Improved OSB Formation Quality Through Production Process Control Using 3D Image Analysis and Laser

1
Improved OSB Formation Quality Through Production
Process Control Using 3D Image Analysis and Laser
Profilometry
Hu (2005) applied laser imaging to measure wood
surface roughness. Among the advantages of the
use of this technique is that no contact is
necessary, it is fast and accurate. Chuanshuang
(2005) applied laser imaging to measure wood
surface roughness. Recent application of laser to
measure fiber angle in industrial environment
have been developed (Chen, 2005). Advantages
No contact necessary, fast, accurate. EXPERIMENTAL
DESIGN Strand Geometry Distribution Study
material consists of Red Maple, Aspen and Fir
strands 0.045 in thickness. Strander
knife/counterknife angles were adjusted
separately in the three species to attain a
target strand width of ½ inch. Angles ranged from
70 to 85 degrees. Disposable knives were used.
The target strand length was established at 6
inches. One pound of randomly selected strands
were sampled and measured following initial
stranding, drying (Koch conveyor drier) and
screening (Trillium roller screen) operations.
Digital images (1296x1016 pixels) of all the
specimens were acquired with a uniform
illumination of 1.2x105 Candelas/m2 at a constant
scale of 0.01 inches/pixel. Dimensions of the
individual strands were measured using a
developed MatLab code for digital image analysis
(DIA). A sub-sample was measured using a caliper
as a control. The following measures were taken
average width (W) average length (L) area (A),
defined as the sum of pixels in binary image
perimeter (P) or pixels in the borderline and
convex hull area (CHA). From the measurements
the following parameters were computed shape
regularity (S), defined as the proportion of the
pixels in the convex hull that are also in the
region, computed as A/CHA eccentricity (E)
defined as the ratio of distance between the foci
of the ellipse and major axis length slenderness
(SL) defined as L/W ratio rectangularity (R)
defined as the proportion of pixels out of the
binary region and Irregularity (I) defined as
the ratio of actual strand perimeter and
equivalent ideal rectangle perimeter in pixels.
Mean, Variance, and Normality tests were
conducted in order to analyze the significance of
changes in the distribution of the strand
geometry through the production process. Strand
Orientation Controlled orientation OSB boards
will be digitally imaged during their formation
creating a matrix of five images to be processed
using a Matlab DIA code. Strands geometry will be
measured using previous methodology and a laser
profile will be obtained form top and bottom
layers of the boards. Measured Parameters will be
Single Strand Geometry x, y Orientation Angle z
axis Orientation Angle and Strands deformation.
INTRODUCTION The anisotropic character of OSB is
directly correlated with the strand alignment,
which has been extensively studied (Shaler, 1991
Barnes, 2002). The role of strand overlap and
out-of-plane strand orientation (waviness) are
not as well defined. Further, the influence of
variable strand geometries on alignment and
packing are known to be important (Dai, 1994),
but their relationship to optimal process
conditions has not been well quantified. Image
analysis have been applied to surface strand 2D
orientation ignoring three dimensional
orientation of flakes and other important
geometrical properties of strands (Wang, 2000).
OBJECTIVES 1. Study the relationship between
single strand geometry distribution and 3D
orientation angle distribution using image
analysis and laser profilometry 2. Study the
relationship between a change in geometrical
distribution in the production line and
orientation angle distribution as a predictor of
mechanical properties of OSB. 3. Develop a tool
for online formation quality control using image
processing and laser profilometry in an
industrial environment. BACKGROUND Several
authors have addressed the effect of strand
alignment on the mechanical properties of OSB
boards (Barnes, 1988 Shaler, 1991). Geometrical
and environmental variables affecting mat
formation had been revealed in simulation as well
as in actual studies (Wang, 2000). Improvements
in resolution acquisition and processing time
have made more and more feasible the use of
digital imaging in the mill (Barnes, 1988).
Barnes (1998) claimed a patent on a measuring
system for the bi-dimensional alignment of
strands based on digital image analysis (DIA).
