Post on 23-Mar-2018
DAVID GRANT
EFFETS DE LA DISTRIBUTION DE L'ORIENTATION DES PARTICULES DE
COUCHES EXTÉRIEURES SUR LES PROPRIÉTÉS ~MÉCANIQUES
DES PANNEAUX DE PARTICULES ORIENTÉES
Mémoire
présenté
à la Faculté des études supérieures
de IZTniversité Laval
pour Pobtention
du grade de maître ès sciences (MSc.)
Département des sciences du bois et de la forêt
FACULTÉ DE FORESTERIE ET GÉOMATIQUE
UNNERSITÉ LAVAL
OCTOBRE 1997
Q David Grant, 1997
National Libmry Bibliothèque nationale du Canada
Acquisitions and Acquisitions et Bibliographie Services services bibliographiques
395 Wellington Street 395, rue Wellington Ottawa ON K1A ON4 Ottawa ON K1A O N 4 Canada Canada
The author has granted a non- L'auteur a accordé une licence non exclusive licence allowing the exclusive permettant à la National Library of Canada to Bibliothèque nationale du Canada de reproduce, loan, distribute or sell reproduire, prêter, distriiuer ou copies of this thesis in microtom, vendre des copies de cette thèse sous paper or electronic formats. la forme de microfiche/fdm, de
reproduction sur papier ou sur format électronique.
The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thése. thesis nor substântial extracts f?om it Ni la thèse ni des extraits substantiels may be printed or othefwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation-
This study demonstrates the effects of the d a c e layer strand alignment distriibution on the
mechanical properties of orîented strand board. Oriented strand boards of a 23/32-inch (18
milluneters) thichess and 38 pounds per cubic foot (610 kilograms per cubic meter) density
were produced with Mace layers composed of three strata. Surface strata were differentiated
by strand size, strand orientation and position within the layer. Mathematical models were built
to desmie the relationship betweai the orientations of the individual d a c e layer -ta and the
unidirectional moduli of rupture and elasticity of the panels.
The models affinned the well-documented positive influence of strand alignment on the bending
moduli. However, the d t s also suggested that there was no marginal retum in regards to the
mechanical properties fkom improvements in strand orientation above a certain threshold.
Furthemore, the constituents' contributions to strength and stBhess was found to diminish fkom
the outer surface stratum inward towards the core. Board density was observed to have a
positive influence on both the modulus of elasticiv and modulus of rupture, even within the
small range in variability measured.
Cette recherche démontre les effets de la distribution de I'orïentation des particdes de couches
extérieures sur les propriétés mécaniques des panneaux dz particules orientées. Des panneaux
de particules orientées d'une épaisseur de 23/32 de pouce (18 minmiètres) et d'une densité de 38
Iivredpieds cube (610 kilo&rammeslmè&es cube) ont été produits. Chacune des surfaces de ces
panneaux renfermait trois couches variant selon la taille et l'orientation des particules de même
que leur position à l'intérieur de la sinface. La conception des modèles mathématiques a
démontré une relation entre I'orientation des différentes couches des surfaces extérieures et Ies
modules de rupture et d'élasticité unidirectionnels.
Les modèles ont confirmé les principes déjà cornus concernant les effets positif!! de l'orientation
des particules sur les modules de flexion, cependant, les résultats expérimentaux indiquent que
I'accroissement de l'orientation des couches au-delà du seuil déterminant n'amène aucun bénéfice
marginal. De plus, l'analyse des rédtats démontre qu'une diminution des effets des matériaux
se concrétise à partir de la d a c e jusqu'au centre des panneaux. Finalement, la densité des
panneaux révèle aussi des effets positifs sur les modules de flexion même si i'étendue des
mesures de variabilité était restreinte.
David Grant
Étudiant
Michel Beaudoin
Directeur
1 would Like to acknowledge the contribution of a number of individuals and organizations
without whose assistance this thesis would not have been possible. 1 would like to thank my
thesis director, Michel Beaudoin, for his generosity and guidance thoughout this rwearch, as well
as his latitude towards rny autonomy. Thanks, aiso, go to Bernard Riedl who made himself
available for any and alI concerns regarding the logistics of the graduate program.
1 would like to thank Forintek Canada Corporation and particularly, Jack Shields, manager of
the composites department, for providing the facilities and materials necessary to undertake this
work. Thanks go to Emest Hsu for his input on composite theory and for assisting in the
development of the experimental design. 1 would especidly IÎke to thank François Grondin for
his assistance with the considerable mathematical content of the research, as well as for his
diligent editing of my thesis dr&. 1 thank Dan Lachance, Luiz Couto, Louis Grave1 and
Frazlcine Côté for their assistance in the material prepmtion, panel production and testing phases
of the work,
I tender my most heart-felt thanks to my father, Peter Grant Sr, for his constant encouragement
and support, and for providùig me with the background and expenence in industy, without
which, this type of applied research would not be possible.
Finally, this work wouId never have been possible without the precious support of my fiancée,
Julie Pouliot Without her patience, encouragement and understanding this work wodd have
long ago been abandoned.
TABLE OF CONTENTS
Page
... PREFACE ............................................................... iii
TAlBLE OF CONTENTS .................................................... iv
........................................................ LISTOFTABLES vii
... LISTOFFIGURES ....................................................... viii
LIST OF SYMBOLS AND ABBREVIATIONS ................................. xi
CHAPTERI INTRODUCTION ............................................... 1
1.1 Effects of over-capacity and the business cycle decline . . . . . . .-. . . . . . . . . . . . . . 2
1.2 Effects of diminished wood supply ..................................... . 4
.................................................. 1.3 Probledchallenge . 6
1.4 Objectives ......................................................... 6
2.1 Enduseapplications ............................................... 10
2.1.1 Stresses ..................................................... 12
2.1.2 Reactions .................................................... 13
2.1.3 Deformation ................................................. 13
2.2 Product design ..................................................... 17
2.3 Production parameters ............................................... 22
2.3.1 Electrostatic alignment ........................................ - 2 3
2.3.2 Mechanicd alignment ......................................... - 2 9
2.3 .2.1 Oscillating-fiame alignment device ........................ 29
2.3.2.2 Rotary disk fomiing machines ............................ 33
2.3.2.3 Vane(chamber) roll alignmentmachines .................... 38
2.4 Methods for measuring and chacterizing strand alignment ................. 41
2.4.1 Direct surface measmement ..................................... 41
2.4.2 Mechanicd property (MOE, MOR) ratios .......................... 48
2.4.3 Stress wave velocityratio ....................................... 48
2.4.4 Sonic velociq ratio ............................................ 48
2.4.5 Electncal capacitance ......................................... -49
2.4.6 Microwave attenuation ........................................ - 5 0
CHAPTER III MATERIALS AND METHODS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - 5 1
3.1 Strand alignment measurement ........................................ 51
3.2 Orientation parameter .............................................. -59
3.3 Strand alignment prediction algorithm .................................. 61
3 .3.1 Experimental design ........................................... 61
3.3.2 Strand generation ............................................ - 6 2
3.3 -3 Production of oriented mats .................................... - 6 3
3.3 -4 Strand alignment mode1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -64
3.4 Experimental panels ................................................ 65
3 .4.1 Experimental design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -67
3 .4.2 Strand generation and drymg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4.3 Adhesive and wax blending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4.4 Foxming oriented strand board mats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -71
3.4.5 Hot pressing ................................................ - 7 2
3 .4.6 Panel evdution ............................................. -76
3.4.7 Response s d a c e methodology .................................. -76
CHAPTER rV RESULTS AND DISCUSSION .................................. 79
4.1 Strandorientation .................................................. 79
4.1.1 Equality of concentration parameters .............................. 79
4.1.2 Homogeneity o f concentration parameters ......................... - 8 5
4.2 Boardtests ........................................................ 90
4.3 Experimental designIrnode1 inputs ..................................... 92
4.4 Modulus of rupture (MOR) mode1 .................................... -95 4.4.1 Mode1 seleetion .............................................. - 9 5
4.4.1.1 SequentiaI mode1 sum of squares .......................... 95
4.4.1.2 Lack of fit ............................................ 96
..................................... 4.4.1.3 Summarystatistics -97
4.4.2 Model diagnostics ............................................ - 9 9
4.4.3 Modelequation .............................................. 102
4.4.4 Significance of mode1 factors ................................... 106
4.5 Modulus of elasticity (MOE) mode1 ................................... 108
4.5.1 Mode1 selection .............................................. 108
4.5.1.1 Sequential mode1 su m of squares ......................... 108
4.5.1.2 Lack of fit ........................................... 109
..................................... 4.5.1.3 Sllmmary statistics 109
............................................ 4.5.2 Modeldiagnostics 1 1 1
........................................... 4.5.3 Model optimization 114
.............................................. 4.5.4 Modelequation 118
................................... 4.5.5 Sipificance of mode1 factors 122
............................................... 4.6 Costhenefit analysis 124
CONCLUSIONS
LIST OF TABLES
Page
Table 1 . Box-Behnken design for study of çtrand aiignment effects in OSB ........... 69
. Table 2 Flaker machine specificaîions and sertings .............................. 71
Table 3 . Target and actual orientation (concentration parameter) of experimental
OSBmats ........................................................ 80
Table 4 . Two-sample tests for equality of concentration parameters (K) .............. 84
Table 5 . Calculations for testing homogeneity of K with R < 0.45 .................. 87
Table 6 . CalcuIations for testing homogeneity of K with 0.45 s R r 0.70 ............ 88
Table 7 . Cdculations for testing homogeneity of K with R > 0.70 .................. 88
Table 8 . Test results for panels (18 mm) with variable strand alignment .............. 91
Table 9 . Factor limits used in modehg efforts .................................. 93
Table 10 . Revised Box-Behnken design for study of sirand alignment effects in OSB .... 94
Table 11 . Sequential mode1 sum of squares for the MOR models ................... - 9 6
Table 12 . Lack of fit tests for the MOR modeis ................................. - 9 7
Table 13 . ANOVA sufnmary statistics of the MOR models . . . . . . . . . . . . . . . . . . . . . . . . . 97
Table 14 . Test of the signincance of the MOR mode1 factor coefficients . . . . . . . . . . . . . 106
Table 15 . Sequential mode1 stun of squares for the MOE models ................... 108
. Table 16 Lack of fit tests for the MOE models ................................. 109
Table 17 . ANOVA summary statistics of the MOE models ........................ 110
Table 18- Cornparison of ~tunmary statistics for original and optimized MOE models . . . 1 14
Table 19 . Test of the signincance of the MOE mode1 factor coefficients .............. 123
Table 20 . Benefit analysis for a reduction in board density ........................ 126
LIST OF FIGURES
Figure 1 . The developrnent cycle of oriented strand board ......................... -9
Figure 2 . SimpIy supported beam with unifonnly distributed load .................. 10
Figure 3 . Typical floor systern with Iumber joists, OSB subfl oor sheathing. and
ahardwoodfloorsurface ........................................... I I
Figure 4 . Deformation (bending) of OSB sheathing under vertical loading . . . . . . . . . . . . 14
. Figure 5 Nomai failme mode of OSB when loaded in bending .................... 15
. Figure 6 Bending moment and force couple endured by the OSB sheathing .......... 15
Figure 7 . Three-dimensional Cartesian axis system .............................. 17
Figure 8 . Axial system as dehed by wood growth structure ....................... 18
Fi- 9 . Geometry of a typical OSB strand with associated grain orientation . . . . . . . . . 20
Figure 10 . Surfie view of a typicd OSB panel .................................. 21
Figure I l . Principles of electrostatic digrment . single cell ........................ 24
Figure 12 . Schematic of a single celi electrostatic alignment device with controlled
. Figure 13 Schematic of a typicd osciIiating-fÏame alignment device ................. 30
. Figure 14 Staggered plate configuration of an oscillating-fiame alignment device ...... - 3 2
Figure 15 . Schematic of a rotary disk forming station ............................. 33
. Figure 16 Disk assembly of the Schenck surface layer foxming head ................ -35
Figure 17 . Distribution of furnsh by strand sîze through the bottom surface layer
of an OSB panel ................................................. - 3 6
Figure 18 . Onenting action of a cross-aligning vane roll ........................... 38
Figure 19 . ALignment groove and flake geometry influence on the effectiveness
........................................... ofthevane.typeorienters 39
Figure 20 . Relationship of the von Mises concentration parameter (K) with the
mean orientation vector (R) ........................................ -46 Figure 2 1 . Schematic of the image analysis system used in the measurement of
.................................................. mdal ignmen t 53
. ........ Figure 22 Image of an onented strand mat within the IA operating environment 55
. Figure 23 11 x 11 grid superimposed on mat image for strand sample selection ....... - 5 6
. Figure 24 Edge delineation of selected strands ................................. -57
. Figure 25 lsolated strand "edges" representing wood grain orientation ................ 58
.......... . Figure 26 Angle measmement of strand orientation to the cardinal direction - 5 9
Figure 27 . Laboratory forming apparatus used for orienting strands in an OSB
............................................................ mat 63
Figure 28 . Weighted distribution of strands by width approxirnates the noma1
distribution ...................................................... 66 . Figure 29 Press profile used in the production of onented Strand boards .............. 75
Figure 30 . Modulus of rupture (MOR) and modulus of elasticity (MOE) for OSB
panels produced with differing levels of strand alignment in the surface
layerstrata ...................................................... 92
Figure 3 1 . N o d probability plot of the residuals for the MOR mode1 .............. 100
Figure 32 . Plot of studentized residuai versus predicted response values for the
.................................................... MORmodel 100
. ............. Figure 33 Plot of Cook's distance of the data points for the MOR model 101
. ...............- Figure 34 Plot of the leverage of the data points for the MOR mode1 101
. ......... Figure 35 Response surface of MOR in relation to coded values of K, and K, 103
. ......... Figure 36 Response surface of MOR in relation to actual values of K, and K, 103
. ......... Figure 37 Response surface of MOR in relation to coded values of K, and K, 104
Figure 3 8 . Response surface of MOR in relation to actual values of K. and K. ......... 104
Figure 39 . Response surface of MOR in relation tu coded values of K. and K. ......... 105
Figure 40 . Response surface of MOR in relation to actual values of K. and ic3 ......... 105
Figure 41 . Normal probability plot of the residuais for the MOE model .............. 112
Figure 42 . PIot of studentized residuai versus predicted response values for the
MOEmodeI .................................................... 112
Figure 43 . Plot of the leverage of the data points for the MOE mode1 ................ 113
Figure 44 . Plot of Cook's distance of the data points for the MOE model ............. 113
Figure 45 . Normal probabiiity plot of the residuais for the new MOE mode1 .......... 116
Figure 46 . Plot of studentized residuais versus predicted responses for the
................................................ newMOErnode1 116
. Figure 47 Plot of the leverage of the data points for the new MOE mode1 ............ 117
. Figure 48 Plot of Cook's distance of the data points for the new MOE mode1 ......... 117
. Figure 49 Response surface of MOE in relation to coded values of K, and K, .......... I l9
. Figure 50 Response d a c e of MOE in relation to actual values of K, and K, .......... 119
Figure 51 . Response d a c e of MOE in relation to coded values of K, and K, .......... 120 . Figure 52 Response surface of MOE in relation to a c t d values of K, and K, .......... 120
. Figure 53 Response surface of MOE in relation to coded values of K~ and K, .......... 121
. Figure 54 Response surface of MOE in relation to actuai values of K, and K, .......... 121
LIST OF SYMBOLS AND ABBREVIATIONS
OSB
UDL
cm
MOE
MOR
0
N
P
Q
DC
AC
FIFO
oriented strand board
unifomily distnbuted load
miIlimeter
centimeter
modulus of ehsticity
modulus of rupture
angle f?um some given reference point
strength of wood at an angle 0 to the grain
strength parailel to the grain
strength perpendicuiar to the grain
meter
constant
direct curent
altemathg curent
first in - first out method for rotating inventory
percent dignment
average of the absolute angles ranging fiom O to 90°
% of flakes within 20" of the cardinal angle (O0)
standard deviation
Pi (3.14)
function with respect to some parameter(s)
probability distribution function
mean angle of a set of anpuiar data ranging fkom O to 360°
4 - sine
Perp
GAI
DSM
IA
kg
SMSS
PRESS
R'
P
RSM
HP
RPM
ft
min
concentration parameter of the von Mises probabiiity distribution function
modified Bessel hct ion of order zero
average sine of the angles
average cosine of the angles
mean vector of a set of angular data ranging fiom O to 360'
orientation coefficient of the truncated normal distribution
error fiinction
kiloHertz
voltage meter
parallel
perpendicular
grain angle indicator
direct surface rneaSuTement
image analysis
kilogram
sequential mode1 sum of squares
predicted residud nim of squares
coefficient of determination or multipiple correlation coefficient
fiee-fa distance
plate gap
sirand widîh
strand iength
F statistic correspondhg to the F distribution
caiculated probability
response surface methodology
horsepower
revolutions per minute
feet
minute
dependant variable associated with a polynomid equation
a coefficient, constant
X independant variable associated with a polynomial equation
ANOVA andysis of variance
MSE
t
H,
Y n
lz
b
C
N(0,I)
u w
x2 v
d
MPa
A
B
C
DF
Ib
Coef
MMSF
MSF
mean squared error
t statistic associated with the t distriiution
nulI hypothesis
alternate hypothesis
sample size
function of the mean vector R constant
constant
standard nonnal distribution
statistic for testing homogeneity of concentration parameten
function of the sample variance
Chi-square distniution
fûnction of the sample size or number of samples
function of the sample size and number of samples
Megapascais
coded factor for the RSM polynomial equations
coded factor for the RSM polynomial equations
coded factor for the RSM polynomid equations
degrees of fÏeedom
pound
coefficient
million square feet @ased on some thickness)
thousand square feet (based on some thiclaiess)
INTRODUCTION
The oriented strand board (OSB) industry faces two impending developments which will cause
serious problems in the upcoming decade. The fkst development is the diminishing wood
supply, the second is the onset of new production capacity. Separately, each development is a
major concem, but together they will threaten the very h v a l of many less efficient OSB
producers.
Both developments will erode profit margins: a decrease in wood supply will increase wood
cos& and hence, production costs; and increased product supply wiU exert downward pressure
on the product price. The cornbined effect will lead to a significant decrease in the profitability
of operatiom.
More indepth analyses of these developments and their impacts are required to underscore the
importance of positive action on the part of industry. The problems will be malyzed separately
to address their different impacts, but with the proposal of one cornmon solution. The analysis
of over-capacity will be conducted in a wholistic sense, such that observations could apply to any
commodity-producing iadustry, but speciûc examples wiU be used to illustrate applicability to
OSB. The analysis of diminished wood supply will be conducted in regards to its impacts on
the OSB indu*.
1.1 E ffects of over-capacitv and the business cvcle decline
As has repeatedly occurred in the past, new capacity coming on-line over a short period of time
will deluge the market, Twenty-one new OSB mills are currently being built in North Amerka
and will begin production in the 19954997 penod (Lowood et al. 1995). This represents an
increase of approximately 4.5 million cubic meters in supply. It is unlikely that a sunicient
increase in demand will arise. This wiU result in a low demandhpply ratio and pnces will
plummet. Inventories will accumulate and orders will decline. With the deterioration in price,
profit rnargins mode away and hi@ cost producers must shut down operations. The threshold
(break-even) price level wiU depend upon the size of inventories and the prevailing market
demand. These two factors wiil also influence the duration of depressed prices.
Commodity-produchg businesses have two essential requirernents for break-even or profitable
performance in a business cycle domtuni. The fkst (and most important) option is to operate
with a differential between production costs (lower) and market price @igher). The second
option is to ensure consistent high standards of quality and service. Commodity products have
an element of customer loydty, such that the best s e h g brands during a downtum are ones that
do not experience repeated claims on faulty and substandard product. In facf a given producer's
quality must be head and shoulders above the rest of the pack to maintain cwtomer loyalty.
With the parity that commodities have in pncing, co~lsumers are loyal to service and quality, not
the brand itself. They (consumers) will not be forgiving when times are tough.
Superior companies address both requirements simultaneously. Continual improvernent in the
production process (leading to hcreased efficiency and lower costs) and consistent high qualiw
output will ensure the sumival of a rnanufacturing Company. Nowhere is this more true than in
the OSB industry.
Lowering production costs can corne h m many areas of the operation. Key process
consunables (ie., hydro and water) can be optimally utilized and prices negotiated, quantity and
performance of additives (ie., min and wax) used can be optimized and pnces negotiated, capital
investment in new, more efficient equipment and optimization of existing equipment w a lead
to both increased efficiency and reduced costs.
Fresh capital investment is critical to ongoing operaîions, but optbkation of existing equipment
offen greater cost-reduction potentiai. Substantial improvernent cm be realized with lower-cost
effort. Deferring the cost of new equipment, while extending the life of existing equipment, is
dways a preferable strategy. Furthemore, the techtucd knowledge ac-d during optimization
eEorts would reduce hîure capital acquisition coçts by linrithg unlaiowns and by identiwg
areas for improvement. Pressure could be placed upon equipment rnanufacturers to meet these
demands.
The fint step to ensuring consistent quality is by understanding the contribution of the different
processes to the final performance of the product. By idenûfjing the performance parameters
(and their optimal conditions) associated with each of these processes, one can define the
requirements for consistent quality production.
The next step wodd be to identw the resource (raw material) and machine (equipment)
parameters associated with the procesç. Once understanding of the process and its key variables
has been established, one can manipulate the operating parameters to create the best possible
conditions for quality production. Ergo, one could rapidly adjust the process to compensate for
changes in operating conditions.
Finally, to take the improvement process one step M e r , one can develop mathematical
relationships between the operating (equipment and reso urce) parameters and the product
performance. Optimization efforts would seek to ensure the best production conditions by
controlhg and rnatching suitable equipment and resource parameters.
To fkther examine the benefits of optimization, we can analyze the problem of diminishing
resource supply and what efforts can be made to mitigate its adverse effects on profitability.
1.2 Effects o f diminished wood sup~lv
Between 1980 and 1995, the North American demand for structurai panels (plywood and
oriented strand board) grew yearly by an average 666 thousand cubic meters. In the I 995 fiscal
year, 29.3 1 m . o n cubic meters of structurai panels were c o m e d in North America Demand
was expected to grow to 3 1.68 million cubic meters in 1996. Long term projections of the
strvcîural panel market predict a demand of 46.77 million cubic meters by 201 1 (RISI 1996).
The onented strand board (OSB) share of the structural panel market in 1980 was 13% in Canada
and 3% in the US. By 1995 OSB had captured 66% of the Canadian and 36% of the US markets.
The overall North American market share for OSB rose fkom 5% in 1980 to 3 7% in 1995. Aç
with the trend in overdl demand, the OSB market share is expected to grow to 45% in 1 996 and
to 83% by 20 11. (EUSI 1996) This represents an enormous increase of OSB production as the
industry stmggles to keep apace with rising demand.
Meeting the soaring demand is beghhg to put a strain upon the once undemtilized species of
North American forests. As with al i sectors of the wood products industxy, OSB producers are
being forced to stretch their wood supply and maximize yield. Wood costs currently represent
39% of the average variable costs in Canada and 42% in the U S (RISI 1996) and are expected
to rise with the decrease in resource supply. Future success of OSB miIls hinges upon the
efficient utilization of available resources.
