Chapter VI Multivariate Modeling of MDF Panel Properties in Relation to Wood Fiber Characteristics

Table des matières

•  Résumé
• Abstract
• 6.1 Introduction
• 6.2 Material and Methods
  • 6.2.1 Material Sampling
  • 6.2.2 Furnish Preparation
  • 6.2.3 MDF Panel Manufacturing
  • 6.2.4 MDF Panel Property Testing and Wood Fiber Characterization
• 6.3 Statistical Analysis and Methods
• 6.4 Results and Discussion
  • 6.4.1 Modulus of Rupture
  • 6.4.2 Modulus of Elasticity
  • 6.4.3 Internal Bond Strength
  • 6.4.4 Linear Expansion
  • 6.4.5 Thickness Swell
  • 6.4.6 Water Absorption
  • 6.4.7 Analysis of Model Residuals
• 6.5 Conclusions
• 6.6 Literature Cited
• 6.7 List of Tables
• 6.8 List of Figures

Authors: Jun Li Shi, Bernard Riedl, and S. Y. Zhang

Résumé

La relation entre les propriétés de panneaux de fibres de densité moyenne et les caractéristiques des bois et de la fibre a été étudiée dans ce travail. Les panneaux de fibres de densité moyenne ont été manufacturés en laboratoire à partir de différentes espèces et types de bois, qui étaient du bois d’épinette noire d’arbres agés de 0-20, 21-40, et plus de 40 ans, ainsi que les billes du haut, de la mi-hauteur, et du bas d’arbres d’épinettes noires, de trois clones de peuplier, de mélèze, et d’un mélange d’épinette, pin, et sapin. Les propriétés mécaniques des panneaux évaluées ont été le module de rupture (MOR), le module d’élasticité (MOE) en flexion, et la cohésion interne (IB), ainsi que la stabilité dimensionnelle évaluée par l’expansion linéaire (LE), le gonflement en épaisseur (TS), et l’absorption d’eau (WA), qui ont été analysées comme variables de réponse dans l’étude. Diverses caractéristiques du bois et de la fibre, telles que la densité du bois, son pH et sa capacité tampon basique, le pH de la fibre et sa capacité tampon basique, la longueur moyenne arithmétique des fibres, leur largeur, le pourcentage de fines des fibres, le pourcentages de fibres réparties dans les tamis de taille 3,240 mm², 0,828 mm², 0,281 mm², 0,017 mm², et sur le tamis 0,017 mm², ont été mesurées, et ont été utilisés comme variables prédictrices. L’analyse de régression linéaire multiple a été utilisée afin de mettre en évidence les rapports fonctionnels entre les propriétés des panneaux et les caractéristiques des bois et de la fibre. Les résultats indiquent que le module de rupture est négativement lié au pourcentage arithmétique moyen de fibres fines dans la pulpe. Le module d’élasticité est affecté négativement par le pourcentage de petites particules (<taille 0,017 mm²) dans la pâte, et également lié au pH du bois et à d’autres caractéristiques du bois et de la fibre qui n’ont pas été mesurées et analysées. La cohésion interne dépend négativement du pourcentage arithmétique moyen de fibres fines, mais est amélioré par un contenu croissant de petites particules (<taille 0,017 mm²). Le pH de la fibre a eu un effet négatif sur le test de tension perpendiculaire. Les résultats indiquent également que la cohésion interne est liée à d’autres caractéristiques inconnues du bois et des fibres qui n’ont pas été incluses dans l’analyse. L’expansion linéaire du panneau s’est avérée être négativement liée à la densité du bois, et reliée à d’autres caractéristiques du bois et de la fibre qui n’ont pas été mesurées. Le gonflement en épaisseur a été négativement affecté par la longueur arithmétique moyenne des fibres. La largeur arithmétique moyenne des fibres a un effet négatif sur l’absorption d’eau des panneaux. Les valeurs des modules de rupture, d’élasticité et de test de tension perpendiculaire des panneaux sont liées au contenu de fibres fines dans la pâte indiquant un rôle significatif du processus de raffinage.

