Wood Science and Technology 35 (2001) 143±156 Ó Springer-Verlag 2001

Numerical and experimental analysis

of a wood drying process
D. MartinovicÂ, I. Horman, I. DemirdzÏicÂ

143
Abstract Experimental investigation and computational analysis were performed
to evaluate the in¯uence of the ambient air parameters during the drying process
on the temperature, moisture and resulting deformations and stresses in wood
samples. The numerical procedure uses the Finite Volume Method to discretise
the equations governing heat, mass and momentum balance and takes into
account the anisotropic nature of wood. The comparison of the numerical and
experimental results shows very good agreements, implying that the proposed
numerical algorithm can be used as a useful tool in designing wood drying
schedules.

Introduction
The wood drying process is an important step in the manufacturing of wood
products. During that process a non-uniform distribution of moisture content
and temperature causes deformation and stresses in the wood and may result in a
deformed and/or cracked end-product. Therefore, during the development of new
drying schedules, the internal stresses that arise in the wood are of major concern,
and their prediction is a pre-requisite for the design of a controlled, ef®cient, low-
cost drying process.
A wood drying process can be described as an unsteady process of heat, mass
and momentum transfer in an orthotropic continuum with variable physical
properties, and its numerical solution requires a very sophisticated algorithm.
There have been numerous attempts to solve this problem with various degrees of
complexity and completeness. The simplest ones are based on the solution of a 1D
diffusion equation for the moisture content by employing the Finite Element
Method (FEM) (Thomas et al. 1980) or the Finite Difference (FD) based solution
algorithm (Bui et al. 1980; Droin et al. 1988; El Kouali & Vergnaud 1991). The
others solve 1D problem of simultaneous heat and mass transfer in wood again
using either FD (Bramhall 1979; Mazjak & Ilkiv 1987) or FE method (Avramidis &
Hatzikiriakos 1995). Further generalisations are 2D solutions of energy and
concentration transport equations by using the FEM (Gui et al. 1994) or the Finite
Volume Method (FVM) (Pang 1996, 1997), and a 3D modelling of the moisture
absorption by FD (Droin-Josserand et al. 1989). Finally, there are methods

Received 31 March 1999

D. MartinovicÂ, I. Horman, I. DemirdzÏicÂ
MasÏinski fakultet u Sarajevu
Vilsonovo ÏsetalisÏte 9, 71000 Sarajevo
Bosnia and Herzegovina

Correspondence to: I. DemirdzÏicÂ

 capable of predicting temperature, moisture, deformation and stress ®elds in 2D
situations by employing FEM (Morgan et al. 1982) or FVM (Martinovic & Hor-
man 1998).
In this paper the development of a computational method based on the ®nite
volume discretisation of the governing equations is outlined and its applicability
to the analysis and design of the process of drying wooden beams is demon-
strated. The FVM was originally developed for ¯uid ¯ow and heat and mass
transfer calculations (Patankar 1980), and later generalised for stress analysis in
isotropic linear and non-linear bodies (DemirdzÏic & Muzaferija 1994; DemirdzÏicÂ
et al. 1997; DemirdzÏic & Martinovic 1993). For the purpose of the present analysis
144 the method described in detail in DemirdzÏic & Muzaferija (1995) is modi®ed to
take into account the anisotropic nature of the wood, and the in¯uence of the
moisture content on the deformation and stresses in the wood samples. It solves
a coupled set consisting of energy, moisture potential and momentum equations,
and gives the unsteady temperature, moisture, displacement and stress ®elds in
the solution domain. Although the governing equations, the constitutive relations,
and the numerical procedure are given for 3D situations, the examples of
application are all 2D.
In order to generate data for comparison, measurements of temperature,
moisture and deformation of wood samples in a laboratory dryer were conducted
for several drying regimes. It is shown that the predicted and measured values
compare very favourably.