Two ways to reduce the impact of increasing wood costs are to decrease the material input per
unit output and to maximize recovery of raw material. The first strategy requires a reduction in
board weight for a given thickness and m u t be accomplished without depd ing board
performance. Maxhizing recovery entails impmvements in the flaking operation and utilization
of Iowa quaiity materiai.
Reducing board weight and to a certain extent, maximizing recovery, can be accomplished by
improving the fonning process. Numerous studies (Snodgrass et al. 1973, Talbott 1974, Geimer
et al. 1975, Geimer 1976, Kieser and Steck 1978, Geimer 1980, Lau 1980, Shaler and
Blankenhom 1990, and McNatt et al. 1992) have estabfished that panel density (board weight)
and strand aligment have positive influences on board strength and stifiess. Logically, one
couid assume that the adverse effects (on performance) of a reduction in board weight could be
compensated by an improved control of strand alignment.
hprovement in wood recovery could be accomplished by the efficient allocation and
distribution of fumish in the product Performance requirements m e r spatidy in the product
and strategic placement of furnish quality (as dehed by strand geometry) wouid dlow
maximum material usage without a significant loss in panel properties.
To recap, the first problem is one that is experienced time and time again. Business cycles are
a natural occmence in the economy. However, surviving a downhim while potentiaily making
a profit is a serious consideration for every business. One d g a t i n g effort put forth was by
optimizing the manufacturing process with the aid of mathematical modehg techniques. The
existence of an optimization program for a aven process would broaden knowtedge, thereby
prompting quicker response time and increased efficiency. Demand for a higher degree of
conml wodd force equipment manufacturers to supply machinery with greater versatility and
ease of adjustment.
The second problem is progressive in nature - it gets worse with time. Diminishing wood
supply has greater consequences to a manufacturer's viabrlity than the cyclical decline in
business. One potentid offset to this problem could be through the optimkation of the forming
process, with regard to the control of strand alignment, strand geometry, and their vertical
distribution in the OSB mat A higher proportion of low-quality materiai could be used and an
overall net reduction in wood usage could be realized.
One operating principle of the surface Iayer forming machines used in indutry is the ability to
establish gradients of strand geometry and strand alignment through the surface layers. The
operation wïii be discussed in more detail in a later chapter, but çuffice to say that the fumish is
usuaily distributed so that the larger material is layered towards the surface and the smaller
material towards the core. The redting orientation of each "stratatt of strands in the surface
layer typically degrades towards the core (a resuit of the interaction between the machine and
resource variables of the fonning operation).
Typically, research efforts for strand alignment improvement have targeted improvement of a
homogenous strand size or mix of strand sizes (Geimer 1976, Harris 1977, Geimer 1979, Geimer
1980, Higgins 1990, and Geimer et al. 1993). This approach has ignored the reaiity of industrial
production. Surface layers are not uniform in composition. Other alignment studies have been
conducted in a very focused rnanner, where homogenous unidirectionally-aligned panels are
produced, evaiuated and modeled (Geimer 1980, Lau 1980, and Higgins 1990). What these fine
works lack is easy applicability to industry - it is difncdt to apply these models to heterogenous
multi-layer panels.
1 -4 Obiectives
The objective of this study is to demonstrate the effect of the surface layer strand alignment
distribution on the mechanicd performance of oriented s-d board. The objective requires:
0 qualification of product design and end-use requirements;
0 qualification of normal industrial operating conditions and process parameters;
0 a system for measuruig strand alignment in OSB;
0 a m d a(ignment prediction algorithm which will enable the production of OSB panels with
controlled aiigmnent in each layer,
O production of strands with controiied geometries;
0 production and evaluation of multi-layered panels with controkd strand alignment;
0 creation of a model to demonstrate the impact of controI1ed strand alignent on panel
p erfo manc e.
WhiIe recogniPng that conditions between the individual industrial production facilities Vary and
that the relationships described by the experimenral dgonthms will not exactiy mimic those
found in industry, the models governing strand alignment and Iayer forming will provide a
powerful tool for optimizing forming operatiom.
LITERATURE REVZEW
A firm grasp of the importance of strand alignment and its distribution in OSB entaiIs a
reviçitation to the basic principles of strucniral engineering and product design. A narrative is
required to provide continuity and cohesiveness between the three phases of OSB development.
This account does not follow a nomai chronology (design production ,- end use) because the
development of a product folIows a repetitive cycle with each phase being influenced by the
others. This concept is illustrated in Figure 1.
The following are descriptions of some aspects of the phase relationships:
1) The design of a product is govemed by its end use. The product must have certain properties
to withstand b a l application conditions. Some of these properties are obtainable with the
proper design and allocation of raw materids. An example of the reverse situation would
be where the product codd not be engineered to rneet the end use requirements and an
alternative application would have to be identined.
DESIGN
END USE 4 PRODUCTION
Figure 1. The development cycle of oriented strand board.
2) The production phase is govemed by the design of the product. The process must be
manipdated to satisfi the design criteria and produce a usable product. A good example
would be the optimi7:ation of forming processes to ensure the best possible strand
alignment and distribution of fumish by quality. The reverse situation codd be where the
production process did not have the capabiliw to achieve sorne design criteria In this
case, the design wodd have to be dtered to better suit the process or new equipment
would have to be acquired.
3) The end use is governed by the production phase or, in reality, by the product itself
Even ifa product had the most innovative engineering, its application couid be limited
by constraints inherent to the production technology. For example, production of high
perfomance OSB would not be possible without a formhg system capable of achieving
the high level of strand aiipnent required by its design. A reverse situation could be
iIlustrated by considering the sweiiing requirements of the application condition. A
portion of the swelling couId be controlled by the design aspect (ie., increased resin
content), but the balance wouid corne about h m proper manipulation of the production
processes (ie., higher press temperature).
The whole concept cm get fairly complex if one continuaIIy attempts to address di of the
intricacies of interconnectivity in the development process. For the sake of sirnplicity, it is best
to concentrate on a linear course of action (the cycle).
2.1 End use applications
The foIlowing section will bnefly examine the hdamental engineering principles involved in
the end use applications. Figure 2 illustrates one of the most basic design concepts used in
building systems.
Unfonnly Distrlbutsd Load
Roller SuppoR Beam /(Al
Fixed Support
> Span
Figure 2. Simply supported beam with unifonnly distributed load. (Turna 1969)
Figure 2 depicts a simply supported beam that is Ioaded uniformly across its span (Tuma 1969).
A uniformly distributed load (LTDL) would occur when the force appiied to the structure is
constant over a given area An example wodd be hardwood flooring instailed on subfloor
sheathing. A point load (not shown in this figure) wodd occur when an appiied force is
concentrated in one place. Examples of point loads wodd be table legs or a stationary person's
feet. This same structure could be used to illustrate panel applications - specifically that of
oriented strand board.
Using Figure 2 as a template, the cornponent names could be replaced by some cornmonly used
building materials. A better impression codd be made by visualinng a ffoor system step-by-
step. Imagine 2" x 10" (50 mm x 254 mm) lumber joists (support components) spaced apart by
a span of 24 inches (6 1 cm) and stretching across the length of the floor. Nexf X-inch-thick (1 9
mm) Tongue & Groove OSB floor sheathing (similar to the beam component) is installed with
the 8 foot (2.2 m) length nuining across the joists. FinaIly, a 2" x 1" (50 mm x 25 mm)
hardwood lumber suiface (ioad component) finisha the floor. Figure 3 illustrates one structural
unit of the fioor systern just descriied. It represents only one small section of the floor, but will
sufnce to demonstrate the system components.
Hardwood Ffaoting
/ Oiisnted Strand Boa& Sheathing
/- FIoar Jobt
f
Span
Figure 3. Typical fioor system with lumber joists, OSB subfloor sheathing, and a hardwood
floor surf'ace.
NOTE: Some individuals wiil undoubtFully argue that the OSB component is simply a skin
and does not contniute much to the overall strength of the system - that is
absolutely hue. In the real world of stmctural design the joist is the beam component
and the OSB is part of the load. This system also does not address property
requirements for other OSB end use requirements such as 1-joist webs or rimboards.
However, the simple beam concept can still be used for the purpose of examining the
physical and mechanical rigours of flexural bending end use applications (ie.,
flooring).
When a structurai systern (as described above) is loaded, whether uniformiy or by point load,
there are numerous effects on the system components. The three effects that will be discussed
are: stresses, reacbons, and defomations-
2.1.1 Stresses
Stresses are the forces and moments exated on a material by a load. They can be thought of as
the forces and moments transmitted internally through the material (Higdon et al. 1967). A force
is an ideaiized description of a load, such that the load has a magnitude and direction, and can
henceforth be expresseci as a vector quantity (Wright 1993). A simple example of a force could
be to consider a man's effect on a floor system. The man weighs (or represents a load of) 75
kilogcams and the force he exerts on the floor is 75 kg x 9.8 meters/sec2 (mas x gravity) = 735
Newtons. He would be exerting the force down or in the negative Y direction of a two-
dimensional Cartesian plane.
A moment is the tendency of a force to rotate about a point (Wright 1993). A simple example
describing the effect of a moment would be to consider the action of a door. If one considered
the hhge to be a point and pushed (applied a force) on the hingeiess side of the door, the door
would swing open (or rotate on its hinge). The moment is a function of the force and the
distance from the point and, like forces, moments have magnitudes and directions (dthough they
are angular directions). The angdar velocity or how quickiy the door wodd open depends on
the size of the moment.
2- 1.2 Reactions
Reactions are forces and moments deveIoped by the system components to counteract the
stresses created by the applied Ioad (Higdon et al. 1967). A body at rest is said to be in
equiIiirium- The requirernent for this state is that aU forces and moments be balanced, such that
the e x t d loads are counterbaianced by reaction forces and moments. Therefore, were there
no net forces or moments acting upon the system (ie. the sum is equal to zero), the system would
not move or change position. The ability of a system or material to achieve equilïbrium is a
function of its shape and strength characteristics. Once the stress becomes greater than the
materid strength, the material fails or breaks. The door example couid be used again to
demonstrate this concept If the dom were locked, it would not open when someone pushed on
it. The lock would generate forces and moments in the opposite directions so there wouid be no
rnovement (ie. there are no net forces or moments). If one pushed hard enough, either the door
or the lock wouid break.
2.1.3 Deformation
Deformation, or strain, is a change in the shape of the material caused by stress, moisture
content, temperature and other conditions (Wright 1993). AU materials have different modes or
phases of deformation. The elastic component of deformation is considered recoverable strain
such that the materiai will retum to its original shape when the load is removed (for example,
think of the way a spring or an elastic band works). Conversely, the plastic component of
defoxmation is considered irrecoverable strain with the material retaining its deformed shape
upon load removal (for example, think of clay and the way you can mold it). Material failure
could be considemi the ultimate plastic strain. Most load systems have a mixture of these two
components, such that the material will regain a portion of its origind shape. Flexural
deformation or "bendllig" is the strain type most considered in simple beam-like applications.
The bending moment produces a relative rotation of the two ends of the OSB so that the panel
"bencis", but in reaiity, the upper haIf of the panel shortens and the bottom halflengthens. Fipure
4 illustrates the bending action.
Rotation - Orfentecf Strand Board Shsathing
/ Rotaüon
Fiaor Joist Delbmed Shaps of OS8 Ffoor k ist
Span
Figure 4. Defornation (bending) of OSB sheathing under vertical loading.
The ability of a material to resist f l e d deformation is termed the modutus of elasticity (MOE).
The MOE is derived hom an equation relating the applied load to subsequent deflection
(deformation). The mie of t h m b is that "the higher the MOE, the M e r or more able the board
is to resist bending". A large MOE is desirable to limit movement of the systern for obvious
reasons (itfs not fun to bounce with every step taken), but it is also desirable to protect other
system components which do not have as great a resistance to deformation (imagine the effect
on plaster if the shealhing bent too rnuch). The ability of a material to resist dhmate
deformation (or failure) is expressed as the moduIus of rupture (MOR). The higher the board
MOR, the greater the loading capacity prior to failure.
The panel in the simply supported beam example couid fail in two ways. It couid fail in
horizontal shear (not a consequence of alignment) or in bending. The typical bending failure
character is illustrated in Figure 5. The farlure originates at the bottom surface and migrates up
through the material to the upper surface and will most kquentiy occur directly under the load.
The maximum bending moment, maximum stress and maximum deflection occur direcdy below
the midsection in both mid-span point loads and full system UDL's. The failure site is highly
dependent on the presence of defects, but OSB, imWce some products, is fke of the normal
defects that plague soiid wood products.
Joist Failun,
Joist
Figure 5. Normal failure mode of OSB when loaded in bending.
The next task is to examine the moment action and the stresses it creates. Bending moments can
be represented by force couples, such that the forces acting in conjunction will cause a rotational
effet. The simplest way of illutmfing the moment or force couphg is to cut the beam in half
and examine the stresses occurring internally- Figure 6 depicts the left half of a bisected beam.
Carnpression OS8 Sheathing Cut Plane Forces
Bending Moment
Tension Forces
Figure 6. Bending moment and force couple endured by the OSB sheathuig.
As demonstrateci in Figure 4, the top halfof the beam will shorten or "cornpress" durkg bendh.
Conversely, the bottom half of the beam wiIl lengthen or "tense" during benduig. Hence, it is
understandable that the force couplkg components are caiïed compression and tension forces.
Compression forces act to compact or reduce the material dimensions and tension forces act to
expand or elongate the material dimensions. The stress block (shown on the cut plane of the
bearn) is useful for illustrating the magnitude and direction of stress forces. The stress block
shown here differs in shape fkom the one which represented the UDL in Figure 2. This half
hougiass shape is characteristic of force couples. There can be only one (compression) or the
other (tension) existing in any one place at a given time due to the additive nature of forces
(oppoates cancel or balance each other). One can see fkom Figure 6 that the location of
maximum stress occurs at the surface and gradually diminishes to zero at the center or neutd
iuas. This axis is not fked and cm sh3t up and down depending upon the loading and rnaterial
character. Considering the stress distri'bution, it could be concluded that the strength requirement
of the matenai would aiso dirninish towards the neutral axis-
The bending moment acts counter-clockwise in this part of the beam. This couid be detemiuied
even without the moment figure. Were the neutrai axis considered a pivot point and the
outennost forces considered to be arrows (attached to the pivot point) travelling in opposite
directions, one would expect the stress diagram to spin iike a windmiIl. Were the stresses
examined on the other surface of the cut plane (ie. the right halfof the bisected beam) one wouid
find the forces acting in the opposite directions and the moment acting clockwise. The
connotation is the same because the compressÏon forces are still acting towards the material and
the tension forces are still acting away fkom the material. The moment direction is simply
governed by the respective action directions of the force couple.
The reason why most materials fail from the tension side is because even when compression
damage occm, it instantly seais by the very nature of the compression action. When a tension
tear occurs, it shih the neutraI axis upward and effectively &ces the vertical dimension of the
beam. The material strength is strongly dependant upon its vertical dimension and a reduction
in size while maintainhg or increasing the stress level will lead to an accelerahg rate of failure.
Therefore, a matenal used in bending is required to be very strong in tension (less so in
compression) and especialiy so at the lower surface.
In summary, some of the stresses at work in a simple ffoor system containhg OSB sheathing
were discussed. These inchded the applied force, the bending moment, and the compression-
tension force coupling. The review covered the typical deformation (bending) occurring in a
beam system and how it is caused by stress. Cornmon representations of strength (iMOR) and
sti&ess (MOE) were defined and fbaUyy a discussion c o v e ~ g the mechanism of bending
failme wrapped up the review. With an awareness of the application conditions and
requirements, the narrative c a . move on to discuss the design of oriented strand board.
2.2 Product design
Wood is anisotropic (orthotropic) in nature; Le., it exhibits different physical and mechanical
properties dong its three major directiod axes. The three axes, differentiated by growth pattern
and spatial distribution, are the radial, tangentid, and longitudinal directions. These axes cm
be envisioned as a the-dimemional Cartesian system comtructed with each direction separated
nom the other by a 90 O angle. Figure 7 iliu~frates the 3-dimensional representation of the mis
system.
Z axis
Y axis
X axis
Figure 7. ïhree-dimensional Cartesian axis system.
In practice, the longitudinal direction (2 axis) can be distinguïshed as the one that ~ i s dong the
grain, or that runs paralle1 to the stem of the tree. The radiai and tangentid directions can be
differentiated by examining the cross-section of a log. The radiai axis extends fiom the center
to the outer surface (or vice versa) and can be considered the X ais. The tangentid axis can be
pictured by taking the tangent of one growth ring and using it as the Y axis. The axial
assumptions are illustrated in Figure 8.
Tangentiel (Y-axis) Longitudinal (Z-exfs)
/ \.
i
1
Figure 8. Axial system as dehed by wood growth structure.
It is impoitant to understand the directional differences of the three axes because of their
significant differences in physicai and mechanicd behavior. This behavior arises fiom the
structure and organization of ceIlulose and hemiceiiulose in the ceil was , the elongated shape
of the wood ce&, and theK longitudinal-radid arrangement resulting fiom the radiai symmetry
of the tree trunk (Panshin and de Zeeuw 1980). As a consequence, compressive and tende
strengths Vary between longitudinal and lateral (radial and tangentid) directions in the wood.
Stifkess cornparisons between the three principal directions can be demomtrated by examining
the ratio of modulus of elasticity (Cooper 1992):
where
MOEL is the longitudinal modulus of elasticity
MOE, is the radial modulus of elasticity
MOET is the tangentid modulus of elasticity
Wood is 20 times stiffer in the longitudinal direction than in the tangentid direction. Stifiess
in the radial direction is only slightly higher than in the tangentid direction. Stifiess usually
has a strong correlation with strength except where there has been substantiai degradation of, or
physical change in, the wood substance (ie., significant changes in the wood could be brought
about by thermal pyrolysis or by s t e m treatment).
Panshin and de Zeeuw (1980) stated that the ratio of compression strength parallel to the grain
versus compression strength perpendicular to the grain cm Vary fiom 4 to 12, depending on
species. The same basic relation holds tme for tension, with wood being stronger in tension than
in compression.
Strength at angles to the grain may be estimateci by the empirical Hankinson equation (Koilmann
and Côté 1968):
w here
N is the strength at an angle 8 to the grain;
P and Q are the strengths p d e l and perpendicular to the grain,
respectively; and
a is an empirically detemllned constant, ranging from 1.5 to 2.5.
h e d with the above reiationships, it is not difficult to conclude that the ideal product
configuration would have the wood grain of the elements ninning parailei to the long dimension
of the product Product designers consistentIy attemp t to maximize the strength characteristics
of their products. In the case of wood products, they achieve this by capitalizing on the
anisotropic character of the material. As previously mentioned, compression and tension
strengths are greatest pardel to grain. Therefore, the strongest product would have the grain
onented paraUeI to the direction of the compressive and tende stresses. Onented strand board,
Iumber, and plywood aii exempw this design feature.
OSB is manufactured by forming a mass of strands into a mat and pressing it under heat and
pressure into a panel. At the beginning of the production process, solid wood is flaked into mail
thin stmnds which somewhat mimic the structure of the tree. Thin (- 0.025-inch or 0.63 5 mm
thick) rectanguiar strands with the grain direction pardel to its long dimension (- 2 - 6 inches
or 50 to 150 mm) and its width (- %-inch or 19 mm) cut radidy or tangentially (depending on
the angle of the cutting knives relative to the log position) are generated during the flaking
operation. Figure 9 illustrates the grain orientation of an OSB strand.
Longitudinal Grain Orientation
/
Strand ". 7 Width \,.
(4 * Stand Length
Figure 9. Geometry of a typical OSB strand with associated grain orientation.
OSB is manuf'hired with the surface Iayers orienteci more-or-less parallel to the long dimension
of the panel (see Figure 10). Standard building procedures call for panel installation of its length
&g perpendicdar to the longitudinal direction of its supports (eg. joists, w d studs, and roof
tnisses). As demonstrated above and in Section 2.1, this orientation offers the maximum
resistive strength to the applied compression and tension stresses (refer to Figure 6).
8 Foot (2.4 m) Length >
4-1 Strand Alignment Direction Shands
4 Foot (1.2 m) Length
Figure 10. Surface view of a typical OSB panel.
It has been demo~l~trated numerous times (Geimer 1980, Geimer et al. 1975, Higgins 1990,
Kieser and Steck 1978, Lau 1980, McNatt et al. 1992, Shaler and Blankenhom 1990, Snodgrass
et al. 1973, and Talbott 1974) that the degree of strand alignment has a strong positive
relationship with the strength and stifnless of the board. One wodd venture that "the better the
degree and control of al iment, the greater the board bending properties in the aligned
direction".
To achieve the maximum possible board strength in the aligned direction, the entue thichess
of the panel wodd have to be aligned in the same direction. This is rarely done in practice
because of hear expansion and stresses occrrrring across the panel width. The simple example
in Section 2.1 did not investigate the stresses in other directions, but rather assumed a two-
dimensional perspective, when, in reality, the world exists in a three-dimensionai state. Thus,
if the whole system was envisioned three-dimensionally (XYZ axis system), there would also
be stresses o c c ~ g in the perpendicular direction. A panel aligned wholly in one direction
would have maximum strength in that direction, but minimum strength in the perpendicular
direction. It is partidy for this feason that the core layer of most commercial panels are cross-
oriented. Cross-orientation is performed primarily for linear expansion reasons (dimensional
stabifity), but also to impart some strength across the board.
Recall h m Section 2.1 that the stress caused by the bending moment decreased Linearly towards
the middle of the board. Because the strength requirement in this area is not as great as it is at
the surface, the core strands can be cross-onented without significant impact on panel
performance in the parallel direction. With cross-orientation, perpendicular strength is irnparted
to the board, but at the cost of pardel strengîh. Depending on the application strength
requirements for the two directions, parallel sfrength could be augmented by increasing the
surface to core layer ratio or by random-orienting the core.
To summarize, the three p ~ c i p l e axes associated with the anisotropy of wood were dehed.
The different physical characteristics of those axes were probed and abundant evidence was cited
to show that maximum wood strength occun dong the grain, ie., in the longitudinal direction.
It was concludeci that the strongest composite wood products in bending could be produced by
ensuring the constituents were aligned with the grain parallel (or opposite) to the acting stresses.
OSB design was show to have the d i c e layers oriented paraIlel to the board axis (to
counteract the beam-like stresses) and the core Iayer cross-onented (to inpart perpendicular
strenDOth and dimensional stability) or random-uriented.
2.3 Production parameters
This phase of the product development cycle considers the actual assembly and production of
oriented strand board. For all intents and purposes, it is the most important phase because it
produces a tangible item -- the product. In the miU environment, process variables are
manipulated to yieId conditions necessary for the achievement of the required board properties.
The forming process alluded to in Section 2.2 is the sector of production controlling strand
aiignment and will be the one on which this review will focus.