Abstract

Properties of medium density fiberboard (MDF) panels in relation to wood and fiber characteristics were investigated. Laboratory MDF panels were manufactured from raw fiber materials of black spruce ( Picea mariana (Mill.) BSP.), three hybrid poplar clones ( Populusspp.), two exotic larch ( Larix gmelinii and Larix sibirica ), and a mix of spruce, pine, and fir (S-P-F) wood chips. The panels were evaluated for modulus of rupture (MOR), modulus of elasticity (MOE), internal bond (IB), linear expansion (LE), thickness swell (TS), and water absorption (WA). These properties were analyzed as response variables. As predictor variables, various wood and fiber characteristics were measured including wood density, pH and base buffering capacity and fiber coarseness. Multiple linear regression analysis was performed to develop the functional relationships between panel properties (response variables) and wood fiber characteristics (predictor variables). Ten dummy variables were created and incorporated into the analysis to examine the effects of wood species or type on MDF panel properties. MOR was negatively related to arithmetic fine percentage. MOE was negatively affected by the percentage of small particles that can pass through the mesh with a size of 0.017 mm² (mesh size smaller than 0.017 mm²) and wood pH. IB strength was negatively related to arithmetic fine percentage and fiber pH, but positively related to the percentage of small particles (mesh size smaller than 0.017 mm²). Wood density affected LE. TS was negatively affected by arithmetic mean fiber length. Arithmetic mean fiber width had a negative effect on panel WA. The presence of the dummy variables in MOE, IB, and LE models indicates that wood fiber characteristics other than those measured in this study affected panel MOE, IB, and LE significantly. The study indicates that refining process can play a significant role in manipulating MDF panel properties.

6.1 Introduction

A number of factors influence the properties of medium density fiberboard (MDF) (Maloney 1993). Pressing schedule and platen temperature are critical since these factors can result in significant differences in the panel vertical density profile (VDP), which, to a large extent, determines panel physical and mechanical properties (Suchsland and Woodson 1974; Kelly 1977; Harless et al. 1987; Winistorfer et al. 1996; Wang et al. 2001; Wang et al. 2004). The VDP can be determined by the pressing schedule and the temperature of the platens used for panel consolidation (Andrews et al. 2001; Wang et al. 2001). The VDP is closely related to strength and physical properties of composite panels (Suchsland and Woodson 1974; Kelly 1977; Harless et al. 1987; Winistorfer et al. 1996; Wang et al. 2001; Wang et al. 2004). The shape of the VDP curve is of importance. High face density and low core density lead to excellent bending properties, but to relatively poor IB strength (Woodson 1976b). Characteristics of raw fiber materials and wood species used for MDF panel manufacturing are also important (Maloney 1993).
Wood density was found to be negatively related to most strength properties of MDF panels (Nelson 1973; Woodson 1976b). It is easier to make good bonded panels at a specific target panel density with low-density (LD) wood than with high-density (HD) wood due to higher compaction ratios (panel density divided by wood density) resulted from the former (Maloney 1993; Hsu 1997).
According to a study by Nelson (1973), strength properties and linear expansion (LE) of hardboards were positively correlated to pulp pH. Water absorption (WA) was also related to pulp pH; linear stability was positively related to fiber length. Yet, Nelson (1973) found that thickness stability was not related to any wood or fiber property studied. Therefore, fiber morphological and chemical properties, fiber composition and fiber strength are considered to be basic characteristics influencing fiberboard panel properties (Jones 1960; Johns et al. 1985; Myers 1987; Suchsland and Woodson 1990; Maloney 1993; Moss and Retulainen 1995; Hsu 1997; Groom et al. 1999).
The influence of fiber strength on fiberboard quality is dependant on panel density. It has little influence in the case of low density fiberboard (LDF), but its influence becomes significant in MDF. Fiber strength can be the predominant factor for high density fiberboard (HDF) (≥900 kg/m³) (Jones 1960). In LDF, failures usually occur in the bonds between fibers. In HDF panels, the contact between fibers is more intimate and thus failures occur within the fiber itself (Suchsland and Woodson 1990).
Fiber length is a critical factor determining the properties of paper (Dinwoodie 1965) and MDF (Suchsland and Woodson 1990). Longer fibers have the tendency to yield higher bulk density. Nelson (1973) found that low bulk density was associated with high panel strength. Fiber length and fiber orientation in the board are also correlated. Shorter fibers are more likely to deviate from the horizontal compared to longer fibers (Suchsland and Woodson 1990). Thus, shorter fibers could be disadvantageous since the contact area between fibers is diminished.
The pH and buffering capacity of the fibers are important parameters for resination. If urea-formaldehyde (UF) resin is used, higher fiber pH and lower buffering capacity are desirable (Nelson 1973; Johns and Niazi 1980; Johns et al. 1985; Maloney 1993; Hsu 1997). A too low pH may eventually cause pre-cure during the initial phase of hot-pressing, while buffering capacity influences the cure rate of UF resin.
Fiber composition refers to the percentages of fines, fiber bundles, whole and broken fibers of the pulp (Myers 1987). Low percentages of fiber bundles and fines are desirable as the former destroy the uniformity of fiber distribution and the latter consume more resin (Woodson 1976a; Moss and Retulainen 1995; Groom et al. 1999; Barnes 2002).
A high level of refining energy spent on wood chips, flakes, or shavings may lead to reduction in fiber length and production of debris (Clark 1978; Myers, 1987; Kure and Dahlqvist 1998; Kure et al. 1999). Steam time and pressure, disc diameter, plate pattern, loading rate, gap between the grinding plates, and rotation speed of the plates also affect fiber composition and size distribution.
The previous reviews show that the impact of the various parameters on panel properties is complex. The objective of this study was to investigate the relationships between MDF panel properties and fiber characteristics. MDF panels were fabricated from eleven raw fiber materials. Fiber samples were taken from the eleven materials for wood characterization. To eliminate the influence of the processing parameters, a constant hot-pressing program was used for the manufacturing of the panels. 

