METHOD FOR IDENTIFYING PROPERTIES OF WOOD BY INFRA-RED OR VISIBLE LIGHT
FIELD OF INVENTION
The invention comprises a method for predicting the behaviour of wood during processing for end use, such as the susceptibility of the wood to form internal checks during drying, or predicting for one or more characteristics of the wood after processing, such as the dimensional stability of the wood during or after drying or in service, or the elastic modulus of the wood after drying.
BACKGROUND OF INVENTION
Radiata pine is used in New Zealand for many end uses. These require appropriate grading and consequent decision making at several stages of processing felled stems or logs towards the end product, with the aim being to extract as much high quality product out of the wood resource as possible, while avoiding overprocessing or inferior or reject product. This requires informed decision making at one or more stages of processing of the wood. The earlier an informed decision can be made, the more money can be saved or made by sending timber towards the correct end use or the processing stream for the best end use for the particular wood. An informed decision requires the ability to assess the quality of the material as early as possible in the processing of the wood. This requires predicting the behaviour of the wood during subsequent processing for end use, or predicting characteristics of the wood which affect the grading or quality of the processed wood products.
Internal checking in wood can occur after drying of the wood. Rapid drying of wood from a higher moisture content to below 18% and typically around 12% moisture content may result in some breakdown of the cellular structure of the wood creating small voids within the wood, which are exposed when the wood is further sawn or planed making the wood or timber pieces unsuitable for further processing to finishing boards or mouldings, or in wood panels, etc. The incidence of internal checking after drying of timber is an increasing problem for new crop radiata pine. Checking occurs in the sapwood earlywood adjacent to the heartwood/sapwood boundary. The incidence of checking can extend for several rings towards the bark, although checking does not develop through the latewood. Currently it is difficult if
not impossible to predict in advance the susceptibility of particular wood pieces or logs to internal check formation during drying.
Another significant issue with the use of framing timber studs in construction is one of dimensional stability. It is generally accepted that this arises from the differences between the moisture content of the newly-dried and seasoned lumber from the wood drying kiln, and the moisture content the wood attains in use. At the completion of the drying and reconditioning cycle, the moisture content of the wood is an average 12%. In use as framing studs, the moisture content of wood can vary from 5% to over 18% with an annual average of 16% expected by NZ Standard 3603. Radiata pine wood is permeable, and therefore the moisture content of the wood can readily change to equilibrate with the changing relative humidity of the environment, with the result that wood, which has the propensity to do so, undergoes distortion.
SUMMARY OF INVENTION
In broad terms in one aspect the invention comprises a method for predicting the behaviour of wood or one or more characteristics during or after subsequent processing of the wood for end use, comprising obtaining an infra-red or visible light reflectance spectra for the wood and predicting the behaviour of the wood during subsequent processing or one or more characteristics of the wood after subsequent processing by reference to the infra-red or visible spectra obtained.
Preferably the method includes subjecting logs or wood pieces to a source of infra-red radiation, detecting the levels of reflected radiation over the infra-red range or at a number of wavelengths in the infra-red range, and analysing the infra-red reflectance spectra relative to stored comparative information on infra-red reflectance data for logs or wood pieces.
By radiation in the near infra-red (NIR) region is meant radiation of wavelength(s) in the range 1100-2500 nm. By radiation in the mid infra-red (MIR) region is meant radiation of wavelength(s) in the range 2500-25000 nm. By visible light is meant radiation of wavelength(s) in the range 400- 1000 nm.
Preferably the method includes assigning a weighting value to the analysis obtained for each log or wood piece and predicting the behaviour of the wood during the
subsequent processing of the wood or the one or more characteristics of the wood after the subsequent processing, by comparing the weighting to a preset dividing value for such weightings between wood known to and known not to exhibit the behaviour to a predetermined extent or known to or known not to exhibit the one or more characteristics to a predetermined extent.
More particularly in one aspect the invention includes a method for predicting the susceptibility of wood to internal checking during or after subsequent processing of the wood comprising obtaining an infra-red or visible light reflectance spectra for the wood and assessing the susceptibility of the wood to internal checking by reference to the infra-red or visible spectra obtained.