At the industry's curent level of technology, there are two aIignment techniques available to
OSB producers: electmstatic alignment and mechanical alignment. The nnt method employs
electncal forces to align wood particles. The technique was developed to use with smaIIer
particles as no suitable mechanical method existed for particles as mid as fibers (Talbott 1974).
The second method empioys mechanicd means to achieve alignment. There are several types
of mechanicd equipment bekg used to achieve alignment, but ail are based on a ke-fdl design
with some device configuration for separation and orientation of the fumish.
Talbon and Stefanakos (1972) described a device which wouid aiign particles through the
application of an electric field. The apparatus was constructed in the fom of a verticaily
oriented open-ended box. The materid was fed fiom the top and passed through an electrical
fieid. The partides rotated durllig free-fa to orient themselves with their longitudinal axis
perpendicular to the electmde Surface or parallel to the field direction. Ensuing experimentation
revealed that high field intensities and slender particles yielded the best aiignment, but required
high moishire contents (>15%) to be effective. Figure 11 illussiftes the p ~ c i p l e s and operation
of a single celi electrostatic alignment device.
The hdamental principles goveming the operation of the device are based on the anisotropic
dieIectnc properties of wood. Wood is normally considered to be an excellent insulating
mataid with a direct-current @C) resistivity of 3 x IOL7 to 3 x lOI8 ohm-centimeters at room
temperature and air-dry conditions (Clark and WiUiams 1933). Water, being a polar molecule,
has a distinct effect on the electrical properties of wood. Wood becomes progressively more
conductive (or less resistive) with increasing moistwe content. Pazlshin and de Zeeuw (1980)
stated that at 16 percent moishne content, the value of resistivity for wood at room temperatures
decfea~es to 108 ohm-centimeters, and at fiber saturation point it became approximately that of
water alone (1 @ to 1 O6 ohm-centirneters). Skaar (1 972) reported that resistivity across the grain
was 2.5 to 8.0 times greater than aiong the grain for hardwoods and 2.3 to 4.5 times greater for
coaifers.
Fall ravity
Random Flakes
Field Unes X
E l ~ c b o ~ n l t l c Polarizad ' Fbld Aligned
and Gravity Flakes
Y Y
Figure 1 1. Principles of electrostatic alignment - single ceIl. (Fyie e t d 1980)
A more popu1a.r measure of wood's electrical behavior can be expressed as the material's
dielectric constant. The dielechic constant is employed as a measure of the insulating capacity
of a material under an alternating curent (AC), but the value expressed is somewhat rnisleading
in that increases in magnitude represent decreases in resistivity. As with DC resistivity, the
dielectric constant varies with moisture content and displays anisotropic differences. In general,
the dielectric constant is 1.3 to 1.5 times greater in the longitudinal than in the transverse
direction (Panshin and de Zeeuw 1980). Dielectric constants for wood range nom about 2 in the
oven-dry condition to 81 at moishue contents above the fiber saturation point (Clark and
Williams 1933).
The anisotropic effects are caused by wood anatomical and chernical fàctors. The wood structure
is relatively unifonn and hornogenous in the longitudinal direction, but is very difTerent
transversely fkom one ceU type to the next (eg. earlywood versus latewood). These Merences
are important to note because of the behavior of wood particles and tibers when subjected to an
electric field. WhiIe wood is generally considered a nonconductor, it is made up of dipolar
molecules (ceUuIose, lignin, water, etc,...). The dipolar aspect of their make-up induces the
wood particles to orient themselves in an applied electrïc field. When the field direction is
reversed, the dipoles will reorient themselves (Panshin and de Zeeuw 1980). The configuration
of the dipolar substances in wood partially accounts for the anisotmpy in the dielectric constant,
ie., cellulose has a predorninantIy longituduial axis and water migrates through wood easier in
the longitudinal direction-.
The TaIbott and Stefanakos (1972) aligmnent device capitalized on the dielectnc anisotropy of
the wood pamcles. The applied field would induce polarkation in the dipolar elements of the
wood. Due to the anisotropic resistivities of the wood substance, polarization would induce a
migration of charges towards the longitudinal extremities of the materid. The appiied electric
field exerts a torque on the particles (or rather the torque is created by the dipolar tendency of
the particle constituents to orient themselves) and the rmhindered particle ro tates to orient itself
parallel to the field lines. Kawai et al. (1987) and others (Pdido et al. 199 1 and Yoshida et al.
1988) have expanded the study of aligning torque on wood particles.
The work of Tdbott and Stefanakos (1972) was continued by Morrison-Knudsen Forest Products
Company and others (Kawai et al. 1982, Kawai et al. 1987, Lang et al. 1982, Pulido et al. 199 1,
Sasaki et al. 1989, Yoshida et al. 1988, Yoshida et al. 1 Yoshida et ni. 1989b). Orighai ly
interested in smaller particle geometnes, Morrison-Knudsen shifted its focus to larger flake-sized
partictes to take part in the growing OSB industry. Fyie et al. (1 980) and Peters (1 983) reported
the companyts progress with the FORCELINEm electrostatic formuig machine. The basic
governing concepts remaineci the same in FORCELINE~, but it entailed more sophistication in
technique.
The primay production parameters encountered by Momson-Knudsen and others which had an
effect upon alignment were flake geometry (Yoshida et al. 1989a), ffake moisture content
(Kawai et al. 1982), field intensity (Kawai et al. 1982, Yoshida et al. 1989a, Yoshida et al.
1989b), field size, fÎee-fa distance9 and h e s p e d Kawai et al. (1982) disputed the
sigdicance of fiake geometry and reported that species did not have any signincant eEect on
alignment.
At a given field intensity and flake moisture content, flakes with high slendemess ratios would
align best The electrostatic alignment device performed well with relatively fine particles, but
Iost its efféctiveness with increasing flake size. Possible compensatory meanires could be to
inmase the flake rnoisture content or the field intensity, but both wodd cause more harm than
good.
Flake moisture content has a positive relatioflship with al ipnent in the case of electrostatic
fonning devices. Impure water is a highly polar substance with an emordinary ability to
conduct electrkity. High moisture contents coupled with the wood fiber structure increases
charge migration (polarization) and augments the orientation effect. However, excess moisture
is anathema to conventional pressing techniques. The excas mat moisture would require higher
applied pressure to increase steam pressure and its temperature, longer press holding times to
ensure the min cure and longer vent times to exhaus the steam. These adjustments would lead
to excessive wood cell collapse f?om the action of moistrne, temperature, and high pressure and
result in a thinner higher-deflsity board. The presence of excessive moisture would interfere with
resin cross-linking and result in weak internal bonds.
The field size between the electrodes will govem the degree to which the flakes will aIign
themselves with the field Obviously, the longer the flake is exposed to the eleceic field during
its passage, the more tune the torque has to act upon it. There are aiso some conditions that can
preempt any field size. These iuclude large flake geometries, low moisture contents, and weak
field intensities.
The free-fdl distance is the distance between the alignment device and the fomiuig line. The
flakes are no longer influenced by the electric field and could be reoriented during the fa11 by
gravitational and other applied forces. The extent of reorientation nom other influences is
govemed by the expanse of the drop. Free-fd distance and degree of alignment share a negative
relationship, where greater heights lead to more variable alignment.
Field intensity is a measure of the strength of the electric field This variable is probably the
easiest to control, but the most costly and hazardous. Adjusting the field intensity would ody
be a matter of tuning a dial, but managing that change would be another matter. Use of a high
voltage field for any purpose presents several potentiai problems. Due to the changing nature
of the fumish passing through the field, there is always the potential for electrical arcs and
grounding. An arc could occur when a conductive material entered the field and acted as a
conduit. Even a weak electrical arc could ignite airborne dust or cause a nasty electricai bum.
Grounding would cause probiems associated with short circuits and skewed field Lues would
Iead to poor orientation efficiency. There are hi& probabilities of both these problems
(grounding and arcing) occuring given the overwhelming presence of metal in the processing
equipment. Fyie et al. (1 980) encountaed such a dilemma with the grounded metai caul plates
used for tramferring the mat into the press. They were able to overcome this obstacle by
devising a controlled transfer of the aligned particles away nom the hi& voltage field and onto
the caui plates. Figure 12 illutrates the FORCELIM? electrostahc alignment device with the
controlled transfer. Incidentally, this controlled transfer wodd eliminate the fkee-fall variable
and its effects on h a 1 flake position. Sasaki et al. (1989) refuted this method because of the
tendency of particles to stick to the lower edge of electrodes and form "bridges" between the
electrodes and between the mat and electrodes. Obviously this "bridging" tendency disturbed
orientation, not to mention d o n n i t y of fomüng.
Yoshida et al. (1988) presented another type of electrostatic aligner which consisted of
polyvinylchloride conveyor belts with the electrodes on the reverse side of the forming belt.
Yoshida et al. (1989b) lata reported that mat height was a significant factor affiéchng alignment
with this device, but that fkee-fd distance was no longer an issue. Sasaki et al. (1 989) M e r
reported on the practical opportunities with this method An industrial scale set-up was proposed
with features: 1) electrodes are located only on the reverse side of the fonning belt so that the
unstable movement of particles are eluninated, 2) a special mat having beîîer orientation towards
the bottom suffixe is fomed, and the top Oess-oriented portion) is shaved off to obtain even
thichess, and 3) two mats are mated together redting in a double mat with better orientation
tow ards the d a c e s -
Random Free Fall v
Electrostatic Field - Alignment
Controlled Transfer
v 1 1
Caul F
Forming Line
Figure 12. Schematic of a single ceU electrostatic aiignment device with controlled transfer to
a cad. (Fyie et.al. 1980)
Line speed is as important to electrostatic forming as it is to mechanicd forming. Increasing iine
speed requires more furnish volume to pass through the orienter in a given time fiame to
maintain mat weight An excess of materid would retard the ability of the fiake to orient itself.
One flake would act as a support for another, which wouid generate reaction forces and impede
flake rotation. Fast line speeds will also produce a bouncing or skidding effect by the flake when
it contaas the Line. Instead of the flake being positioned the way it fell, it codd bounce or get
knocked askew.
In summary, the main machine parameters influencing the efficiency of electrostatic alignment
are field inteflsity, field size, fke-fa distance, and line speed. Significant resource parameters
are particle geometry and moisture content In general, the optimal conditions for alignment are
high field inteflsities, large field sizes, short free-fall distances, and slender hi&-rnoisture-
content particles. F d e r study and improvement of the electrostatic method has been performed
by numerous mearchers in Japan (Kawai et al. 1982, Kawai et al. 1987, Lang et al. 1982, Pulido
et al. 1991, Sasaki et a l 1989, Yoshida et al. 1988, Yoshida et al. 1989a, Yoshida et al. 1989b).
However, the problems inherent to the method when producing OSB prechde its practical use
in indutry.
2.3.2 Mechanical alignment
There have been numerous studies investigating production variables and their relationships with
strand alignmertt. Studies have dso been conducted to compare the performance of dif5erent
mechanical alignment methods. Mechanical a l i v e n t devices include rotary disks, vane roils,
vibrating fins, gravity chutes, reciprocating bars, and pairs of adjacent chahs or beits traveling
in opposite directions (Geimer 1976). The following review will describe three types of
aligiment devices and discuss the production parameters affecthg alignment.
This type of aiignment device has many different configurations, dependhg on which facility
you visit. The oscillating-firame orienter is preferred by the research c o m m u n i ~ for its
simplicity of construction and superior control of alignment. The apparatus is generally
confïgured with vertically oriented plates which form rectangular slots. Flakes are metered
verticaily onto the h m e and fa11 through the slots onto the forrning belt aligned in the long
direction of the slots. Dining operation the fkne osdates back and forth to facilitate fl&e
entry to the slots. Figure 13 illustrates the side and top views of a typicd tiame-type orienter.
S p ring
SlDE VlEW
S p ring
Alignment Slot
TOP VlEW
Figure 13. Schematic of a typical oscillating-fiame alignment device. (Zhou 1989)
Zhou (1 989) studied the influences of the four main factors govemuig the operation of an
o s d a î h g orientery ie. the distance between two neighbouring plates (slot width), the oscillating
fkquency, the length of the slot, and the falling distance (distance from the orienter to the mat).
Zhou (1 989) reported that dot width (plate gap) and fkee-fd distance had significant innuences
on the orientation of strands (in accordance with other reports (Geimer 1976, Geimer 1 980, Lau
1980), slot length only had a marginal effect, and oscillating kequency exerted iittle or no
- 3 1 - #
influence. Numerous researchers (Alexopoulos 1990, CarlI and Link 1988, Canadido et al. 1988,
Geimer 1976, Geimer 1980, Kajita 1987, L ~ u 1980, Suzuki and Sekino 1982, Yoshida et al.
1989, Zhou 1989, Zhou 1990) reported the use of such devices in studies of onented strand
board.
Lau (1980) examineci the machine factors controllkg &and alignment with a similar oscillating
orienter- Lau (1980) likewise reported that the main machine factors affecting alignment were
plate spacing and f k - f d distance. Lau (1980) also concluded that oscillation frequency, whiIe
not overly aEecting the quality of aligment, had a significant effect on the material flow-
through capacity of the machine. The relatiomhip was determined to be positive with increased
Eequency allowing more material to flow through over a given time period.
Using the same alignment device as Lau (1980). Aïexopoulos (1990) discovered horizontal
density pro blems as a result of superior control of strand alignment The strands were essentidy
stacked one on top of another resulting in altemating longitudinal strips of high and low
densities. The lack of overlapping between two adjacent strips wodd create a density gap and
result in planes of weakness dong the panel. Alexopoulos (1990) solved the density problem
by moving the plate assembly lateraiiy during forming to effect some overlap.
This srpe of alignment device is timited alrnost exclusively to research applications because of
its low productivity. The oscillating actian of the fiame does not provide enough force to
separate a large mass of flakes and results in slow flow-through rates and heterogeneous
distribution of flakes. The assembly would require a vertical oscillation component to be
effective in influencing strand entry to the alignment slots.
Another means to compensate for the tendency of the flakes to remain on the top of the firame
is to stagger the top heights of the dot plates (Geimer 1976). This way, the fumish is partially
distributed and funneled into the slots. Figure 14 illustrates the staggered configuration required
to improve flake distribution and flow-through rates.
-32-
l ; Flakes metered in Sùaggemd Plats Helghts
**- Pîate Spadng
-> Lateral Movement
of Asaembly
I
;
Fi,pre 14. Staggered plate configuration of an oscillating-he a l i m e n t device.
One other deficiency of the oscillating-fhne orienter is that there are no means to veizically
distribute the flakes by size. The stress character and strength requirements of the panel were
discussed previously and stipulated that the panel requires the highest material mength on the
surfaces to withstand applied stresses. It was also stipulated that higher strength could be
derived by better strmd alignment. However, machine parameters are set to accommodate the
largest strand size in the size distniution. Therefore, a mix of strand sizes will guarantee a Iess
than optimum distribution of strand a l imen t through the surface layers, lacking any method
for segregating and class$mg strands by sue. Even with the efficient screen classification
systems currently in use, variability in flake geometry wilI still be present in the process.
In sl~rnmary, the main production variables affecthg alipnment with an oscillating-fkame device
are plate spacing, fke-fidl distance, h e speed, and fl ake geometry. The most irnpractical feature
of this type of alignrnent device is the oscillation requirernent to efficiently produce a
homogeneously-aligned mat. The system requires oscillation in three dinerent directions to
ac hieve passable flow-through rates and homogenous horizontal distribution of matenal.
2.3 -2.2 Rotary dÏsk forming machines
Rotary disk forming machines are used in the indusûy to form and &gn the panel surface layers.
The gross components of the forming station feature a metering bin, feed chute, and fomiing
head Figure 15 illustrates the components of a typicd disk-type fomiing head used in industry.
/- Rake back
- \
\ ,<
\ r b ' j J
Guide flap >\ 1
Live bottom l
ivid Guide
A/- Disk roll
Figure 15. Schematic of a rotary disk forming station.
The overhead metering bin holds a store of blended flakes and ensures a contuluous supply of
furnish for the forming head. Some pre-fomiing is undertaken in the metering bin to render a
homogenous spread of material across the width of the h e . F u . s h is introduced to the top of
the pile near the output end of the metering bin by a reciprocating conveyor system. The rake
back conveyor located at the top of the bin acts longitudinally (fiont to back) on the fumish pile
to drag the uppermost material towards the back of the bin. This action not only contributes to
a FIFO (nrst in - k t out) rotation of the materid, but also provides additional horizontal
smoothing to minimize differences in buik density. The Live bottom belt operates in the opposite
direction to the rake back conveyor by moving the whole pile towards the rotating picker rolls.
The belt speed govems the mass ffow of matenal to the heads and is synchronized with the
fomiing line to ensure a steady mat weight. The chuming action of the picker roUs dislodges a
diffuse volume of fumish kom the pile which then f d s down the feed chute to the forming head.
The presence and action of the picker rolIs also works to break up clumps and discourage
avalanching of the furnish (ie. surge of a convoluted mass of flakes). The fi akeç encounter a
level of dividing rolls which act to separate strands by size. The next level of rolls, the
dissolving rolls, tend to distribute the furnish more evenly and M e r separate strands by size.
Strands f a f?om the dissolving rolis on to the &gning bed of disk rolls.
Figure 16 illustrates the disk assembly of a Schenck-type fomiing head. The assembly consists
of rows of disks mounted on shafts with each row's disks staggered laterally fkom the next. The
staggering eliminates low density zones by overlapping adjacent aligned flakes and it inhibits
strands fÏom onenting perpendicular to the forming direction (by fdling between adjacent disk
shafks). The spacing distance between disks is detemineci by the largest flake width (-2-inch
or 50 mm). One aspect of this setup is that strand alignment diminishes with the smaller s m d
sizes .
! ! i ; ! , , 1 <
: 4 A ( / b) FRONT VlEW
Disk Alignment Gaps
c ) TOP VlEW
Figure 16. Disk assembly of the Schenck surface Iayer fomiing head.
The disks are equipped with teeth on their perimeters to improve contact with the Eumish. FFlake
edges are caught in the gaps between teeth and the rotating disk changes the horizontal
positioning of the flake, making it easier to faü into the gaps between two adjacent disks, or
"flings" the flake in the direction of disk rotation. The probability that the flake will be flung is
dependant on its geometry, the disk spacing and whether the teeth grab hold On any given shaft,
larger flakes would have a higher probability of being "flung" than smaller flakes.
It is the mechanical ability of the system to identify ffake geometry which enables the
classification and distribution of flakes by size through the thickness of the mat layer. For
example, consider the bottom surface layer of the board. Highly aligned flakes (larger flakes)
are desirable on the bottom face of the Iayer to counteract the maximum applied b e n h g
stresses. Placing the smaller material towards the core (with its diminished strand alignment)
is less detrimental to board performance because of diminishing strength requirements (Figure
6). This graduation by size is illu~frated in Figure 17. The dividing, dissolving and disk rolls
wodd be configured to h g the Iarger flakes towards the back of the forming head. This would
ensure the 1- ones are placed on the bottom of the Iayer with progressively higher quantities
of çmaller materid being mixed in towards the core. A small pick roll would be staggered above
the last shaft (with the largest gaps) and rotate in the opposite direction to reposition the large
flakes which did not pass through the gaps. The degree of ~Iassification by flake size is
dependant on the disk rotational speed, disk spacing, and variability in flake geornetry.
Large strands
nds
Face
Figure 17. Distribution of fumish by strand size through the bottom surface layer of an OSB
panel.
A number of studies (Geber 1976, Geimer 1980, Kieser and Steck 1978, and McNatt et al.
1992) reported that the machine parameten rnost mfluencing alignment were disk-spacing and
k-fd distance. Line speed had some effect on alignment and disk rotational speed had Little,
if any, influence. Flake geometry (a resource parameter) also had a significant effect on
alignment.
Disk spacing governs how much leeway a flake has to Vary around the principle alignment
direction. This is entirely dependant on flake geometry, in that the smdler flakes have a larger
anpuiar range to position themeIves as they pass b u g h the disk gaps. Graduated disk spacing,
where each successive shaft has a d e r spacing betwea adjacent disks, would be a better way
of controlkg the alignment of hmish in each of layer sirata. Siempeikamp-type formes
incorporate this design. However, optimization of disk spacing requires the flake geometry
distribution to be known.
Free-fa distance is the distance fkom the bottom of the disk assembly to the fomiing line. It is
over this distance that a flake can fkeely rotate while it f d s to the line. Air turbulence and
mechanical energy imparted tu the flakes as they f a between the disks is free to act when there
are no impedùnents to movement Rotational energy that was transferred to the flakes from the
disks acts to spin the flakes in fke-fd and reposition them at different alignment angles. A nile-
o f thumb is to ensure the fkee-fd distance not be greater than the nominal length of the flakes
and if possible, operate with the lowest Eee-fa distance possible.
As with the other types of orientation devices, the line speed has an effect on the final position
of the flakes. Faster line speeds apply a directional force to the ff ake when they make contact.
This force can knock the flake into a new horizontal orientation.
To summarize, the production parameters affecting alignment with rotary disk orïenting
machines are disk spacing, fiee-fa distance, line speed, and flake geometry. Disk-type for-g
machines are used exclusively for the surface layers of panels. Their operation enables a
qualitative (size) distribution of flakes through the thickness, thereb y i m p d g maximum
bending strength and stiffuess to the panel. Disk orienters provide the best alignment for all
sizes of fumish and have superior flow-through rates (McNatt et al. I992).
2.3 -2.3 Vane (chamber) roll alignment machines
Vane-type fomers are employai in industry for the cross-aliment of the mat core Iayers. The
forming station is very similar to the rotary disk orienter fomiing station, but with vane roUs
instead of disk shah. The roiis have a series of laterally arranged vanes projecting fkom the
shaft and have the appearance of a large cylindrical sprocket. Fumish fdIs on the roll and is
caught in the depressions between the vanes. The rotation of the roll deposits the flakes on the
mat at the bottom point of the revolutio~ Figure 18 depicts the orienting action of the vane rolls.
Figure 18. Onenting action of a cross-aligning vane roll.
McNatt et al. (1992) compared the aligning performivlce of disk and vane formllig machines and
conchded that disk-type orienters aligneci flakes better. This conclusion was based on
cornparisons of board strength (MOR) and stifkess (MOE) values of the finished panels. To
M y understand this assertion, the operation and orientkg effect of the vane orienting device
The m h fds into the groove between two vanes and becomes oriented by the combination
of gravity, the angular force caused by the rotation of the vane roll, and the groove geometry.
The groove is configureci to ensure its base is no wider than the nominal flake width, ie. with a
nominal 314-inch (19 mm) flake width, the vanes would be spaced 3/4-inch (19 mm) apart on
the shaft. Figure 19 iUustrates the geometrical configuration of the roll grooves and the
positionhg of Fumish.
Alignment Axis
Figure 19. Afignment groove and flake geometry uifluence on the effdveness of the vane-type
orienters.
The width of the p o v e is graduated throughout the depth of the groove. This represents a
progressive loss in control of aligmnent towards the roll perimeter, even with a perfectly
homogenous flake size. This variability in aIignrnent is transfened to the mat in each subsequent
rotation and deposît On visual inspection, there would be distinct discontinuihes in mat
alignment quality. Sharp delineations in alignment levels because of patterns of degeneration
of alignrnent would appear in the IongituRinal (forming) direction. Further constricting groove
depth in order to improve a l i m e n t would have a negative effect on the matend flow-through
capacity of the device.