Theory

Governing equations
A wood drying process is governed by the following energy, mass and momentum
balance equations which, when written in a Cartesian tensor notation, read:

o oqj
…qcq T† ˆ ‡ qsq …1†
ot oxj

o om_ j
…qcm M† ˆ ‡ qsm …2†
ot oxj
 
o oui orij
q ˆ ‡ qbi …3†
ot ot oxj

In these equations, t is time, xi is the Cartesian coordinate, q is the density, cq and

cm are the speci®c heat and the speci®c moisture, T is the temperature, M is the
moisture potential and ui is the displacement, sq and sm are the heat and mass
source, bi is the body force, and qj , m_ j and rij are the heat and mass ¯ux vector,
and the stress tensor components, respectively.

Constitutive relations
In order to close the system of Eqs. (1)±(3) the constitutive relations for heat and
mass ¯ux based on the theory of Luikov (1966) which takes into account both the
Soret and the Duffort effect, together with the constitutive relation for an elastic
body are used:
 q oT q oT oM
qj ˆ kjl ‡ e r m_ j ˆ …kjl ‡ e rdkm
jl † e rkm
jl …4†
oxl oxl oxl
oM oT
m_ j ˆ km
jl d km
jl …5†
oxl oxl
 
1 ouk oul
rij ˆ cijkl kl aij DT bij DM ˆ cijkl ‡ aij DT bij DM …6†
2 oxl oxk
q
Here kij and kmij are the heat and mass conduction coef®cient tensor components,
respectively, e is the ratio of the vapour diffusion coef®cient to the coef®cient of 145
total diffusion of moisture, r is the heat of the phase change, d is the temperature-
gradient coef®cient, ij are the strain tensor components, cijkl are the elastic
constant tensor components, aij are the coef®cients of thermal expansion, bij are
the shrinkage (contraction) coef®cients, DT ˆ T Tu , DM ˆ M Mh and Tu is
the temperature at an undeformed state and Mh is the moisture potential at the
®ber saturation point. For an orthotropic material and the coordinate axes
aligned with the symmetry axes, Eqs. (4)±(6) can be written in the following
matrix form:
2 3 2 q 32 3 2 3
q1 k11 0 0 oT=ox1 m_ 1
4 q2 5 ˆ 4 0 kq 0 54 oT=ox2 5 e r4 m_ 2 5 …7†
22
q
q3 0 0 k33 oT=ox3 m_ 3
2 3 2 m 32 3 2 m 32 3
m_ 1 k11 0 0 oM=ox1 k11 0 0 oT=ox1
4 m_ 2 5 ˆ 4 0 km 22 0 54 oM=ox 2
5 d4 0 km 22 0 54 oT=ox2 5
m
m_ 3 0 0 k33 oM=ox3 0 0 km
33 oT=ox3
…8†
2 3 2 32 3
r11 A11 A12 A31 0 0 0 11 a11 DT hb11 DMi
6 r22 7 6 A12 A22 A23 0 0 0 7 6 22 a22 DT hb22 DMi 7
6 7 6 76 7
6 r33 7 6 A31 A23 A33 0 0 0 7 6 a33 DT hb33 DMi 7
6 7 6 76 33 7
6 r12 7 ˆ 6 0 0 0 A44 0 0 76
7 12 7
6 7 6 6 7
4 r23 5 4 0 0 0 0 A55 0 54 23 5
r31 0 0 0 0 0 A66 31
…9†

where the terms in hi brackets are `active' only for M < Mh , while the nine
non-zero orthotropic elastic constants Aij are related to the Young's moduli
Ei , the Poisson's coef®cients mij and the shear moduli Gij by the following
relations:

1 1 m23 m32 1 1 m13 m31 1 1 m21 m12

A11 ˆ ; A22 ˆ ; A33 ˆ ;
J E2 E3 J E1 E3 J E2 E1
1 m12 ‡ m32 m13 1 m31 ‡ m21 m32 1 m23 ‡ m21 m13 …10†
A12 ˆ ; A31 ˆ ; A23 ˆ ;
J E1 E3 J E2 E3 J E1 E2
A44 ˆ 2G12 ; A55 ˆ 2G23 ; A66 ˆ 2G31 ;
 1 m12 m21 m23 m32 m31 m13 2m21 m32 m13
Jˆ
E1 E2 E3

Note that the pair of constitutive Equations (4) and (5) can be extended to take
into account the effect of the pressure gradient on the heat and mass transfer.

