nir spectroscopy for the rapid measurement of wood …nir spectroscopy for the rapid measurement of...
TRANSCRIPT
NIR spectroscopy for the rapid measurement
of wood properties
L. Schimleck, C.-L. So, D. Jones, L. Groom
R. Daniels, A. Clark, G. Peter and T. Shupe
Overview
• Description of NIR spectroscopy
• Application of NIR to whole-trees
• Application of NIR to lumber
• Application of NIR to short clears
• Application of NIR to cores
NIR spectrum of a typical wood sample
1100 1300 1500 1700 1900 2100 2300 2500
Log
(1/R
efle
ctan
ce)
0
0.1
0.2
0.3
0.4
0.5
0.6
Wavelength (nm)
Bands arise from rotations, vibrations of bonds in molecules
O-HN-H
S-H
N-H
O-H N-H
C-H vibration throughout spectrum
Methodology
• Estimation of a parameter involves the following steps:
• Collect spectra of calibration samples
• Develop a calibration (regression)(y = B0 + X1*B1 + X2*B2 + ………..+ XN*BN)
• Collect NIR spectra of test (or unknown) samples
• Estimate parameter of interest for test set samples using the calibration
Partial least squares regression
CALIBRATION MODEL
Measured CuO retention (pcf)-0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
Pred
icte
d C
uO re
tent
ion
(pcf
)
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Calibration SetValidation Set
Wavelength (nm)1000 1200 1400 1600 1800 2000 2200 2400
Rel
ativ
e In
tens
ity
-300
-200
-100
0
100
200
300
400
MOE
REGRESSION COEFF.
DECOMPOSEMATRICES
TEST SET
Wavelength (nm)
Abs
orba
nce
PREDICTED PROPERTIES
MOEMOR MFASG
MC
n x pmatrix
CALIBRATION SET
Wavelength (nm)
Abs
orba
nce
nsa
mpl
es
p variables
n x qmatrix
MEASURED PROPERTIES
MOEMOR MFASG
MC
nsam
ples
q variables
Laboratory determined pulp yield (%)
NIR
fitte
d pu
lp y
ield
(%)
45
50
55
60
45 50 55 60
n = 1334 factorsR2 = 0.88SEC = 0.84
Pulp yield range= 45.6 to 57.1 %
Pulp yield calibration for Tasmania
Within-tree property variation
MicrofibrilAngle
Within-tree property variation
Extractives Lignin
Hemicellulose Cellulose
Application of NIR to whole-trees
• Many studies, since late 1980’s
• Pulp yield, cellulose, lignin, extractives
• Based on whole-tree composite chips
• Examination of within-tree variation of PY
• Studies have shown that breast height cores provide similar calibration statistics to composite chips for whole-tree properties
• = nondestructive estimation of whole-tree properties
Within-tree variation of pulp yield
• Little is known about the within-tree variation of
pulp yield. NIR predictions of pulp yield can be
used to obtain maps that show the variation
47 50 53 56 59 62
Whole-tree chip versus core calibrations
•Core and whole-tree calibrations were similar for basic density, pentosans, specific cons and total lignin
• Core calibrations could be used to rank trees
• 1.30 m identified as the most suitable sampling height
0
0.2
0.4
0.6
0.8
1.0
BasicDensity
SodaCharge
Pulpyield
Totallignin
SpecificCons.
Pento-sans
R2
Wood property calibrations Whole-tree
0.65 m core
1.30 m core
Schimleck et al. (2005). Estimation of whole-tree wood quality traits using near infrared spectra collected from increment cores. Appita J. (in press)
Application of NIR to lumber
• Meder et al. (2003)
• 185 P. radiata cant centers scanned by NIR in mill scale trial
• Aim to ID corewood stiff enough to be graded as MGP 8 (lowest structural grade)
• MGP 8 worth $80/m3 more than non-structural
• Data from 409 boards available for regression
• Calibration R2 = 0.54 (big logs)
• Calibration R2 = 0.57 (small logs)
• Sufficient for economic segregation of cants
Bruker Matrix-F scanning a cant
Scanning speed approx. 2 m/s
Picture courtesy A. Thumm and R. Meder, Forest Research, New Zealand
Application of NIR to lumber
• Meder et al. (2003) cont.