Further loss of aligcunent could occur ditring the deposition of flakes on the line. The rotational
action of the roll generates a centfigai force which acts to propel the flakes away fkom the
center of rotation (si&). The combination of the centrifbgd force and gravity induces migration
of the flakes nom the bottom of the depression out towards the perimeter and contact with the
sides of the vanes in egress could change the orientation of the flakes. The magnitude of the
centrifbgal force is dependant on the rotational speed of the r d , (ie. faster roii rotation generates
more cen*gal force), and this relates directly to the amount of force transferred during flake
contact with the roll. Loss of a l i p e n t is only deerned a possibiiity because the oppoate effect
on alignment could occur. Forced contact with the vanes during egress could possibly reorient
poorly aligned flakes and improve o v d alignment of the fumish. Whether a loss or gain
would be experienced would depend upon the volume of flake flow-through and the roll
rotational speed. Rotational speed also has a . e f f i on final potitioning by goveming with how
much force the flake will impact the forming line.
As with the preceding two examples of mechanical alignment techniques, the kee-fall distance
has a significant effect on aLignment. The residual forces and rotational tendency imparted by
the action of the device are f?ee to work on the flake during its fall.
Line speed would have more effect on final alignment position with the vane-type orienter than
with any other alignment method discussed. With the flakes oriented cross-wise, the directional
force applied by the forming h e would be able to act over a greater area Recall fiom Section
2.1 that the applied moment is a function of the application distance. Force application on the
Iong dimension of the flake wouid result in a greater tendency for rotation and possible loss of
orientation-
The vane, in itself. does not provide a segregation and spatial distribution by fïake quality. The
firmish is homogeneously deposited on the line throughout the mat layer regardles of Bake size.
It is possible to engineer a graduateci layer through the action of the dividing and dissolving rolls,
but much Iess efficiently than with the disk-type former. It is for this reason (and the superior
control of orientation with disk formers) that the vane orienter is not used for surface layer
forming. Consequently, the overall performance of the core Iayer wodd be better with a
homogeneous rnix of ff ake geometries. There is a diminished requirernent for bending strength
in the core, but an increased need for bond strength and a homogeneous density distrîibution. The
random mWng of flake &es tends to minimize the occurrence of localized regions of density
variabili~ by fi lhg voids and providing a greater surface area for bonding.
In summary, the production parameters affecting alignment with the vane cross-onenting
machine are: vane spacing (groove base width), groove depth, roll rotational speed, fiee-fall
distance, line speed, and flake geometry. The lower control of alignment with the vane rolls is
not overly detrimental to panel performance because their use is limited to the core layers.
2.4 Methods for measuring and characterizing strand a l i m e n t
2.4.1 Direct surface measurement
The direct d a c e meanirement method involves a visual assessrnent of fiake angles with regard
to some reference orientation on the d a c e of an OSB mat or panel. In most instances, this was
performed by a hand-held protracthg device.
Geimer (1976) was the £irst to employ a direct surface measurement technique for measurïng
ffake alignment. To determine the extent of alignment, flake angles were measured at the 300
intersections of a '/-inch (12.5 mm) by 3%-inch (87.5 mm) grid superimposed on the upper
surface of a 24 by 28-inch (60 cm x 70 cm) panel. Fiake angle was dehed as the absolute angle
(ranging h m O0 to 90') between the axk of the cardinal (or aligning) direction and the long axis
of the fl ake. The numencal average of the m e a d angles was computed and alignment percent
was cdcuiated according to:
where
%A is the alignment percent; and
8 is the average alignment angle.
In a labour-saving effort, Geimer (1976) reported that the average alignment angle couid be
estimated with the percent of fl akes aligned within 20' fiom the cardinal alignment direction.
Equation [3] illustrates the least squares equation developed to estimate the average alignment
ange.
where
y is % flakes within * 20" of the cardinal angle; and - 0 is the average alignment angle.
Geimer (1976) determined that flake dispersion around the cardinal angle decreased as the
percent of alignment increased. From this relation, the standard deviation (or flake dispersion)
could be estimated using the least squares equation:
where
s is the standard deviation; and - 0 is the average alignment angle.
He concluded that average aligmnent angle could also be estimated f?om the ratio of MOE
paralle1 to MOE perpendi~uiar~ and was well correlated with linear expansion perpendicular to
the aligned direction. Consequentlyy the report noted that smaU increases in degree of alignment
created substantial increases in bending strength and stifhess in the aligned direction.
Numerous other studies have employed this method for measurïng and describing strand
alignment (Geimer 1979, Lang et al. 1982, Kawai et al. 1982, Laufenberg 1983, Kajita 1987,
Canadido et al. 1988, Sasaki et al. 1989,Yoshida et al. 1989b, Pulido et al. 199 1)
Lau (1980) and (198 1) employed a sampling method similar to Geimer (1976). Nmety-one (91)
flake alignment angles per 21-inch x 24-inch (52.5 cm x 60 cm) mat were meanired at the
intersection points of a superimposed plastic grid (3- x 1 '/-inch or 75 mm x 37.5 mm squares).
He characterized the extent of alignrnent by a normal distriiution, where standard deviation was
the definitive statistic. Equation [5] illustrates the normal distribution.
where
0 is the absoiute flake angle (O0 to 90' range); and
s is the esthated standard deviation.
The average absolute angle was related to the standard deviation by:
where
8 is the average absolute angle; and
se is the distribution standard deviation.
Harris (1977) and Harris and Johnson (1982) reported the use of directional distribution
hctions to describe flake orientation in experimental Meboard. The von Mises probability
distniution fimction @df) and a tnmcated nomal distribution were reported as quite nmilar, but
the truncated normal was discontinuous at the interval limits whereas the von Mises was
contllruous. This particular distribution was distinctively useful for chamterking angular
measurements. The von Mises pdf is shown in Equation [7].
where
g(û,p,~) is the strand ange pdf, a two parameter probability
distniution function;
p is the angle between the board axis direction, set at zero degrees, and
any chosen reference (e-g. axis angle of loading on bias-cut test
composites);
IC is the orientation parameter of strand angular spread; this is a
measure of angular concentration; and
I,(K) is the modified Bessel fûnction of order zero aven by the
polynomid approximation formula of Abrarnowitz and Stegun
(1965).
A rnarked departure h m the normal distniution methods presented by Geimer (1976) and Lau
(1980) rests in the expression of anguIar data by trigonometric moments and vectonal
representation. Axial &ta (O0 to 180') was coiiected and transformed to the 2x radians interval
by doubling the angles. The multiplication factor of two was required because the cumulative
fkequency of the von Mises distribution of a range of R radians is 0.5, rattrer than 1.0 (Shaler
1991). The distribution of anguIar data can be represented by a mean vector with angle p and
magnitude
The mean angle, p, is derived according to:
w here - sin8 is the average sine of the angles; and
cos0 is the average cosine of the angles.
The mean vector, i, is derived according to:
The von Mises concentration parameter, K, is non-luiearly related to the mean vector An
estimate of K is determined when R is h o w n , fiom the tabulations of Mardia (1972), in
Appendix 2.3, or Table B, found in Batschelet (1965). The relationship is illustrated in Figure
20.
A method for random measmement of flake angles was developed by Harris (1977) using direct
surface rneasurement principles. Photographie dides of both surfaces of the flakeboards were
prepared and projected onto a screen with 50 points randornIy distributed in the background.
The points served as sampling locations, and the orientation of the particles at each point was
measured to the nearest degree by placing the rotatable straight edge of a drafting machine
paraIlel to the longitudinal (grain) direction and reading the mguiar deviation h m a scaie at the
base of the straight edge. Data was collected on both sides of the boards to obtain a cornparison
of the two sides, yielding a total of 100 sarnple points for each board. Parameters characterizing
the extent of orientation for a variety of boards with pre-specified degrees of alignment were
v d e d using the measurement procedure. This sample size (100) was found to be sufficient to
spec* the stmnd orientation in the range of practical orientability (by hand or machine).
0.00 O. 10 0.20 0.30 0.40 0.50 0.60 O. 70 0.80 0.90
Mean Orientation Vector (R)
Figure 20. Relatiomhip of the von Mises concentration parameter (K) with the mean orientation
vector (R) (Data fkom Mardia 1972).
Suzuki and Sekino (1982) used the truncated normal distribution function proposed by Harris
(1 977) in a study of the effects of specific gravity and strand alignment on the elasticity of
oriented flakeboard The orientation angle of almost every visible flake on both surfaces of 1 1 -
x 1 1 -inch (27.5 cm x 27.5 cm) boards were measured. Equation [1 O] iIIustrates the tnincated
normal distribution bction.
where
f (8) is the orientation distribution fiuiction;
8 is the orientation angle of the M e ;
p is the orientation coefficient; and
ERF(Pd2) is the error hct ion (Abramowitz and Stegun 1965).
Higguis (1990) used the von Mises distribution to characterize s m d alignment in his
construction of bending moduli prediction models. Candidate strands were identified by
superimposing a clear plastic g i d with horizontal slots (paralie1 to the reference angle) on the
test specimens, randoxniy selecting sampling locations, and penrnarking strands through the dots.
Grain angles of rnarked strands were measured through a large viewing lem with a protractor.
Angles were taken fiom each side of the five replicate test boards f o e g a treatment (100 data
points).
Shaier and Blankenhorn (1990) used the von Mises distribution to describe the extent of
orientation in the developrnent of a mode1 to predict the fiexu~d modulus of elasticity of one-
layer onented flakeboards. Strand angles were measured at 100 randorn locations as specified
by Hmis (1977). The angular data was tested with a multi-sample Watson-Williams test and
the lever of orientation was found to be consistent among dl boards.
Subsequent shidy by Shaler (1 99 1) compared the von Mises and Geimer methods for rneasuring
strand alignment. One fïnding of Shaler's work was that the concentration parameter, K, and
percent alignment for a given average angle p were related nonhearly. Shder (1 99 1) offered
a table and a cornputer program to convert the von Mises concentration parameter (K) to percent
alignment.
2.4.2 Mechanical property (MOE, MOR) ratios
Geimer (1 976) reported a high correlation of stand alignment with the MOE and MOR of
onented strmd boards. Therefore, he conchded, ratios of rnoduli in the parallel and
perpendicular directions could be used as a qualitative representation of the degree of alignment.
Kieser and Steck (1978) used the ratio of MOR across the length and width of the boards as an
indication of the degree of orientation of which 3: 1 was a stated objective for onented panels.
The study reported that orientation did not influence vertical density profile and a cornparison
of mechmical properties could be based upon orientation aione (aU else being equai). In a study
of alignment with electrostatic foxmers Fyie et al. (1980) reported that an orientation index
de- by the ratio of MOE in the parallel direction to MOE in the perpendicular direction could
indeed be used to characterize aligment efficiency. Geimer (1 986) later descnbed a Iogarithmic
relation which enabled conversion of the MOE ratio to alignment percent. McNatt et al. (1 992)
employed the MOE ratio and Geirner's conversion equations in a study of aligning methods and
layer alignment combinations.
2.4.3 Stress wave velocity ratio
Use of mess wave velocities to dinerentiate aligment in wood products demonstrated another
effective utiiization of wood anisotropy. Stress wave velocities are highly correlated with
bending stifiess and could be used to nondestructively measure the modulus of elasticity of
materials. Peters (1983) used the Meîriguard Mode1 239A Stress Wave Timer to measure
parallel and perpendicular MOE of fully aIigned flakeboard and employed the MOE ratio to
characterize alignrnent.
2.4.4 Sonic velocity ratio
Wood exhibits anisotropic behavior with regard to acoustics. This implies that there are
differences in the speeds that s o d is transmitted in different directions, ie. longitudinal
transmission is superior to tramverse. Given known velocities in specific directions, the signal
çtmigth measured dong a certain direction wodd be representative of the degree of alignment.
Geimer (1979) reported the development of an equation using sonic velocity as an indicator of
alignment in a one-Iayer uniform-density board Fïake alignment was measured by the direct
d a c e measurement technique and by the velocity of a sonic wave. A 50 lrHz sound at intervals
of 0.1 second with a 1-to-100 on:off ratio was produced by a James V-rneter through a 3 by 9
in. (76 by 229 mm) section of the bending specimens. Measurements were taken in the
directions pardel and perpendicular to alïgnment. The technique of measuring aake alignment
by the percent of flake angles within * 20° of the cardinal direction was confirmed (Geimer,
1976). Again the relationship between and the standard deviation (s) was found to be
different £kom a nomal distribution. The sampling variance was found to increase with
increasing specific gravity and with increasing alignment The sonic velocity ratio of parallel
to perpendicular meaSuTements was highiy correlated with MOE ratios (padperp) and could be
desm'bed by a non-linear fiuiction. Velolocities rnûasured in the direction of testing (bending) had
more weight in determining the strength or stifhess values. Geimer mployed the souk velocity
ratio method in subsequent studies (1980, 1982 and 1986) of alignment effects on bending
properties and dimensional stability. Bucur (1992) expanded the study of this technique by
empioying dtrasonic fkequencies and measuring acoustic emissions.
Wood is more resistive to electrical conduction in the transverse direction than in the
longitudinal direction. According to this principle, applying a charge to wood substance in
different orientations to the grain direction wodd result in different signal magnitudes in a
receiver. Geimer et al. (1993) presented a new technique for detemiining flake alignment by
measuring the average grain angle of a panel. The Metriguard Mode15 10 Grain Angle ùidicator
(GAI) is an electrical capacitance-type device which uh&es the dielecrric properties of wood
to infer the degree of alignment in a board. The principle goveming its effdveness is that the
dielectnc constant of wood is greater dong the grain than across the grain, and the sensed
voltage signal changes as the sensor is rotated relative to the wood. Measurements generated by
the GAI were compared to measufements obtained using bending (MOE and MOR) ratios, sonic
velociw ratios, and direct surface measurements for severd target alignments. Alignment
measurements by the GAI and sonic velocity ratios were consistently higher, while those by
MOE ratios were consistently lower, than the direct d a c e measurement for targets of random,
30% and 50% alignment. The papa reported that, uniilce the other methods previously
described, the GAI was not iïmited to sUTface measurements and the depth of field penetration
was controllable. However, it may be more sensitive to board density, moisture content, and
sensor proximity to the board than angular rneasurements (direct surface).
2.4.6 Microwave attenuation
This method of strand aliment measurement is related somewhat to the electrical capacitance
method Once again capitiilizing on the anisotropy of wood, an applied electric field will cause
the wood substance to attenuate or "give off' microwaves. The quantity emitted by the wood
depends on the direction of the appiied field as defhed by the grain orientation. Musial (1988)
reported the use of rnimwave attenuation for characterization of the degree of flake alignment
in oriented strand board. The proposed technique utilized the dielectric properties of the wood
such that the anisotropy of microwave attenuation, given an applied electric field of constant and
known intensity, would Vary according to the anisotropic properties of the wood The measured
attenuation was reported to be closely correlated with the orientation of the flakes.
To nimmarize Section 7.4, six (6) methods for measuring or estimahg strand a l imen t were
identified: 1) direct surface mea~u~ement; 2) mechanical property ratios; 3) stress wave velocity
ratio; 4) sonic velocity ratio; 5) electrical capacitance; and 6) micro w ave attenuation. Direct
surface measurement was deemed to be the most accurate method. Several variations in the
sampling method were reported for orientation data collection. The von Mises directional
distribution fiinction was identifid as the best method for treatment of orientation data.
MATERIALS AND METHODS
3.1 Strand ali-gment measurement
Of all the methods avaiIable for rneasuring and describing strand alignment, the direct surface
measUrexnent @SM) technique was considered the most accurate. This wodd be the case in
panels produced with a homogenous rnix of strand sizes through the thickness. However, this
is not the case with industriai production It was demonstrated in the preceeding chapter that
industrial forming machines establish a gradient of dirninishing strand size through the surface
layers, fiom the surface towards the core. Efficacy of the DSM technique is questionable in such
a case. For example, a typicd 23/32-inch- thick (18 mm) panel might have 13 strata of
individuai strands in each surface layer, and a single scan of the uppermost stratum may not be
indicative of the alignment of the other 12 strata. Al1 is not lost though, the technique may still
be exploited by forming the strata (represented by homogenous strand sizes) separately and
scanning each layer prior to fomiing the next stratum.
Past efforts to characterize çtrand alignment by direct d a c e measmement (Geimer 1976, Lau
1980, Harris 1977 and Higgins 1990) were very time-consuming. Strand angles were
p-gly measured by hand according to some sampling criteria In an effort to overcorne
these hurdles, a Forintek initiative (Grant 1 W6a) was dedicated to the development of an
automated strand alipmat mc'ilsur~~ment systern based on DSM principles and image analysis
(IA) technology.
One product of the IA development efforts was a semi-automated system, as depicted
schematicdy in Figure 21. The system can be broken down into 4 main modules: 1) the
camera, lem, lighting and hegrabber form the image aquisition module; 2) the computer,
monitor and video card comprise the main contrd processing module; 3) the Empix Imagine
NORTHERN EXPOSLJREa software coIlSfitutes the image analysis module; and 4) the
Microsof!t@ EXCEL@ spreadsheet software forms the staîistical analysis module.
In any LA application's operation, the camera digitizes the image and routes the signal to the
fknegrabber (the camera lem and lighting component control the quality of the acquired
image). The fiamegrabber, a card installed in the maidtame, "captures" the digital signal and
transforms it into a pixel-based image. This image appears on the monitor within the
NORTHERN EXPOSUREa window. Once acquired, the image may be manipulated in aoy
number of ways and various image characteristics cm be measured with the IA software. These
quantifications can be stored in a simple text file or exported directly to the EXCEL'
spreadsheet. The repetitive operations can be programmed into a macro so that the computer
would perform operations automaticdy at the touch of a button. Upon completion of a sarnple
set, another macro is ran in the EXCEL' environment to automatically perform the necessary
staîisticai andysis and report generation. Further analysis could be performed at the operator's
discretion and leisure, with data archived in EXCELm files.
1
NORTHERN EXPOSURE, EXCEL
Figure 21. Schematic of the image analysis system used in the measurement of strand
alignment.
Figures 22 through 25 illustrate the main operations of the strand alignment measurement
system. Figure 22 illustrates the NORTHERN EXPOSURE@ interface (window) with an
acquired image of the top surface of an oriented strand board mat The toolbox in the upper right
quadrant of the window houses a series ofbuttons which control various image manipulation and
analysis operatioos. The box in the lower right quadrant of the window houses a series of
buttons, which when activated, laimch automated rnacros. The tMatt' button Launches the strand
digrnent analysis procedure.
In keeping with the samphg techniques developed by Geimer (1976) and Lau (1980), an image
of a grid (1 1 by 1 1, with 1 O0 internal intersections) was superirnposed onto the mat image
(Figure 23). Strands occurring at the intersection points of the grid were chosen for
measurement- Lines were drawn by the operator dong the long edge of the selected strands
(Figure 24). Figure 25 iliustrates the lines isolated fkom the background image. This "tracing"
was assumed to represent the grain ar@e orientation, given the propensity of strands to break
into their final width geometry dong the grain. However, it shouid be noted that there are grain
angle deviations in some strands generated during the typical flaking process. Fortmately these
amounts are so negligible as to be statisticaily insignincant. This "line drawing" tool has a
feature which measures the orientation of the line with respect to some reference angle, in this
case - the fomiing direction (Figure 26). The orientation data obtained Erom each "strand
orientation" was automatically exported to the EXCELm spreadsheet. The NORTHERN
EXPOSURJ? macro has an irnbedded loop which r e q k s 100 measurements, coincidentally,
the same number of intersections of the grid and sample size required for accurate estimation of
strand aligment (as reported by Hanis 1976). When the NORTHERN EXPOSURE@ macro
finished running its programming, the EXCEL@ statistical analysis macro was launched and
numerous measures of orientation (mean angle, standard deviation, concentration parameter,
etc, ...) were derived fiom the data
Figure 23. 11 x 11 grid superimposed on mat image for strand sample selection
Figure 24. Edge delineation of selected strands.
Figure 25. Isolateci strand "edges" representing wood grain orientation.
Forming Direction
Figure 26. Angle measmement of strand orientation to the cardinal direction.
3 -2 Orientation parameter
Given the Înability to distinguish between angles separated by x radians (ie., 120° and 300° are
the same), the angles form a data set falling between O and 1 80". Data in the O to 1 80" interval
are cded axial data To use directional statistics, the axial data must be transfomied to a 2x
radian intemal. This required the angles to be doubled. The angles cannot be meaningfully
averaged arithmetically, but the associated complex points can be averaged (Higgins 1990). The
mean vector, &, was computed f?om the transformed data according to Equation [9].
The R vector has a mean angle and magnitude. The mean angle should always be O" (except
for random orientation) because of syrnmetcy about the cardinal angle (direction). The
magnitude, mging fkoom O (random) to 1 (perfect orientation), is a measure of concentration
around the mean angle.
The mean vector has been shown to be a suitabIe estimator for the von Mises concentration
parameter (Mardia 1972). Harris (1977) reported that the von Mises probability distribution
function (pdf), with concentration parameter K, was suitable for describing the dispersion of
strand angles about the most probable orientation angle. Harris (1977) and Higgins (1 990) both
demonstrated the usefidness of using the von Mises pdf in m o d e h g strength expectations of
oriented strand boards. Whiie this type of modelling will not be employed in the curent study,
the concentration parameter will be used to descnie strand orientation, so that future work may
incorporate the data into pro bability-based models.
The IC concentration parameter is related to the R vector nonlinearly (see Figure 20). Several
polynomid approximations of r have been offered in the Iiterature, but a plot of the relationship
(ftom Appendix 2.3 in Mardia (1972)) v e m the polynomid regressions showed that the
approximations were satisfactory ody for certain ranges of In light of this, the relationship
between R and IC was broken into three segments with separate regression equations for each
segment. Polynomid approximations were derived for segments 1 (Equation [Il]) and 2
(Equation [12]). Segment 3 was approximated by Mardia (1972) with Equation [13].
Equations [Il], [12] and [13] yielded t h e figure accuracies.
Several notable hdiugs resulted fkom the work by Harris (1977). The first was that a sample
of 100 strand angles was sufEcient to estimate the concentration parameter of oriented strand
board within the practical range of orientability. This ranged fiom K = 1.1 for a commercial
panel to IC = 9.0 for the most oriented lab panel. A second contribution was the validation of the
accuracy of eshating the mean angle p and concentration parameter K. This was determined
by produchg panels of hown K and applying the method to estimate K. A paired t-test revealed
no significant difference between the calibrated and estimated values at the 95% significance
level.
Kiggins (1990) determined the 95% confidence limi,ts for evaiuating the repeatibility of the
determination of the orientation parameter. In addition, he assessed and venfied the goodness
of fit.