In the identification of wood prone to internal check formation the wavelengths of strongest correlation are about 1340 and 1930 nm in the case of NIR and bands in the region of about 1520, 1670, 1920, 2295, 2375 and 2475 cm-i in the case of MIR.
Processes governing the development of internal checks during drying are perfectly understood. Internal checking is often referred to as being the result of drying stresses. It may be caused by changes in cell wall composition and particularly in the way that water is bound in the cell walls. Nuclear magnetic resonance relaxation time studies support the difference between the nature of water in cells of wood prone to checking and that which is not.
More particularly in another aspect the invention includes a method for assessing the susceptibility of wood to dimensional instability after subsequent processing comprising obtaining an infra-red or visible light reflectance spectra for the wood, and assessing the susceptibility of the wood to dimensional instability by reference to the infra-red or visible spectra obtained.
Light in the visible spectrum and particularly in the 570 - 620 nm band (yellow) shows particularly strong correlation with dimensional stability. Energy with the 1330 - 1400 nm, 1670 - 1700 nm and 1900 - 1950 nm wavelength bands in the NIR region showing strong correlation.
Traditionally the likely dimensional stability of wood over time has been considered difficult to assess or predict, because it is not related to any particular measurable
wood property or mix of wood properties and is normally referred to as being the release of growth stresses during drying (natural or otherwise). Dimensional stability may be related to, or dimensional changes caused by, complex and fundamental changes in chemical wood composition, and by recording spectral data of wood proved to be unstable and wood proved to be stable (during processing and use) we have developed chemical fingerprints that are directly correlated with stability performance in wood. The method of the invention is useful for identification of wood or wood products which are prone to changes in dimension due to twist, bow, crook or cup after initial drying of the material or while in service due to changes in moisture content or humidity.
More particularly in a further aspect the invention includes a method for predicting the elastic modulus of wood after subsequent processing, comprising obtaining an infra-red or visible light reflectance spectra for the wood, and predicting the elastic modulus of the wood pieces by reference to the infra-red or visible spectra obtained.
Light in the 1530 - 1540 nm, 1670 - 1700 nm 1900 - 1950 nm and 2080-2100 nm wavelength bands in the NIR region in particular show strong correlation.
The method enables the elastic modulus of dry wood to be predicted before drying, when the wood is at a higher moisture content and may be in an undried or green state.
Preferably in any case the method includes analysing the resulting spectra using a principle components analysis methodology or a projection to latent structures regression methodology.
Preferably the method includes assigning a weighting value to the analysis obtained for each piece of wood and predicting the likelihood of internal checking, or of dimensional stability or instability, or the elastic modulus for the wood piece by comparing the weighting to weightings for wood known to be and known not to be susceptible to internal checking or dimensional instability, or of known elastic modulus.
The subsequent processing of the wood or logs or wood pieces may include drying the wood to a lower moisture content such as less than 30% by weight of the wood,
less than 18% by~weight of the wood, or in the range 8 to 12% by weight of the wood. Before drying the wood may be green wood.
DESCRIPTION OF DRAWINGS
The invention is further described with reference to the accompanying drawings and by way of example, wherein:
Figure 1A is a schematic view showing monochromatic radiation being selected by a monochromator and directed via a fibre optic to the wood surface with the reflected monochromatic light being directed via a fibre optic to the detector system,
Figure IB is a schematic view showing monochromatic radiation being selected by a monochromator and directed via a fibre optic to the wood surface with the reflected monochromatic light being detected directly by a detector system,
Figure 1C is a schematic view showing polychromatic light being applied to the surface from an external source which may or may not be simply sunlight and the reflected light being directed via a fibre optic to a monochromator selecting monochromatic radiation to be detected by the detector system,
Figure 2 schematically shows the steps required to develop and implement a PLS model,
Figure 3 shows the prediction of propensity for internal check formation of samples based on mid infra-red spectra obtained from ground samples,
Figure 4 shows the prediction of stability of samples based on mid infra-red spectra obtained from ground earlywood,
Figure 5 shows the prediction of stability of samples based on near infra-red spectra obtained from ground earlywood and latewood,
Figure 6 shows the prediction of stability of samples based on near infra-red spectra obtained from solid samples of wood, and
Figure 7 is a diagram showing the steps used to produce a laminated veneer lumber joist for reference testing of elastic modulus.
