This report documents the methodology, results and conclusions of the one year, 2009/2010 project “Vis/NIR Scanning for Hem-Fir Species Separation at the Lumber Stage” by FPinnovations, Forintek Division and funded by Forest Innovation Investment (FII).
The aim was to undertake the technical research required to implement new Visible/Near Infrared spectrometry technology in sawmills to enable hem-fir producers to separate Western hemlock and Pacific Silver fir lumber. Separating hem-fir has several potential opportunities for value-added, specialty products and improved commodity processing efficiency. These include enhanced treatability, greater uniformity in finishing of appearance grade products, and enhanced lumber drying efficiency. Milota et al estimated that pre-sorting hem-fir could save the hem-fir lumber industry US $46M annually by reducing warp and splits.
Vis/NIR spectroscopy has been successfully implemented as an on-line, over-the-conveyor, process analyzer for various widespread applications from agricultural applications such as wheat and grain, to food processing, mining, and pharmaceutical applications. FPInnovations - Paprican have also demonstrated the use of Vis/NIR for the measurements of wood chip properties including hem-fir species proportions as well as for measurements of pulp fibre brightness and kappa (residual lignin) number.
Western hemlock and Pacific Silver fir cannot be reliably distinguished at the lumber stage. The anatomical differences are subtle and can only be reliably identified by a trained wood anatomist using microscopic analysis. This is a problem for producers that want to sort species to capitalize on their unique properties. Manual species sorting (by mill graders) has been shown to be less than 60% accurate, making it necessary to develop a rapid, on-line or online technique for species separation.
The experimental plan involved three main steps, including:
1. Setup of the scanning system hardware and software to distinguish between species for rough, planed, green and kiln-dried lumber conditions. This included collecting spectra from each group as training sets for developing calibration models. Lumber was collected from five coastal mills.
2. Laboratory and pilot plant testing to determine the sorting accuracy for both static, contact measurements and non-contact, on-line conveyor scanning at 150 ft/min.
3. Mill demonstration of the technology.
For both the offline (static) and online tests, measurements were made along the top centre of 8-foot sample boards. Static measurements were performed by using the contact probe fitted with a hand grip to manually scan 1x1-cm lumber surface areas along the top centre length of the board. For online testing, custom software was developed for automatically displaying and tabulating the species ID results from each board scanned. The software also incorporated the means for identifying and disqualifying knots and other lumber defects based on the Mahalanobis Distance (M-Dist) which quantifies the measurement’s “goodness of fit” to the spectrometer calibration model. The online setup also used a board detection sensor for triggering the scanning sequence.
The results clearly showed that Vis/NIR spectrometry can discriminate between hemlock and Pacific Silver fir with = 90% prediction accuracy for all sample lumber groups tested in this study with the exception of green/rough lumber which averaged 80%. The accuracy results of both green/planed and green/rough lumber were consistently lower, by approximately 10%, compared to their kiln-dried counterparts and this can be attributed to the spectral interference caused by strong absorption from water in the NIR region. Prediction accuracy was also consistently better, by approximately 10%, with planed versus rough lumber. This can be explained by the increased light scattering from the unevenness of rough sawn surfaces.
Online tests conducted at Forintek simulated mill conditions and the results showed that prediction accuracy was diminished by 7-8% compared to static, offline measurements, yet still yielded measurements with greater than the 90% prediction accuracy and successfully demonstrated the practicality of implementing Vis/NIR spectrometry for automated lumber sorting application for hemlock and Pacific Silver fir.
A mill demonstration carried-out at Western Forest Product’s Mid-Island Reman mill in Chemainus, B.C. was a success with a correct prediction accuracy of 97% as validated by anatomical species identification. The offline spectrometer system, while slower and not fully automated provided the highest accuracy readings in this study and proved that it can be a valuable tool for fast and accurate hem-fir species identification. The equipment for the offline scanning unit is available “off the shelf” and can be readily loaded with a hem-fir calibration model from FPInnovations for immediate use.