The LA method for measwing strand angles was tested with a two-sample test developed by
Mardia (1972). Severai images with differing 1eveIs of alignment (K= 1,3 and 5) were analyzed
by two different operators and tested to determine whether there were any ciifferences between
sample concentration parameters. The nd i hypothesis of equality was not rejected at the 95%
significance level in each case. Concentration parameters for several mats were derived from
manual measurements (with a protractor) and with the IA system. Verification against manual
measurements likewise prompted the null hypothesis (no difference) not to be rejected at the
95% sigdicance level.
Strand a l i m e n t prediction akorithm
The objective of the curent study is to investigate the effects of the surface layer strand
alignment distribution. This required some method of controlhg the strand aiignment in each
layer. Isolating the strata by building the paneI, layer by layer, is easily accomplished. However,
controllhg strand alignment is more of a task. It was poçhilated that alignrnent could be
controlled by building mathematical models of the forming process which could be used to
predict the strand alignment resulting fkom variable operating parameters. The results of a
Forintek study (Grant 1997) has yielded such a model. The folIowing subsections provide an
overview of the F o ~ t e k study.
3.3.1 Experimental design
Plate gap, fiee-fd distance, strand length and strand width were identified in a Forintek report
(Grant 1995a) as the critical forming parameters controlling strand alignment in the production
of oriented strand board Response d a c e methodology was employed with a four (4) variable
Box-Behnken design experiment to study the effects of these controlling parameters on strand
alignment. The Box-Behnken design was useful because of its optimal design space coverage
and discrete variable Ievels (A more indqth description of experimental design appem in
Section 3.4.1).
The variable levels and uni& initially used were:
Strand width: {OS, 1.0, 1.5} (inch) (12.5,25,37.5} (mm)
Strand length: {2,4,6) (inch) {50,100,150} (mm)
Free-fd distance: {0.5,3.5,6.5} (inch) (13-5,87.5, 162.5} (mm)
Plate gap: {1 ,2 ,3 ) (ratio of gap width to strand width)
Subsequent analysis yielded a rimitecl range of achievable orientations. Therefore, a second triai
was perfomed with an extended range of plate gaps and nee-fall distances. The variable levels
of the new trial were:
Plate gap : (3.4, 5 )
Free-fdl distance: {6.5,9.5, 12.5} (inch) (162.5,237.5,3 12.5) (mm)
3.3 -2 Strand generation
Aspen roundwood was cut to 12-inch (30 cm) lengths and çawn Ien-&Wise to either L/t, 1, or 1 %-
inch (12.5, 25, 37.5 mm) in thickness. The 12-inch (30 cm) pieces were aimmed to remove
wane and sawn to either 2,4, or 6 inch (50, 100, 150 mm) lengths. The prepared blocks were
fl aked to 0.025-inch-thick (0.635 mm) strands with a CAE 6/36 lab disc flaker. Blocks were
positioned in the feed conveyor so that the flaker knives cut paraiiel to the grain, producing
strands of 1/21 1, or 1%-inch (12.5,25,37.5 mm) in width and 2,4, or 6-inch (50, 100, 150 mm)
in length. A suitable number ofblocks were ffaked to produce approximately 1 kg of dry funiish
for 9 dinerent strand sizes (as defined by combination of strand width and length). The strands
were spread diffusely on the floor and air-dried for severai days. The dry strands were Tyler-
screened to rernove matenal smaiIer than the treatrnent sizes.
3 -3 -3 Production of oriented mats
Seventy-four (74) 25-inch x 25-inch (62.5 cm x 62.5 cm) mats were produced with a Iaboratory
fomiing machine (Figure 27). Sample treatments differed by combination of strand length,
strand width, plate gap and kee-faII distance. Five hmdred (500) grams of stnmds were loaded
into the hopper and mbber belts conveyed the furnish to the orienting apparatus. The material
was evenly distributed across the beIt during conveyage by the action of two sets of scalping
rolls. A constant feed of hrnïsh was deposited by the conveyor onto the plate assembly, where
the reciprocating action of the plates facilitated materiai flow and deposition onto the forming
cade The caul was moved back and forth at a constant speed under the reciprocating plate
assembly to ensure a homogenous layering of the fumish. An image of the mat surface was
taken after forming and çtrand alignent was determined with the procedures outlined in
Sections 3.1 and 3.2,
Cornera - - Co mputer
AIignment apparatus top view
l
Mat top view A lignm en t apparatus f i0 n t view
Figure 27. Labontory fomiing apparatus used for onenting strands in an OSB mat. 3.3 -4 Strand alignment mode1
The orientation data was analyzed with the Stat-Easem DESIGN EXPERP statistical analysis
software. The mode1 for predicting the K concentration parameter was best descnbed by a
quadratic polynomial. This conclusion was based on differing performance in sequential model
sum of squares (SMSS) and lack of fit tests, the coefficient of determination (Et2), and predicted
residual sum of squares (PRESS). Lower p-vaIues for SMSS, smaller PRESS, higher p-values
for lack of fit and Iarger R2 signifieci a better mode1 (a more indepth description of these statistics
will follow in the next chapter). Diagnostic testing revealed that transforming K by the square
root improved mode1 performance. Insignificant terms @-value > 0.05) were detemiuied
through a t-test and omitted fYom the mode1 equation. The reduced quadratic polynomial is
shown in Equation [14].
where
FF = kee-fall distance (inches);
PG = plate gap;
W = strand width (inches); and
L = Strand length (inches).
The SMSS of the quadratic mode1 was tested and found to be significant with p c 0.00 1 (F =
86.53). There was no sipifkant lack of fit (p = 0.765, F = 0.77) and the mode1 had a coefficient
of detemination w) of 0.93.
One interesthg relationship described by the model and independently verified, was that more
slender strands (smailer width) were more highiy orientable. It was hypothesued and visually
verified that the wider sh-ands tendeci to have a more turbulent fiee-fdl and were more affected
by the action of the aiigning plates. The reciprocating action of the aligning plates and aK
resistance tended to cause the wider strandç to fa on edge and at an angle askew to the normal
(not straight down). Impact with the mat surface was chaotic and alignment was M e r thrown
off. Conversely, more slender strânds tended to Boat straight down and were deposited Batwise
(not on edge). This wouid appear to fly in the face of the industry practice of putting the larger
strands on the d a c e and the SmaUer strands towards the core. It is moa Wsely that this
relaîionship has never before been investigated, and is therefore, not hown. This subject will
be discussed M e r in the next chapter.
3.4 Experimentai panels
One main concem of the study was to mùnic industrial conditions and processes as much as
possible so that resuits codd b e more easily applied to industrial production. As previously
stated, industrial OSB mats are constructed with a gradient of sûand geometry through the
d a c e layer of the mat. The gradient is estabiished with the largest strzxnds at the outer surface
and progressing to the smakst strand &es at the inner strata of the d a c e Iayers. The method
proposed to ensure this gradient involves a multi-strata surface layer. However, given that there
are approximately 13 individuai strand strata in each surface layer of a 2302-inch-thick (1 8 mm)
panel, producing 13 different strand sizes and forming each strand Iayer would be costly and
impractical. A simplified approach was required.
The product of the flaking operation is a population of strands with relatively good control of
strand length and thickness, but with randorn strand widths. Results of a Forintek study (Grant
1996) has shown that the weighted distribution of strands by width approximates the normal
distriibution, The weighted average strand width will Vary, given prevaîling flaking conditions,
but tends to be approximately I -inch in normal circumstances. Likewise, the dispersion of strand
widths around the average varies with flaking conditions, but retains its normal distribution. A
graphical representation of the weighted strand width distribution is depicted in Figure 28. A
three-size distribution with 25 percent of the furnish having a %-inch (12.5 mm) strand width,
50 percent with a l-inch (25 mm) width and 25 percent with a 1 K-inch (37.5 mm) width was
chosen to approximate the normal stmnd width distribution. A 4-inch (100 mm) strand length
and 0.025-inch (0.635 nmi) mand tbickness, as typicdy employed in industry, were chosen as
geometric constants.
O 0.25 0.50 0.75 î .00 1.25 i.50 1.75 2-00 (in) O 6-25 12.50 18-75 25.00 31.25 37-50 43.75 50.00 (mm)
Strand Widîh
Figure 28. Weighted distribution of strands by width approximates the n o d distribution.
The experhental panels were designed with three (3) strata in each surface layer and 1 core
layer, making a seven (7) fayer board. The face-to-core layer ratio was 1 : 1 by weight. The strata
ratio for each suffie layer would follow the population distniution, with a 25 percent by suface
layer weight outer stratum of 1%-inch-wide (37.5 mm) strands, a 50 percent by surface layer
weight middle stratum of 1-inch-wide (25 mm) strands, and a 25 percent by surface layer weight
h e r stratum of %-inch-wide (12.5 mm) straads. The core layers were manufactured with a
homogeneous mix (no Iayering by strand size) of the three strand &es in proportions mimicing
the overall population (25:50:25).
The critical forming parameters seIected were: concentraiion parameter (K,) of lm surface laye1
stratum (outermost surfiace layer stratum), concentration parameter ( ~ 3 of 2"6 surface laye1
stratuni (middle surface layer stratum) and concentration parameter (K,) of 3" d a c e laye1
stratum. These variables were asçumed to be independent and not corrdated with one another.
3.4. I Experimentai design
Response surface rnethodology (RSM) was used to design a three (3) variable Box-Behnken
experirnent It (RSM) is used to quant@ relationships between one or more rneasured responses
and a number of input factors.
The Box-Bebnken design type uses three levels for eacb fàctor, a Iowa level (-l), a middle level
(0) and an upper level (+l). Coding (as represented by the integers in parentheses) reduces the
range of each factor to a common scale, regardless of its relative magnitude. Factor levels are
coded to facilitate data processing and caiculations, especially when comp uting squared t ems
and interactions. Actual variable leveIs were as follows:
Stratum 1 concentration parameter (KJ: {O, 1.25,2.5)
Stratum 2 concentration parameter (K& {O, 1.5,3.0)
S tratum 3 concentration parameter (15): {O, 2.3, 4.6)
The zero (O) concentration parameter denotes randorn orientation. The rrpper limit concentration
parameters appmximates the maximum orientability of that strand N e achievable with Equation
[14]. Note that the maximum orientability increases with a decrease in strand width.
The RSM program constructs designs in "Observation Order". For the identical design, the
observation order would always be the same. Each point was additionally given a "Design
Nimiber". Were a design point replicated, its design number would not change. To perform the
experiments, a random order, called the "Run &der1', was assigned by the program. Blocking
affects this run order. The order will be randomized within each "BIock", but in the current case,
one block was u s d
The RSM produces a mathematical model which could be used to predict a response. The
fouowing design features must be provideci to build a good mode1 (Stat-Easem 1992):
Enough uniaue design points to esfimate al1 the terms in the postulated model: linear,
quadratic or cubic. The terms multiply as the number of factors are increased in the
expairnent
Exîra unique design points, above what is needed for esfimahg the model and pure error,
to test how the model fÏts the data These points must be at locations in the design space that
are different fkom the model points. They are used in a "Lack of Fit" test for the model. At
least four of these extra points shodd be specifïed to give an adequate statisticd test.
Replicates of some design points to estirnate the experimentd, or pure, error. This is the
error expected in the response were the experiment repeated fÎom scratch. Typicaily, the
center point of the design is repeated, o h four or more times. This gives an adequate
estimate of the variation of the response and provides the number of degrees of fkeedorn
needed for an adequate statistical test of the rnodel. Other points in the design may be
duplicated should better estimates of the response be desùed at those areas in the
experimental space.
A full factorial design, giving complete coverage of the design space, wodd require 33 = 27
unique design points. With the 4 extra repetitions stipulated in 3. above, the total samples
required would be 3 1 - this is the ideal design. However, t h e and materid cost constraints
prornpted a more optimum design. The Box-Behnken procedure creates designs with desirable
statistical properties, but, most importantly, with only a hction of the experiments required by
other design types. The experimental design is illustrated in Table 1. Seventeen (1 7) design
points were specitied by the program. Two additional ones, representing the upper ({K,, ic,, K3 J
= {H, +l, +l )) and lower ({K,, K, k3} = (-1, -1, -1)) k t s of orientability each strata, were
included for direct empirical cornparison (rather than predicted values) and greater mode1
precition.
Table 1.
Box-Behnken Design for Study of Strand Alignrnent Effects in OSB
DSN
Obs indicates the observation number
ûrd indicates the nui order
Blk indicates the block number ' DSN ID indicates the design number
3 -4.2 Strand generation and drying
Green rough-sawn Cinch by rlinch (100 mm x 100 mm) Aspen lumber was obtained &om a
sa& situated just outside of Quebec City. The 4 x 4 lumber was ripped lengthwise into L/2,
1, or lXinch (12.5, 25, 37.5 mm) thicknesses and sawn to 4-inch (100 mm) Iengths. The
prepared blocks were flaked to 0.025-inch-thick (0.635 mm) strands with a CAE 6/36 Iab disc
flaker (see Table 2 for specincations and settings). The scoring knives and reactor bars were
removed and a non-aggressive comteriaiSe angle was used to minimize damage to the strands
durkg flaking. Damage tends to promote the splitting of strands into ander sizes, which leads
to a loss of geometry control. Blocks were positioned in the feed conveyor so that the flaker
laiives cut parailel to the grain, producing strands of %, 1, or 1%-inch (12.5,25,37.5 mm) in
width and 4 inches (100 mm) in length. The strands were spread diffusely on the noor and air-
dried for several days. The dry sh'ands were Tyler-screened to remove material smaller than the
treatment sizes and stored in 2-ply plastic bags. Immediately pnor to blending, requisite
quantities of strands were dried to 6% moisture content for the suface Iayers and 4% for the core
layer with a forced-air drier.
3.4.3 Adhesive and wax blending
Four different blends were necessary to produce each panel (although enough funiish was
prepared to produce four panels at a time). The three dinerent strand sizes for the surface layers
had to be blended separately. The core fumish was rnixed together in the drying process in the
requisite quantities (25:50:25). Molten slack wax was applied to the strands on a 1.5% oven-dry
wood weight basis. Following wax application, a quantity of powdered phenol-formaldehyde
resin, equal to 2.5% oven-dry wood weight, was introduced into the blender. The blender was
ailowed to nui for 5 minutes to ensure sufficient mixing and good resin coverage. Wax is
sometimes omitted when there are no dimensional stability considerations. However, this is only
done when the adhesive is in a liquid form (ie., Iiquid phenolic or isocyanates). Resins in a solid
form requin some additive or "wetting agent" to improve the initial adherence to the wood
surfaces or rather, to give something for the resin to stick to. The Neste BD 804 powder resin
was used in the surface Iayers and the Neste BD 951 powder resin was used in the core Iayer.
Resin types dinered in their speed of cure.
Table 2.
Flaker Machine Specifications and Senings
Disk diameter
Power
Disc RPM Feed rate
Number of laiives
Length of W e
Cutting length
Scoring W e setting
Wedge ( M e ) angle
Counterknife angle
Reactor bar angle
Pressure lip angIe
Knife protnision
Distance between laiife and counterknife
Gap (between bife tip and pressure lip)
(175 mm)
(150 mm)
0.025 inches (0.635 mm)
0.236 inches (6.0 mm)
0.140inches (3.5mm)
3.4.4 Fonning oriented strand board mats
The apparatus depicted in Figure 27 was used to form 21-inch by 24-inch (52.5 cm x 60 cm)
oriented strand board mats. The strands were orimted parallel (except for randorn orientation)
to the 24-inch (60 cm) axis of the panel. The plate gap and k - f a l l distance were set according
to Equation [14], which, in combination with the strand size, would ensure the strand orientation
required by the experimentai design, The first layer of 1%-wide (37.5 mm) strands was formed,
with several adjustments of the forming caul platform height to maintain a constant f?ee-fall
distance. An image of the layer was taken and the strand alignrnent was measured according to
the procedure outlined in Sections 3.1 and 3 2 (as verificatîon). The plate gap and fke-fall were
re-adjusted to specifications for the second layer of Linch (25 mm) strands, and later for the
third layer of %hch (12.5 mm) strmds. The core layer, being random oriented, did not require
the use of the forming apparatus and was felted randomly by hand. The final three layers were
formed in similar fashion to the first three Iayers, but in the opposite order (%, 1, 1 %-inch or
12.5,25,37.5 mm strands). Surface strata in the panels designated as random (IC = O) were also
felted randomly by hand.
3.4.5 Hot pressing
The pressing process is one of the most important steps in panel production. Perfect controol of
processing parameters pnor to this stage of panel production wouid be moot were the press
"cycle" not properly designed, The conventional hot press serves two basic functions: 1) to act
as a source of heat for resin cure; and 2) to bring the wood material into close contact so the
cured resin will hold the panel shape d e r pressing. This may seem like a simple process, but
it is not. There are a number of factors to take into account. However, a detailed andysis of
pressing parameters is not required by the objectives of this study, so only two requirements will
be discussed.
The first point of discussion is the vertical density profile of the h a 1 product. Wood is a
viscoelastic material. This means that part of the deformation resdting form applied heat and
pressure is irreversiile. During pressing, the combination of heat and moisture soften the wood
material and make it plastic. Applied pressure will cause the wood cells to coIlapse, and the
wood material becornes compacted, or "densified". The cured resin wiU hold the wood in that
densified form. This densincation is not necessarily a bad thing. In fact, it increases the strength
of the nnal product. Comrnon OSB manufacturing practices calls for a compaction ratio of
around 1.5 (ratio of £inal deLlSity to initial wood density). Additionally, the proper manipulation
of closing speed, face moistirre contait and press temperature will remit in a nonruiifom densi&
profile through the thickness of a panel with higher density faces and a lowa density core ('HSU
1995). This strategic "densification" follows the strength requirements of the panel as outlined
in Section 2.1. It is not, however, the goal of this study to opfimize the pressing parameters. The
purpose is to investigate the effects of strand alig~~ent. The effects of the density profile on the
strength and s M h e s s properties may mask or misrepresent the actual contributions of strand
alignment, For instance, Ge- (1979) found that the impact of strand alignment on MOE and
MOR v&ed non-Iinear1y with panel density. Therefore, the press cycle employed in thk study
will seek to rninimirre the creation of a vertical density pronle.
The creation of a density profile has other significant drawbacks. Too steep of a profile will
r e d t in increased thickness swelling of hi& density Iayers. Additiondy, there may not be
enough contact pressure between core strands to effect good stsand-to-strand bonding durùlg
resin cure. This leads us to the second requirement of our press cycle. Good strand-to-strand
contact is desired at the precise time that the temperature at the contact plane reaches the resin
cure temperature. Optimum contact pressure for each strand-to-strand intenace would require
very little differential densification through the thiclmess. This again necessitates a uniform
vertical density pronle.
It was mentioned earlier that the proper manipulation of closhg speed, face moisture content and
press temperature would result in the creation of the verticaI density pro file. Conversely, their
proper manipulation would also result in a more irnifonn density profile. A slow closing speed,
represented by a lower applied pressure, would not cause as much coilapse of the wood materid.
The moistue would be driven towards the core and the d a c e material would dry out and iose
much of its plasticity. The cure temperature of the resin would be reached and the strands would
be glued in their uncolIapsed state. Of course, a certain degree of coilapse would occur, but not
as much. A slow closing speed would also slow the temperature nse in the panel because the
distance for heat transfer would be greater (Smith 1982).
A lower face moisture content would inhibit the creation of a steep density profle. The wood
wouid "dry" out more WcHy and a Iowa quantity of "plasticizef (steam) would be present to
promote coiiapse. The Iower face moi- content would also r d t in a lower vapour pressure
différentia1 between the face and core materid (which partially drives the migration of the seam
and promotes collapse).
F d y , a lower press temperature wodd slow down the transfer of heat through the board. The
lower tramfer of heat would also slow down the vaporization of the water, which has an effect
on the plasticity of the wood The m a s (steam) and heat transfer processes can, by themselves,
cause the wood celk to collapse with a minimum of applied pressure. Therefore, it is a good
strategy to reduce their rate of migration. Too slow of a tram fer of heat and mas, on the other
hand, is not good for the panel either. Productivity, while not a great concem in research, would
be very Iow with the long press times required to heat the center of board to resin cure
temperature. Given OUT desire to emulate industrial practices, this strategy must be optimized.
A long press time wouId also result in excessive thermal degradation to the outer surface
material. Therefore, a compromise must be made. Some degree of density dinerential between
the Iayers of the panel must be endured to allow reasonabIe press times and minimai thermal
damage.
A m e r constraint is reproductibility of the press cycle. Control by pressure is not feasible
because mat bdk density and moisîure variations could cause excessive collapse in some boards.
There would be dissimilarities in the density profiles between boards and this could skew the
results of mechanical tests. Therefore, the press cycle must be controiled by position, so that
closing speed would be ngidly controlied. Slight density variations may occur within the panels
themselves, but these will be marginal if the cycle were designed with a number of srnail
position steps.
Several trials were ran with a computer-controlled 24 x 24-inch (60 cm x 60 cm) hot press to
determine the optimum closing speed and press time for 23/32-inch-thick (1 8 mm) panels with
face moime content of 6 percent and core moisture content of 4 percent The press cycle was
initiated when the dayiight, or position, of the press reached I 15 percent of the final thichess.
The fast close to 1 15 percent of final thickness was approximateIy 3 seconds. The cycle had a
2 minute closing time (to thickness h m 1 1 5 percent of h a 1 thickness), 3 minute holding time
(at thichess) and a 1 minute ventllig/decornpression step. Appiied pressure, mat thickness and
core temperature were recorded throughout the cycle. A plot of the pressing profile, with
changes in appiied pressure, mat thiclmess and core temperature, is illustrated in Figure 29.
Figure 29. Press profile used in the production of oriented strand boards.
Seventeen (1 7) 23132-inch-thick (1 8 mm) panels were produced with d a c e layer strata oriented
according to the experimental design. Two additional panels were produced with surface layer
strata randomly-aligned and fully-aligned. A screen was used on the bottom surface to facifitate
venting (decompression) of the hot gases (steam).
3 A.6 Panel evaluation
Flexural propertïes were determined according to the Canadian Standards Association CSA
043% 1-93 test method and the American Society for Testing and Materiais ASTM D 1 O3 7-93
standard test method.
The standards specined a specimen length of 2 inches (50 mm) plus 24 times the nominal
thickness, or 19.25 inches (489 mm) for 23/32-inch-thick (18 mm) board. Specimen width was
3 inches (75 mm). Six (6) test specirnens were extracteci fiom each panel, with the surface layer
wood grain running parailel to the long axis of the specimens. Specimens were allowed to cool
and condition for one &y in a climate-contrdled room at 20°C and 65% relative humidity. Due
to delays in preparation of the hot press and time comtmhts for completion of the study, the
specimens were not conditioned to the equilibnum moisture content. The specïmens were tested
in static bending with an Instron 1 120 £2 (Revised) test machine. A crosshead speed of 0.345
i.ch/minute (8.8 d m i n ) and test span of 17.25 inches (439 mm) were used.
Moisture content was detemineci &om a 3-inch by 6-inch (75 mm x 150 mm) sarnple taken fkom
each test specimen, and the specific gravity was computed fkom the dimensions and weight of
the bending test specimen at time of test and moisture content.
