New York: John Wiley & Sons. 653. Olsson, R.J.O., Optimizing data- pretreatment by a factorial design approach, in Near Infra-Red Spectroscopy, K.I. Hildum, et al, Editor. 1992, Ellis Horwood Limited: Chichester. p. 103-107) to obtain the linearising combination which give optimal results for each property and parameter of the samples.
Multivariate analysis of absorbance or reflectance data identify spectral fea¬ tures, so-called principal components, which are then used to model qualitative
and quantitative properties of the samples. When a sample set is obtained for a number of samples with a representative spread in desired qualitative and quantitative properties, chemometrics is used to construct a calibration set and to predict unknown samples. There are several useful chemometrical methods. PCA (Principal Component Analysis) (ASTM, Standard E131-90, Def ofPCA, in Standard Definitions of Terms and Symbols Relating to Molecular Spectroscopy. De- vaux, M.F., et al., Application of multidimensional analyses to the extraction of dis¬ criminant spectral patterns from NIR spectra. Appl. Spect., 1988. 42(6): p. 1015- 1019. Devaux, M.F., et al, Application of principal component analysis on NIR spec- tral collection after elimination of interference by a least square procedure. Appl. Spect., 1988. 42(6): p. 1020-1023. Geladi, P. and B.R. Kowalski, Partial Least- Squares Regression: A Tutorial Anal. Chim. Acta., 1986. 185(1-17): p. 1-32. Mar¬ tens, H. and T. N ES, Multivariate Calibration. 1989, John Wiley & Sons. Wold, S., K. Esbensen, and P. Geladi, Principal Component Analysis. Chemometrics and Intelligent Laboratory Systems, 1987. 2: p.37-52)
PLS (Partial Least Squares regression) (Geladi, P. and B.R. Kowalski, Partial Least-Squares Regression: A Tutorial. Anal. Chim. Acta., 1986. 185(1-17): p. 1-32. Haaland, M.D. and E.V. Thomas, Partial Least-Squares Methods for spectral Analy- ses. 1. Relation to Other Quantative Calibration Methods and the Extraction of Quali¬ tative Information. Anal. Chem., 1988. 30(11): p. 1193-1203. Martens, H. and T. NJES, Partial Least Squares Regression (PLSR), in Multivariate Calibration. 1989, John Wiley & Sons: p. 112-165).
The chemometric algorithms preferred in this invention are PCA for qualita¬ tive results and PLS for quantitative results.
PCA is a powerful transformation technique which projects a multidimen¬ sional data set with correlated variables to a smaller data set with non corre¬ lated variables. The purpose of this transformation is to rotate the coordinate system so that maximal amount of information is obtained on fewer axis than in the original arrangement. This smaller and non correlated amount of data which retains almost the same amount of information as the original data-set thereafter can be used to calculate predictive model. Generally 2-4 principal components explain up to 99% of the variance of the data. The principal com- ponents which can be related to desired properties of the samples are then used to determine these properties i.e. qualitative results.
PLS is a development of PCA having a regression step between information from the principal components and a desired property of the samples, i.e. PLS gives quantitative results.
The invention describes a method for the determination of quality of differ¬ ent material originating from wood, which we hereafter have given the collec¬ tive name of cellulose fibre products. The invention will now be described in detail. The procedure according to the invention is described for discrete wood, saw dust, chips and pulp samples which have been scanned by NIRR in a probe cell in laboratory environment, but the procedure is not limited to samples measured in this manner, measurements in situ can also be made. For the latter measurements fibre optics can be used in any standard configuration i.e. the detector and/ or source of light can be placed remote from the measured object, the actual cellulose fibre product. The detector obtains its analyte-beam from the light source via one or more optical fibres by which in some way passes via the sample.
During the calibration phase absorbance data, preferably in the wavelength region between 200 cm"1 and 15400 cm"1, are collected for samples with known chemical concentrations and physical properties.