DESCRIPTION OF PREFERRED EMBODIMENTS
A preferred arrangement for the acquisition of the spectral data is illustrated in Figure 1. The apparatus comprises a source 1 of polychromatic radiation in the visible and/ or infrared region of the electromagnetic spectrum, a monochromator for generation of monochromatic radiation 2, a fibre optic cable to guide the radiation to and/ or from the wood surface 3, and a detector system 4.
The source for infra-red radiation may be a tungsten halogen lamp in the case of visible light and near infrared or a glowbar in the case of mid infrared, for example. The detectors may be lead-sulphide based for example for visible and near infrared or alternatively a diode array detector, and DTGS for mid infrared.
The method of the invention may however be carried out in any NIR or MIR or visible light scanning arrangement. Preferably the whole of the wood pieces are scanned to obtain an average spectrum along the length of each wood piece. A source and detector may move over the wood piece or wood pieces or a stationary source and detector may be connected to a gantry-mounted moving head via a fibre optic cable system as outlined above for example, or alternatively the wood pieces may move past a source and detector in a production of flow for example.
The data resulting from the scan, comprising reflectance data over the wavelength range or bands used for the particular application is then analysed and compared to comparative spectra information and memory of a data processing computer or control system, on infra-red reflectance data for wood subsequently found to have suffered or not suffered internal checking or subsequently found to be dimensionally stable or unstable, or of known elastic modulus, after drying for example. The statistical analysis may be carried out using a principle components analysis methodology or a projection to latent structures regression methodology for example.
Principal component analysis (PCA) is a well documented procedure (Wold S, Ebensen K and Geladi P (1987) Principal Component Analysis. Chemom. Intell. Lab.
Syst. 2:37-52, and Martens H and Nses T (1991) Multivariate Calibration, John Wiley
& Sons, Chichester. 419 pp.), by means of which the variation in a set of multivariate data is described in terms of a set of uncorrelated linear combinations of the original, uncorrelated variables. The information in a multivariable matrix X, can be described as the product of two smaller matrices T (the scores matrix) and P' (P- transpose, the loadings matrix) such that the original X matrix can be described as the product of the score an loading matrix according to
X = TP' + E ; where E is the x-residuals matrix
These new variables, or principal components, can be used as explanatory (regressor) variables in multiple regression; they have the considerable advantage of linear independence and so they avoid the problems associated with multicollinearity.
Principal components are derived in decreasing order of size and variance, so that the first few components usually account for most of the variation in the original data. It has been suggested that where this is so, the first few components may be used to summarise the data and the smaller components ignored as they contain little information. The dimensionality of the data, and the number of regressor variables can thus be reduced.
Projection to latent structures regression also results in a linear prediction equation and the development and implementation of a PLS model is shown schematically in Figure 2 wherein the sample 1 is measured for a particular property or properties using a reference method 2 and its corresponding spectrum obtained 3 such that the PLS modelling software 4 can produce a calibration model 6. The measured property may be recorded after subsequent processing. Spectra obtained from unknown samples can then be compared against the calibration model developed and a prediction made as to their property performance during or after processing. The general form of the PLS model for is:
X = TP' + E and Y = TB + f ; where B is the regression matrix and f is the y-residuals matrix
The PLS vectors are derived one at a time and so computation time can be reduced when there are many latent variables and most contribute little information about the dependent variable.
The method of the invention is further illustrated by the following examples.
EXAMPLE 1 - Prediction of Internal Checking
Sections of kiln dried boards were sampled. The boards were cross-cut at 50 mm intervals and the total number of checks recorded and summed per board. This number was used as the regressed variable in the Y matrix of the PLS model. Subsamples were cut from rings in which the checking occurred. Where checking did not occur, samples were cut from the rings adjacent to the heart/sapwood boundary. Subsequently the samples were ground to pass through a 60 mesh sieve in the discharge.
Separate earlywood and latewood subsamples for the 20 boards were recorded using a commercial infra-red spectrometer according to established in-house protocols. The recorded wavelengths are 400 - 4400 cm-i (64 scans, resolution 4 cm 1). Near infrared spectra were recorded on a commercial spectrometer equipped with a remote reflectance system (intact solid samples) or a sample cup (ground wood meal). Spectra were recorded between 400 and 2500 nm in 2 nm increments.