The apparent modulus of elasticity WOE) and moddus of nip ture (MOR) were detemiined for
each test specimen. The average was computed for each treatment and used as the response for
the modeling effort.
3.4.7 Response surface methodology
Response d a c e methodology, via the DESIGN-EXPERT@ program, was used to relate the
study parameters (concentration parameters of the 3 sutface layer -ta) to the responses (MOE
and MOR). The program fits linear, quadratic, and cubic polynornials to the data. The rnodel-
fitting algorithm in the DESIGN-EXPERT@' program uses Householder transformations to h d
a QR factorkation of the design ma&. AU computations were canied out on a standardized
(coded) version of the design matrix to avoid numerical instabilities as much as possible (Stat-
~ase@ 1992). The three orders of polynomials fit to the data were of the form:
Linear rl = a, + 2 ay,
DESIGN-EXPERP provides a nunmary analysis of variance (ANOVA) table and other useful
statistics to compare the fitted polynomials. Six main statistics were evaluated in determining
the optimum polynomial to use in the m o d e h g efforts. The sequential model sum of squares
(SMSS) and lack of fit tests were conducted and the root mean squared error (MSE), coefficient
of determination (R2), adjusted coefficient of detennination (Adjusted R3 and predicted residual
sum of squares (PRESS) were computed for the hear , quadratic and cubic polynomials fitted
to the responses. The SMSS demonstrated how te= of inmeashg complexity contributed to
the total model. Models were not rejected if the probability of such a large F statistic was less
than 0.05 (signincant at the 95% significance level). The "Lack of Fit Tests" compared the lack
of fit error to the pure error fiom replicated design points. Were there a significant lack of fit,
as demo~l~trated by a low probability @ c 0. l), then the model would not be used as a predictor
of the response. Root MSE estimates the standard deviation of the error in the design - it should
be srnall. The R2 and Adjusteci R2 gave the fkaction of total variation that was attnbutable to the
model, or more sïmply pu& the percentage of the response variation which could be explained
by the mode1 variables. Values doser to mity signifïed a betîer model. The PRESS was a
measure of how a particular model fit each point in the design. A low PRESS value was desired.
Several diagnostic tests were performed to determine w hether the statistical assump tions
underiying the andysis of variance were satisfid A normal probability plot of the studentized
residuals was used to indicate the nonnalcy of the error term. Departure fkom a straight iine
would indicate non-normaiity. A pIot of the studentized residuais against the predicted values
was also a good indicator of the validity of the ANOVA model. The plot should exhibit a
random scatter around the zero he. A plot of Cook's distance, which is a measure of the effect
of each point on the model, indicated whether there were any significant departures from the
model relahonship. Data points with hÏgh Cook's distance relative to the other data points should
be deleted fkom the experirnent. F d y , a plot of the leverage, which is a measure of how each
point influences the model fit, was performed to check for any overly influentid points.
Leverages of 1 mean that that point controls the mode1 - leverages of 1 should be omitted fkom
the model or more data points should be inciuded.
The significance of the factors for the chosen model were determined with a t-test, which verses
whether the coefficient was different fÏom zero. The associated p-values (Rob > t) are
interpreted as the probabib of getting a coefficient as large as that observed, when the tnie
coefficient equals zero. A factor was considered signifïcant if the p-value was l e s than (c) 0.05
(95% significance level).
To summarize, the methodology for s tmd alignment meaSuTement and treatment of orientation
data was detailed. A strand aiignment prediction algofithm, created by Forintek Canada
Corporation (Grant 1 997), was used to determine the forming machine parameters required to
achieve target strata strand orientations. Both the experimental design for the onented saand
boards and procedures for the production and evaluation of the experimental panels were
reported Finally, details were given conceming the rnodeling and statistical testhg of those
modeis for relating the çhidy parameters (strata orientations) to the responses (MOE and MOR).
RESULTS AND DISCUSSION
4.2 Strand orientation
Strand orientation for each of the three d a c e strata were controued by £king the plate gap and
fiee-fall distance of the strand alignment machine. A predictive model (discussed in Section 3 -3)
which relates key fomiing parameters (plate gap, hee-fall distance, strand Iength and width) to
strand orientation was used to determine the optimum settings. The strategy was to minimize
the changes in plate gap settings, as this alteration was very rime-consurnîng, and to instead rely
on varying the fkee-falI distance for each treatment- With optimal se-s determined, a number
of test runs were paformed for each strand size and orientation level. The measured orientations
were tested to see if they were consistent h m one nin to another and if they were significantly
different fkom the targets.
4.1.1 Equality of concentration parameters
Onenteci strand mats were constructed in triplicate (3 repetitions) with target orientations equal
to the middie (0) and highest (+l) levek of orientation for each of the 3 stratum specified in the
experimental design. Mats were not produced with the lowest levels (-1) of aiignment as they
were formed by hand and did not make use of the model. The strand orientation for each
m e n t (a c o m b ~ o n of strand size and orientation) was determineci by methods reported in
Section 3.1. The average orientations of the heatments, represented by the concentration
parameter (K), are compareci with target orientations in TabIe 3. There appeared to be several
instances where the discrepancies were quite Luge (ie., 6.39 for the highest leveI (+1) of
digrment for the %-inch-wïde (12.5 mm) strands). However, Harris (1 977) showed that Iarge
Merences in K, when K is large, should not be viewed with alarm because of the exponentid
dope of the cuve (Figure 20) when R > 0.80. Statistical tests may be used to determine
whether the target and actuaI vaiues were si@cantIy different.
Table 3.
Target and Actual Orientation (Concentration Parameter) of Experimental OSB Mats
Strand Size %-inch (12.5 mm) 1 -inch (25 mm) l %-inch (37.5 mm)
ALignment Level O 1 O f O 1
Target K 2.30 4.60 1 .50 3 .O0 1.25 2.50
A two-sample test for equality of concentration parameters was developed by Mardia (1972).
It can be used to evaluate whether there were any significant difference between target and actual
values. This test has different forms depending on where the mean orientation vector (E) f d s
within the O to 1 range. Cnticd boundaries were identified at K = 1 and 2, or R = 0.45 and
0.70. The problem was considered by testing:
The R for the test nui replications were averaged and the resultuig R was used to compare
with the target.
Case L For R c 0.45, trader K,, w e use:
where
and
The critical region for Equation [18] consists of the equal tails of the standard normal
distribution,
Case II. For 0.45 s R s 0.70, under H, we use:
where
and
The critical region for Equation [20] consists of the equai tails of the standard normal
distn'bution,
Case III. For k > 0.70, under % w e use:
The critical region for Equation [22] consists of the equal tails of the F-distribution.
The complexity of these tests of e q d t y merit demonstration. Examples are provided for each
test case.
Example 4.1. la The R of the middle orientation (0) for the 1 %-inch-wide (37.5 mm) strand
straturn feu in the R r 0.45 range. The test case O illustrated in Equation [l8] was appropriate
for dis range. It was found that:
g,(ga,d = 0-4161,
The value in the denominator of Equation [18] was
g,(li,,,) = 0.7064
0.2500. Consequentiy, the cntenon was
2.3228, which is greater than 1.96, the 95th percentile of the standard normal distriibution (Kvanli
1988). Therefore the null hypothesis of equaiity was rejected This would impiy that the mode1
was not valid for this particular combination of forming parameters.
Example 4.1.1 b. The middle orientation (0) for the 1 -inch-wide (25 mm) strmd stratum fell
within the 0.45 s R ?; 0.70 range, and
Since 0.45 s R s 0.70, we use test case II (Equation [20]), and h d that:
The value of the denominator in Equation [20] was 0.1283. Consequently, the value of the
criterion was 0.2846, which is less than 1-96, the 95th percentile of the standard normal
distribution. Therefore the null hypothesis of equality was not rejected.
Example 4.1. lc. The remahder of the orientation treatments fell within the R > 0.70 range.
For the highest (+l) orientation of the %-inch-wide strand stratum, we have:
where
Since R > 0.70, we use test case III (Equation [21]), and calculate an F-statistic of 0.4174. The
2.5% value of F,,, is 1.51 (Kvanli 1988). Shce F,, < F ,, , we do not reject the null
hypothesis of equality. The remaining orientation treatments were tested in similar matter. .
Table 4 dispiays the results of the tests for equaliry between the actual and target orientations.
The results of the equality tests verify that the empirical model is valid for controlling strand
alignment Only one case proved to be significdy different h m the mode1 predictions. It was
noted in Section 3.3.4 that orientability decreased with increasulg strand width and that control
over the deposition of wide strands was chaotic. However, this anomalous behavior should have
been accounted for with the non-hearity of the polynomial equation. One possible explmation
could be the incomplete coverage of design space inherent to optimal experimental designs (Box-
Behnken). Were an extraordinary combination of factors, resulting in a significant departue
fiom the mathematicaily-established trends, not included in the experimental design, the
h m the mathematicdy-established trends, not included in the experimentai design, the
modehg effort wouid have missed or extrapolated past it. Verincation with the design points
did indeed show that the combination of factors used Ïn the present experiment were not inchdeci
in the construction of the strand alignment prediction model.
Table 4,
Two-sample Tests for Equaiity of Concentration Parameters (IC)
Strand Size %--inch (12.5 mm) 1-inch (25 mm) 1 %-inch (3 7.5 mm)
ALignment Lever O 1 O 1 O 1
Test Case a
Distribution
Criteria
S taîistic
do not do not do not do not do not rejec t
reject reject reject reject rej ect
" specines the critena as defked by range of R specifÏes which distribution type the critena foilow: F signifies the F-distribution; N
signifies the standard normal distribution.
Another possible explanation codd have been that the strand orientation of the cument test trials
were improperly measured. This possibility was summanly dismissed when re-analysis of the
mat images yielded in the same strand orientations.
There was one other possible explanaiion for the departure nom the model values. The strand
alignment model was built with trials conducted with constant strand sizes. The wider strands
were all of the same width (1 %-inch or 37.5 mm). It was duly noted that the strands used for the
lrst stratum (1%-inch-wide (37.5 mm) strands) in the curent experiment did contain a
significant amount of smaller materid (ie., Sinch-wide (12.5 mm) and 1 -inch-wide (25 mm)
strands). The volume of material required for the experimentaI panels was such that perfect
screening was not possible. The strand alignment mode1 predicts that, at the same length and
machine parameters, '/-inch (12.5 mm) strands would have a IC parameter of 0.79 and 1-inch (25
mm) strands would have a IC parameter of 0.92. It is not S c d t to see that inclusion of smaller
material would decrease the overd degree of orientation.
4.1.2 Homogeneity of concentration parameters
Mardia (1972) developed a set of tests to determine the homogeneity of concentration
parameters. Again, two criticai points were stipulated at R = 0.45 and 0.70. Consider the
composite hypothesis:
H , : K ~ = - - - = ~ c ~ = K , where~isunknown
Case 1. For R < 0.45, the criterion was defined as:
where
and g,(@,) was calculateci as in Equation [1 91.
Case II. For 0.45 r R s 0.70, under % we use:
where now
and g,(q) was calculated as in Equation [21].
Case m. For R > 0.70, under H, we use:
where
and
As with the tests for equality of concentration parameters, the tests for homogeneity are cornplex
and merit demonstration. Examples are given for each test case.
Example 4.1.2a Since R < 0.45 for the middle orientation (0) of the 1 %-inch-wide (3 7.5 mm)
strand siratun, we use the U, test (Equation [23]). Table 5 shows the relevant calcutations.
Table 5.
Calculations for Testing Homogeneity of K with R c 0.45
Total 384 160.1461 67.7238
Using the totals for Xwi , Xw, g,2, and Zwi g,, we calculate:
U = 67.7238 - (160. 1461)2 /384 = 0.9353
The 5% value of is 5.99 (Kvanli 1988), which is greater than U, therefore the concentration
parameters for the middle orientation level (O) of the l %-inch-wide (37.5 mm) strand stratum
may be regarded as homogenous. Given the significant diffaence between the actual and target
values, the actud K value of 0.70 wiU be used in the modeling effort.
Example 4.12b. Since 0.45 5 R s 0.70 for the middle orientation (0) of the 1-inch-wide (25
mm) strand stratu~, we use the U, test (Equation [25]). Table 6 shows the relevant calculations.
Table 6.
Calculations for Testing Homogenei~ of K with 0.45 s R s 0.70
Total 364.7073 498-0563 682.3532
Again, the 5% value of X: is 5.99, therefore the concentration parameters for the rniddle
orientation level (O) of the 1-inch-wide (25 mm) strand stratum may be regarded as homogenous
and the target K value of 1.50 wiU be used in the modeling effort.
Example 4.12~. Since 5 > 0.70 for the midde orientation (O) of the %-inch-wide (1 2.5 mm)
çtrand stratum, we use the U, test (Equation [27]). Table 7 shows the relevant calculations.
Table 7.
Calculations for Testing Hornogeneity of IC with R > 0.70
-- -
Total 2.296 0.3069
Using Equation [28], d was -0.00 12, and Equation [27] gives:
Again, the 5% value of X? is 5.99, therefore the concentration parameters for the middle
orientation level (O) of the %-inch-wide (12.5 mm) stratum may be regarded as homogenous.
AIthough the discrepancy between the target K (2.3) and actual K (2.5) was not large, the actual
value was chosen for use in the mode- effort.
The upper level (+l) orientations of aiI three strata had an R > 0.70. Testing with Equation
[27l showed that ail were homogenous. The achial K values were chosen for the upper Ievels
(+l) of the '/-inch-wide (12.5 mm) strand ( K = 11.0) and 1-inch-wide strand ( E = 4.3) strata
and the target K (2.5) was chosen for the upper level (+ 1) of the l '/-inch-wide (1 2.5 mm) strand
stratum in the modeling effort.
Establishg the homogeneity (or consistency) of strand orientation, resuiting nom certain
operating setcings, served three purposes. The k t was that it supported the existence of a
predictive model for controhg strand alignment. Inconsistency wouid thoroughly invalidate
the use of such a model. hcreased control through the use of such a mode1 wouid enable
researchers to expand the study of the impact of straud a l imen t on panel performance (as was
the case with the present study). It would dso enable OSB rnanufacturers to optimize forming
operations in pursuit of higher quaiity products and lower production costs.
Secondly, the greater degree of control would enable panels to be produced for this experiment
with orientations required by the experimental design. Certain design featwes must be provided
to build good models. A description of those design requirements c m be found in Section 3 -4.1.
Lastiy, estabiishing consistency of strand alignment for the chosen operating conditions (plate
gap, fke-fa distance, strand width and length) would elllninate the necessity for measuring the
alignment of each stratum in each panel. Aithough the strand aiignment measurement system
did substantially decrease the time required for evduation of a given mat Iayer, it still
represented a signincant hindrance to progress. As such, a s i g d c a n t cost saving was reaIized
by eliminating this procedure kom the subsequent panel prodtlction phase.
4.2 Board tests
Panel densities (Table 8), at moisture content, mged fkom 3 7.8 to 3 8 -5 Ib/@ (605 to 6 1 7 k9/m3).
Panel thicknesses were al l about 0.020 inches (0.5 mm) too thick, but this phenomenon was not
unexpected. Stop bars of 23B2-inch-tbickness (18 mm) were used to ensure that the panels were
not over-pressed during the pressing operation. Out-of-press "sp~gback" is a normal
occmence with OSB panels. The cellulosic matend has a tendency towards regaining its
natural fom (ie., cellular). Densification during pressing results in collapsed wood ceus, giving
the cells an ellipticd or "squashed" form. Removal of applied pressure wodd allow the wood
c e k Eeedom to resume their nahual circular form. Ambient rnoisture also promotes springback
as it is absorbed by the wood. The wood c e k will retain rnuch of their assumed form (squashed)
through the r e w g action of the cured adhesive and through permanent plastic deformation,
but some springb ack does occur. This "springback" exp lains the lower-than-target density .
Although the target density was 3 8 Ib/p (6 10 kgkd ), it was the target "basic" (oven-dry)
density. The density at moistue content (typically 1 - 2% for same or next day testing) is
usudy 0.5 - 1 Ib/ft3 (8 - 16 kg/m3) greater than the basic density. Moishue contents of the test
panels ranged h m 1.3 to 1.8% when the bending tests were perfomed, thus basic density of the
panels were roughly 1 lb /e (1 6 kg/m3) too light (again due to the increased final thickness).
Static bending tes& yielded evaluations (Table 8) of the modulus of rupture WOR) and apparent
modulus of elaçticity (MOE). MOR performance ranged kom 27.8 MPa for the totally random-
oriented panel to 55.8 MPa for the my-aligned panel (highest level of alignment in each surface
stratum). MOE performance ranged fkom 4635 MPa for the totally random-oriented panel to
7875 MPa for the Mly-aligned panels. There were several panels with lower degrees of
alignment that had comparable or higher MOE than the fWy-aligned panels. This phenomenon
will be discussed in detail M e r on.
Table 8.
Test Results for Panels (1 8 mm) with Variable Strand Alignment
' Results reported are an average of 6 test specirnens.
Figure 30 austrates the upward trends in MOR and MOE pdormance of OSB with increasing
surface layer alignment. Pan& wÎth =dom (-1, -1, -1) and fUy-aligned {+1, +1, +1} surface
Iayers represent the &grunent extremes in this case. Aligning each surfiace layer stratum to their
maximum lirnit of orientability d t e d in a 100% increase in MOR and a 70% incfease in MOE
&om the random-oriented condition.
{-ln 4. -t] {+ln +t, +t) Random Fulfy-Allgn ad
MOR
{-t, 4.43 {+f, +f, +i] Random Fully-A IJgnmd
Fipure 30. Modulus of rupture ('OR) and modulus of elasticity (MOE) for OSB panels
produced with difZering leveh of strand alignment in the d a c e layer strata,
Experimental desiwmodel inputs
As was discussed in Section 4.1.2, certain target orientations were replaced by the achiai
orientations. T h i s required an alteration of the experimental design (changing mode1 input
values). The new ranges for the factors are presented in Table 9.
Table 9-
Factor Limits Used in Modeling Efforts -
Factor -1 Level +l Level
The Box-Behnken design stipulates a O level, which in the factors' cases, wouid require 1.25 for
K,, 2.15 for K,, and 5.5 for K, . However, in lieu of M e r trials with the alignment model to
determine and ver@ the right settings to achieve these orientations, the actual orientations of the
test nins were used in the experimental design. For K,, the O coded level(1.25) was changed to
0.70. This also changed the coding of this factor level to -0.44. nie 1c2 rniddle level (0) was
changed fiom 2.15 to 1.5, thus changing the coded level to -0.30. The middle level (O) of the K,
factor was changed fkom 5.5 to 2.5, changing the coded level to -0.55. The actual factors used
were:
{O, 0.7,2.5}
{O, 1.5,4.3}
{O, 2.5, 11.0)
The change in the middle factor level could be perceived as a less than optimum combination
of factors for model development However, the new input values were sufficiently different
fkom the -1 and +l levels as to not compromise the validity of the model. The new experimental
design is shown in Table 10.
Table 10,
Revised Box-Behnken Design for Study of Shand Alipment EEects in OSB
Run DSN
Obs indicates the observation number
Ord indicates the nui order
Bk indicates the block number
DSN ID indicates the design number
4.4 Modulus of rupture NOR) mode1
Model-fitting and statistical anaiysis was pafomed with the DESIGN EXPERT@ (Stat-Easem
1992) software. While the program perfomed dl of the cdculations, the wer was responsible
for choosing the correct model, evaluating model performance and interpreting the results.
Linear, quaIlratic and cubic polynomials were fiîted to the data in an effort to relate the design
factors (K~. K 2 y K~ ) to the response (MOR). Foms o f these polynornials were displayed in
Equations [15], [16] and [17]. There were several tests which, in cornparison of the different
model performances, would assist in selecting the best model.
4.4.1.1 Sequential mode1 sum of squares
Table 1 1 displays the "sequential model sum of squares" (SMSS) summary table. This table
shows how temis of increasing complexity contribute to the total model (Stat-Easb 1992). The
most important rows are the ones headed by Linear. quadmtic and cubic. The "linear" row reports
the sequentid sum of squares for the lin= tennç (ie., A, B). The F value tests the significance
of adding linear terms to the intercept and the block effects. A small p-value (Prob > F) would
indicate that a d h g the linear terms impmved the model. The model was considered valid if the
p-value fell below 0.05 (95% significance level). The "quadratic" row shows the sequential sum
of squares for the quadratc terms (ie., A2, B2y AB). The F value tests the significance of adding
quadratic terms to the linear model. Again, a smaU p-value (< 0.05) would indicate that adding
quadratic tenns improved the model. The "cubic" row reports the sequential s u m of squares for
the cubic terms (ie., A3, B3, C3. ABC ). The F value tests the significance of adding cubic terms
to the quadraîic model, Again, a small p-vdue (< 0.05) would indicate that adding cubic terms
improved the model. The DF column provides the degrees of needom for each source.
Table I l .
Sequentid Model S u . of Squares for the MOR Models
Source Sum of Squares
Mean Square F-Value Prob > F
Mean 38277.3 1 3 8277.3
Linear 660.2 3 220.1 12-02 < 0.001 Quadratic 135.0 6 22.5 1.45 0.296
Cubic 58.8 5 11.8 0.58 0-719
Residual 80.9 4 20.2
Total 39212.1 19
The andysis of the SMSS enables a selection of the best degree of polynomial to describe the
relationship. The goal was to select the highest degree model with a p-value lowa than the 0.05
sigdicance level. In this case, the linear polynomial @-value c 0.001) was the only significant
model,
4.4.1.2 Lack of fit
The second phase of the model selection e n a s a comparison of tests of the fitness of each
model. Table 12 displays the "lack of fit" tests which diagnose how well each of the full models
Oower degree terms included) fit the data. As with the SMSS analysis, the most important rows
were the linear, quadratic and cubic ones. The nul1 hypothesis of this test was that the model did
not fit the &ta Therefore, a small F-value and large p-value (> 0.05) were desired in order to
reject the null hypothesis.
Both the linear and quadratic models displayed good fitness @value > 0.05), however, the hear
model proved to be the only siguikant one in the SMSS analysis. The cubic model did not fit
the &ta at alI because there was not enough unique data points to determine all of the terms in
the cubic model. Cubic models with three (3) factors will only be valid with 65 data points (this
experimental design had oniy 19 data points).
Table 12.
Lack of Fit Tests for the MOR Models
Sum of Mean Source DF F-Value Prob > F
Squares Square -
Linear 193.7 11 17.6 0.87 0.6 16
Quadratic 58.8 5 11-8 0.58 0.719
Cubic 0.0 O
Pure error 80.9 4 20.2
4.4.1.3 Surnmary statistics
The third and ha1 phase of the model selection entails a cornparison of other important model
statistics. Table 13 displays the analysis of variance (ANOVA) summary statistics of models
fit to the data. The important staîistics to scrutinize were the Root MSE (mean square error), RZ
(coefficient of determination), Adjusted R2 (R2 adjusted for the number of coefficients in the
model relative to the number of points in the design), and PRESS (predicted residual sum of
squares). Descriptions of these statistics are included in Section 3.4.7. In comparing models,
lower values of Root MSE and PRESS and higher values (closer to 1) of R2 and Adjusted R2
were considered superior. The cubic model was disqualified for reasons stipulated in the
p receeding paragrap h
Table 13.
