You are on page 1of 11

Apparent thermal diffusivity estimation of the banana during

drying using inverse method
Viviana Cocco Mariani a,*, Antonio Gilson Barbosa de Lima b, Leandro dos Santos Coelho c
a

Mechanical Engineering Graduate Program, Pontifical Catholic University of Parana´ – PUCPR, Imaculada Conceicßa˜o,
1155, Prado Velho, Zip code 81215-901, Curitiba, PR, Brazil
b
Department of Mechanical Engineering, Federal University of Campina Grande – UFCG, P.O. Box 10069, Zip code 58109-970, Campina Grande,
PB, Brazil
c
Industrial and Systems Engineering Graduate Program, Pontifical Catholic University of Parana´ – PUCPR, Imaculada Conceicßa˜o, 1155, Prado Velho,
Zip code 80215-901, Curitiba, PR, Brazil

Abstract
A new approach for the estimation of apparent thermal diffusivity of foods at different drying temperatures was explored, analysed
and discussed in this work. Temperature versus time was obtained numerically at the center of the food (banana, ‘‘nanica˜o” variety)
using the 1D Fourier equation with drying temperatures in the range between approximately 17–65 °C and moisture content in the range
between 0.01 and 3.43 (dry basis). The solution of the partial differential equation is made with a finite difference method coupled to an
optimization technique of Differential Evolution used in inverse method. The mathematical model proposed considered the effects of
shrinkage and convective heat transfer at surface of fruit. Parameters of two functions, the first dependent of the moisture content
and the second dependent also of the temperature were obtained by inverse method modelling the apparent thermal diffusivity. Such
parameters that provide the best least square fit between the experimental and predicted time-temperatures curves are presented in this
work. This study demonstrated that a small change in the temperature and moisture content of banana cause an abrupt change in the
apparent thermal diffusivity, which decrease with the decreasing of the moisture. Statistical analysis shows the excellent agreement
between reported and estimated curves.

Keywords: Thermal diffusivity; Banana; Inverse problem; Differential Evolution; Food

1. Introduction
Drying, wetting, heating, cooling and freezing are
important parts of food processing operation. Interest in
transport properties of foods (thermal conductivity, heat
capacity, density, mass and thermal diffusivity and heat
and mass transfer coefficient) appears due to the importance to predict heat and mass transfer rates during pro-

*

Corresponding author. Fax: +55 041 3271 1345.
E-mail addresses: viviana.mariani@pucpr.br (V.C. Mariani), gilson@
dem.ufcg.edu.br (A.G. Barbosa de Lima), leandro.coelho@pucpr.br
(L. dos Santos Coelho).

cessing, preservation and optimal design of processing
equipment.
The main objective of present work is to estimate the
apparent thermal diffusivity of bananas during the drying.
There are numerous methods to measure the thermal diffusivity proposed in the specialized literature. Nevertheless,
most of them need relatively complex instrumentation or
experimental assemblies and demand an expertise of the
thermal phenomena. Several papers present such methods
and results of thermal diffusivity of different foods, some
are cited as follow. Choi and Okos (1983a, 1983b) propose
a line heat-source thermal conductivity probe with auxiliary
thermocouple to determine thermal conductivity and thermal diffusivity, simultaneously. Sweat (1986) recommends