Initial and boundary conditions

In order to complete the mathematical model of a wood drying process, initial
and boundary conditions have to be speci®ed. As initial conditions, the
temperature, the moisture potential, and the displacement and velocity
146 components have to be speci®ed at all points of the solution domain. For a
convective drying process, the following boundary conditions are normally
appropriate:

oT
q
kjl nj ‡ hq …T Ta † ‡ …1 e†rhm …M Ma † ˆ 0
oxl
oM oT …11†
km
jl nj ‡ hm …M Ma † ‡ d km
jl nj ˆ 0
oxl oxl
rij nj ˆ fsi

where hq and hm are the (convective) heat and mass transfer coef®cients,
respectively, fsi is the surface traction, and all quantities are calculated at the
solution domain boundary, except for those with subscript a which correspond
to the ambient air.

Numerical method

Generic transport equation

Before the construction of a numerical algorithm is started, it is important to notice
that the governing Equations (1)±(3) when combined with constitutive Equations
(7)±(9), can be written in the form of the following generic transport equation:
 
o o / o/
…qB/ † Cjl S/ ˆ 0 …C/jl ˆ 0 for j 6ˆ l† …12†
ot oxj oxl

which can be integrated over an arbitrary solution domain V bounded by the

surface S, with unit outer normal vector nj to yield:
Z Z Z
o o/
…qB/ †dV C/jl nj dS S/ dV ˆ 0 …C/jl ˆ 0 for j 6ˆ l† …13†
ot oxl
V S V

The generic variable / stands for T, M or ui and the meaning of the coef®cients
B/ , C/jj and S/ is given in Table 1.

Finite volume discretisation

As all numerical methods, the present one consists of time, space and equations
discretisation. The time interval of interest is subdivided into a number of
subintervals dt, not necessarily of the same length. The space is discretised by a
Table 1. The meaning of B/ , C/jj and S/ in Eqs. (12) and (13)
/ B/ C/11 C/22 C/33 S/
q m q m q m
 
T cq T k11 ‡ e rd k11 k22 ‡ e rd k22 k33 ‡ e rd k33 o oM
e rkjlm ‡ q sq
oxj oxl
m m m  
M cm M k11 k22 k33 o oT
d kjlm ‡ q sm
oxj oxl
     
u1 o u1 A11 A44 A66 o o u2 o u3 o A44 ou2 o A66 ou3
A12 ‡ A31 ‡ ‡
ot 2 2 o x1 ox2 ox3 ox2 2 ox1 ox3 2 ox1
o 
A11 …a11 DT ‡ hb11 DMi† ‡ A12 …a22 DT ‡ hb22 DMi†
ox1

‡ A31 …a33 DT ‡ hb33 DMi† ‡ q b1
     
u2 ou2 A44 A22 A55 o A44 ou1 o ou1 ou3 o A55 ou3
‡ A12 ‡ A23 ‡
ot 2 2 ox1 2 ox2 ox2 ox1 ox3 ox3 2 ox2
o 
A12 …a11 DT ‡ hb11 DMi† ‡ A22 …a22 DT ‡ hb22 DMi†
ox2

‡ A23 …a33 DT ‡ hb33 DMi† ‡ q b2
     
u3 ou3 A66 A55 A33 o A66 ou1 o A55 ou2 o ou1 ou2
‡ ‡ A31 ‡ A23
ot 2 2 ox1 2 ox3 ox2 2 ox3 ox3 ox1 ox2
o 
A31 …a11 DT ‡ hb11 DMi† ‡ A23 …a22 DT ‡ hb22 DMi†
ox3