Statistical analysis was executed using The Unscrambler™ Version 7. 1 (CAMO A/S, Norway). This detects systematic differences between the samples based on spectra or a set of physical data and to summarise them in a few variables called latent variables. The method used to create the prediction models was projection to latent structures regression.
The statistical analysis was executed in the following steps:
1. Principal component analysis of the separated earlywood was performed. This was considered valid, as checking does not occur in the latewood band. This was performed for two independent sample sets.
2. The earlywood samples of the two sample sets were combined into one sample set. From this set 16 samples were removed randomly as a test set for a later prediction (3 with and 3 without checking from first set, 5 with and 5 without checking from the second set). The remaining 30 samples were used for the
development of a prediction model via PLS against which the test set was predicted.
The results of predicting the susceptibility of the 16 samples in the test set are shown in Figure 3. All samples with moderate to severe checking have been predicted correctly.
EXAMPLE 2 - Prediction of Dimensional Stability
Fresh, green sapwood lumber framing studs of dimensions 50 x 100 mm x 4.8 m selected from similar clonal material were conventionally kiln-dried and cut to 2.4 m lengths. All the studs were then weighed and measured for dimensions and initial conformation (crook, bow, twist and cup). These data served as the "baseline" for measuring distortions induced by the environment.
The studs were transferred to climate controlled rooms. The studs were stood near- vertically side-by- side with sufficient room to allow any distortion to freely take place. The treated studs were submitted to a temperature and humidity regime over the course of 5 weeks cycling from 60-90% RH and including "rain events" to fully saturate the samples. At the end of each conditioning period, the studs were removed from the climate room and measured. The studs were kept out of the climate rooms for as short a time as possible to enable the measurements to be obtained. In all, six sets of measurements were taken (including the controls at week zero) .
Spectral data was recorded before humidity cycling using two different methods. In the first instance, offcuts (20 cm) from the dried samples were separated into earlywood and latewood sub-samples with hammer and chisel. Then each sub- sample was separately ground and sieved to a fraction size of 50 - 160 μm. The samples were recorded in the mid infra-red and near infra-red regions were acquired on commercial spectrometers.
In the second instance, the dried samples were machined to size to fit into the NIR sample cell (90 x 40 x 10 mm). In the preferred embodiment fibre optic cable systems would be used to either direct light onto and record light reflected from the surface of the wood or detect the light recorded from the surface when using external sources of
radiation or direct light onto the surface of the wood which is reflected and received by detectors housed at or near the sample under test.
The samples were inserted in a NIR sample holder which was inserted in the NIR transport system. This system moves the sample past the source/detector while several scans are recorded, thus yielding an average spectrum across the whole length of the sample (- 100 mm).
All measurements were entered into Excel spreadsheets for data processing and subsequently exported to The Unscrambler™ software for data analysis and prediction modelling. This program has the ability to detect systematic differences between the samples based on spectra or a set of physical data and to summarise them in a few variables called latent variables. The method used to create the prediction models was Projection to Latent Structures regression (PLS). The final number of latent variables selected for incorporation into the prediction model was subsequently chosen by the software as being the principal component that gives the first minimum Y-residual variance when using full cross-validation.
The spectral data from the FTIR were baseline corrected and normalised to remove offsets. The spectral data from the NIR were transformed to their second derivative to remove offsets and sloping baselines prior to data analysis.
To generate the PLS regression model, samples that were determined as behaving in a stable manner were assigned the value 1, while samples that were determined as unstable were assigned the value 0. When using binary descriptors (1 and 0) to describe the stability the consequence is that those samples with a predicted y value greater than 0.5 are predicted as stable while samples with a predicted y value less than 0.5 are predicted as unstable.
FTIR spectra obtained from ground earlywood are able to differentiate between stable and unstable radiata studs using a PLS regression model based on two latent variables with a regression of R2 = 0.79 with no outliers (Figure 4).
Near infra-red spectra of ground samples is also able to identify unstable radiata studs using 7 latent variables, giving a coefficient of determination R2 = 0.86 with no outliers. This is valid for the earlywood subsamples as well as for the latewood
subsamples. The regression used four latent variables and gave a R2 = 0.81 with one outlier (Figure 5).