6.2 Material and Methods

6.2.1 Material Sampling

Sixty black spruce ( Picea mariana (Mill.) BSP.) trees were collected in July 1999 in Réserve Ashuapmushuan (Québec, Canada) from a second-growth natural stand. The trees of 70-80 years of age were bucked to 8-foot-long butt, middle and top logs. The debarked butt logs were cut into 1-foot-long sections. On the cross section of each 1-foot-long section, growth rings were counted and marked with three different zones according to cambial age: 1-20 year (juvenile wood), 21-40 year (mostly mature wood), and over 40 year (mature wood). The transition age from juvenile to mature wood occurs in 20-25 year (Alteyrac 2005). A band saw was used to separate the three types of wood along the marked lines from the short sections. Moreover, top, middle, and butt logs were collected to make three samples. The logs were debarked and chipped (a portable chipper was used). Thus, there were six samples from black spruce: 1) 1-20 year old wood, 2) 21-40 year old wood, 3) over 40 year old wood, 4) top logs, 5) middle logs, and 6) butt logs.
Hybrid poplar ( Populus spp.) material came from a clonal trial established by the Forest Research Branch of the Québec Ministry of Natural Resources in St-Ours, Southern Québec, Canada in 1993. The poplar trees were planted at a 1.5×3.5 m spacing in this site that is a part of the Champlain marine deposit with rich salty-clay soil (40 % clay). A systematic thinning was carried out in the spring of 1996 and the thinning led to a spacing of 2.5×3.0 m (Zhang et al. 2003). Four trees from each of hybrid poplar ( P. maximowiczii and P. balsamifera ) clones 915303, 915311, and 915313 were harvested from this site in December 2002 at the age of 10 years.
Larch material was a mix of two individual species ( Larix gmelinii and Larix sibirica ) with an approximate proportion of 4:1. The trees were collected from an exotic larch plantation grown in Northern Ontario, Canada through a commercial thinning. The trees were 10-15 years old, and thus were composted of juvenile wood. In total, seventy sample trees were collected and transported to Forintek Canada Corp. for this study as well as a lumber study.
As MDF mills in eastern Canada usually rely on mixed chips of black spruce, jack pine and balsam fir (S-P-F), a mixed sample of the S-P-F wood chips was collected as a reference from a composite panel mill in Québec, Canada. 

6.2.2 Furnish Preparation

Wood chips of the eleven samples in sequence were fed into a pressurized disc refiner at the MDF pilot plant of Forintek Canada Corp. (Québec City). Refining was carried out without wax and resin injection. The refining parameters are given in Table 6-1. The refining energy and the clearance between the two grinding plates were adjusted and then fixed when fiber bundles were visually observed to be minimized. 

6.2.3 MDF Panel Manufacturing

Fibers were dried in a laboratory-scale dryer until the moisture content (MC) was reduced to about 2-3 %. Fibers were then passed through a hammer mill to completely disperse the fibers. A laboratory-scale blade blender was used to blend resin and wax with the fibers. 10 % (by weight of dry fibers) Borden 302 UF resin (65 % solid content) and 0.5 % wax were slowly sprayed onto the fibers. Following resin and wax blending, the fibers were passed through the hammer mill again to disperse the fiber balls. All panels were manufactured to a target density of 740 kg/m³, and panel size was 610×610×12 mm. Mats were hand formed in a frame. The MC after resin and wax blending is presented in Table 6-1. Three replicate panels were made for each sample. A constant hot-pressing program without press stops was applied to all samples. The two platens were slowly closed over 160 s. Panels were kept pressing for another 160 s and the platens were gradually opened in 40 s. The temperature of the platens was set at 135 ^oC. The above mentioned pressing scheme was adopted since we aimed to produce panels with flat density profiles. A flat density profile can reduce variations in panel properties that could result from different density gradients along panel thickness.

