b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/issn/15375110

Research Paper

Determination of diffusion and convective transfer

coefficients in food drying revisited: A new
methodological approach

Francisco J. Arranz a, Tatiana Jim
enez-Ariza b, Bel
en Diezma c,

Eva C. Correa d,*
a
Grupo de Sistemas Complejos, Departamento de Ingenierı́a Agroforestal, ETSI Agrono
mica Alimentaria y de
Biosistemas, Universidad Polit
ecnica de Madrid, Av. Puerta de Hierro 2, 28040 Madrid, Spain
b
Facultad de Ciencias Ba
sicas e Ingenierı́a, Universidad San Buenaventura, Carrera 8H # 172 e 20, Bogota
,
Colombia
c
Laboratorio de Propiedades Fı́sicas y T
ecnicas Avanzadas en Agroalimentaci
on (LPT_TAGRALIA), Departamento de
Ingenierı́a Agroforestal, ETSI Agron
omica Alimentaria y de Biosistemas, Universidad Polit
ecnica de Madrid, Av.
Puerta de Hierro 2, 28040 Madrid, Spain
d
Laboratorio de Propiedades Fı́sicas y T
ecnicas Avanzadas en Agroalimentaci
on (LPF_TAGRALIA), Departamento de
Quı́mica y Tecnologı́a de Alimentos, ETSI Agron
omica Alimentaria y de Biosistemas, Universidad Polit
ecnica de
Madrid, Av. Puerta de Hierro 2, 28040 Madrid, Spain

article info
It is usual to determine diffusion and convective transfer coefficients using classical
Article history: approximate relationships involving dimensionless numbers. These approximate relation-
Received 3 January 2017 ships were developed in the past, in order to avoid the difficulties associated with calculations
Received in revised form of large series expansions and its corresponding expansion coefficients given by transcen-
9 July 2017 dental equations. However, the development of improved computing techniques has
Accepted 24 July 2017 removed these difficulties, such that currently such calculations can be easily performed. An
Published online 12 August 2017 iterative methodology is proposed that takes advantage of current computational capabil-
ities, avoiding to use approximate relationships. The proposed methodology is applied to
Keywords: generated data and also to experimental data from carrot drying. Additionally, a MATLAB®
Mass Biot number implementation of the proposed iterative methodology, along with the input data files cor-
Mass Fourier number responding to the results presented in this paper, is provided as supplementary material.
Iterative numerical methodology © 2017 IAgrE. Published by Elsevier Ltd. All rights reserved.
Carrot

conditions, the correct and accurate determination of the

1. Introduction physical parameters involved is a central issue. Especially
important in food engineering are heat and mass transfer
In the development of suitable models for complex processes processes, which are characterised by the corresponding
involving foods, which are applicable to a broad set of external diffusion and convective transfer coefficients.