Sladoje et. al. (2003) developed a methodology
to measure perimeters and area in low resolution
images. In mills with high speed processing,
low-res images are more likely to be obtained
than clear lab photos. El-Sombaty and M.A. Ismail
(2003) studied the matching of objects partially
occluded. The algorithm allows to match lines in
the object in order to recognize its shape even
if it is not completely visible.
Group of strands binary image and the isolated
corresponding measured strand
5-layer image matrix used to study 3D orientation
angle
Measurements on 5-layer pictures
Diagram of image acquisition set up
PRELIMINARY RESULTS Width, area, convex area,
shape regularity and slenderness exhibit high
variability. Rectangularity, irregularity and
eccentricity appear less variant for all the
species. Statistical analysis indicated that
distribution of all geometric parameters was
non-normal. Strand geometry has a statistically
significant influence on distribution changes
after processing of strands (drying and
screening). High correlation was found between
mean rectangularity and mean width. The initial
distribution of strand geometry affects
length/width ratio distribution. PRELIMINARY
CONCLUSIONS Image Analysis can be applied to the
study of strand orientation and geometry studies
for automated acquisition of data.
Non-parametric statistics must be applied to
study the geometrical distribution of strands.
The extraction point for the sample is
statistically significant in the distribution of
the geometrical variables. Rectangularity and
Irregularity have high influence on change of
width and length distributions. FUTURE
WORK Study Strand Geometrical Distribution using
non parametric statistics. Laser Profilometry.
Set up and experimental design. Controlled OSB
images analysis. Application to Industrial
Environment. Design of experiments, application
procedures and equipment. ACKNOWLEDGEMENTS USDA/
CSREES Special Grant Award F 2003-34158-14006
Binary image of actual strand and the
corresponding Convex Hull Image
PRELIMINARY RESULTS Influence of Species on
strand geometry Distribution
analysis Correlated variables
Actual Profilometer
STRANDER SCREENER
CONVEYOR DRYER
BLENDER FORMATION
LINE
Pilot plant used in this research. AEWC Center
Facility. University of Maine at Orono

References Barnes, Derek. A model of the
effect of strand length and strand thickness on
the strength properties of oriented wood
composites. Forest Products Journal 51(2)
36-46 Barnes Derek, John E. Ens., United States
Patent 5764788, 6-9-1988. Research Plan. Chen,
Siguo. http//www.arc.ab.ca/Index.aspx/ARC/4367,
2005-2006 Alberta Research Council. Dai,
Chunping and Paul R. Stainer, Spatial structure
of wood composites in relation to processing and
performance characteristics Part 3. Modeling the
formation of multi-layered random flake mats,
Wood Science and Technology, Volume 28, Number 3,
March 1994, pp 229 - 239 El-Sombaty Yasser,
and M.A. Ismail. Matching Occluded Objects
Invariant to Rotations, Translations,
Reflections, and Scale Changes. In Proceedings of
the 13th Scandinavian Conference on Image
Analysis, SCIA 2003, July 2003. pp 836-843. Hu,
C. and M. Afzal, Automatic measurement of wood
surface roughness by laser imaging. Forest
Product Journal, Vol. 55 No. 12, 2005. pp
158-163. Sladoje Natasia, Ingela Nystrom, and
Punam K. Saha. Measuring Perimeter and Area in
Low Resolution Images Using Fuzzy Approach. In
Proceedings of the 13th Scandinavian Conference
on Image Analysis, SCIA 2003, July 2003. pp
853-860. Shaler, S.M. Comparing two measures of
flake alignment. Wood Science and Technology.
1991. Vol. 26(1) 53-61. Wang, Huaijun.
Creating Virtual Wood Particulate Composites,
Doctor of Philosophy Thesis (in Forest
Resources). The University of Maine, December,
2000. Wang, Siqum and Paul M. Winistorfer.
Fundamentals of vertical density profile
formation in wood composites. Part II Methodology
of vertical density formation under dynamic
conditions. Wood and Fiber Science. 32(2), 2000
pp. 220-238.