6.2.4 MDF Panel Property Testing and Wood Fiber Characterization

All panels were conditioned in a chamber at 22 ^oC and 65 % relative humidity (RH) for four weeks until they reached equilibrium moisture content. For each type of panel, nine static bending (338×75 mm), thirty IB (50×50 mm), six LE (305×76 mm), and six TS and WA (152×152 mm) specimens were prepared. The procedures and methods described in ASTM D 1037-99 (2001) and ANSI A 208.2-2002 (2002) were followed for the determination of modulus of elasticity (MOE), modulus of rupture (MOR), IB, LE, TS and WA. The IB specimens were sanded to remove 1.5 mm from both surfaces before they were glued to blocks. X-ray density profiles were performed on the panel specimens to verify that the flatter profiles were achieved. Panel properties are listed in Table 6-2.
ASTM (D 2395-02) method B (2004) was followed for wood density determination. Wood chips collected from each sample were oven-dried at 103 ± 2 ^oC until constant weight was achieved. The oven-dry weight of the wood chips was taken using a scale. Then the oven-dried chips were soaked in cold water until saturated. The volume of the chips was determined by volume displacement. Density of the wood chips was based on oven-dry weight of wood chips divided by green volume. A large number of wood chips was measured for each sample to ensure that the value was representative. Basic wood density of the eleven samples is listed in Table 6-3.
Wood base buffering capacity was calculated by multiplying the ml of NaOH solution required to raise the starting pH of the wood extract to pH 7 by the normality of the solution (Johns and Niazi 1980). First, wood chips from each sample were ground to powder. Then, 25 g of oven-dried wood powder was added to 200 ml of distilled water for refluxing (20 min). The mixture was filtered through a filter paper using a vacuum and washed several times. The extract was diluted to 500 ml and cooled down before titrating. 50 ml of extract solution was pipetted into a 150 ml beaker and the pH was measured. The extract solution was then titrated to pH 7 with nominal 0.025 N NaOH and a titration curve in 0.1 ml steps was created. The means of four measurements are presented in Table 6-3.
For fiber pH and base buffering capacity, 10 g oven-dried fiber was refluxed and the same procedure as described above was applied. Table 6-3 presents the average of four measurements.
Fiber quality was evaluated using a HiRes Fiber Quality Analyzer (FQA) at the ‘Centre specialisé en pates et papiers’ (Université du Québec à Trois-Riviéres, Canada). Fiber quality parameters evaluated for each sample include fiber coarseness, arithmetic mean length, length weighted mean length, arithmetic mean width, arithmetic fine percentage and length weighted fine percentage are given in Table 6-3.
The Bauer-McNett Classifier was used to determine fiber size distribution according to the procedures described in Tappi T 233 cm-95 (1995). Four screens of mesh size 3.240 mm², 0.828 mm², 0.281 mm², and 0.017 mm² were used. 10 g (oven-dry weight) of fibers were poured into the first compartment. The water flow rate was 11.355 L/min. A 20 min motor running period was employed. The fiber pads were collected and dried at 105 ^oC to a constant weight. The measurement was replicated three times for each sample and the values were averaged (Table 6-3).

6.3 Statistical Analysis and Methods

Based on an assumption that the density profiles of all panels were flat and equal, multiple linear regression was performed to model the properties of MDF panels made from different fiber samples in relation to various wood and fiber characteristics. All the predictor and response variables are presented in Table 6-4. Ten dummy variables z₁, z₂, …, z₁₀ were created and the eleven wood samples were identified by combining the values of these variables (Table 6-1). There were twenty seven predictor variables in the wood fiber characteristic matrix including the ten dummy variables (Table 6-4). Panel bending properties (MOR and MOE), IB strength and dimensional stability (LE, TS, and WA) were response variables and incorporated into the panel property matrix (Table 6-4). All variables were examined for normal distribution, and the results show that the data sets of wood density, wood pH, base buffering capacity of wood chips, length weighted mean fiber length, arithmetic mean fiber width, and percentage of fibers passed through the mesh (0.281 mm²) but retained on the mesh (0.017 mm²) were not normally distributed (Table 6-4). So the data sets of these variables were transformed until they met normal distribution requirement. The transformation function and resulting variables are shown in Table 6-4. The transformed forms together with other variables in the wood fiber characteristic matrix were analyzed as predictor variables for multiple linear regression.
Dummy variables were created as follows:
z₁ = 1if subject is in black spruce 1-20 age zone