* Corresponding author.
E-mail address: evacristina.correa@upm.es (E.C. Correa).
http://dx.doi.org/10.1016/j.biosystemseng.2017.07.005
1537-5110/© 2017 IAgrE. Published by Elsevier Ltd. All rights reserved.
 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9
Nomenclature
Bi Biot number
D Effective diffusion coefficient (m2 s
1)
Di Effective diffusion coefficient calculated at the i-th
iteration (m2 s
1)
F Mass flow (g [water] (m
2 s
1)
Fo Dimensionless time (Fourier number)
M Dry basis moisture (g [water] g
1 [dry matter])
M Position averaged moisture (g [water] g
1 [dry
matter])
M0 Initial uniform moisture (g [water] g
1 [dry
matter])
Meq Moisture at equilibrium (g [water] g
1 [dry matter])
M[ Moisture at food sample-air interface (g [water]
g
1 [dry matter])
SD Generalised significant figures in diffusion
coefficient
Sh Generalised significant figures in mass transfer
coefficient
T Sample temperature ( C)
a Intercept in the fitting to the straight line y ¼ a þ bt
b Slope in the fitting to the straight line y ¼ a þ bt
(s
1)
h Convective mass transfer coefficient (m2 s
1)
hi Convective mass transfer coefficient calculated at
the i-th iteration (m2 s
1)
[ Half thickness of the sample (mm)
[y Half width of the sample (mm)
It is noteworthy that for industrial purposes, as is pointed
out by Zhao et al. (2014), foods are usually dried to remove the
moisture up to a certain level at which microbial spoilage and
deteriorative chemical reactions are minimised, and accurate
determination of diffusion and convective transfer co-
efficients is required in order to facilitate accurate designs.
Different methods have been described in the literature to
obtain experimental values for these coefficients. Zhao et al.
(2014) used a trial-and-error method based on a comparison
of analytic and experimental drying curves, tuning the value
of the diffusion coefficient until the matching of both curves is
reached approximately. In similar fashion, Białobrzewski
(2007) obtained both diffusion and convective transfer co-
efficients by comparing numerical (as opposed to analytical)
and experimental drying curves. Note that analytical solu-
tions are available only for some simple systems, whilst nu-
merical solutions are available for any system. Conversely,
analytical solutions have arbitrarily high accuracy, but the
accuracy of numerical solutions is limited by numerical
approximation, especially in the case of partial differential
equations. Also, in order to obtain the convective transfer
coefficient, the reaction engineering approach has been
applied by considering drying as a competitive process be-
tween condensation and evaporation (Compaore et al., 2017).
However, the most widely used method is the fitting of the
experimental drying curves, in logarithmic form, to a straight
line. For simple one-dimensional geometries, such as for sli-
ces or cylinders, by assuming the first term approximation in
31
[z Half height of the sample (mm)
t Time (s)
ti The i-th time measured in the drying process (s)
v Air velocity induced by the extractor fan (m s
1)
w0 Initial sample weight in the drying process (g)
wdm Sample dry matter weight (g [dry matter])
weq Sample weight at equilibrium in the drying
process (g)
wi Sample weight at each time ti in the drying process
(g)
x Position across thin layer sample thickness (mm)
yi The i-th ordinate value corresponding to the
average dimensionless relative moisture at each
time ti in the drying process
Di Convergence of first term approximation for the
i-th iteration
F Average dimensionless relative moisture
FN The N terms series expansion approximation to
the average dimensionless relative moisture
mn The n-th positive root of the characteristic
equation mntanmn ¼ Bi
r Sample dry matter density (g [dry matter] m
3 )
s2D Variance of the diffusion coefficient (m2 s
1)
s2h Variance of the mass transfer coefficient (m s
1)
s2i Variance corresponding to the i-th ordinate value
yi in the drying process
s2w Variance corresponding to the instrumental error
of the weighing device (g)
the series expansion solution, diffusion coefficient can be
obtained from the corresponding slope (Karim & Hawlader,
2005; Mghazlia et al., 2017; Sampaio et al., 2017; Srikiatden &
Roberts, 2006). Additionally, by using approximate relation-
ships involving dimensionless numbers, in particular the Biot
number, convective transfer coefficient is also obtained
(Dhalsamant, Tripathy, & Shrivastava, 2017; Guine
, Cruz, &
Mendes, 2014; Tripathy & Kumar, 2009). Note that, in order
to obtain the correct values of the coefficients, it is important
that the experimental conditions correspond to the mathe-
matical conditions under which the analytical solution was
obtained. However, it is not common to find in the literature
details explaining the drying experimental methodology
which could fulfil such conditions.
Moreover, it is important to determine the accuracy of the
resulting coefficients and the corresponding significant fig-
ures. Note that the instrumental error of the experimental
data will propagate in the calculations when the coefficients
are obtained, such that the significant figures in the initial
experimental data will decrease (or at best remain the same)
in the calculated coefficients. In this regard, as is well known,
the calculability of diffusion and convective transfer co-
efficients depends on the Biot number, therefore this issue
should be analysed in any method for the determination of
diffusion and convective transfer coefficients.
In drying experiments, since the mass diffusion coefficient
(and also convective mass transfer coefficient) depends on
temperature, in order to calculate the mass diffusion
32 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9
coefficient (and convective mass transfer coefficient) by fitting
to an analytical solution of mass diffusion equation where the
diffusion coefficient (and convective mass transfer coefficient)
is assumed to be constant, it is mandatory to maintain the
sample with uniform temperature in its whole volume. In this
regard, as was shown by Srikiatden and Roberts (2006) when
studying potatoes and carrot drying, when the initial tem-
perature of the sample and target temperature differ, the
maintenance of surrounding air temperature at target tem-
perature is not sufficient to achieve uniform temperature,
with temperature gradients occurring inside the sample. It is
necessary firstly to heat the sample, using a microwave device
which leads to uniform temperature, up to target tempera-
ture, and then maintain the surrounding air temperature at
the target temperature throughout the drying assay. The likely
reason for wrong determination of the mass diffusion coeffi-
cient is due to existence of temperature gradients inside the
sample, as was pointed out in the early literature (Chen &
Johnson, 1969; Chirife, 1983; King, 1968; Vaccarezza,
Lombardi, & Chirife, 1974).
Additionally, aside from temperature, other conditions must
be controlled in order to correctly obtain mass diffusion and
convective mass transfer coefficients. Thereby, sample volume
is usually assumed constant in obtaining analytical solutions of
diffusion equation, so that shrinkage should be controlled in
food samples (e.g., by taking drying data over the initial period,
when shrinkage is negligible). Besides, the asymptotic moisture
value (i.e., sample moisture at infinite time) is also usually
assumed constant, so that the relative humidity of the sur-
rounding air should be maintained invariable.
On the other hand, in the fitting of the corresponding
analytical solution to obtain mass diffusion and convective
mass transfer coefficients, some approximations are usually
made. Firstly, in the infinite series expansion of the analytical
solution, the first term approximation is taken, that is, solu-
tion is approximated by taking only the first term in series
expansion. As is well known, this is a very accurate approxi-
mation only for large enough time (Crank, 1975; Luikov, 1968).
Time threshold is well defined in terms of dimensionless time,
but dimensionless time depends on the diffusion coefficient,
so that some iterative process should be used. Secondly, the
logarithmic form of first term approximation leads to a
straight line, so that the fitting process is straightforward.
Nevertheless, direct calculation of diffusion and convective
transfer coefficients from fitting intercept and slope involves
the solving of a transcendental equation, so that different
approximate relationships involving dimensionless numbers
were developed in order to avoid this difficulty (Luikov, 1968;
Pflug & Blaisdell, 1963). Despite that these difficulties being
easily avoidable, due to the development of computers and
numerical methods, the approximate classical schemes are
still used (Guine
et al., 2014; Karim & Hawlader, 2005;
Srikiatden & Roberts, 2006; Tripathy & Kumar, 2009).
In order to determine diffusion and convective transfer
coefficients from drying data, an iterative methodology is
presented that takes advantage of current numerical and
computational capabilities, avoiding to use the approximate
classical schemes, and numerically solving the corresponding
transcendental equations. As an additional feature, from the
range of obtained parameters involved in this numerical
solving, it can be identified the mismatching of experimental
conditions and mathematical conditions, and so, the unsuit-
ability of used data. As will be explained in detail below,
experimental data are first transformed to an asymptotic
linear form and then fitted to a straight line. The mathemat-
ical conditions under which the analytical solution is obtained
lead to a very narrow range for the intercept of this asymptotic
straight line. So that, a fitting intercept value out of range
implies that experimental data have been obtained under
experimental conditions that differs from mathematical
conditions, hence is unsuitable for the determination of
diffusion and convective transfer coefficients.
The proposed methodology is applied to generated data, in
order to test its behaviour, and also to experimental data from
carrot drying at different temperatures, including an unsuitable
data set corresponding to non-isothermal drying, included in
order to illustrate the detection of the unsuitability of data set.
Additionally, a MATLAB® implementation of the proposed
iterative methodology, along with the input data files corre-
sponding to the results presented in the paper, is provided as
supplementary material.
2. Materials and methods
2.1. Determination of diffusion and convective transfer
coefficients
2.1.1. Mathematical model
It is assumed that, for a thin layer sample, food drying process
is well described by means of the one-dimensional diffusion
equation