‡ A33 …a33 DT ‡ hb33 DMi† ‡ q b3
147
 number of contiguous, non-overlapping hexahedral control volumes (CV), with
the computational points at their centres (Fig. 1). Then the integrals in generic
equation (13) are calculated by employing the midpoint rule, the gradients are
evaluated by assuming a linear variation of the dependent variable / between the
computational points, and a fully implicit temporal scheme is employed. As a
result a non-linear algebraic equation of the following form for each CV is
obtained:
X
aP /P aK /K ˆ b …K ˆ W; E; S; N; B; T† …14†
K
148
where the coef®cients aK and b are de®ned as:

Se
aE ˆ …C/11 †e and similar expressions for other cell faces
dxe
 
V B/ o
aoP ˆ q …15†
dt / P
X
aP ˆ aK ‡ aoP
K

b ˆ S/P V ‡ aoP /oP

where the subscripts P and e denote values at the centre of the CV and at the
centre of the east cell-face, respectively, Se is the area of the east cell face, V is the
volume of the CV, dxe is the distance between points P and E, and all quantities
refer to the current time level, except for those with the superscript o which refer
to the previous, `old' time level.

Solution algorithm
After assembling Eqs. (14) for all CVs and for all transport equations, ®ve (four in
2D case) sets of N mutually coupled non-linear algebraic equations are obtained,

Fig. 1. A typical control volume and the

compass labelling scheme
where N is the number of CVs. Those equations are solved by employing the
following segregated iterative procedure.
First, all dependent variables are given their initial values. Then the boundary
conditions which correspond to the ®rst time step are applied, and the sets of
equations for each individual dependent variable (T, M, ui ) are linearised and
temporarily decoupled by assuming that coef®cients aK and source terms b are
known (calculated by using dependent variable values from the previous iteration
or the previous time step), resulting in a system of linear algebraic equations of
the form:

A/ / ˆ b/ …16† 149

for each dependent variable, where A/ is an N  N matrix, vector / contains

values of dependent variable / at N nodal points and b/ is the source vector.
Systems (16) are then solved sequentially in turn until a converged solution is
obtained. The procedure is assumed converged when the following conditions are
satis®ed for all ®ve (four in 2D case) sets of equations:

X
N X
jaP /P aK /K bj < pR/
iˆ1 K …17†
j/ni /in 1 j < qj/ni j

where p and q are typically of the order 10 3 , R/ is a suitable normalisation factor

and superscripts n and n 1 denote values at two successive iterations.
In the next time step the whole procedure is repeated, except that the initial
values are replaced by the values from the previous time step.
The present discretisation procedure ensures that the matrix A/ has the fol-
lowing desirable properties: it is seven (®ve in 2D case) ± diagonal, symmetric,
positive de®nite and diagonally dominant, which makes Eq. (16) easily solvable
by a number of iterative methods which retain the sparsity of the matrix A/ . Note
that it does not make sense to solve Eq. (16) to a tight tolerance since its coef-
®cients and sources are only approximate (based on the values from the previous
iteration/time step). Normally, reduction of the absolute residuals for one order of
magnitude suf®ces.
The segregated solution strategy employed enables re-use of the same storage
for the matrix A and vector b for all dependent variables /, thus requiring only
8N storage locations (6N in a 2D case). It is also important to mention that the
fully implicit time differencing used, avoids stability-related time step restric-
tions. In principle, it allows any magnitude of the time step to be used, and in
practice it is limited only by the required temporal accuracy.