1. otherwise

z₂ = 1if subject is in black spruce 21-40 age zone

1. otherwise

z₃ = 1if subject is in black spruce over 40 age zone

1. otherwise

z₄ = 1if subject is in black spruce top logs

1. otherwise

z₅ = 1if subject is in black spruce mid logs

1. otherwise

z₆ = 1if subject is in black spruce butt logs

1. otherwise

z₇ = 1if subject is in hybrid poplar clone 915303

1. otherwise

z₈ = 1if subject is in hybrid poplar clone 915311

1. otherwise

z₉ = 1if subject is in hybrid poplar clone 915313

1. otherwise

z₁₀ = 1if subject is in the two individual larch species
0 otherwise
Then the eleven wood species or types can be identified by the values of the dummy variables z₁, z₂, …, z₁₀.
z₁ = 1, z₂, z₃, …, z₁₀ = 0: black spruce 1-20 year-old wood
z₂ = 1, z₁, z₃, …, z₁₀ = 0: black spruce 21-40 year-old wood
z₃ = 1, z₁, z₂, …, z₁₀ = 0: black spruce over 40 year-old wood
z₄ = 1, z₁, z₂, …, z₁₀ = 0: black spruce top logs
z₅ = 1, z₁, z₂, …, z₁₀ = 0: black spruce mid logs
z₆ = 1, z₁, z₂, …, z₁₀ = 0: black spruce butt logs
z₇ = 1, z₁, z₂, …, z₁₀ = 0: hybrid poplar clone 915303
z₈ = 1, z₁, z₂, …, z₁₀ = 0: hybrid poplar clone 915311
z₉ = 1, z₁, z₂, …, z₁₀ = 0: hybrid poplar clone 915313
z₁₀ = 1, z₁, z₂, …, z₉ = 0: two individual larch species
z₁, z₂, …, z₁₀ = 0: S-P-F wood chips
The multiple linear model for the relationships between response and predictor variables took the form:
Y = b₀ + b_ix_1~17 + b_jz_1~10 + b_kx_1~17z_1~10 + … + ε
where: Y is the response variable, MOR, MOE, IB, LE, TS, or WA;
x₁, x₂, …, x₁₇, and z₁, z₂, …, z₁₀ are the predictor variables;
b₀ is the intercept, b_i, b_j, and b_k are the regression coefficients (i = 1, …, 17; j = 18, …, 27; k = 28, …, 197);
ε is the error term.
Statistical Analysis System (SAS) (1990) software was used for data analysis. Since there were a number of predictor variables, the multivariate model candidate selection was based on two criteria, stepwise selection and model R square. Thus, preliminary predictor variables selected by the stepwise procedure were listed. These predictor variables were then screened by model R square selection criterion.
The predictor variables entered in the model might be highly correlated (multicolinearity), which can result in unstable estimates and high standard errors (Draper 1980). To avoid this, the selected predictor variables were checked for multicolinearity using syntax “MODEL response variable=predictor variable/VIF COLLIN”. The calculation of the variance inflation factor (VIF) is to regress a specific predictor variable on other predictor variables in order to check multicolinearity of that variable. Generally, if the VIF value is higher than 10, attention should be paid to this variable (Belsley et al. 1980).

6.4 Results and Discussion

6.4.1 Modulus of Rupture

The summary of stepwise model selection for response variable MOR is shown in Table 6-5. The predictor variable, arithmetic fine percentage, entered into the model, and could explain 53.2 % of the total variation. The model for MOR can be written in the following form according to the parameter estimates shown in Table 6-6.
MOR = 39.8 - 0.509 arith_fine
The MOR model indicates that arithmetic fine fiber percentage has a negative effect on panel bending strength. If the pulp contains a high proportion of fines and the resin content remains constant, the fines consume more resin due to their large surface area, and resin available for the whole fibers is diminished. As a result, and as explained in the introduction, the fiber to fiber bonding system of the panel is consequently weakened. 

6.4.2 Modulus of Elasticity

There were nine variables that entered into MOE model by the stepwise procedure (Table 6-5). The percentage of fine fibers that can pass through the screen with mesh size of 0.017 mm² (<0.017 mm²) is the first variable entered to the model and it can explain 46.8 % of the variation in MOE. With only the variable, percentage of fine fibers (<0.017 mm²) in the model, it has a R square of 0.468. Adding variables tangent wood pH and z₅ into the model, the model R square was improved to 0.850. This means that the three variables, percentage of fine fibers (<0.017 mm²), tangent wood pH, and z₅ can account for 85.0 % of the variation in MOE. Thus, these three variables were selected to describe MOE and multiple linear regression was performed using the three variables as predictor inputs. The values of VIF calculated for the three selected predictor variables were fairly small compared with the multicolinearity cut-off value of 10. This indicates little correlation between the three predictor variables. Hence, the MOE model can be given by the following equation based on the parameter estimates (Table 6-6).
MOE = 2.51×10³ – 409 _200_ + 45.6 tanwood_pH + 479 z₅
The MOE model indicates that very small particles that can pass through a 200 mesh screen had a negative effect on panel MOE. MOE was also positively related to tangent wood pH. According to the feature of the tangent function, wood pH would affect MOE positively if wood pH fell in the ranges of 1/2π – π, π – 3/2π, 3/2π – 2π or 2π – 5/2π. π, 3/2π, 2π, and 5/2π are critical values for wood pH in determining its effect on MOE. If wood pH is equal to π or 2π, there is no effect of wood pH on MOE; if wood pH approaches to the values of 3/2π or 5/2π, the effect of wood pH on MOE becomes instantly great. In addition, the presence of the dummy variable z₅ indicates that an unknown wood fiber characteristic significantly influenced panel MOE and it belongs to black spruce mid logs. 