ANOVA Summary Statistics of the MOR Models
Source Root MSE
Adj usted R2 PRESS
Linear 4.28 0.7062 0.6475 407.62
Quadratic 3.94 0.8506 0.7013 708.12
Cubic 4.50 0.9 135 0.60 16
Root MSE estimates the standard deviation of the error in the design. Consequently, we would
prefer a model with less variation. The quadratic model had a smaller Root MSE compared to
the linear mode1 and was, therefore, superior in performance with this statistic. However, we
must keep in mind that addition of the quadratic terms (SMSS - Table 11) did not improve the
modeI-
In cornparison of the abdute R2 and Adjusted R2 values, the quadratic model once again proved
superior to the Iinear model. The quadratic model explained 14% more of the response variation
(85% vernis 71%) in terms of the R2, however when adjustments were made for number of
coefficients, the differential decreased to 5% (70% versus 65%). C o m p a ~ g the effects of the
adjustrnents, the linear model Iost 6% of its explanatory power while the quadratic lost 15%.
Once again, we rnust keep in mind that the addition of the quadratic te rms did not improve the
modeI.
The PRESS is an evaluation of how well the rnodel fits each data point. The coefficients of the
model were calculatecl without the nrst point and the new mode1 was used to estimate the
misskg point. The residual was caldated and the procedure was repeated for ali the points in
the design. The PRESS is the srmunation of the squareci residuals. SmaLler values of the PRESS
indicate a better fit. The linear mode1 had a lower PRESS than the quadratic and was therefore
considered mperior for descniing the relationship between the model factors and the response.
To sumrnarite, the linear model had superior performance in the SMSS and PRESS tests and los
less expIanatory power (R2 - Adjwted R2) when adjusted for the number of coefficients in the
model. The quadratic mode1 had superior performance in the Iack of fit and Root MSE tests and
had higher overall vaiues of R2 and Adjusted R2. However, the insignificance of the quadratic
terms (as identified in the SMSS test) and higha PRESS value eciipsa its superior performance
in Root MSE, R2 and Adjusted R2. Therefore, the linear model was deemed to be the best model
for descrïbing the relationship between the rnodel factors (K,, q, 15) and the response (MOR).
4.42 Mode1 diagnostics
Diagnostic tests were performed to verify whether the statistical assumptions underlying the
analysis of variance were satisfied. Several useful diagnostic tests are illustrated in Figures 3 1
through 34.
A normal probability plot of the studentized residuals is illustrated in Figure 3 1. Departue fkom
a straight line wodd indicate non-nordty of the error term. Fortunately, that was not the case
with the current rnodel-
Figure 32 displays a plot of the studentked residuals aga& the predicted values. Absence of
any significant problems is indicated by a randornized scatter of points around the zero line -
as was the case with the current rnodel.
Figure 33 displays a plot of Cook's distance, which is a measure of the effect each point has on
the model, or more pointedly, a measure of how much the regression equaîîon would change
were a point deleted. A data point which has a very high distance relative to the other data points
may be an outlier, or one that "sticks out" fkom the others. The points were weil distributed,
indicahng that there were no problems with any one given point.
m .
A plot of the leverage associated with each point is illustrated in Figure 34. The leverage is a
measure of how each point influences the model f i t A value of 1 requires the model to go
through that point (that point controls the model). Leverages dose to 1 were not desired. As
with Cook's distance, Ieverages were weU distributed and none too close to 1.
DESIGN-EXPERT Plot Model: Linear
- A U U -
Response: MOR
-2.07 -1.48 -0.89 -0.31 0.28 0.87 1.45
Studentized Residual
Figure 3 1. Normal probabiliq plot of the residuals for the MOR modei.
DESIGN-EXPERT Plot Model: Linear
Response: MOA
' 35.8 39.5 43.3 47.0 50.7 54.4 58.1
Predicted as MOR in MPa
Figure 32. Plot of studentized residual versus predicted response values for the MOR model.
DESIGN-EXPERT Plot Model: tinear
Response: MOR
Run Number
Fiawe 33. Plot of Cook's distance of the data points for the MOR model.
DESIGN-EXPERT Plot Model: iinear
Response: MOR
Run Number
Figure 34. Plot of the leverage of the data points for the MOR model.
4.4.3 Model equation
The finaI equation in tams of actual factors was:
- MOR - 35.82 + 2.521~, + 2.555 + 0 . 4 6 ~ ~ ~ 9 1
Given the different ranges of factors, a direct cornparison of the effects of each factor is not
possible with the actual equation. This problem, however, can be rectified by expressing the
equation in terms of coded factors:
MOR - - 46.98 + 3.25A + 5.48B + 2.52C i301
where
A is the coded equivalent of K,;
B is the coded equivalent of K,; and
C is the coded equivalent of K,.
Both equations will give the same predictions, but the size of the coefficients in the coded
equation relate directly to the observed change in MOR According to Equation [30], the
orientation of the 2nd stratum had the most effect on the MOR, having 75% more of an impact
than the lrst stratum and 120% more of an impact than the 3rd stratum. The Irst stratum had
more of an impact than the 3rd All three of the factors had positive effects on the MOR. Plots
of the response surfaces in t e m of actual and coded factors are illustrated in Figures 35 through
40.
DESIGN-EXPERT Plot Response: MOR Model: tinear
Coded factors: X = K I Y =K2 Coded constants: K3 = 0.00
I I P o œ I . - I -0- v -*-a-
Fi,gure 35. Response surface of MOR in relation to coded values of K, and K,.
DESIGN-EXPERT Plot Model: tinear
Actual factors: X = K I Y =K2 ActuaI constants: K3 = 5.50
Response: MOR
Figure 36. Response surf"= of MOR in rekîtiotion to acîual values of K, and K,.
DESIGN-EXPERT Plot Model: linear
Coded factors: X =KI Y = K 3 Coded constants: K2 = 0.00
- 104 -
Response: MOR
Figure 3 7. Response surface of MOR in relation to coded values of K, and K,.
DESIGN-EXPERT Plot Modet: Linear
Actual factors: X =KI Y =K3 Actual constants: K2 = 2.15
Response: MOR
Figure 38. Response suditce of MOR in reIation to actuai values of r, and ic,.
DESIGN-EXPERT Plot Model: Linear
Coded factors: X = K . Y = K 3 Coded constants: KI = 0.00
- 105 -
Response: MOR
Figure 3 9. Response surface of MOR in relation to coded values of K, and K,.
DESIGN-EXPERT Plot Model: Linear
Actual factors: X = K 2 Y =K3 Actual constants: K I = 1.25
Response: MOR
Figure 40. Response surface of MOR in relation to achial values of IC2 and K,.
4.4.4 SigniGcance of mode1 factors
Relative size of the coded coefficients does not signw whether the coefficients (and factors)
themselves were signincant. This problem cm be addressed by testing the coefficients with a
t-test (displayed in Table 14).
Table 14.
Test of the Significance of the MOR Mode1 Factor Coefficients
Factor Coefficient
Estimate
Standard t for Ho :
Error Coef = O Prob > Itl
Intercep t 46.98 1 1 .O6 44.3 1
A (KI) 3.15 1 1.34 2.36 0.032
B ( ~ 2 ) 5.48 1 1.37 4.00 0.00 1
c 0%) 2.52 1 1.3 1 1.93 0,073
The t-test examines the null hypothesis of whether the coefficients were different Eom zero. A
zero coefficient would signify that a factor had no effect. With a 95% significance Level, oniy
the coefficients of the 1st and 2nd strata were significantly different f?om zero. Therefore, at
this significance level the 3rd strahim had no effect upon the MOR of the panel.
This fïnding does not corne as any great surprise if we were to consider the strength character
of a panel and the mechanism of faiiure. The MOR rneasures the maximum stress that a panel
may withstand in bending. Recali the description of stress distribution fkom Section 2.1
(illustrated in Figure 6). The strength requirements of a panel follow a haif-hourglass shape,
with maximum strength required at the d a c e and diminishing to zero at the neutral axis
(tenter). The 3rd stratum, being closer to the neutral axis, would contribute much less to the
strength of the panel than the lrst and 2nd strata. Furtbermore, the MOR would not depend
solely on the lrst stratum strength because of the mechanism of board failure. The outermost
stratum of wood material rnay fail and the area across which the stress acts would decrease, but
only by the amount of the surface fdure. The rest of the board materiai would continue to resist
the load (which wodd continue to increase according to the test procedure). It would not be
mti1 a sufncient load was reached, or until the cross-sectional area was reduced enough, that the
panel would fail catastrophicdy. Therefore, in a typical panel, the maximum strength, as
mcaçured by the test procedure, would be measiired at some depth below the actuai d a c e . In
our case, it wodd be somewhere within the 2nd stratum. The 3rd stratum would not matter in
the case of MOR because the cross-sectional area would have been too srnall to resist the Ioad.
There was another factor which had a very signincant impact on bending strengta One that was
not included as a mode1 factor, as evidenced by the Iess-than-perfect RZ value. The 70% R2 value
indicated that 30% of the variation in MOR performance of the panels was attributable to one
or more omitted factors. Some other factors which have a lmown impact on bending strength
are the average board density, vertical density profile and moishire content (Hsu 1995), to name
a few. The common siring to each of these variables are their relation to density. The average
board demis. is self-explanatory, the density profile is indicative of the density at the surface
(where the strength is most important), and moisture content has a counteracting effect on
strength. This supposition is easily tested with some of the replicated data points in the
experimental design - recd nom Section 3.4.1 that the Box-Behnken design called for 4
replications of the center point. Correlation analysis of the MOR of these panels with their
measured densities yielded a R value of 0.98, or an R2 value of 96%. A vaIue this close to unity
does not Ieave much room for debate on the impact of density.
Given such srnalI differences in density (ie., 0.2 lb/ft-' or 3 kg/m3) and such a strong correlation
with MOR, it is likely that the overall impact of density would, were it varied to a greater degree,
be a more influentid factor than strand orientation on the MOR What this observation really
underscores is the need for a more comprehensive study incIuding density and density
distribution as mode1 factors. Especidy since the underlying reasoning for this work was to
discover ways of reducing density without signincant impact to mechanical properties.
To summarize, dl statisuicd tests pointed to there being no problems with the underlying data
and we couId suppose that there was no correlation between the study parameters (strata
orientations).
4.5 Modulus of elasticitv (MOEI mode1
4.5.1.1 Sequentid model sum of squares
Table 15 displays the SMSS m a r y table for the MOE modeling. The importance of the
SMSS was descnied in Sections 3.4.7 and 4.4.1.
Table 15.
Sequential Mode1 Sum of Squares for the MOE Models
Source Sum of Squares
Mean Square
F-Value Prob > F
Mean 9.4203 x 108 1 9.4203 x 108
Linear 1.1046 x 10' 3 3.6819 x IO6 13.53 < 0-001 Quadratic 3.6523 x IO6 6 6.0872 x IO5 12.72 < 0.001
Cubic 2.4657 x 1 O* 5 4.93 14 x 10' 1 .O7 0.487
Residual 1.8397 x IO5 4 4.5992 x 10'
Total 9.5717 x 108 19
Again the goal with the SMSS was to select the highest degree model with a p-value lower than
the 0.05 significance level. In this case, both the linear and quadratic polynomials @-values <
0.001) were signincant, therefore we would choose the quadratic model because of its higher
degree (2" versus l0 for linear).
4.5.1.2 Lack of fit
Table 16 displays the "Iack of fit" tests for the MOE models. The importance of the lack of fit
tests was descrîbed in Sections 3.4.7 and 4.4.1.
Table 16.
Lack of Fit Tests for the MOE Models
Source S m of Squares
Mean Square F-Value Prob > F
- - -
Linear 3-8989 x 106 11 3.5444 x 105 7.72 0.032
Quachtic 2.4657 x 105 5 4.9314 x 10' 1-07 0,487
Cubic 0.0 O
Pure error 1.8397 x 105 4 4.5992 x IO4
The lin- model showed significant lack of fit @-vaiue < 0.05) and the null hypothesis was not
rejected (the nuil hypothesis was that the model did not fit the data). The quadratic mode1
displayed good fitness @-value > 0.05) and the null hypothesis was rejected. As was
encountered with the MOR model, the cubic model did not fit the data because there was not
enough unique data points to detamine all of the terms in the cubic model. Based on this test,
the quadratic mode1 would be the only acceptable one to use.
4.5.1.3 Summary statistics
Table 1 7 displays the analysis of variance (ANOVA) summary statistics of models fit to the
MOE data The importance of the surnxnary statistics was desm'bed in Sections 3.4.7 and 4.4.1.
As with the MOR models, the important statistics to smtÙ1i7:e were the Root MSE (mean square
enor), R2 (coefficient of detennination), Adjusteci R2 (R2 adjusted for the number of coefficients
in the model relative to the nirmber of points in the design), and PRESS (predicted residual sum
of squares). In comparing modeh, lower values of Root MSE and PRESS and higher values
(closer to 1) of R2 and Adjusted R2 were coflsidered superior.
Table 17.
ANOVA Summary Statistics of the MOE Models
Source Root MSE
Adjusted R2 PRESS
- -
Linear 521.7 0.7301 0.676 1 6.4276 x 1 O6
Quadratic 218.7 0.9715 0.943 1 1.9678 x 106
Cubic 214.5 0.9878 0.9453
The quadratic mode1 had a much smaller Root MSE (approximately %) compared to the h e a r
model and was, therefore, superior in performance.
In cornparison of the absoIute R2 and Adjusted Et2 values, the qIiiidratic model once again proved
superior to the linear model. The quadratic model explainecl 24% more of the response variation
(97% versus 73%) in terms of the R2, however when adjustments were made for number of
coefficients, the differential increased to 26% (94% versus 68%). Comparing the effects of the
adjustments, the linear model lost 5% of its explanatory power while the quadratic loa only 3%.
The quadratic model had a much lower PRESS (3 times less) than the linear mode1 and was
therefore considered superior for descniing the relationship between the model factors and the
response.
To sumrnarize, the quadratic mode1 had superior performance in aU the applicable statistical tests
and was considered to be the best model for describing the relationship between the model
factors (K,, K~, KJ and the response (MOE).
4.5.2 Mode1 diagnostics
Diagnostic tests of the MOE models are illustrateci in Figures 41 through 44. The normal
probability plot of the studentized residuals Îs ïüustrated in Figure 41. The scatter was
approximately hear and did not indicate any problems with the data
Figure 42 displays a plot of the studentized residuals against the predicted values. The scatter
of points around zero was random (without any recognizabie pattern) - a M e r indication of
data stabiliîy.
A plot of the Ieverage associated with each point is illustrateci in Figure 43. The points were well
distn'buted, although the upper limit of the range was high (0.77). The high leverage indicated
that including more data points would improve the model.
Figure 44 displays a plot of Cook's distance. The points were faVly well-distriiuted within the
range, with exception of one (0.542). Deleting this point does improve certain features of the
model (SMSS, lack of fit, R2, Adjusted R~, Root MSE and PRESS), however leverage of each
of the remaining points increased. A cornparison of the original and "optimized" models will
be detailed in the next section to choose the more significant model.
DESIGN-EXPERT Plot Model: Quadratic
Response: MOE
Studentked Residual
Figure 41. Normal probability plot of the residuals for the MOE model.
DESIGN-EXPERT Plot Model: Quad tatic
Raspanse: MOE
Predicted as MO€ in MPa x 102
Figure 42. Plot of studentized residual versus predicted response values for the MOE model.
DESIGN-EXPERT Plot Model: Quadratic
Response: MOE
DESIGN-EXPERT Plot Model: Quadratic
Fi-gire 43. Plot of the leverage of the data points for the MOE model.
Responsa: MOE
Run Number
Figure 44. Plot of Cook's distance of the data points for the MOE model.
4.5.3 Model optimi7ation
Optimization was p d o m e d by ornithg the data point idendfied by the large Cook's distance.
This modification did not change the model selection- Table 18 shows a comparison of the
summary sîatistics for both quaciratic modek
Table 18.
Cornparison of Summary Statistics for Original and Optimized MOE Models
Model @-value)
Lack of Fit (p-value)
Root MSE R2 Adjusted R2
PRESS
The optimized mode1 outperformed the original in every statistical comparison. Both models
had p-values < 0.001 for the model ANOVA, but the actual F-value of the optimized model was
larger than the original (37.60 vasus 34-14), indicating that it (optimized rnodel) was more
significant. The larger p-value in the Iack of fit tests indicate that the new model fits the data
better. The Root MSE (standard deviation) decreased by approximately 5% through
optimization. The gains in R2 and adjusted R2 were not great, being on the order of 0.5% and
0.8%, respectively, however it still represented a gain in explanatory power. The greatest gain
nom optimization came with a halving of the PRESS, which indicates that the new model berter
fit each data point (than the original).
Diagnostic testing was once again performed and r d t s for the optimized MOE models are
show in Figures 45 through 48.
The noma1 probability plot of the studentized residuals is iUustrated in Figrne 45. One can see,
when cornparing the scattering of points in this figure with that of Figure 41, that a more iinear
dispersion and tigher scattering of the error terms resulted b r n the optimi7ation effort.
Figure 46 displays a plot of the studentized residuals against the predicted values. The scatter
of points around zero was again randorn (without any recognizable pattern). There was
essentially no signifïcant difference between the results for the original and optimized models.
A plot of the leverage associated with each point is illustrated in Figure 47. The points were
again well distributed, however the upper limit of the range increased to 0.87. While lower
leverage was desired, there were no points with leverage equd to 1, therefore the new mode1 was
valid.
Figure 48 displays a plot of Cook's distance. The points were again weil-distributed within the
range, with no outliers. One cm e d y see the improvement when Figure 48 and Figure 44 are
compare&
DESIGN-EXPERT Plot Model: Quadtatic
flesponse: MOE-OPT
1 -
-1.86 -1.31 -0.75 -0.20 0.35 0.90 1.45
Studentized Residual
Figure 45. Normal probability p b t of the residuds for the new MOE model.
DESIGN-EXPERT Plot Model: Quadratic
Response: MO€-OPT
Predicted as MO€-OPT in MPa x 102
Figure 46. Plot of sîudentized residuals versus predicted responses for the new MOE model.
DESIGN-EXPERT Plot Model: Quadratic
Response: MO€-OPT
' i 3 5 7 9 i 13 15 i i
Run Number
Figure 47. Plot of the leverage of the data points for the new MOE model.
Model: Quad ratic
Response: MOE-OPT
Figure 48. Plot of Cook's distance of the data points for the new MOE model.
0.29t 4
0.1 9-
0.1 4"-
0.1 O--
0.05--
0.00
f
+ + +- -t
4 +
i + + + + +
T
1 3 5 7 9 ft 13 15 17
Run Number
4.5.4 Model equation
The final equation in te- of achial facto= was:
MOE -
The hal equation in terms of coded factors was:
MOE - - 8078.4 + 502.9A + 794.6B + 157.7C - 1 2 9. f A'
- 508.8B2 - 548.1C2 - 312.6A.B - 188.OAC - 25.8BC
where
A is the coded equivalent of K,;
B is the coded equivalent of K~; and
C is the coded equivaient of K,.
According to Equation [32], the orientation of the 2nd stratum (B, B3 had the most effect on the
MOE. The Irst (A) and 3rd (C2) strata dso had large impacts on MOE, relative to the other
fxtors, but not to the same degree as the 2nd stratum. The contributions of the quadratic tems
(ie., A2, B2 and AB) served to reduce the impact of the individual h e m te- (ie., A). This was
readily apparent fkom the sign (-ve) of the quadratic coefficients. The quadratic polynornial
gives the relationship a parabolic form when conceptualized 2-dimensionally. Three (3)
dimensional rendering may take a more complex form and to illustrate this, plots of the response
surfaces in terms of actual and coded factors are shown in Figures 49 through 54.
DESIGN-EXPERT Plot Model: Q uad ratic
Coded factors: X =KI Y =K2 Coded constants: K3 = 0.00
- 119 -
Response: MOEQPT
Figure 49. Response d a c e of MOE in relation to coded values of IC, and K,.
DESIGN-EXPERT Plot Model: Quadratic
Actuai factors: X = K I Y = K 2 Actual constants: K3 = 5.50
Response: MO€-OPT
Figure 50. Response SUrf'e of MOE in relation to actual values of K, and K,.
DESIGN-EXPERT Plot Response: MOE-OPT Model: Quadratic
Coded factors: X =KI 9.OEi03 Y =K3 Coded constants: 7.SEi03 K2 = 0.00
6.OEt03
4.5Ei03
Figure 51. Response d a c e of MOE in relation to coded values of K, and K,.
DESIGN-EXPERT Plot Model: Quadratic
Actual factors: X =Kt Y = K 3 Actual constants: K2 = 2.15
Response: MO€-OPT
Figure 52. Response surface of MOE in dation to acnial values of K, and K,.
DESIGN-EXPERT Plot Model: Quadratic
Coded factors: X =K2 Y =K3 Coded constants: K I = 0.00
- I L L -
Response: MOEQPT
Figure 53. Response d a c e of MOE in relation to coded values of K~ and K,.
Model: Quadratic
Actual factors: X = K 2 Y =K3 Actual constants: K I = 1.25
Response: MOE-OPT
Figure 54. Response surface of MUE in relation to actud values of ic2 and K,.
It shouId be noted that the response "dips" at combinations of maximum orientation in each
stratum. The actual test r e d t s of the experïmental panels showed that maximum MOE was
achieved not at full orientation in each stmhm (+l, +1, +l } , but rather at maximum orientations
in the lrst and 2nd strata and medium orientation in the 3rd stratum (+1, + 1, O}. Furthemore,
a MOE comparable to the maximum orientability (+l , +1 , +l ) was also found at (O, +l , + 1 } .
Two other observations were noted at this point The first was that the panel with the highest
measwed MOE (+1, +1, O} was denser (38.1 versus 37.8 Ib/ft3; 6 1 1 versus 606 kg/m3) than the
one with maximum onentability {+l, +1, +1}. The second observation was that the panel with
lower orientation {O, +1, +l } and the panel with maximum orientability {+1 , +1, +l } , both with
comparable performances in MOE, also had the same density (37.8 lb/p or 606 k#m3). These
observations wodd suggest that at some orientation threshoId the marginal gain in bending
stiffbess nom improvements in alignment wodd be limited and that discrepancies could be
attnbuted to dinerences in density. This hypothesis, while supported circumstantidly b y these
observations, is by no means a proven fact The differaices may have been caused by interaction
effects between the measured factors or some other h o w n influence. It is near impossible to
prove at this t h e without M e r experimentation. As with the MOR model, this question of
density effects underscores the need for a more comprehensive study including density as a
model factor.
4.5.5 Significance of mode1 factors
Relative size of the coded coefficients does not sipi@ whether the coefficients (and factors)
themselves were significant. This problem can be addressed by testing the coefficients with a
t-test (displayed in Table 19).
Table 19.