The important wavelengths for this model are between 490 nm and 590 nm, which is well in the visible range and indicating the differentiation is based in the yellow region of the spectrum. The unstable samples have a significantly darker colour. The regression required three latent variables and gave a coefficient of determination of 0.67 with two outliers.
Using only the visible range (400 - 1 100 nm) gave a differentiation between the stable and unstable samples but with slightly inferior results in the regression (four latent variables, R2 = 0.55, three outliers).
The distinct separation of the samples into stable and unstable classes by a multivariate analysis based on visible and/ or infra-red spectral data shows that there must be a strong chemical difference between this groups. This is of course emphasised by the fact that the ground samples can be even distinguished in a visual inspection.
The fact that solid samples also results in a good prediction (Figure 6) gives this method high potential to be used in a mill before a stabilising treatment is applied.
EXAMPLE 3 - Prediction of Elastic Modulus for 20 x 20 x 300mm Clearwood Test Pieces
486 samples of dry radiata pine (Pinus radiata D. Don) sapwood machined to 20 x 20 x 300 mm (longitudinal direction), were equilibrated at 12% moisture content. All samples were clear and knot-free selected randomly.
Length, breadth and depth, as well as weight were then recorded in order to determine the density and exact moisture content of each sample at test. Static bending tests were performed on a combined Instron/Daytronic test machine with the load being applied to the radial face of the specimen. The load-deflection curve was recorded.
The modulus of elasticity (MoE), was calculated from data taken from the graph using a ruler to draw a line that coincided to the greatest possible extent with the initial curve the major part of which was usually almost straight. The point at which the trace and the line diverge is known as the "Proportional Limit" (Pp). The number of chart squares from the point at which the line crosses the X-axis to a point directly below the Pp is the deflection at proportional limit. The MoE was then calculated as:
Pp . i
MoE(MPa) = (L = span; Δ = deflection; b = breadth; d = depth).
4 - A - b - d
Near infrared spectra were acquired on a commercial NIR spectrometer using a remote reflectance accessory covering both the visible and near infrared range (400- 2,500 nm). A Teflon mount with a ca. 20 mm gap was fitted to the housing to act as a transport guide for the test samples. Average spectra along the entire length of the sample were recorded separately on one tangential and one radial face by moving the samples across the surface of the remote reflectance head with a reciprocating arm. Spectra were acquired in reflectance mode using 32 sample scans and 32 background scans (against an internal ceramic standard).
Table 1 shows the results of acquiring the spectral data on either the tangential or radial face of the wood with the prediction based on spectra recorded on the radial face being superior to that based on spectra recorded on the tangential face. The root mean square error of prediction (RMSEP) is less for the prediction based on the spectra recorded on the radial face. The samples had the load applied to them on the radial face and so there is a more direct correlation with the spectra. Also both earlywood (springwood) and latewood are exposed on the radial face, such that the resulting spectrum is a better representation of the total wood characteristics than a spectrum of the tangential face which in this instance has predominantly earlywood exposed.
Table 1. Comparison of PLS predictions based on NIR spectra acquired on radial and tangential faces (raw data) .
No. latent variables R2 RMSEP used
Radial direction
(moving) 6 of 10 0.71 1356
Tangential direction
(moving) 10 of 10 0.62 1527
The prediction ability for the full range and the separate NIR (1, 100 - 2,500 nm) and visible light (400 - 1, 100 nm) only ranges can be seen in Table 2. All predictions are based on the first derivative of the spectra from the full set of samples. The model based only on the visible range is inferior to the model based on the full range but still shows a potential for prediction. The model based on the NIR range only is slightly superior to the model based on the full range.
Table 2. Comparison of PLS predictions based on visible/NIR, NIR only and visible only spectral ranges.
No. latent R2 RMSEP variables used
Full range 6 0.71 1352
Visible only 8 0.59 1570
NIR only 6 0.72 1298
EXAMPLE 4 - Prediction of Elastic Modulus for Large Scale 100 x 50 mm x 2.4 m Studs
Mixed grade lumber (100 x 50 mm) was selected in a green state from commercial sawmill operations to provide 140 samples of 2.4 m length.