Experiment

Material
50  50 mm thick, 600 mm long beech-wood beams are exposed to the (uniform,
unsteady) ¯ow of the hot air in a laboratory dryer with an automatic control of
the ambient air parameters. The temperature and/or moisture dependent physical
properties of the wood, obtained by ®tting available experimental data, are given
 in Table 2. The others are considered constant and are given in Table 3. The
timber is known to be cylindrically orthotropic. However, the wood samples used
in this study are taken from the outer region of a cylindrical timber log and the
rectilinear isotropy of samples is a reasonable assumption.
For the given data the anisotropy ratios, de®ned as

2A44 2A44
a1 ˆ ; a2 ˆ …18†
A11 A12 A22 A12

have the values a1 ˆ 3:16±10.7, and a2 ˆ 0:6±1.96, depending on the temperature

150 and the moisture content.

Apparatus and experiment

The temperature, moisture and deformation (displacements) of the wood samples
are measured at 24 points (Horman 1999). Temperature was measured at eight
points by the Ni±CrNi thermo-elements which were connected to a multi-channel
thermometer LSB36 (Linseis) and registered every 600 s. The accuracy of the
instrument was 0.01  C.
The moisture was measured at six points corresponding to two symmetry
directions by an FMD hybrid sensor, based on measurements of the wood electric
resistance. The measurements were taken every 6 h at the initial stage of drying
and every 12 h at the later stages. The accuracy of the FMD hygrometer was 0.1%.

Table 2. Temperature and/or moisture dependent physical properties of wood (C ˆ cm M

(%) is the moisture content)

Property C < 30% C  30%

 
E1 (Pa) 7
6:69 4:66 e 1:1
10 C
6:30
…1:8 0:02T †108 2:05…1:8 0:02T †108
 
E2 (Pa) 6
13:22 9:30 e 2:5
10 C
5:75
…1:8 0:02T †108 4:04…1:8 0:02T †108
 
E3 (Pa) 6
81:11 57:0 e 2:5
10 C
5:75
…1:8 0:02T †108 24:8…1:8 0:02T †108
 
q (kg/m3) 559…100 ‡ C† C
559 1
100 0:47…30 C† 100
cq (J/kg K) 467‰C…100 ‡ T †Š0:2
q
k11 (W/m K) 1:36…0:088 ‡ 0:000709T ‡ 0:00181C†
q q
k22 (W/m K) 1:15k11

Table 3. Constant physical properties of wood

Property Value Property Value Property Value

r (J/kg) 2:3
106 m12 0.36 a11 (1/K) 37:6
10 6

cm (kgm /kg°M) 0.01 m21 0.71 a22 (1/K) 28:4
10 6
m
k11 (kgm /m s °M) 4:5
10 9 m13 0.043 a33 (1/K) 4:16
10 6
m
k22 (kgm /m s °M) 1:15k11 m31 0.52 b11 (1/°M) 36:8
10 4
G12 (Pa) 3
108 m23 0.073 b22 (1/°M) 18:0
10 4
d (°M/K) 2 m32 0.45 b33 (1/°M) 1:8
10 4
 The deformations of wood samples were determined by measuring the
displacements at 10 points on the sample surface by employing a measuring
transducer (type kHz TF MessverstaÈrker KWS 3073). The accuracy of the reading
was 10 3 mm.

Results and discussion

At the beginning of the drying process the wood samples had a uniform
distribution of temperature, moisture, displacement and velocity:

T ˆ 21  C; M ˆ 75 M; ui ˆ u_ i ˆ 0 for t ˆ 0:
151
The coef®cients of convective heat and mass transfer, based on the ambient air
velocity of va ˆ 2 m/s and moisture of Ma ˆ 10:5 M, were taken as:

hq ˆ 40 W/m2 K; hm ˆ 1:8
10 6
kg/m2 s M;

while the ambient air temperature and the ratio of the vapour diffusion coef®cient
to the coef®cient of total diffusion of moisture were assumed to vary during the
drying process according to the following schedules:
8 8
< 28; 0  t < 10 min < 0:1; 0  t < 60 min
Ta ˆ 0:35t ‡ 24:5; 10  t < 70 min ; e ˆ 0:5; 60  t < 3660 min
: :
49; t  70 min 1:0; t  3660 min