6.4.3 Internal Bond Strength

Table 6-5 lists the predictor variables that entered in and then were removed from the IB model by the stepwise procedure. There were nine variables left in the model. Arithmetic fine percentage can explain 43.9 % of the total variation; together with the percentage of fine fibers (<0.017 mm²), z₇, and fiber pH, the model accounts for 91.7 % variability (Table 6-5). Adding the other variables listed in Table 6-5 such as z₁, z₉, z₆, fiber base buffering capacity, and z₃ into the model did not significantly improve the R square. Thus, the relationship between IB and wood fiber characteristics can be described by the four variables, namely, arithmetic fine percentage, percentage of fine fibers (<0.017 mm²), z₇, and fiber pH. The estimated parameters for IB model are shown in Table 6-6. A model for IB strength can be written as follows.
IB = 3.48 – 0.0509 arith_fine + 0.405 _200_ + 0.304 z₇ – 0.383 fiber_pH
Internal bond strength is linearly and negatively related to arithmetic fine fiber percentage. However, it is positively related to the percentage of very small particles (<0.017 mm²) of the pulp. Moreover, IB is negatively affected by fiber pH. This seems to be contradictory with the previous findings (Johns and Niazi 1980; Johns et al. 1985; Maloney 1993; Hsu 1997). The negative correlation between IB and fiber pH might be attributed to the relatively low temperature for panel consolidation used in this study. The dummy variable z₇ appearing in the model implies that other unknown wood fiber characteristics belonging to hybrid poplar clone 915303 influenced IB strength significantly. 

6.4.4 Linear Expansion

Through stepwise selection, nine variables, z₁₀, z₁, z₆, fiber coarseness, z₂, compaction ratio, quadratic wood density, z₉, and z₄, entered to the LE model (Table 6-5). Of predictor variables, however, only dummy variable z₁₀ could explain as much as 71.4 % of variability in the data. This indicates that wood fiber characteristics of larch significantly affected panel LE. Quadratic wood density could account for a proportion of 15.8 % of the observed variation. Since as much as 87.2 % of the total variation could be explained by these two variables, a model for LE can be given by the estimated parameters (Table 6-6).
LE = 0.209 + 0.0899 z₁₀ – 0.244 quadwood_density
The model indicates that LE was negatively related to wood density. Nelson (1973) found that linear stability is positively related to fiber length. Here, arithmetic mean fiber length was not selected for the model as it was found not to be related to linear expansion. 

6.4.5 Thickness Swell

Using the stepwise method, variables that entered into the TS model were arithmetic mean fiber length, cubic arithmetic mean fiber width, tangent wood pH, and z₉, respectively (Table 6-5). Only arithmetic mean fiber length was kept in the final model since it could explain as much as 87.2 % of the total variation observed. Parameter estimates for TS model are shown in Table 6-6. Thus, the TS model can be written as follows.
TS = 44.8 – 38.7 arith_length
The TS model indicates that there is a negative and linear relationship between panel thickness swell and arithmetic mean fiber length. 

6.4.6 Water Absorption

A summary of the stepwise model selection for WA is shown in Table 6-5. The variable, cubic arithmetic mean fiber width, accounted for 92.4 % of the total variation. Therefore, the relationship between panel WA and cubic arithmetic mean fiber width can be described with the following equation according to the estimated parameters (Table 6-6).
WA = 158 – 0.00530 cubfiber_width
The predicted relationship between WA and arithmetic mean fiber width is curvilinear. The overall effect of arithmetic mean fiber width on panel WA is negative meaning that larger fiber width is associated with smaller panel WA.

6.4.7 Analysis of Model Residuals

The residuals were analyzed for each response variable, and the results show that the residuals for MOR, MOE, IB, LE, TS, and WA models were normally distributed. The residual dispersions for the models were randomly distributed, and for each model the residuals were about evenly split and close to the average value of that response variable (Figure 6-1). The absence of a systematic pattern in residuals ensures that the model found is indeed the best possible model with respect to that response variable.

6.5 Conclusions

This study shows a significant effect of wood fiber characteristics on MDF panel properties. Modulus of rupture was negatively affected by the arithmetic fine fiber percentage.Modulus of elasticity was negatively related to the percentage of small particles (<0.017 mm²), and also related to wood pH. Internal bond strength was negatively dependant on arithmetic fine fiber percentage and fiber pH, but positively related to the percentage of small particles (<0.017 mm²). Linear expansion was negatively influenced by wood density, while TS was negatively related to arithmetic mean fiber length. Arithmetic mean fiber width had a negative effect on WA. The presence of the dummy variables in MOE, IB, and LE models suggests that wood fiber characteristics other than those measured in this study had significant effects and should be analyzed in order to correlate them with MDF panel properties. As for different raw materials, the refining parameters should be set differently and appropriately in order to produce high performance fibers with less fine fiber content. 