vM v vM
¼ D ; (1)
vt vx vx
with uniform initial condition
Mðx; t ¼ 0Þ ¼ M0 ; (2)
and boundary conditions


vM
¼ 0; (3)
vx x¼0
and


vM h 
¼
M[
Meq ; (4)
vx x¼[ D
where M is the dry basis moisture, t is the time, x is the position
across thin layer sample thickness (with origin x ¼ 0 at internal
centre), [ is the sample half thickness, and D and h are the
physical drying parameters, namely, effective diffusion and
convective mass transfer coefficients, respectively. Besides,
M[ ¼ M(x ¼ [, t) is the moisture at food sample-air interface, and
Meq ¼ Mðx ¼ [; t/∞Þ is the corresponding moisture at equilib-
rium. Note that symmetric external conditions, i.e. the same air
conditions at both sides of the thin layer, have been assumed.
Since initial condition in Eq. (2) is also symmetric, the whole
system is symmetric, so that the mass flow F ¼
rD(vM/vx) given
by Fick's law, where r is the dry matter density, will be null at the
internal centre, leading to the boundary condition in Eq. (3).
 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9 33

Moreover, if food sample temperature is maintained con- FðFo; BiÞ in Eq. (9) must be calculated with sufficient accuracy,
stant during drying process, then the diffusion coefficient D i.e, by taking a large enough number of terms N in the series
can be assumed to be constant. Additionally, if the air velocity expansion leading to a high accuracy approximation
is constant, the convective transfer coefficient h can also be FN ðFo; BiÞ. For this purpose, the corresponding first N roots of
assumed constant. Finally, if the relative humidity of air is the transcendental equation in Eq. (10) need to be calculated
maintained constant, then the moisture at equilibrium Meq with high accuracy, too. However, since the roots are clearly
will be constant. Under these additional conditions, the so- bounded by the inequalities in Eq. (11), there exist different
lution of diffusion equation in Eq. (1) under initial condition in numerical algorithms that are able to calculate them very
Eq. (2), and boundary conditions in Eqs. (3) and (4), is given by efficiently. Specifically, the algorithm of Forsythe, Malcolm,
the following series expansion (Crank, 1975; Luikov, 1968) and Moler (1977) have been used, which is implemented in
 x
 the MATLAB® function fzero.
Mðx; tÞ
Meq X ∞
2sinmn Dt
¼  cos mn exp
m2n 2 ; Moreover, it is well known that the series expansion in Eq.
M0
Meq n¼1
mn þ sinmn cosmn [ [
(9) converges rapidly and monotonically for high enough
(5)
values (depending on Biot number value) of the Fourier number
being mn the positive roots of the equation (Crank, 1975; Luikov, 1968), so that, as time increases, moisture
h[ curve tends asymptotically to the first term in the series, that is
mn tanmn ¼ ; (6)
D
lim FðFo; BiÞ ¼ F1 ðFo; BiÞ: (12)
ordered such that 0 < mn < mnþ1. The experimentally measur- Fo/∞

able overall moisture content M of the sample is obtained by In other words, for high enough values of the Fourier
averaging over the whole sample volume number, moisture time evolution FðFo; BiÞ is well described by
the first term approximation F1 ðFo; BiÞ, that expressed in
Z[
1 logarithmic form
MðtÞ ¼ Mðx; tÞdx; (7)
[
2 
0 2sin m1 m2 D
logFzlog
12 t; (13)
m1 ðm1 þ sinm1 cosm1 Þ [
so that, the overall content moisture corresponding to series
expansion in Eq. (5) will be takes the form of a straight line y ¼ a þ bt with intercept



MðtÞ
Meq X ∞ 2 2
2sin mn Dt 2sin m1
¼  exp
m2n 2 : (8) a ¼ log ; (14)
M0
Meq n¼1
mn ðmn þ sinmn cosmn Þ [ m1 ðm1 þ sinm1 cosm1 Þ

As is well known, the solutions of diffusion equation can be and slope

expressed in terms of dimensionless magnitudes (Crank, 1975;
m2 D
Luikov, 1968). In this way, the series expansion in Eq. (8) can be b ¼
12 : (15)
[
expressed as follows
So that, by taking a set of drying data {w0,w1,…,weq}, the
X
∞
2Bi
2
  corresponding ordinate values can be calculated1
FðFoÞ ¼  2  exp
m2n Fo ; (9)