Zero surface fractions are assumed and boundry conditions (11) are applied.
For the purpose of the numerical calculations the problem is considered to be a
2D plane strain problem. Due to the double symmetry, only one quarter of the
cross-section is taken as the solution domain, shown in Fig. 2 together with the
reference points. For all calculation presented in this study, a uniform numerical
mesh consisting of 20  20 CV was employed, while the time step was varied from
10 to 100 min (®rst seven time steps of 10 min, 31 time steps of 30 min, and
®nally 140 time steps of 100 min). These results are found to be grid and time
independent by performing a systematic grid and time-step re®nement (differ-
ence between the results on the 20  20 CV mesh differ from ones obtained on a
40  40 CV mesh for less than 1%, while the results obtained with dt ˆ 3 h
practically coincide with results obtained with dt ˆ 1:5 h).

Numerical analysis
During the initial phase of drying (0  t < 2 h) the moisture content is above the
®ber saturation point and the deformation is a consequence of the thermal
stresses only. Figure 3 shows the calculated ®elds at t ˆ 70 min. One can see that
an increase in temperature (Fig. 3a) causes the expansion of wood sample
(Fig. 3b) and that the outer region is subjected to compressive and the inner
region to extensive stresses (Fig. 3c, d).
During the period of intensive drying (60  t < 190 h) the deformation
and stresses due to hygroscopic loads dominate. Figure 4 shows that at t ˆ 166 h
the moisture content has fallen below the ®ber saturation point (Fig. 4a) and that
this causes the shrinking of the wood sample (Fig. 4b). Around t ˆ 100 h the
stresses reach their maximum values and are extensive in the outer region and
compressive in the interior of the sample (Fig. 4c, d). By comparing the values of
152

Fig. 2. Solution domain and reference points

Fig. 3. Temperature (a), displacement (b), and normal stresses (c) and (d) at t ˆ 70 min

stresses at t ˆ 70 min and at t ˆ 166 h, it can be seen that the thermal stresses are
around 200 times smaller than the stresses caused by the drop in the moisture
content below the ®ber saturation point.
If one plots the contours of the effective stress at t ˆ 166 h, when it is at its
maximum (Fig. 5), one can see that the effective stress is greater than the yield
stress (ry ˆ 10 MPa at 10% moisture; ry ˆ 6 MPa at 30% moisture) only in a
very narrow surface region (1 mm deep), which indicates that the plastic defor-
 153

Fig. 4. Moisture content (a), displacement (b), and normal stresses (c) and (d) at t ˆ 166 h

Fig. 5. Effective stress at t ˆ 100 h

mation did not take place in the interior of the sample, and that the drying
schedule is well designed.
At the end of the drying process (t ˆ 250 h), the moisture content in the sample
varies from 10.8 to 13.8% (Fig. 6a), while Fig. 6b illustrates the anisotropy of the
wood sample (the contraction is 1.3 mm in the x and 0.6 mm in the y direction).

Comparison with experiments

In order to con®rm the validity of the FV predictions, the calculated temperature,
moisture and displacements are compared with experimental data at reference
points (shown in Fig. 2). Figures 7 and 8 show temperature and moisture content
154

Fig. 6. Moisture content (a) and

displacement (b) contours at the
end of the drying schedule
(t ˆ 250 h)

Fig. 7. Temperature history at reference points A and B

histories at two reference points. One can see a good agreement between calcu-
lations and experiment: maximum difference for both temperature and moisture
was 8%, and the average difference was less than 2%.
 Figure 9 shows how the displacements at two points on the surface of the
sample vary during the drying process. One can see very little deformation during
the initial phase (t  1000 min) and a considerable shrinking of the sample
afterwards, and that predictions closely follow experimental data (maximum
difference 15%, average difference 5%).