6.6 Literature Cited

Alteyrac, J. Influence de la densité de peuplement et de la hauteur dans l’arbre sur les propriétés physico-mécaniques du bois d’épinette noire (Picea Mariana (Mill.) B. S. P.) . Ph.D. Thesis, Université Laval, Québec, Canada. 2005.
Andrews, C. K., Winistorfer, P. M. and Bennett, R. M. The influence of furnish moisture content and press closure rate on the formation of the vertical density profile in oriented strandboard . Forest Prod. J, 51(5): 32-39; 2001.
American National Standards Institute (ANSI). Medium Density Fiberboard (MDF) for Interior Application . ANSI A 208.2-2002. Composite Panel Association, Gaithersburg. MD. 2002.
American Society of Testing and Materials (ASTM). Standard test methods for specific gravity of wood and wood-based materials . ASTM D 2395-02. Vol. 04.10. ASTM, West Conshohocken, PA. pp. 357-364; 2004.
American Society of Testing and Materials (ASTM). Standard test methods for evaluating properties of wood-based fiber and particle panel materials . ASTM D 1037-99. Vol. 04.10. ASTM, West Conshohocken, PA. pp. 141-170; 2001.
Barnes, D. A model of the effect of fines content on the strength properties of oriented strand wood composites . Forest Prod. J. 52(5): 55-60; 2002.
Belsley, D. A., Kuh, E. and Welsch, R. E. Regression Diagnostics: Identifying influential data and source of collinearity. New York, John Wiley. 1980.
Clark, J. A. Pulp Technology and Treatment for Paper . Miller Freeman Publication Inc. San Francisco, California. 1978.
Dinwoodie, J. M. The relationship between fiber morphology and paper properties: a review of literature . Tappi. 48(8): 440-447; 1965.
Draper, N. R. Applied Regression Analysis . John Wiley & Sons Inc. 1980.
Groom, L. H., Mott, L. and Shaler, S. M. Relationship between fiber furnish properties and the structural performance of MDF . 33^rdInternational Particleboard/Composite Materials Symposium Proceedings. April 13-15, Washington State Univ., Pullman, Washington, USA. 1999.
Harless, T. E. G., Wagner, F. G., Short, P. H., Seale, R. D., Mitchell, P. H. and Ladd, D. S. A model to predict the density profile of particleboard .Wood Fiber Sci. 19(1): 81-92; 1987.
Hsu, W. E. Wood quality requirements for panel products . Proceeding of CTIA/IUFRO Inter. Wood Quality Workshop, Timber Management toward Wood Quality and End-product Value. Forintek Canada Corp. Québec City, Canada. I-7/10; 1997.
Jones, E. J. The relation of fiber and pulp properties to the properties of structural fiberboard products . Tappi. 43(6): 600-602; 1960.
Johns, W. E. and Niazi, K. A. Effect of pH and buffering capacity of wood on the gelation time of urea-formaldehyde resin . Wood Fiber Sci. 12(4):255-263; 1980.
Johns, W. E. Rammon, R. M. and Youngquist, J. Chemical effects of mixed hardwood furnish on panel properties . Proceeding 19^thInternational Particleboard/Composite Materials Symposium, Maloney, T. M. Ed. Washington State Univ., Pullman, WA, pp. 363-377; 1985.
Kelly, M. Critical literature review of relationship between processing parameters and physical properties of particleboard . Gen Tech. Rep. FPL-10. USDA Forest Serv., Forest Prod. Lab., Madison, WI. 1977.
Kure, K. A. and Dahlqvist, G. Development of structural fiber properties in high intensity refining . Pulp Pap. Can. 99(7): T233-T237; 1998.
Kure, K. A., Sabourin, M. J., Dahlqvist, G. and Helle, T. Adjusting refining intensity by changing refiner plate design and rotation speed-effects on structural fiber properties . Proceeding 85^th Annual meeting PAPTAC, B59-B65, Montreal. 1999.
Maloney, T. M. Modern Particleboard & Dry-process Fiberboard Manufacturing . Updated edition. San Francisco. Miller Freeman Inc. 1993.
Moss, P.A. and E. Retulainen. The effect of fines on fiber bonding: cross-sectional dimensions of TMP fibers at potential bonding sites . In: proceeding of the International Paper Physics Conference Niagara-on-the-lake, Ontario. Tappi: 97-101. 1995.
Myers, G. C. Characterization of fiberboard pulp . Forest Prod. J. 37(2): 30-36; 1987.
Nelson, N. D. Effects of wood and pulp properties on medium density, dry formed hardboard . Forest Prod. J. 23(9): 72-80; 1973.
SAS Institute, Inc. SAS/STAT User’s guide . Cary, N.C. 1990.
Suchsland, O. and Woodson, G. Effect of press cycle variables on density gradient of medium density fiberboard . In proceeding 8^thparticleboard symp. pp. 375-396; 1974.
Suchsland, O. and Woodson, G. Fiberboard manufacturing practices in the United States . Forest Products Research Society. 262p; 1990.
Tappi Standard. Fiber length of pulp by classification . T 233cm-95. 1995.
Wang, S., Winistorfer, P. M., Young, T. M., Helton, C. Step-closing pressing of medium density fiberboard, Part I: Influences on the vertical density profile . Holz-als-Roh-und-Werkstoff. 59(1-2): 19-26; 2001.
Wang, S., Winistorfer, P. and Young, T. Fundamentals of vertical density profile formation in wood composites. Part III. MDF density formation during hot-pressing . Wood Fiber Sci. 36(1): 17-25; 2004.
Winistorfer, P. M., Young, T. M., Walker, E. Modeling and comparing vertical density profiles . Wood Fiber Sci. 28(1): 133-141; 1996.
Woodson, G. E. Effects of bark, density profile, and resin content on medium density fiberboards from Southern hardboards . Forest Prod. J. 26(2): 39-42; 1976a.
Woodson, G. E. Properties of medium density fiberboard related to hardwood specific gravity . Proceeding 10^th Washington State University Particleboard Symposium. Pullman, WA. pp175-192; 1976b.
Zhang, S.Y., Yu, Q., Chauret, G., and Koubaa, A. Selection for both growth and wood properties in hybrid poplar clones . Forest Sci. 49(6): 1-8; 2003.

