Imaging NIR spectroscopy for investigation of wood and applications on wood materials Sven-Olof Lundqvist Thomas Grahn Lars Wallbäcks MeMoWood, Nancy, October 4, 2013 Outline of presentation Technique – Measurements with NIR spectroscopy – Single and multiple spot NIR measurement – Hyperspectral NIR imaging Examples from measurements on wood: o Radial variations from pith to bark, high spatial resolution o Wood discs, medium spacial resolution o Identification of compression wood o Classification of wood of different tree species o Composition of paper Illustrations from the projects Bio4Energy, Trees4Future, a Swedish - South African collaboration on eucalypts, and various pre-studies at Innventia Summary and our next steps Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 2 Lundqvist, Grahn, Wallbäcks NIR spectroscopy - an indirect measurement techique with many applications Radiation of different wavelengths in the near infrared (NIR) range interacts with different chemical bonds. Therefore, NIR absorption spectra may provide information about the chemical composition of materials and related features The method is indirect and models are needed to estimate the contents of chemical compounds or other properties from the NIR spectra The models are developed using multivariate statistical methods applied on sets of spectra and data from analyses with reference methods The spectra can be used also without models to classify different types of objects and to investigate spectral variations NIR spectroscopy is today used for charcterisation of materials and objects in a multitude of applications: On wood and materials from wood, in food, medicin and chemical industries, in agriculture and mining, etc. Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 3 Lundqvist, Grahn, Wallbäcks Single and multiple spot NIR measurement Single position NIR measurement Stepwise single position measurement on translated sample for information about linear variation, i.e. from pith to bark Linear translation Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 4 Lundqvist, Grahn, Wallbäcks Example of stepwise singe spot measurement: Innventia’s NIR Wood Scanner - spectra from pith to bark on “SilviScan strips” Probe Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 5 Lundqvist, Grahn, Wallbäcks Property estimation from NIR Wood Scanner data Example: Content of cellulose Radial sequence of NIR spectra Prediction of cellulose content from pith to bark on “SilviScan strip” MeasuredYVar(Cellulose) cellulose (mg/g) 450 y=1*x-1.702e-005 R2=0.8298 67 23 60 1 69 75 6663 70 64 359 45 4 2 62 88 46 6144 76 47 3335 41 74 87 42 4 3 7 28 82 12 37 848 5152 65 2457 68 2689 4939 20 71 50 53 73 27 81 32 9 787772 56 79 55 16 30 54 2983 25 11 91 40 58 86 6 85 17 3122 5 10 440 430 420 410 400 390 380 19 13 3634 84 370 Norway spruce and Scots pine 14 360 18 15 80 90 38 21 350 350 360 370 380 390 400 410 420 430 440 450 YPred[2](Cellulose) Predicted cellulose (mg/g) SIMCA-P+ 12 - 2008-12-11 14:23:07 (UTC+1) Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 6 Lundqvist, Grahn, Wallbäcks Hyperspectral NIR measurement Single position hyperspectral NIR measurement Array of detector points (320 points array) Pixel 1 Pixel 2 … Pixel 3 Pixel 4 … Pixel 320 Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 7 Lundqvist, Grahn, Wallbäcks Hyperspectral NIR imaging Stepwise hyperspectral NIR measurement on translated sample for information about two-dimesional spatial variations Array of detector points (320 points array) Linear translation Pixel 1 Pixel 2 Pixel 3 Pixel 4 … … … … … … Pixel 320 Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 8 Lundqvist, Grahn, Wallbäcks Hyperspectral NIR imaging: Example: Wood from pith to bark, high resolution Sample of larch Photo * Images of the 3 first principal components of all spectra from the sample high value PC1 PC2 PC3 Predicted variations in lignin concentration with a preliminary PLS model * Sample size: 10 mm x 130 mm Moving sample measured with 100 Hz, Whole sample measured in 42 seconds, low value Spatial resolution: 31 µm x 31 µm resulting in 100 x 320 = 32000 spectra/sec resulting in 1.34 million spectra from the sample * The first principal component is the linear combination of absorptions for all wavelengths which describes most of their variation. The second is the linear combination ortogonal to the first one which describes most of the remaining variation, etc. A PLS model combines a limited number of the principal components to estimate a property. Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 9 Lundqvist, Grahn, Wallbäcks Illustration of spatial resolution pith bark Sample strip of Norway spruce, previously analysed on SilviScan NIR PC 1 Spatial resolution/ /pixel size: 30 µm x 30 µm Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 10 Lundqvist, Grahn, Wallbäcks First large scale application: Characterisation of samples from 6000 spruces trees Samples from 6000 spruce trees will be analysed within the Swedish project Bio4Energy, aiming for genetic investigations and development of genetic markers Our intention is to develop this into a platform for high throughput phenotyping in tree breeding programs Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 11 Lundqvist, Grahn, Wallbäcks Illustration of spectral differencens between and within 3 eucalypt species Eucalypt species A Eucalypt species B Close to top of tree Breast height Variation in the 1st principal component, blue = low value, red = high value Trees from South African stands of 3 eucalypt species were sampled at several heights. Wood strips from pith to bark were imaged with NIR. A Principal Component Analysis was performed on all the millions of spectra from all samples together. Above, spectral differences between and vatiations within two of the trees representing different species are illustrated with maps of the first principal component. Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 12 Lundqvist, Grahn, Wallbäcks Illustration of spectral differencens between and within 3 eucalypt species Maps of variation in the 1st principal component blue = low value, red = high value Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 13 Lundqvist, Grahn, Wallbäcks Illustration: NIR imaging of wood discs Wood used for illustration: Scots pine, fresh Infected on one side, bark lost, deterioration Dot of syntetic chain oil applied Heartwood Sapwood Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 14 Lundqvist, Grahn, Wallbäcks Images of Principal Components high value low value PC 1 PC 2 Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 15 PC 3 Lundqvist, Grahn, Wallbäcks Spatial resolution / pixel size: 1 mm x 1 mm PC 3 Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 16 Lundqvist, Grahn, Wallbäcks Example of tasks worked upon: Identification of compressionwood Norway spruce, preliminary results Photo NIR PC1 The aim is to develop a model for classification of compression wood Work in cooperation with Urszula Zajaczkowska, University of Life Sciences, Warsaw, within the project Trees4Future Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 17 Lundqvist, Grahn, Wallbäcks Example of task worked upon: Classification of tree species Samples NIR based classification Eucalypt Spruce Birch Pine Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 18 Lundqvist, Grahn, Wallbäcks Content of lignin in wood A large number of containers with wood meal were scanned automatically within the EU project Trees4Future. Average NIR spectra were calculated for each sample. Reference data for ligning content were available from previous analyses by Manfred Schwanninger, BOKU. A model for estimation of lignin content with R2 value 0,89 was developed. Four of the containers and NIR spectra Measured vs predicted lignin content 34 0,50 Analysed Klason Lignin, % 0,45 0,40 0,35 0,30 0,25 0,20 0,15 0,10 R2: 0.89 RMSEP: 0.38 32 30 28 26 Validation set Calibration set 24 0,05 Linear ( ) 0,00 22 800 1 300 1 800 2 300 Wavelength, nm 2 800 22 Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 19 24 26 28 30 Predicited Klason Lignin, % Lundqvist, Grahn, Wallbäcks 32 34 Example of task worked upon: Variations in composition of paper Interpretation from principal components: Variations in gramage (g/m2) of Fibres Fillers Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 20 Starch Lundqvist, Grahn, Wallbäcks Summary and our next steps Hyperspectral imaging offers many posibilities beyond conventional spot measurements with NIR It has many potential applications on wood and materials from wood Innventia has recently installed a new instrument for imaging NIR, specially designed for large flexibility to work in different scales and level of detail, as well as in different set-ups The instrument is now used in research on various types of materials We will also use it extensively as a development platform for new measurement methods and applications Imaging NIR spectroscopy on wood MeMoWood, Nancy, 4 Oct 2013 www.innventia.com © 2013 21 Lundqvist, Grahn, Wallbäcks
© Copyright 2025 Paperzz