n¼1 m2n Bi þ Bi þ m2n wi
weq
yi ¼ log ; (16)
where F ¼ ½MðtÞ
Meq =ðM0
Meq Þ is the average dimension- w0
weq
less relative moisture, Fo ¼ Dt/[2 is the dimensionless time where w0 is the initial sample weight at t0 ¼ 0, wi is the sample
known as Fourier number, and Bi ¼ h[/D is the Biot number, weight at each time ti(i ¼ 1,2,…), and weq is the sample weight
which determines the relative importance between diffusion at equilibrium. Then, intercept a and slope b can be obtained
and convective transfer processes. Notice that, in dimen- by means of the least squares fitting to a straight line of the
sionless form, the coefficients mn are defined as the positive data pairs {ti,yi}(i ¼ 1,2,…).
roots of the equation It should be noted that, by using the logarithm function in
the ordinate values definition yi, the corresponding variances
mn tanmn ¼ Bi: (10)
s2i will increase with time, so that, a weighted least squares
If diffusion process is much faster than convective transfer fitting should be used in the obtaining of slope and intercept
process (external problem), then Bi/0 and the positive roots values. Indeed, assuming the same variance s2w for all weight
of Eq. (10) are given by mn ¼ (n
1)p. Conversely, if convective data wi, which is determined by the weighing device used
transfer process is much faster than diffusion process (inter- (instrumental error), the variances for the ordinate values
nal problem), then Bi/∞ and the corresponding roots are yi(i ¼ 1,2,…) are given by2
given by mn ¼ ð2n
1Þp=2. So that, when both processes are
simultaneously relevant (0 < Bi < ∞), the positive roots of Eq. 1
Note that, in order to calculate the average dimensionless
(10) are bounded by the two extreme cases, that is, relative moisture F from drying data, only weight data w are
needed, and no average moisture values M should be calculated,
ðn
1Þp < mn < ð2n
1Þp=2: (11) M
M
since M0i
Meq
w
w
¼ w0i
weq :
eq eq
2
Given a function f(x1,…,xN) of variables xk with variances
2.1.2. Iterative numerical methodology s2k ðk ¼ 1; …; NÞ, error propagation theory provides the following
In order to obtain the diffusion and convective transfer co-
2
P vf
formula for the variance s2 of the function f.s2 ¼ N k¼1 vxk s2k :
efficients from drying data, the dimensionless moisture
34 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9
1 1
s2i ¼ 2s2w  2 þ   þ
wi
weq wi
weq w0
weq
The weighted least squares fitting was carried out by using
the algorithm of Strang (1986), which is implemented in the
MATLAB® function lscov.
After fitting process, the first root value m1 is calculated
from Eq. (14). Notice that, as in the calculation of the roots of
Eq. (10) discussed above, it is a transcendental equation, so
that the algorithm of Forsythe et al. (1977) has also been used.
Then, diffusion coefficient D is calculated from Eq. (15),
convective transfer coefficient h is calculated from Eq. (6), and
Biot number Bi from Eq. (10).
However, the coefficients thus obtained could be wrong, since
the first term approximation F1 ðFo; BiÞ is valid for high enough
values (depending on the Biot number value) of the Fourier
number, and both Fourier and Biot numbers depend on diffusion
and convective transfer coefficients. So that, in order to ensure
the correctness of the coefficients calculated, an iterative pro-
cess is mandatory. Thus, iteratively, the (i
1)-th data pair will be
deleted and the corresponding coefficients Di and hi recalculated
by fitting of remaining data pairs, until the convergence of first
term approximation within data accuracy, namely
 