. A2. thermal conductivity and thermal diffusivity. to minimize the objective function (Carvalho & Silva Neto. Faistel. Laraqi. counter of generations in DE L length of the banana (m) n dimension of the vector solution M size population (individuals) Nu Nusselt number Pr Prandtl number r coordinate in the radius (m) R radius in the banana cross-section (m) Re Reynolds number R0 initial radius in the banana cross-section (m) randb random number generation rnbr index chosen randomly between {1. Wood. Tao. and Le Bail (2003) determine the thermal properties of the gelatin gel during thawing using artificial neural networks. 2002). A4 parameters used to obtain thermal diffusivity. (2006) estimate thermal diffusivity of the olive oil during treatment high-pressure and Baı¨ri. particularly stochastic approaches. specific heat and mass density. Eqs. 1999. Boillereaux. i. based on gradient information. temperature in position r+1 equivalent the position r  Dr (°C) TR temperature in position r = R (°C) T ji theoretical temperature of the banana at node central. Other techniques based on . Celso Filho. Araga˜o. & Jarny. 2. Mendoncßa. their main advantage lies in their good convergence rate.K) Def diffusion coefficient (m2/s) f objective function fm mutation factor kmax maximum number of generations h heat transfer convective coefficient (W/m2 °C) hfg latent heat of vaporization of water (J/kg) hm mass transfer convective coefficient (m/s) i index j index represents time interval k thermal conductivity of the banana or air (W/ m °C). tÞdr (°C) T dimensionless average temperature in the section sji experimental temperature of the banana at termocouple central. which usually supply a good solution or until the global optimum.Nomenclature A1. Rao. and time indicator j (°C)   R 1 R T average temperature. r3 integers R2 Pearson multiple correlation coefficient index t drying time (h) T temperature (°C) T0 initial temperature (°C) Te equilibrium temperature (°C) Tr temperature in position r (°C) Tr1 temperature in position r1 equivalent the posittion r + Dr (°C) determination of food thermal diffusivities from experimentally obtained values for thermal conductivity. and Banerjee (1994) study the thermal properties of the wheat. and Garcı´a de Maria (2007) determinate thermal diffusivity of foods using 1D Fourier cylindrical solution.. and Laurindo (2002) estimate the effective thermal diffusivity of mortadella using data of the cooking process. . Carciofi. Kuba´sek et al. Many papers in the literature involving inverse problems use deterministic methods. For examples. New optimization methodologies are being used to solve inverse problems. ¼ R 0 T ðr. Bryden. and Liuzzo (1994) analyse the thermal properties of two varieties of rice bran. . t) local theoretical moisture content (kgw/kgdm) X0 initial moisture content (kgw/kgdm) Xe equilibrium moisture content (kgw/kgdm) X theoretical average moisture content in the section (kgw/kgdm) X theoretical dimensionless average moisture content in the section z mutant vector Tr+1 Greek symbols a thermal diffusivity of the banana (m2/s) b step length used in BFGS evaluated by Armijo or index for indicates vectors in DE d recombination mechanism e tolerance value q air density (kg/m3) l air viscosity (Pa s) / index that indicates the method for selecting of the parent chromosome Dt time step (s) Dr spatial mesh step (m) Growing interest has recently been evidenced in the analysis and solution of inverse problems of heat transfer. Shyamal. These researchers conclude that thermal diffusivity decrease linearly with moisture content. Bailleut. the computational time required generally exceeds that of deterministic methods (Suram. Chakraverty. Khachf. 1996). Cadet. i. r2. (20) and (21) (m2/s) CR crossover or recombination rate cv specific heat of vapor of water (J/kg. however. A3. and da Silva (2005) and Simpson and Corte´s (2004) use the inverse method to estimate thermophysical properties of foods. 2005. n} r1. and time indicator j (°C) u trial vector x vector solution X(r. & Ashlock. Although such methods can lead to local rather than global minima. specifically the specific heat.

Mujumdar. Def (m2/s) is the effective mass diffusivity. 2004. the two first works use other methodology to obtain the apparent thermal diffusivity while the last work to solve the mass transfer determining diffusion and convective coefficients. 0Þ ¼ X 0 8r.. Eqs. X ðr. few references concern about thermophysical properties during drying. Mikki et al. The linear system of equations proposed with associated initial and boundary conditions for the modelling of such physical problem involving the energy conservation equation. (iv) The banana is considered homogeneous. which requires knowledge of the thermal properties of foods. Queiroz. Lima. As initial condition. To simplify the problem the following hypotheses were considered: (i) The whole ripe banana is represented in the geometric shape of an infinite cylinder of length L (m) and radius R  0. ot r or or   oX 1 o oX ¼ rDef . 2004). Direct problem The physical problem involves a one-dimensional medium initially at uniform temperature and uniform moisture content. thus. (1) and (2) based on Kulasari and Samarasinghe (1996). Kacimov. analyse and validate a new transient measurement methodology using inverse method for determination of apparent thermal diffusivity in the range between 20 and 70 °C for the drying temperature. (3) and (4).artificial intelligence field. 1999. ð3Þ ð4Þ Boundary conditions:  oT ¼ 0. Eqs. & Campo. 2. Furthermore.015 (m) defined between [0. where R  L (see Fig. Karlik. (5) and (7). for such materials the Luikov (1966) equations to capillary-porous bodies can be simplified as Eqs. (ii) The thermal diffusivity was considered dependent on temperature and moisture content during drying. Thus the objective of this work was to study heat and mass transfer aspects of the banana during drying process and use transient temperature and moisture content fields to estimate the apparent thermal diffusivity. Eqs. A second objective was to explore.. Sablani. Null flux (symmetry) conditions were considered at the banana geometric center. using Differential Evolution as optimization technique for obtain parameters of two functions through of inverse method. Many references about fundamentals of banana drying were found in the literature. T (°C) is the internal temperature. such as genetic algorithms and artificial neural networks. as Lima (1999). it was considered that initial temperature and moisture of the food are uniform. based on Fourier’s law and mass transfer equation described by Fick’s unidirectional diffusion equation (Crank. ¼ hðT R  T e Þ þ qs Dr k or r¼R ot ð6Þ . r (m) is the transfer direction and t (s) is the time. 1. R]. This paper presents a simple procedure to estimate apparent thermal diffusivity of banana as a function of the moisture content and temperature from a range of numerical/experimental temperatures. a r L b Air flow r r =R r= 0 r r-1 r+1 Air flow Insulated Δr r=0 center r-1/2 r r1/2 r =R surface Fig. Pe´rez (1998) and Queiroz and Nebra (2001). The convective effect heat transfer and moisture at surface. 2003.(6) and (8). such as the analysis of transient heat transfer during the drying. 1975) as follows:   oT 1 o oT ¼ ra . Radial representation of the banana. was considered. & Tandiroglu. However. have been used for the solution of inverse problems (Ayhan. 1a). The problem considered here is relevant in food processing operations. cooling or freezing of fruits and vegetables in continuous systems. 1b. the longitudinal heat and moisture transfer were neglected and the axial symmetry was considered. One of the boundaries is in contact with the surrounding air thus resulting in a convective boundary condition for both the temperature and the moisture content as illustrated in Fig. Perret. Initial conditions: T ðr. 0Þ ¼ T 0 8r. respectively. ð5Þ or r¼0   oT oX  hfg þ cv ðT R  T e Þ . and Nebra (2002). ot r or or ð1Þ ð2Þ where a (m2/s) is the thermal diffusivity. Therefore. X (kgw/kgdm) is the moisture content (dry basis). (iii) For materials with small critical dimensions in terms of transfer mechanisms and long drying periods. Boillereaux et al. the conduction term dominates the temperature regime of the material.