• Based on stiffness, 50% of central boards could be upgraded to MGP 8
• Further calibration expected to increase percentage of upgraded boards
• Many upgraded boards were unstable
• Both stiffness and stability (twist) must be predicted for NIR to be useful for segregating radiata pine structural timber
• Calibration for twist investigated (R2 = 0.26)
Application of NIR to short-clears
• Several studies reported, different species and approaches
• Hoffmeyer and Pedersen (1995) P. abies
• Gindl et al. (2001) L. decidua
• Thumm and Meder (2001) P. radiata
• Schimleck et al. (2001) E. delegatensis, Schimleck et al. (2001) P. radiata
• Via et al. (2003) P. palustris
• Kelley et al. (2004) 6 softwood species
• Density, MOE and MOR examined
Measuring Mechanical Properties
IncrementCore
(SilviScan)
BendingSpecimen(Instron)
NIRNIR
Stiffness (Bending Specimens)
Calibration (313 spectra)
Calibration (313 spectra)
Prediction (156 spectra)
Prediction (156 spectra)
NIR
-MO
EN
IR-M
OE
Instron-MOEInstron-MOE
0 1e+6 2e+6 3e+6 4e+60
1e+6
2e+6
3e+6
4e+6
0 1e+6 2e+6 3e+6 4e+6 5e+60
1e+6
2e+6
3e+6
4e+6
5e+6
8 factorsR2=0.82SEC=0.4 Mpsi
R2=0.79SEP=0.4 Mpsi
Strength (Bending Specimens)N
IR-M
OR
NIR
-MO
R
0 10000 20000 30000 400000
10000
20000
30000
40000
0 10000 20000 30000 400000
10000
20000
30000
40000
Instron-MORInstron-MOR
8 factorsR2=0.80SEC=4 Kpsi
R2=0.75SEP=4 Kpsi
Calibration (313 spectra)
Calibration (313 spectra)
Prediction (156 spectra)
Prediction (156 spectra)
Application of NIR to cores
Radial strips cut from cores used but core surface OK for
NIR spectroscopy
Properties examined :
Tracheid length (FQA)
Cellulose, sugars, lignin (wet chemistry)
Air-dry density, MFA, stiffness, tracheid properties (SilviScan)
Georgia-wide calibrations
• P. taeda grown in 3 regions in Georgia
• Three sites selected as being
representative of each region
• Selection based on site index
• Ten trees selected per site representing a
range of breast height diameters
• Pith-bark breast height samples obtained
Georgia-wide calibrations
• Strong calibrations for density, MFA, stiffness and tracheid coarseness, length and wall thickness
• These calibrations performed well on the separate test set
• Strong calibrations for cellulose, lignin, glucan, arabinan, mannan and xylan
• Moderate prediction accuracy - possibly due to the small number of samples
MFA – 729 MSC treated spectra
5
15
25
35
45
5 15 25 35 45
Fitted MFA (degrees)
Mea
sure
d M
FA (d
egre
es)
Factors = 8R2 = 0.90SEC = 2.33SECV = 2.38RPD = 3.11
Predicted MFA (225 spectra)
0
5
10
15
20
25
30
35
40
45
50
0 10 20 30 40 50
Predicted MFA (degrees)
Mea
sure
d M
FA (d
egre
es)
Factors=8R2= 0.84SEP= 3.12RPD= 2.34
What resolution??
• Increasing resolution = decrease in
calibration accuracy
• Management of spectra becomes
difficult owing to large number of
spectra
• Fiber optic probes with a small spot
size provides options
2 mm MFA calibration (4156 spectra)
8 FactorsR2 = 0.75
SEC = 4.05SECV = 4.10RPD = 1.48
0
10
20
30
40
50
60
0 10 20 30 40 50 60
Fitted MFA (degrees)
Mea
sure
d M
FA (d
egre
es)
2 mm MFA prediction
0
5
10
15
20
25
30
35
0 10 20 30 40 50 60 70 80 90
Distance from pith (mm)
MFA
(deg
)
2 mm MFA prediction
05
1015202530354045
0 10 20 30 40 50 60 70 80 90
Distance from pith (mm)
MFA
(deg
)
Wood property calibrations – green versus dry wood
0.10
0.15
0.20
0.25
0.30
0.35
1100 1300 1500 1700 1900 2100 2300 2500
T - greenRL - green T - dryRL - dry
Wavelength (nm)Wavelength (nm)
Log
(1/R
)Lo
g (1
/R)
Stiffness (green wood spectra)
0
5
10
15
20
25
30
0 5 10 15 20 25 300
5
10
15
20
25
30
0 5 10 15 20 25 30
CalibrationCalibration PredictionPrediction
SS2-
stiff
ness
SS2-
stiff
ness
NIR-stiffnessNIR-stiffness
5 factorsR2=0.88
SEC=1.8 GPaR2=0.81
SEP=3.0 GPa
Stiffness (dry wood spectra)
0
5
10
15
20
25
30
0 5 10 15 20 25 300
5
10
15
20
25
30
0 5 10 15 20 25 30
CalibrationCalibration PredictionPrediction
SS2-
stiff
ness
SS2-
stiff
ness
NIR-stiffnessNIR-stiffness
6 factorsR2=0.95
SEC=1.1 GPaR2=0.92
SEP=1.9 GPa
NIRVANA – Near Infrared Visual & Automated Numerical Analysis
• Automated spectra collection
• High resolution video camera
• Real time property predictions
• Ideal for process monitoring &
QC applications
MOE VariationM
OE
(GPa
)M
OE
(GPa
)
Core Length (mm)Core Length (mm)
0.E+00
1.E+06
2.E+06
3.E+06
4.E+06
5.E+06
6.E+06
7.E+06
8.E+06
9.E+06
-250 -200 -150 -100 -50 0 50 100 150 200 250
Conclusions – NIR spectroscopy
• Determination of within-tree variation
• Estimation of whole-tree properties
• Applicable to milled wood, short-clears, increment cores
• Automatic scanning of cores possible
• Important to improve resolution
• Green wood can be examined