The samples collected reflects the actual span of the parameters or properties of the product. The number of samples which in this way are gathered for the calibration should be more than 30 and they must also be representative for the parameters or properties to be measured. Standard methods (e.g. TAPPI meth- ods) which are common in wood, paper and pulp industry are used to obtain mentioned reference concentrations and properties. Absorbance data, in the above mentioned spectral region, of corresponding samples are then analysed with the mentioned multivariate statistics with the objective to build a predict¬ ing model.
EXAMPLES
For all examples following below the absorbance measurements where made in a "high-fat/moisture test cell" with a diffuse Near Infrared Reflectance (NIRR) spectrophotometer, NIRSystems 6500, between 400 and 2500 nm and with a resolution of 2 nm. An exception is pulp I, where the instrument used was a technicon Infra Analyser 500 in the spectral range of 1100-2500 nm and a resolution of 4 nm.
Example 1.
Figure 1 "Qualitative analysis of wood chips" show the possibilities to iden¬ tify wood chips from different tree species and trunk sampling location. The samples are from spruce, pine, maple, birch and are divided in bark and wood respectively. In the case of pine a separation of splint and core was made. Every quality are represented by 5 samples of each. Obtained spectral data were ana¬ lysed by PCA. Very good possibilities for identification are present as shown in figure 1.
Example 2.
To obtain optimal quality in board manufacturing the charged saw dust must have passed through a certain storing/ageing. The storing process is af¬ fected by external factors, such as temperature, moisture, season, biological degradation, quality and origin of ingoing saw dust. Especially large problems are present in winter time. The extent of how far the storage procedure has pro- ceeded is today determined by sensoric human estimation. The present inven¬ tion correlates the qualitative judgement of the status of the storage procedure, so-called processability, by spectral data.
Obtained absorbance data on 26 stored samples which processability has been classed sensorically, were predicted by PLS. Figure 2 "predicting of processability of saw dust" show that the descriptor processability can be well correlated. In the figure the 'true' value is plotted against that predicted by the model which means that if the points are on the 45°, the model is predicting exactly the true value. The invention is applicable on classing of saw dust before manufacture of boards, as well as during the process itself for adjustment of different process parameters, as for instance the amount and quality of the glue.
The statistical measures in the figures are defined as follows:
SEP = (n- l)-1∑(c1 - c1 - (c - c))2 (4) i=l
RMSEP = VMSEP (5)
where: n = number of samples
£• = corresponding value predicted by the model c- = traditionally measured parameter or measure of quality i = sample number (i=l:n)
Example 3.
Characterisation of pulp traditionally is difficult and time consuming as the pulp is chemically and physically complex. The use of NIRR and multivariate analysis yield a substantial time gain, as you with only one spectral measure¬ ment can determine several chemical and physical parameters and properties simultaneously.
I. Chips were taken from birch, stored during 1.5 years, dried and sieved. The chips were boiled in an autoclave in laboratory scale and 46 pulp samples were taken out during different intervals to obtain a varying degree of delig- nification. The samples were dried and ground before analysis. Figure 3 a-f "characterisation of pulp I" show how yield, kappa number and concentrations of lignin, glucose, xylose and uronic acid were predicted by PLS.
II. 25 samples of pulp from normal production at the paper mill were ground each at 0, 500, 1000, 2000 and 4000 rpm, which resulted in 125 samples which were scanned by NIRR. Figure 4 a-h "Characterisation of pulp II" show how degree of beating, tensiieindex, tearindex, burstindex, light scattering coef¬ ficient, modulus of elasticity, density and bending value number be predicted by PLS. In several of the figures there is a visible difference between the differ¬ ent degrees of grinding.
Example 4
One of the most important parameters to be determined in timber is mois¬ ture. In the example below moisture was determined on 50 samples of spruce. The samples are sorted after area of growth, different sampling locations on the trunk. The samples were obtained during various periods from January through April. Samples from all these groups are used in the regression model. The samples were accuired as saw dust from above mentioned spruce logs which subsequently was submitted to NIRR analysis in the wavelength interval of 400-2500 nm. The spectra obtained have first been pretreated by the second derivative according to Savitzky et. al. Then a multivariate regression was made with all wavelengths using the PLS regression, for prediction results see figure 5.
Reference analysis of moisture content was determined according to "Tappi test methods".