FN ðFoi ; Bii Þ
F1 ðFoi ; Bii Þ  si ;
where the values of Fourier and Biot numbers are recalculated
at each iteration from fitting coefficients Di and hi. When the
condition in Eq. (18) is fulfilled, the corresponding coefficients
Di and hi can be considered the correct coefficients. Observe
that, in order to verify the convergence of first term approxi-
mation to exact values, we have used a high accuracy
approximation FN ðFo; BiÞ that can be considered as exact
within data accuracy.
Finally, notice that by applying the inequalities in Eq. (11) to
the first root m1, the intercept a and the slope b of the fitted
straight line must necessarily fulfil the following conditions
8 symmetric external conditions (i.e., the same air conditions at
log < a < 0;
p2
and
b < 0:
So that, the non-fulfilment of conditions above, when
fitting a given data set, implies that the experimental setup
breaks the model assumptions, and then the fitting co-
efficients obtained will be wrong. In this regard, condition in
Eq. (19) is particularly relevant since it establishes a very
narrow range, a2(
0.21, 0), for checking the suitability of
experimental data set.
A MATLAB® implementation of the proposed iterative
methodology, along with the input data files corresponding to
the results presented in this paper, is available in the
supplementary material.
2.2. Experimental setup
The carrot samples were obtained from entire carrots (Daucus
carota L. cv. Nantes) purchased from local market, which were
!
1 manually water washed, peeled, and lengthwise sliced with a
2 : kitchen slicer (Cooking, CASA International nv, Itegem, Ant-
w0
weq
werpen, Belgium) in order to obtain constant thickness. The
(17)
samples were sliced from central part of carrots, and even-
tually cut into rectangular pieces. The dimensions of each
carrot sample were measured by means of a digital caliper
gauge (DigitCal® Classic 05.30035, Swiss Instruments Ltd.,
Mississauga, Ontario, Canada).
Before drying process, each carrot sample was heated in a
microwave oven (PTMW 700, Electrodome
sticos Taurus S.L.,
Oliana, Le
rida, Spain), with the aim of achieving a uniform
temperature in the whole sample. With the purpose of
determine diffusion and convective transfer coefficients at
different temperatures, distinct heating periods were consid-
ered to reach temperatures around 40, 50, 60, and 70  C. In
order to avoid the start of drying process throughout the
heating period, a small bowl with water was introduced into
microwave oven along with each sample.
After heating, sample temperature was measured in each
case by using a type T thermocouple thermometer
(HH2001TC, Omega Engineering Inc., Stamford, CT, USA),
giving temperatures of 40, 55, 64, and 72  C. Before each
sample was introduced into the weighing device, it was
sandwiched between two metal grids in order to keep the slice
shape. The setpoint temperature, corresponding to each
(18) measured temperature at which the parameters will be esti-
mated, was established in each case in the weighing device
(AMB 300 moisture balance, Inscale Measurement Technology
Ltd., Bexhill, East Sussex, UK), and the sample weight
recording every 20 s was started. As shown in Fig. 1, the pro-
grammable weighing device comprises a chamber, with the
weighing pan inside, equipped with two halogen heaters, and
also a RS-232 interface for connection to a computer, so that
weighing data are automatically recorded in a file. The
instrumental error of the weighing device, which is required
by the proposed iterative methodology in Eq. (17), is
sw ¼ 0.001 g. Additionally, at the top of the chamber, there is a
ventilation grill. Observe that, in order to achieve the required
(19)
both sample sides), the sample was raised on four feet above
the sample pan.
Air temperature and also air relative humidity were
(20)
monitored by means of an electronic logger (iButton® Hygro-
chron™ DS1923, Maxim Integrated Products Inc., San Jose
, CA,
USA) located into the weighing device chamber. In order to
prevent the increase of relative humidity in the sample sur-
roundings, an extractor fan (EVERFLOW F128025BU, SilverStone
Technology Co. Ltd., New Taipei, Zhonghe, Taiwan) was
mounted on the ventilation grill of the weighing device
chamber [see Fig. 1(a)]. Extracted air velocity was measured by
using a multifunctional precision manometer with Prandtl's
tube (DC 100pro, Wo € hler Messgera € te Kehrgera€ te GmbH, Bad
Wünnenberg, Paderborn, Germany).
Lastly, each carrot sample was entered into a drying oven
(Conterm Poupinel, JP SELECTA S.A., Abrera, Barcelona, Spain)
at 105  C until steady weight was achieved, and thereafter
sample dry weight was determined with a precision balance
(ADP 720/L, Adam Equipment Co. Ltd., Kingston, Milton
Keynes, UK).
 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9
Fig. 1 e (a) Overview of the weighing device with the mounted extractor fan. (b) Section view of the weighing device with
the sample raised on four feet above the sample pan. (c) Detail of the sample sandwiched between two metal grids with the
four feet.
3. Results and discussion
3.1. Determination of coefficients from generated data
In order to test the behaviour of the proposed iterative
methodology, the series expansion in Eq. (9) with original
coefficients D, h, and other conditions described in Table 1,
was used to generate 18 sample weight data {w1, w2,…,w18}
along drying process. By taking N ¼ 14 terms in series
expansion, an accuracy of 0.001 g in the worst case Fo ¼ 0 was
achieved. Gaussian noise was introduced into generated data,
so that the final data set with weighing accuracy sw ¼ 0.01 g
was obtained. Pseudorandom values for Gaussian noise were
calculated by using the algorithms of Marsaglia (Marsaglia &
Tsang, 2000; Marsaglia & Zaman, 1991) which are imple-
mented in the MATLAB® function randn.
The successive iterations corresponding to applying the
proposed iterative methodology to generated data have been
depicted in panels (a)e(g) of Fig. 2. At each iteration, the first
data point is deleted, and the remaining data fitted to a
straight line leading to the corresponding values of diffusion
Di and convective transfer hi coefficients. Note that, in the first
iterations, the original fitting data points and the corre-
sponding exact points obtained from Eq. (9) with the fitting
coefficients Di and hi differ significantly. As iterative process
progresses both data sets get closer, and finally, when
convergence is reached at i ¼ 7, they match within weighing
Table 1 e Original and calculated diffusion and convective
transfer coefficients in the determination of coefficients
from generated data. Note that calculated coefficients
correspond to those in Fig. 2(g) rounded to two significant
figures.
Coefficients D (m2 s
1) h (m s
1)
Original 4.76$10
11 2.74$10
7
Calculated 4.8$10
11 2.7$10
7
Data generation conditions: [ ¼ 2 mm, w0 ¼ 6.47 g, weq ¼ 0.77 g,
sw ¼ 0.01 g.ti ¼ {0.5, 1, 1.5,…,9 h}.
35
accuracy sw. Notice that convergence condition in Eq. (18) is
graphically represented in panel (h) of Fig. 2.
It is worth noting that we have verified in a heuristic
manner that, as is shown in Fig. 2 and summarised in Table 1
for the present case, diffusion and convective transfer co-
efficients are obtained roughly with one significant figure less
than the weighing values. Thus, the generated data set starts
with w1 ¼ 5.86 g and ends with w18 ¼ 3.01 g, corresponding to
three significant figures, and diffusion and convective transfer
coefficients are obtained with two significant figures.
The number of significant figures obtained depends on the
Biot number, as well as the features of the experimental setup,
mainly the accuracy of the weighing data. In order to establish
the number of significant figures obtained in the calculation of
diffusion and convective transfer coefficients under typical
conditions in drying experiments, we have carried out several
series of simulations by generating weighing data and fitting
generated data to obtain the coefficients, in a similar way to
the example above. Sample half thickness [ ¼ 2 mm, weighing
device accuracy sw ¼ 0.001 g with 4 significant figures, and
range 10
9
10
12 m2 s
1 for the order of magnitude of the
diffusion coefficient are assumed as typical conditions in
drying experiments. Additionally, for each Biot number
Bi ¼ {0.001,0.01,0.1,1,10,100,1000} a set of 10 diffusion co-
efficients are randomly obtained, and for each random diffu-
sion coefficient 10 sets of weighing data are generated at
Fourier numbers Fo ¼ {0,0.1,0.2,…,1}. By comparing original
diffusion and convective transfer coefficients with those
calculated from generated data sets, the value of the general-
ised number of significant figures SX is obtained. The gener-
alised number of significant figures in X is defined by means of
the logarithmic difference
SX ¼ 1 þ log10 ðXÞ
log10 ðsX Þ; (21)
where X ¼ D, h is the value of the corresponding coefficient at a
Bi
given Biot number, and sX is the standard deviation of the
11.5
coefficient from the different simulations (10 simulations in
11
our case) at that Biot number. Notice that, for the same Biot
number but different random diffusion coefficient, somewhat
different values of SX will be obtained, which will be
36 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9

Fig. 2 e Iterative process for obtaining diffusion and convective transfer coefficients applied to generated data described in
Table 1. At each iteration (i ¼ 1,2,…,7), the logarithm of average dimensionless relative moisture F versus Fourier number Fo
has been depicted for fitting data (open circles), fitted straight line corresponding to first term approximation F1 ðFoi ; Bii Þ
(thick line), exact drying curve corresponding to high accuracy approximation FN ðFoi ; Bii Þ (thin line), and the points over
exact curve corresponding to Fourier numbers of fitting data (filled circles). (a) Initial fitting with all data. (b)e(g) Consecutive
fitting iterations, where the first datum is deleted at each iteration until reaching convergence. Fitting coefficients Di
(10¡11 m2 s¡1) and hi (10¡7 m s¡1) and also c each iteration are indicated. (h) Decimal logarithm of the difference
 