Conclusions
An experimental investigation and numerical predictions of the wood drying
process have been performed.
Development and application of a FV based computational method for
prediction of the distribution of temperature, moisture content, deformation 155
and stresses in wood during a drying process is presented. The method is
shown to be simple, unconditionally stable and conservative, and its accuracy was
demonstrated by comparing numerical results with experimental data.

Fig. 8. Moisture content history at reference points C and D

Fig. 9. u Displacement at reference point E and v displacement at reference point F

 The segregated iterative solution procedure employed makes the method both
memory and CPU time ef®cient, and enables a fast analysis and design of wood
drying schedules.

References
Avramidis S, Hatzikiriakos SG (1995) Convective heat and mass transfer in nonisothermal
moisture desorption. Holzforschung 49: 163±167
Bramhall G (1979) Mathematical model for lumber drying ± I. Principles involved, and II.
The model. Wood Sci. 12: 14±31
Bui X, Choong ET, Rudd WG (1980) Numerical methods for solving the equation for
diffusion through wood during drying. Wood Sci. 13: 117±121
156 DemirdzÏic I, Martinovic D (1993) Finite volume method for thermo-elasto-plastic stress
analysis. Comput. Methods Appl. Mech. Engrg. 109: 331±349
DemirdzÏic I, Muzaferija S (1994) Finite volume method for stress analysis in complex
domains. Int. J. Numer. Methods Engrg. 37: 3751±3766
DemirdzÏic I, Muzaferija S (1995) Numerical method for coupled ¯uid ¯ow, heat transfer
and stress analysis using unstructured moving meshes with cells of arbitrary topology.
Comput. Methods Appl. Mech. Engrg. 125: 235±255
DemirdzÏic I, Muzaferija S, Peric M (1997) Benchmark solutions of some structural analysis
problems using ®nite-volume method and multigrid acceleration. Int. J. Numer. Methods
Engrg. 40: 1893±1908
Droin A, Taverdet JL, Vergnaud JM (1988) Modelling the kinetics of adsorption by wood.
Wood Sci. Technol. 22: 11±20
Droin-Josserand A, Taverdet JL, Vergnaud JM (1989) Modelling the process of moisture
absorption in three dimensions by wood samples of various shapes: cubic, parallelepipedic.
Wood Sci. Technol. 23: 259±271
El Kouali M, Vergnaud JM (1991) Modeling the process of absorption and desorption of
water above and below the ®ber saturation point. Wood Sci. Technol. 25: 327±339
Gui YQ, Jones EW, Taylor FW, Issa CA (1994) An application of ®nite element analysis to
wood drying. Wood Fiber Sci. 26(2): 281±293
Horman I (1999) Finite volume method for analysis of timber drying. PhD Thesis,
University of Sarajevo (In Bosnian)
Luikov AV (1966) Heat and Mass Transfer in Capillary Porous Bodies. Oxford: Pergamon
Press
Martinovic D, Horman I (1999) Finite volume method for analysis of hygro-thermal elastic
stresses. Strojarstvo 41(1,2): 39±48 (In Croatian)
Mazjak ZY, Ilkiv IN (1987) An optimisation of a simple drying chambers wood dryer with a
variable drying regime. Lesnoj zurnal 5: 69±74 (In Russian)
Morgan K, Thomas HR, Lewis RW (1982) Numerical modeling of stress reversal in timber
drying. Wood Sci. 15: 139±149
Pang S (1996) Moisture content gradient in a softwood board during drying: simulation
from a 2-D model and measurement. Wood Sci. Technol. 30: 165±178
Pang S (1997) Relationship between a diffusion model and a transport model for softwood
drying. Wood Fiber Sci. 29: 58±67
Patankar SV (1980) Numerical Heat Transfer and Fluid Flow. New York: McGraw-Hill
Thomas HR, Lewis RW, Morgan K (1980) An application of the ®nite element method to
the drying of timber. Wood Fiber Sci. 11: 237±243