Test of the Sigdïcance of the MOE Model Factor Coefficients
Coefficient Standard t for& : Factor DF Prob > 1 tl
Estimate Error Coef = O
With a 95% significance level, only the A, B, B2, C and AB factors were significant. The
coefficients of the C, A2, AC and BC factors were not significantly di"erent h m zero (no effect
on MOE) at this çignincance Ievel. These factors, with the exception of C, may be omicted kom
the model. The C factor must be retained because of the model hierarchy (ie., the effect of C
would become negative ifonly the C2 term was included). On the other han& ail C factors may
be removed (ie., C, C2, AC, BC), but this resuits in a loss of model significance. The F-value
of the SMSS decreased from 37.60 to 32.15, the p-value of the lack of fit test decreased f?om
0.546 to 0.160, the Root MSE increased fkorn 208.1 to 325.5, the R' decreased nom 97% to
91%, the adjusteci RZ decreased nom 95% to 88%, and the PRESS increased Eom 1.046 x 106
to 3.060 x IO6 when ail C factors were ornitted Tt is usually a good idea to leave al l the factors,
regardless of sigdicance, in the equation so as not to compromise the o v e d tignificance of the
model (as was just demonstrated).
The low significance and impact of the C (5) terms did not corne as a surprise. As with the
MOR, the material contri%ution to bending stiffness (MûE) foilows the half-homglass charactm
show in Figure 6. The contribution to the MOE of the panel was highest at the surface and
gradualiy diminished to zero at the neutrd axis (core). The third stratum (represented by K,) did
not contribute significantly to the overd MOE of the panel. As with the MOR, the 2nd stratum
(KJ had the greatest impact and was the most significant factor. This phenornenon was more
likely due to the proportion of the materiai in the 2nd stratum. Had the proportions been
%:%:%, the In t and 3rd strata would have contributed more significantly to the h a l model.
This obsewation poses the question: what impacts to MOE could be expected were the strata
ratio varied fkom the 25:50:25 proportions by weight used in the current experiment?
In summary, ail statistical attributes of the MOE model suggesî that there were no underlying
problems with the data. One could suppose that the model variables were independant and not
correlated with one another.
4.6 Costhenefit analvsis
The p h a r y benefits of this study were the advancement of knowIedge regarding strand
alignment eff'ects on mechanical properties of OSB and the identification of areas for hture
research. There was a secondary benefit, albeit one not brought to immediate attention. This
research also provides a powerful tooi for optimizing industrial forming operations. Using the
models built in this study and others (referenced), mill personnel could simulate new operating
conditions O ff-line and evaluate optïmization efforts prior to implementation (trial-and-error
optimization is expensive!). We could build a fictionai scenario to demonstrate these benefits.
Were we to r e m to o u primary justification for thÏs study, that of hding ways of reducing
panel inputs (density), we may derive an estimate of the cost savings accmed through the
implementation of this research in our fictional OSB mill. Certain assumptions must be made
concerning production capacities and costs. Consider a typical OSB mill with an annual
production capacity of 360 MMSF (3/8-inch-thick basis), 190 MMSF (23/32-inch-basis), or
approximately 3 18,600 m3. The annual capacity is based on 350 days per year of production.
The raw material cost (ie., wood, resin, wax) attntbuted to 1 MSF of 23132-inch-tEck production
is $131-00 or $78.00 for I m3.
The mill utilizes Schenck-type foming machines with a plate gap of 2.5 inches (62.5 mm) and
fiee-fidl distance of 1.0 inch (25 mm). Strands were produced at a constant length of 4 inches
(100 mm) and random width. The weighted strand width distribution follows our assumption
of 25 percent of 1 %-inch-wide (37.5 mm) strands, 50 percent of 1 -inch-wide (25 mm) strands,
and 25 percent of %-inch-wide (12.5 mm) strands. Using Equation [14], with the given forming
machine and strand geometry parametes, we predict u, = 1 -9 for the 1 rst stratum, K, = 1 -9 for
the 2nd stratum and K, = 2.5 for the 3rd stratum. Using these orientations as inputs for Equation
[3 11, we predict an MOE of 8010 MPa for the panels. Considering that the data fkom which
these models were built represent ideai conditions and that milI conditions are decidedly not
ideal, we will consider OUI target standard as 8010 MFa (The actual CSA 0437.0 standard for
MOE measured parallel to the forming direction was 5500 for the 0-2 grade).
A min strand alignment improvement initiative planned to modify the forming heads to a
Siempelkamp-type design with graduated plate spacing dong the length of the fomiing head
according to the strand width distribution. Twenty-five (25) percent of the disk rolls were to be
set to a plate gap of 1.75 inches (43.75 mm). These rolls, located at one end of the folming head,
would be responsible for orienting the larger strands (3 7.5 mm or 1 %-inch in width). Fifty (50)
percent of the disk r o k were to be set to a plate gap of 1.25 inches (3 1.25 mm). These rolls,
located in the middle of the fonning head, would be responnMe for onenting the medium-sized
strands (25 mm or 1-inch in width). Lastly, 25 percent of the disk rolls were to be set to a plate
gap of 0.75 inches (1 8-75 mm). These mils, at the opposite end of the forming head (from the
large-gapped rolls), wodd be responsible for orienting the srnail strands (12.5 mm or %-inch in
"dth). Once again using Equation [14] and the new operating parameters, we predict K, = 2.3,
I C ~ = 2.9 and and K, = 4.2. Inputting these orientations into Equation [3 11 gives an MOE of 8513
MPa As a result of the improvement initiative, the bending stifibess would be increased by 503
MPa
Ernest Hsu (1995) described a h e a r relatiouship between board density and MOE. The
relatiomhip stipulated that a 1 lb/@ (16 kglm3) reduction in board density would translate k t0
a reduction in MOE of approxkately 200 MPa Were we to discount the interaction effects
between board density and strand orientation (ie., assume that there were not any), we could
reduce the board density by 2 Ib/P (fiom 38 to 36 lb/ft3) or 32 kglm3 (610 to 578 kg/m3) and
still exceed our target MOE (8513 - 2*200 = 81 13 MPa > 8010 MPa).
The bottom ihe of this optimization scenario, given our assumed production capacity and raw
material costs, is sun1~11arized in Table 20. Daily production would not change because it is
based on volume, not weight. However, raw rnaterial costs do depend on weight and can be
adj usted to reflect the new lower density - the unit cos& would be only 3613 8 (57816 1 O), or
94.7%, of the original c o s The new production cost at the 36 Ib@ (578 kg/m3) density wodd
be $IX/MSF ($73 .84/rn3), for a reduction of $7lMSF ($4.1 6/m3). Savings in raw material costs
would be $3801 per day, or $1,330,350 per year - a pretty fair return relative to the capital
investment (cost of the for-g machine modifications).
Table 20.
Benefit Andysis for a Reduction in Board Density
Board density (lb/ft3)
W m 3 >
Production (MsF/da~)
(m3/day)
Raw material cost ($/MW
($/m3)
Savings on raw materials ($/MSF)
(%lm3)
Savings on raw materials ($1 06/year)
To summarize, it was demonstrated that a reduction board density by 2 lb/P or 16 kgh? (or
appmximately 5 percent), achieved through a compensatory improvement in strand a l iment ,
wodd Save a typical OSB d $1.33 miIlion per year in raw matenal costs. The actual cost
savings could amount to much higher were otlier production costs (ie., energy) included-
CONCLUSIONS
This study demonstrated the effect of the surface layer strand a l i m e n t distniution on the
mechanical properties of oriented strand board. Thick (23/32-inch or 18 mm) oriented strand
boards were produced with SUfface layers of three strata which were difkentiated by strand size,
strand aIipnment and position within the layer. Mathematical models were built to describe the
relationship between the orientations of the individual d a c e layer -ta and the unidirectional
rnodulus of rupture and rnodulus of elasticity of the panels. Several notable conclusions were
drawn fkom this research.
The image analysis-based method for measuring strand alignment proved to be a quick and
accurate alternative to the pre-existing direct surface measurement techniques. However, due
to its Iimitation to d a c e scans, aIipment information at depth is excluded. This disadvantage
is highlighted in the foilowing paragraph.
The excellent performance of the models support the importance of sirand aIignment through the
thickness of the panels. Traditiondy, strand alignment has been expressed on the basis of a
simple surface measurement or as an average of the whole panel thiclmess (through panel
propem ratios). The present models refbte the accuracy of using those strand aiignment
indicaton for building property-factor relationships. This finding underscores the need for a
strand alignment measurement method which accormts for the variabiiity of aIignment through
the thickness of a panel.
The predictive model employed to set the forming machine parameters and effectively control
the resuIting strand alignrnent aIso proved very useful. While the model was very basic in
p""ple, its predictive nature still provides a powerful tool for process optimi7ation in industry
and for subsequent research efforts. Furthemiore, it caLls for a more comprehensive effort to
model the i n d d a l fomiing operation. A predictive model which incorporated the strand size
disiribution (instead of constant geometric feahie sizes), classification effects of the former (in
relation to establishg gradients of strand size through the thickness) and resulting distribution
of strand aliment through the thickness would significantly improve the control over final
properties of oriented strand board.
In general, the results afkned the weU-documented positive influence of strand alignment on
the unidirectional b e n h g moduli. However, the results also suggested that there was no
marginal r e m in regards to the mechanical properties h m improvements in strand orientation
above a certain threshold-
There are several limitations goveming the usefùlness of the mechanical property models which
could be addressed in fiture research. The first was alluded to in the preceeding paragraph. A
continuous funetion describing the distribution of strand digrunent through the thickness would
be a better model factor than the discrete variables ernpioyed in the present effort. Indusmal
panels only have signincant discontinuities at layer separations (ie., surface or core). There is
no rapid change in s-d alignment within a given layer. Employing a strand alignment
dismiution fuaction, nich as the output of the mode1 suggested in the preceeding paragraph,
would be a suitable indicator.
The second limitation was the study of only one board thiclaiess, a fked layer ratio (surface to
core), and h e d geometry sizes and proportions. A model expanding the ranges of these
variables would significantly improve its applicability.
A third limitation was the omission of the average density and density profile parameters. Board
density and its distributon through the thickness of a panel have signifiant effects on the finai
board properties. Again, a new mode1 includùig these parameters wodd signincantly improve
its predictive power, scope and appricability.
FÏnally, ody the moduli in one direction were evaiuated and analyzed in the present study.
Evduating and building models for other important properties of the hished panel wodd
expand the usefbess of this research. It is recommended that hture research be expanded to
uiclude the moduli in the perpendicdar direction, the dimensional stability in bo th directions,
and other signifïcant physical and mechanicd properties.
Significant cost savings could be achieved through a two-pronged approach to process
optimization. The models predicted that improving the strand alignment wodd result in
increased mechanical prophes of the panels. It was likewise shown that reducing the density
of the boards would cause a corresponding decrease in mechanical properties. Therefore,
cornbining both modifications would cause no net change in the mechanical properties, but a
significant reduction in raw materiai costs per unit of production would be enjoyed. A very
attractive proposition for oriented strand board manufacturers, indeed!
1. Abramowitz, M.; Stegun, LA- Handbook of Mathematical Functions. Dover, NY; 1965.
2. Alexopoulos, J. Development of a High Performance OSB Panel. Internai progress
report. Ottawa, ON: Forintek Canada Corporation; 1990.
3. ASTM. Standard Test Methods for Evaluating Properties of Wood-Base Fiber and
Particle Panel Materiak; ASTM D 1037-93. Philadelphia, PA: Amencan Society for
Testing and Matenals; 1993.
4. Batschelet, E. Statistical methods for the analysis of problems in animai orientation and
certain biological rhythms. Washington, DC: American Institute of Biologicical Sciences;
1965.
5. Bucur, V. Anisotropy characterization of structural ffakeboards with uitmonic methods.
Wood and Fiber Sci. 24(3):337-346; 1992.
6. CS A Test Methods for OSB and Waferboard; CSA 0437.1 -93. Rexdale? ON: Canadian
Standards Association; 1993.
7. Canadido, L.S.; Saito, F.; Suniki, S. Effect of particle shape on the orthotropic
properties of oriented strand board. Mokuzai Gakkaishi. 34(l):Z 1-27; 1988.
8. C d , CG.; Link, C.L. Tende and compressive MOE of fiakeboards. For. Prod. J.
38(1):8-14; 1988.
9. Clark, JD.; WÏÏams, LW. The electrical conductivity of commercial dielectrics and its
variation with temperature. J. Phys. Chem. 37(1): 1 19- 13 1; 1933.
10. Cooper, P.A. Physical and Mechanical Properties of Wood; FOR 336F - Lecture Senes.
Toronto, ON: University of Toronto; 1992.
11. Fyie, I.A.; Henckel, D.J; Peters, T.E. Electrostatic orientation for efficiency and
engineering composition panel properties. Proceedings of the WSU 14b Annual
International Particleboard Symposium. Pullman, WA. 14:26 1-280; 1980.
12. Geimer, RL. Flake Ahement in Particleboard as Mected by Machine Variables and
Particle Geornetry. Madison, WI: USDA For. Serv. RP FPL 275; 1976.
13. Geimer, EL. Data basic to the engineering design of reconstituted flakeboard.
Proceedings of the WSU 13" Annuai International Particleboard Symposium. Pullman,
WA. 13:105-125; 1979.
14. Geimer, RL. Predicting flakeboard properties: Impmvements in bendings properties by
aligning a mixture of flakes. Proceedings of the WSU 1 4 h u a l International
Particleboard Symposium. Pullman, WA. 14:59-76; 1980.
1 5. Ge+, RL. Dimensional stability of flakeboards as affiected by board specific gravity
and flake alignment. For. Prob J. 32(8):44-52; 1982.
16. Geimer, RL. Mechanical Property Ratios: A Measure of Flake Alignment. Madison,
WI: USDA For. Serv. RP FPL 468; 1986.
Geimer, EL.; McDonaid, KA.; BechteI, F K ; Wood, XE. Measurement of Flake
Alignment with Grain Angle uidicator. Madison, WI: USDA For. Serv. RP FPL 5 18;
Geimer, R.L.; Montrey, H.M.; Lehmann, W.F. Effécts of layer characteristics on the
properties of three-layer particleboards. For. Prod. J. 25(3): 19-29; 1975.
Gran4 D R Analysis of Fonning Equipment Used in the Production of Oriented Strand
Board. Sainte-Foy, PQ: Canadian Forest Senrice and Forintek Canada Corp.; 1995a
Grant, D.R. Measurement of Strand Aligxment in Onented Strand Board. Sainte-Foy,
PQ: Canadian Forest Service and Forintek Canada Corp.; 199%.
Grant, D.R Strand Alignment Measurement by 'Image Andysis. Sainte-Foy, PQ:
Canadian Forest Service and Forintek Canada Corp.; 1996a
Grant, D.R Optimization of the Flaking Operation. Phase 1 - Aspen Resource
Parameters. Sainte-Foy, PQ: Canadian Forest S e ~ c e and F o ~ t e k Canada Corp.;
1996b.
Grant, D.R. Strand Alignment Factors in the Production and Performance of OSB.
Sainte-Foy, PQ: Canadian Forest SeMce and Forintek Canada Corp.; 1997.
Harris, RA M e a s h g Particle Alignment in Particleboard and Predicting Selected
Mechanical Properties of Oriented Boards; Phd. Thesis. Blacksburg, VA: Virginia
Polytechnic M t u t e and State University; 1977.
Harris, RA.; Johnson, J.A. Characterization of flake orientation in fl akeboard by the von
Mises probability distrîbution hction. Wood and Fiber. 14(4):254-266; 1982.
Higdon, A; Ohlsen, E.H.; Stiles, W.B.; Weese, J - k Mechanics of Matenals; 2" Ed.
New York, NY: John Wdey & Sons, Inc.; 1967.
Higgins, E.D. The Strength of Bonded Wood Strand Composites; Phd Thesis.
Vancouver, BC: University of British Columbia; 1990.
Hsu, W.H.E. Opb'mization of the Pressing Parameters. Sainte-Foy, PQ: Canadian Forest
Service and Forintek Canada Corp.; 1995.
Kajita, H. Oriented particleboard with Sugi thinliings (Cgptomeria japonica) 1: EEects
of degree of particle alignment and board density on physical and mechanicd properties.
Mokuzai Gakkaishi. 33(11):865-871; 1987.
Kawai, S.; Lang, L.L.; Sasaki, H.; Ishihara, S. Production of oriented board with an
electrostatic field 1: Factors aEecting the alignment of wood particles. Mokuzai
Gakkaishi. 28(5):295-300; 1982.
Kawai, S.; Sasaki, H.; Norimoto, M. Aligning torque generated on wood partictes in an
electrostatic field 1: Effects of various factors on the aligning torque. Mokuzai
Gakkaishi. 33(11):872-878; 1987.
Kieser, J.; Steck, E.F. The influence of flake orientation on the MOR and MOE of
strandboards. Proceedings of the WSU 12" Annual International Particleboard
Symposium. Pullman, WA. 12:99-122; 1978.
Kollmann, F.F.P.; Côté, W.A. P ~ c i p l e s of Wood Science and Technology. 1. Solid
Wood. New York, NY: Springerlag; 1968.
34. Kvanli, AH. Statistics - A Computer Integrated Approach. S t Paul, MN: West
Publishing Company; 1988.
Lang, L.L.; Kawai, S.; Sasaki, H.; hhihara, S. Production of oriented board with ar
electrostatic field II: Roperties of oriented particleboard. Mo kuzai Gakkaishi.
28(5):3O 1-306; 1982.
Lau, P. W-C. Properties of ûriented Waferboard and a Model to Predict the ModuIus
of Elasticity; MSc. Thesis. Toronto, ON: University of Toronto; 1980.
Lau, P.W-C. Niunerical approach to predict the modulus of elasticity of oriented
waferboard. Wood Sci. 14(2):73-85; 1981.
Laufenberg, T.L. Characterizing the nonlinear behavior of ffakeboards. Wood and Fiber
Sci. IS(l):47-58; 1983.
Lowood, J.D. OSB - The emerging world class structural wood panel. Proceedings of
the WSU 29" Annual International Particleboard Symposium. Pullman, WA. 29: 1 23 - 128; 1995.
Mardia, RV. Statistics of Directional Data. London, England: Academic Press Inc.
Ltd,; 1972.
McCormac, K. Structural Analysis; 3"1 Ed. New York, NY: htext Educational
Publishers; 1975.
McNatt, J.D. Basic Engineering Properties of Particleboard Madison, WI: USDA For.
Serv. RP FPL 206; 1973.
43. McNatt, J.D.; Bach, L.; WeUwood, RW. Contribution <if flake alignment to
performance of strandboard. For. Prod. J. 42(3):45-50; 1992.
44. Musial, M.W. A non-destructive method for determining the degree of flake orientation
in OSB. Wood Sci. Tech. 22:371-378; 1988.
45. Panshin, A.J.; de Zeeuw, C. Textbook of Wood Technology; 4Lh Ed. New York, NY:
McGraw-Hill PubIishing Company; 1980.
46. Peters, TE. The use of electrostatic orientation in the production of fully aligned
flakeboard for COM-PLY joist and t m s Iumber. Proceedings of the WSU 17Lh A M U ~
IntemationaI Particleboard Symposium. Pullman, WA. 1 7: 1 19- 134; 1983.
47. hilido, O R ; Yoshida, Y.; Wang, Q.; Kawai, S.; Sasaki, H. Aligning torque generated
in wood particles by an electrostatic field V: Torque due to dipole polarization and its
effect on the orientation of wood particles. Mokuzai Gakkaishi. 37(8):711-718; 199 1.
48. RISI. Wood Products Yearbook. Eugene, OR: Resource Monnation Systems, hc.;
1996.
49. RISI- Wood Products Review. Eugene, OR: Resource Information Systems, Inc.; July
1996.
50. Sasaki, 8; Kawai, S.; Pulido, 0.R; Pengprecha, N. Production of oriented board with
an electrostatic field III: Semi-strand board having better orientation toward both
surfaces. Mokuzai Gakkaishi. 35(8):725-730; 1989.
51. Shaler, SM. Comparing two measures of flake alignment. Wood Sci. Tech. 26(1):53-
61; 1991.
52. Shaler, S.M.; Blankenhom, P.R. Composite mode1 prediction of elastic moduli for
flakeboard. Wood and Fiber Sci. 22(3):246-26 1; 1990.
53. Skaar, C. Water in Wood. Syracuse, NY: Syracuse University Press; 1 972.
54. Smith, D.C. Waferboard press closing strategies. For. h o d J. 32(2): 40-45; 1982.
55. Snodgrass, ID.; Saunders, RJ.; Syska, A.D. Particleboard of aligned wood strands.
Roceedings of the WSU Annuai Internationa.1 Particleboard Symposium. Pullman,
W A 7:415-448; 1973.
56. Stat-Ease. DESIGN-EXPERT Software Version 3.0 User's Guide for Response Surface
Methodology and Mixture Experimentç. Minneapolis, MN: S tat-Ease, Inc.; 1 992.
57. Suniki, M.; Sekino, N. Anisotropic elasticity of oriented flakeboard. Mokwai
Gakkaishi. 28(2):91-96; 1982.
58- Tdbott, J.W. Electncaliy alignai particle and fibre board. Proceedings of the WSU 8'
Annual International Particleboard Symposium. Pullman, WA. 8: 153-1 82; 1974-
59. Talbott, J.W.; Stefanakos, EX. Aligning forces on wood particles in an elecaic field.
Wood and Fiber. 4(3): 193-203; 1972.
60. Tuma, J.J. Theory and Problems of Structural Anaiysis; Schaum's Outhe Series in
Engineering. New York, NY: McGraw-Hi11 Book Company; 1969.
6 1. Wright, P.M. Logic of structures; ARC 25 3 F - Lecture Series. Toronto, ON: University
of Toronto; 1993.
62. Yoshida, Y.; Pulido, OX.; Kawai, S.; Sasaki, H. Aligning torque generated in wood
particles by an electrostatic field Ei: DistrÏbution and inteI1Sification method of aligning
torque in the space surrounding electrodes. Mokuzai Gakkaishi. 34(5):401-408; 1988.
63. Yoshida, Y.; Pulido, OR; Kawai, S.; Sasaki, H. Production oriented partideboard 1:
Structure of particle mat fo& with onented particles. Mokuzai Gakkaishi. 3 5(3) :Z7-
233; 1989a
64. Yoshida, Y.; Pulido, OR; Kawai, S.; Sasaki, H. Production oriented particleboard II:
Aligning rnethods and board properties. Mokuzai Gakkaishi. 35(8):718-724; l989b.
65. Zhou, D. A study of oriented structurai board made Eom hybrid poplar Eff i t s of some
factors of mechanical forming installation for orientation effectiveness. Holz-als-Roh-
und-Werksto ff. 47:405-406; 1889.
66. Zhou, D. A study of onented stmctural board made nom hybrid poplar: Physicd and
mechanical properties of OSB. Holz-als-Roh-und-Werkstoff. 48:293-296; 1990.
APPLIED IMAGE. lnc 1653 East Main Street - -- - - Rochester, NY 14609 USA -- -- - - Phone: 716/482-0300 -- -- - - Fm 716i2ûü-5989