Di ¼ FN ðFoi ; Bii Þ
F1 ðFoi ; Bii Þ at convergence. Values corresponding to deleted data points (open circles) and used fitting
data (filled circles) are depicted. Data error at convergence s7 is represented in dotted line.

statistically characterised by the corresponding mean and

standard deviation.
The results of these simulations are shown in Fig. 3, where
the mean values and the standard deviation values of the
generalised significant figures corresponding to each generated
data set are represented as filled circles and error bars, respec-
tively. Observe that, as expected, optimal results for both co-
efficients jointly are obtained in the range 1  Bi  10. Moreover,
as Biot number increases the accuracy of diffusion coefficient
also increases, whilst the accuracy of convective transfer coef-
ficient falls quickly. Even so, at Bi ¼ 100 the accuracy of transfer
coefficient is not too low. On the other hand, as Biot number
decreases the accuracy of diffusion coefficient falls quickly, and
the accuracy of transfer coefficient also falls, although it de-
creases slower than the diffusion coefficient accuracy. This is
not a surprising behaviour, since the transfer coefficient value is
calculated through Eq. (6) by using the previously obtained
diffusion coefficient. However, until Bi ¼ 0.01 the accuracy of
calculated transfer coefficient is very good.
The iterative process shown in Fig. 2 can be reproduced,
and additional details obtained, by running the MATLAB® pro-
gram with the input data file corresponding to generated data
available in the supplementary material.
Fig. 3 e Generalised significant figures, SD and Sh, obtained
3.2. Determination of coefficients from experimental in the calculation of diffusion (thick line) and convective
data transfer (thin line) coefficients, respectively, as a function
of the Biot number Bi for different simulations performed
The proposed iterative methodology has been applied to under typical conditions in drying experiments (see text
experimental data corresponding to carrot drying for details).
 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9 37

experiments. In order to establish the temperature depen-
dence of diffusion and convective transfer coefficients,
isothermal drying at different temperatures was performed.
The final fitting at convergence is depicted for each tempera-
ture in Fig. 4 (top panels) along with the corresponding
graphically represented convergence conditions (bottom
panels). The obtained diffusion and convective transfer co-
efficients are summarised in Table 2 in addition to some
sample parameters as Biot number Bi, thickness 2[, width 2[y,
height 2[z, initial weight w0, equilibrium weight weq, and dry
matter weight wdm. Note that, in this case, weighing data have
four significant figures, so that three significant figures have
been assumed for the obtained coefficients. By means of
simulations as those depicted in Fig. 3, the accuracy of the
obtained diffusion and convective transfer coefficients has
been estimated, namely sD ¼ 0.03$10
9 (m2 s
1) and
sh ¼ 0.01$10
6 (m s
1), respectively.
Since the main goal of this section is to illustrate the appli-
cation of the proposed methodology, for the sake of clarity only
one data set at each temperature has been considered. However,
in order to obtain a better estimation of the coefficients accu-
racy, accounting correctly for the experimental error, several
data sets at each temperature should be regarded, and the cor-
responding error calculated in standard manner.
As is expected, both diffusion coefficient and convective
transfer coefficient mostly grow as temperature increases.
Nevertheless, as is shown in Table 2, diffusion coefficient
decreases very slightly from the value at T ¼ 55  C and the
value at T ¼ 64  C. We assume that this effect is due to the
different drying behaviour of core and cortex parts in carrots
and the slightly different core-cortex ratio in each sample.
Notice that, in order to achieve the one-dimensional behav-
iour for the thin layer samples, it was not possible to separate
core and cortex parts.
In this regard, Srikiatden and Roberts (2006) obtained slightly
higher values of the diffusion coefficient for core tissue than for
cortex tissue in carrots (D. carota L. cv. Red Core Chantenay) cut
into cylindrical samples. Specifically, they found at tempera-
tures of 40, 50, 60, and 70  C, the values of 1.01, 1.32, 1.90, and 2.34
(10
9 m2 s
1) for core tissue, and values of 0.76, 0.92, 1.06, and
1.36 (10
9 m2 s
1) for cortex tissue, respectively. Taking into
account that carrot samples of Srikiatden and Roberts corre-
spond to a different cultivated variety than those used in the
present work, it can be considered that their results and our
results are in good agreement. Additionally, Markowski (1997)
found the value h ¼ 1.37$10
7 (m s
1) for convective transfer
coefficient of carrots (cv. not specified) at T ¼ 60  C and natural
convection. Considering that our results correspond to forced
convection, albeit rather at a low air velocity, it seems reason-
able that our results are somewhat higher.
Moreover, it is important to highlight that, since an
analytical solution of the diffusion equation under certain
conditions is used, in order to obtain physical coefficients of
carrots rather than unphysical best fit parameters, these
conditions should be approximated in drying experiments, as
much as possible. Thus, one-dimensional behaviour for the
thin layer samples is achieved by taking thin layer samples
with width and height one order of magnitude greater than

Fig. 4 e Iterative process for obtaining diffusion and convective transfer coefficients applied to experimental data from carrot
slices drying summarised in Table 2. For different temperatures, the logarithm of average dimensionless relative moisture F
versus the Fourier number Fo has been depicted for fitting data (open circles), fitted straight line corresponding to first term
approximation F1 ðFoi ; Bii Þ (thick line), exact drying curve corresponding to high accuracy approximation FN ðFoi ; Bii Þ (thin
line), and the points over exact curve corresponding to Fourier numbers of fitting data (filled circles). (a)e(d) Final fitting at
convergence for temperatures T ¼ 40, 55, 64, 72  C, respectively. (e)e(h) Decimal logarithm of the difference
 