6.7 List of Tables

Table 6-1 Refining parameters, fiber moisture content and creation of dummy variables.
Table 6-2 Panel bending properties, internal bond strength and dimensional stability.
Table 6-3 Measured wood and fiber characteristics.
Table 6-4 Predictor/response variables and transformed predictor variables.
Table 6-5 Summary of stepwise model selection for MOR, MOE, IB, LE, TS, and WA.
Table 6-6 Multiple linear regression model parameter estimates for MOR, MOE, IB, LE, TS, and WA.

Table 6-1 Refining parameters, fiber moisture content and creation of dummy variables.

Refining speed, cooker pressure, retention time and temperature were 2500 rpm, 7.5 bar, 3 min and 160 ^oC for all types of material.
* Moisture content after resin and wax blending. 

Table 6-2 Panel bending properties, internal bond strength and dimensional stability.

Values in parenthesis represent standard deviations.
LE was determined by the linear variation with changing RH from 50 % to 80 % as the percentage change in length based on the length measured at 50 % RH. TS and WA were obtained by calculating the variation in thickness and weight following 24 h water soaking divided by the thickness and weight measured on the specimens equilibrated at 22 ^oC and 65 %.

Table 6-3 Measured wood and fiber characteristics.

Values in parentheses represent standard deviations.
Numbers in columns ‘>3.240 mm²’, ‘0.828-3.240 mm²’, ‘0.281-0.828 mm²’, and ‘0.017-0.281 mm²’ were the percentages of fibers retained on the screens with mesh sizes of 3.240 mm², 0.828 mm², 0.281 mm², and 0.017 mm². Numbers in column ‘<0.017 mm²’ were the percentages of fibers passed through the screen with mesh size of 0.017 mm².

Table 6-4 Predictor/response variables and transformed predictor variables.

Table 6-5 Summary of stepwise model selection for MOR, MOE, IB, LE, TS and WA.

Table 6-6 Multiple linear regression model parameter estimates for MOR, MOE, IB, LE, TS, and WA. 

Response Variable	Predictor Variables in Model	Parameter Estimates	Variance Inflation Factor	Model R2	Adjusted Model R2
MOR	intercept arith_fine	39.8 -0.509	0 1.00	0.5324	0.4805
MOE	intercept _200_ tanwood_pH z5	2.51×103 -409 45.6 479	0 1.21 1.35 1.21	0.8501	0.7859
IB	intercept arith_fine _200_ z7 fiber_pH	3.48 -0.0509 0.405 0.304 -0.383	0 3.52 2.00 1.68 3.03	0.9170	0.8617
LE	intercept z10 quadwood_density	0.209 0.0899 -0.244	0 1.18 1.18	0.8723	0.8403
TS	intercept arith_length	44.8 -38.7	0 1.00	0.8720	0.8578
WA	intercept cubfiber_width	158 -0.00530	0 1.00	0.9237	0.9152

6.8 List of Figures

Figure 6-1 Residual dispersion of MOR, MOE, IB, LE, TS, and WA models.

Figure 6-1 Residual dispersion of MOR, MOE, IB, LE, TS, and WA models.

For further details log on website :
http://archimede.bibl.ulaval.ca/archimede/fichiers/23592/ch05.html