Each 2.4 m board was scanned in the green state using the scanner of the previous example over the full range 400 - 2,500 nm (Visible and NIR ranges). The boards were scanned on the two 100 mm faces and one of the 50 mm edges. Spectra were acquired in reflectance mode using 32 sample scans and 32 background scans (against an internal ceramic standard).
The stiffness of the dried 2.4 m studs was determined by 3-point bending over a 2.0 m span.
Table 3 shows the results of acquiring the spectral data from either the green or dried wood elements. The data was taken from either two faces (for the first line in the table) or on the radial face of the wood element. The spectral data was either used in a raw form or the first derivative of the spectral data was used. The number of latent variables required in the PLS analysis varied between 2 and 5. The calibration regressor ranged between 0.38 and 0.79 and the validation regressor ranged between 0.22 and 0.77. These regressor values range between 0 and 1 with the higher values indicating the greatest correlation between the data and the predicted response. Table 3 shows that the best results were obtained from the raw data scanned on an average of two faces. The analysis of this data required 5 latent variables, had a calibration regressor of 0.79, a validation regressor of 0.77 and a root mean square error of prediction of 1.1.
Table 3. Summary PLS model data for MOE prediction of 2.4 m studs using NIR spectra on green and dry material. n Data No. latent R2 R2 RMSEP
Manip. variables Calib. Valid. reqd
Green, average 71 Raw 1 100-2200 5 0.79 0.77 1.1 of 2 faces
Green, radial 67 Raw 1100-2200 3 0.35 0.23 2. 1
Dry, radial 70 Raw 1100-2200 5 0.40 0.28 2.2
Green, radial 68 lrt deriv. 4 0.76 0.22 2.4 1100-2200
Dry, radial 67 1s deriv. 2 0.38 0.25 2. 1
1 100-2200
EXAMPLE 5 - Prediction of Elastic Modulus for Veneer Sheets
Freshly dried radiata veneers (< 6.5 % moisture content (MC)) were collected from a commercial veneer plant. A total of 140 veneers were batch selected over a 10 week period.
Figure 7 shows the steps used to produce mini laminated veneer lumber (LVL) panels used to prepare mini- LVL joists for reference MoE testing. The 2.4 x 1.2 m veneer sheets 1 were cross-cut to produce two sheets of 1.2 x 1.2 m 2. Each of the half sheets 2 was scanned (fullwidth) each across the tight face (ie across the direction of the original 1.2 m width), shown by arrow 3, using NIR at ca. 1 m s-1 and the average spectrum for the veneer sheet 1 calculated.
For each of the veneers selected, four 6-ply LVL mini-panels 4 were prepared from the 200 x 600 mm sections by using A1-A6 in one panel, B 1-B6 in the next and so on. A 20 mm wide mini- LVL joist similar to a small clear test piece 5 (18 x 20 x 600 mm) was machined from each panel and the stiffness tested as joists over a span of 280 mm with the load applied at 5 mm/min. The mean panel stiffness for each veneer sheet was then calculated for regression with the mean NIR spectra for each veneer sheet.
The results, displayed in Table 4, show potential for assessing the stiffness of whole sheets of veneer based on the stiffness of final LVL product produced. For each of the veneer sheets four mini-LVL panels were prepared and a test joist machined from them. It can be seen that there is good fit between predicted and measured values particularly considering the inherent error in the MOE measurement. The model is significantly improved when the first derivative of the spectra is used. The average NIR spectra and the average stiffness for each sheet were then regressed for the sheets. The results of this show a coefficient of determination of R2 = 0.96 for the calibration model and R2 = 0.74 for the validation model (Table 4).
Table 4. PLS modelling data for veneer sheet stiffness (based on averaged mini-LVL joist stiffness) and average NIR spectra for the veneer sheet.
n Data Optimum number Outliers R2 R2 RMSEP Manip. of latent variables Calib. Valid.
56 raw 9 5 0.77 0.59 1.2
1100-2200 56 1* deriv. 6 5 0.96 0.74 0.9 1100-2200
The data in Tables 1 to 4 indicates it is possible to use visible light and/ or infrared spectroscopy to predict elastic modulus.
The foregoing describes the invention including preferred forms thereof. Alterations and modifications as will be obvious to those skilled in the art are intended to be incorporated within the scope hereof, as defined in the accompanying claims.