Di ¼ FN ðFoi ; Bii Þ
F1 ðFoi ; Bii Þ at convergence corresponding to temperatures T ¼ 40, 55, 64, 72  C, respectively. Values
corresponding to deleted data points (open circles) and used fitting data (filled circles) are depicted. Data error at
convergence for each temperature si (i ¼ 39,34,78,18) is represented in dotted line. Fitting coefficients Di (10¡9 m2 s¡1) and hi
(10¡6 m s¡1) and also the Biot number calculated at convergence are indicated.
38 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9

Table 2 e Diffusion and convective transfer coefficients obtained for carrot slices and additional sample parameters.
T ( C) D (m2 s
1) h (m s
1) Bi 2[ (mm) 2[y (mm) 2[z (mm) w0 (g) weq (g) wdm (g)

10
40 7.27$10 6.54$10
7 1.28 2.85 32.80 61.17 4.795 0.662 0.593
55 1.55$10
9 1.20$10
6 1.45 3.75 27.40 62.48 5.450 0.890 0.845
64 1.53$10
9 1.29$10
6 1.21 2.87 29.51 62.96 4.986 0.733 0.702
72 2.95$10
9 2.65$10
6 1.76 3.92 23.43 52.13 3.986 0.479 0.456

Obtained coefficients accuracy: sD ¼ 0.03$10
9 m2 s
1, sh ¼ 0.01$10
6 m s
1. Weighing device accuracy: sw ¼ 0.001 g. Air velocity: v ¼ 0.71 m s
1.

thickness, as is shown in Table 2. Homogeneous initial

moisture M0 can be assumed for fresh carrots preserved in
moisture insulated atmosphere. Additionally, diffusion D and
convective transfer h coefficients must be constant. Since both
coefficients depend on temperature, carrot sample tempera-
ture should be uniform and constant. Sample temperature
homogeneity is achieved by using a microwave device in the
heating process, and it is kept constant by controlling sur-
rounding air temperature, which was monitored by means of
an electronic logger. Notice that convective transfer coeffi-
cient depends also on surrounding air velocity and carrot
sample shape, both conditions being kept constant as is
explained below. Besides, sample moisture at equilibrium Meq
must also be constant, which is achieved by controlling rela-
tive humidity of surrounding air. In order to avoid the relative
humidity increase due to drying process, an extractor fan at
very low velocity was used (generating constant air velocity),
and relative humidity was monitored through an electronic
logger. Finally, shape and size (volume geometry in diffusion
equation) must be constant along drying process. Carrot slices
were sandwiched between two metal grids in order to keep
Fig. 5 e Iterative process applied to unsuitable (non-
the slice shape. Nevertheless, shrinkage is an unavoidable
isothermal) experimental data from carrot slices drying.
effect in carrot drying, which changes size and also shape,
The logarithm of average dimensionless relative moisture
albeit most shrinkage occurs at the initial stages of drying
F versus time t has been depicted for fitting data (open
process, since it is strongly correlated with the moisture
circles), and for the corresponding fitted straight line
decrease. However, the proposed methodology discards initial
logðFÞ ¼ a þ bt (thick line). The unsuitability of data set is
data, where most shrinkage occurs.
detect at first step of process, since fitting intercept a is out
Additionally, a data set within a narrow moisture range has
of range, i.e., a;ð
0:21; 0Þ.
been considered, so that shrinkage can be neglected. Notice
that, as verified, weighing data corresponding to a non-
negligible shrinking period lead to a convex (instead of
concave) drying curve, as in the case of non-isothermal drying analytical solution on which our methodology is based is a
described below, being detected by the proposed methodology concave curve (positive second derivative). As a consequence
as unsuitable data. of this feature, fitting intercept a is out of the required range
The sample temperature homogeneity is a remarkable (
0.21, 0) in Eq. (19), being a positive value a > 0, and so
condition in order to obtain correct diffusion and convective detecting the unsuitability of data set.
transfer coefficients from drying data fitting. As it has been The iterative process corresponding to each experimental
shown by Srikiatden and Roberts (2006) in drying experiments data set at different temperatures shown in Fig. 4, as well as
with carrots and also with potatoes, convective hot air drying that corresponding to the unsuitable data set, can be repro-
with constant air temperature leads to non-isothermal drying, duced, and additional details obtained, by running the MATLAB®
since a temperature gradient is induced inside samples, giving program with the corresponding input data file available in
experimental drying curves that do not fit the corresponding the supplementary material.
analytical solution of diffusion equation with constant diffu-
sion and transfer coefficients. In this regard, an example of
checking of suitability of data set provided by the proposed 4. Conclusions
iterative methodology in Eq. (19) is shown in Fig. 5. Drying data
in Fig. 5 correspond to a carrot slice sample initially at room Physical drying parameters, namely, mass diffusion and
temperature (~20  C) that is drying at constant air temperature convective mass transfer coefficients, have a significant rele-
of 65  C. Observe that the drying curve given by experimental vance on proper determination of optimal drying conditions in
data is convex (negative second derivative), whilst the convective hot air food drying, in order to achieve the aims of
 b i o s y s t e m s e n g i n e e r i n g 1 6 2 ( 2 0 1 7 ) 3 0 e3 9
food industry. In this paper, a new methodological approach has
been presented for the determination of diffusion and convec-
tive transfer coefficients from drying data, which takes advan-
tage of current numerical and computational capabilities versus
the classical scheme, based on approximate relationships
involving dimensionless numbers, that is usually applied.
As in classical schemes, data fitting to asymptotic first term
in series expansion solution is used, but in our method the
Fourier number threshold is iteratively calculated, since the
Fourier number depends on diffusion coefficient. Also, by
using standard numerical methods, involved transcendental
equations are accurately solved, so that series expansion so-
lution can be obtained with arbitrarily high accuracy, and also
diffusion and convective transfer coefficients can be directly
obtained from fitting parameters. As an additional feature,
from the range of obtained parameters involved in the nu-
merical solving, it can be identified the unsuitability of used
data, avoiding calculation of wrong coefficients.
The proposed methodology was tested by applying to
generated data, resulting in obtaining diffusion and convec-
tive transfer coefficients, for Biot numbers roughly within the
range 1e10, with only one significant figure less than the
weighing data values.
Additionally, the new methodological approach has been
applied to experimental data from carrot drying at different
temperatures, showing an adequate behaviour and giving
reasonable results. Also, our method was applied to non-
isothermal drying data, showing the feature of automatic
detection of unsuitable data.
Finally, it is noteworthy that although our method has
been specified for the infinite plate one-dimensional diffusion
system, the extension of the proposed iterative methodology
to other typical one-dimensional diffusion systems with
analytical solution, as those corresponding to cylindrical or
spherical geometries, is straightforward. Notice that, in these
cases, the same diffusion coefficient should be obtained for
the different geometries (assuming that the diffusion coeffi-
cient is a physical property of food tissue), but different
convective transfer coefficients will be obtained for each ge-
ometry, since it is assumed that the convective transfer co-
efficient depends on solid shape, among others factors.
Acknowledgments
The funding of this work has been covered by Comunidad de
Madrid and European Union through S2013/ABI-2747 (TAVS-
CM) project.
Appendix A. Supplementary data
Supplementary data related to this article can be found at
http://dx.doi.org/10.1016/j.biosystemseng.2017.07.005.
references
Białobrzewski, I. (2007). Determination of the mass transfer
coefficient during hot-air-drying of celery root. Journal of Food
Engineering, 78(4), 13881396.
39
Chen, C. S., & Johnson, W. H. (1969). Kinetics of moisture
movement in hygroscopic materials. I. Theoretical
considerations of drying phenomena. Transactions of the ASAE,
12(1), 109e113.
Chirife, J. M. (1983). Fundamentals of the drying mechanism
during air dehydration of foods. In A. S. Mujumdar (Ed.),
Advances in drying (Vol. 2). New York, US: Hemisphere
Publishing Corp.
Compaore, A., Dissa, A. O., Rogaume, Y., Putranto, A., Chen, X. D.,
Mangindaan, D., et al. (2017). Application of the reaction
engineering approach (REA) for modeling of the convective
drying of onion. Drying Technology, 35(4), 500e508.
Crank, J. (1975). The mathematics of diffusion (2nd ed.). London, UK:
Oxford University Press.
Dhalsamant, K., Tripathy, P. P., & Shrivastava, S. L. (2017).
Moisture transfer modeling during solar drying of potato
cylinders considering shrinkage. International Journal of Green
Energy, 14(2), 184e195.
Forsythe, G. E., Malcolm, M. A., & Moler, C. B. (1977). Computer
methods for mathematical computations. Englewood Cliffs, US:
Prentice-Hall.
Guine
, R. P. F., Cruz, A. C., & Mendes, M. (2014). Convective drying
of apples: Kinetic study, evaluation of mass transfer
properties and data analysis using artificial neural networks.
International Journal of Food Engineering, 10(2), 281e299.
Karim, M. A., & Hawlader, M. N. A. (2005). Mathematical
modelling and experimental investigation of tropical fruits
drying. International Journal of Heat and Mass Transfer, 48(23e24),
4914e4925.
King, C. J. (1968). Rates of moisture sorption and desorption in
porous dried foodstuffs. Food Technology, 22(4), 165e171.
Luikov, A. V. (1968). Analytical heat diffusion theory. New York, US:
Academic Press.
Markowski, M. (1997). Air drying of vegetables: Evaluation of mass
transfer coefficient. Journal of Food Engineering, 34(1), 55e62.
Marsaglia, G., & Tsang, W. W. (2000). The ziggurat method for
generating random variables. Journal of Statistical Software, 5(8),
1e7.
Marsaglia, G., & Zaman, A. (1991). A new class of random number
generators. Annals of Applied Probability, 1(3), 462e480.
Mghazlia, S., Ouhammoua, M., Hidara, N., Lahninea, L.,
Idlimamb, A., & Mahrouz, M. (2017). Drying characteristics and
kinetics solar drying of Moroccan rosemary leaves. Renewable
Energy, 108, 303e310.
Pflug, I. J., & Blaisdell, J. L. (1963). Methods of analysis of
precooling data. ASHRAE Journal, 5(11), 33e41.
Sampaio, R. M., Neto, J. P. M., Perez, V. H., Marcos, S. K.,
Boizan, M. A., & Da Silva, L. R. (2017). Mathematical modeling
of drying kinetics of persimmon fruits (Diospyros kaki cv.
Fuyu). Journal of Food Processing and Preservation, 41(1), e12789.
Srikiatden, J., & Roberts, J. S. (2006). Measuring moisture
diffusivity of potato and carrot (core and cortex) during
convective hot air and isothermal drying. Journal of Food
Engineering, 74(1), 143e152.
Strang, G. (1986). Introduction to applied mathematics. Wellesley, US:
Wellesley-Cambridge Press.
Tripathy, P. P., & Kumar, S. (2009). A methodology for
determination of temperature dependent mass transfer
coefficients from drying kinetics: Application to solar drying.
Journal of Food Engineering, 90(2), 212e218.
Vaccarezza, L. M., Lombardi, J. L., & Chirife, J. (1974). Kinetics of
moisture movement during air drying of sugar beet root.
International Journal of Food Science & Technology, 9(3), 317e327.
Zhao, D., An, K., Ding, S., Liu, L., Xu, Z., & Wang, Z. (2014). Two-
stage intermittent microwave coupled with hot-air drying of
carrot slices: Drying kinetics and physical quality. Food and
Bioprocess Technology, 7(8), 2308e2318.