 tþDt   T r  T tr 2a  tþDt : ¼ 2 T trþ1  T tr þ T tþDt rþ1  T r Dr Dt ð10Þ To boundary condition at surface. h (W/m2 °C) is the heat transfer convective coefficient. can be calculated as follows: h i   qs Dr2 X tþDt X t hDr T tþDt  c T Þ ðh fg v e R1 þ k T e  k Dt h  i T tþDt ¼ : ð11Þ R qs Dr2 cv X tþDt X t hDr 1þ k þ k Dt The discretization in Eq. Dr is the spatial mesh step.6 . sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 j j j¼1 ðs0  T 0 ðaÞÞ f ¼ .21 2. Under this comments the model used in this work is more realistic physically.1 1913.31 4.9 49. 1985) can be used rather than the finite element method or finite volume method for the solution of these partial differential equations.38 0.96 3. cv (J/kg. are known. thus at the food center (Fig. The objective the direct problem is to determine the temperature and moisture content fields. hfg (J/kg) represents the latent heat of vaporization of water obtained by air dry conditions. Mathematically. thus determining the transient temperature and moisture distribution in the food. (1) can be described and approximated in the following terms. If any of these magnitudes or a combination of them is unknown.85 3.9 72. In most of the techniques developed to solve inverse problems.1428 0.0211 0.56 4.39 2430750 2407180 2383140 2357850 2357100 2337310 1902.94 4. If the latent heat of vaporization in Eq.oX or  ¼ 0.50 4. but experimental data are available on the temperature measured inside and/or on the external surface of the food.3 27. t) and X(r. Inverse problem Knowing the food geometry and physical properties.4 3. In this work is used Crank– Nicolson implicit method.0 1907.0 40.0426 0.96 7. This type of problem is called a direct problem. the simpler finite difference technique (Smith. such values are presented in Table 1. Eq.0121 0.37 0.0664 0. t). 3. respectively. The solution of Eqs. The values of hm and Def used in this work were obtained of the work of Queiroz and Nebra (2001) that used the same experimental data of this work. it is desired to minimize the objective function.0 1911. and n is the number of samples. The superscript j above refers to the time when the measurements are taken.9 60. Using this numerical method. in the media. So. since Eq. Thus.33 0.8 27.0579 0. r = 0. qs = 1970 (kg/m3) is the dry solid density.9 39. (1) and (2) is referred as direct problem when initial and boundary conditions.04 2. as the boundary and initial conditions.48 1. (2) is omitted here due to its analogy with Eq.96 6. provided those data contain sufficient information. (1) at r = 0 can be replaced by ot ¼ 2 or a or .K) is the specific heat of vaporR of water (the two last R presented in Table 1). (6) is neglected then the coupling between heat and mass transfers is erroneous. enables one to solve Eqs. in this work is desired to minimize the difference between experimental and predicted temperatures. (6).8 35.90 3. tÞdr (kgw/kgdm) is the average moisture content in the section and hm (m/s) is the mass transfer convective coefficient. 1b) symmetric condition was considered where Tr+1 = Tr1.  tþDt  T r  T tr a   t tþDt r1=2 T rþ1  T tr þ T tþDt ¼ rþ1  T r 2 2rDr Dt   ð9Þ r1=2 T tr  T tr1 þ T rtþDt  T tþDt r1  oT oT o Eq. For the estimation of such parameter. Due to the characteristics of the mathematical problem (one-dimensional and homogeneous medium). X ¼ R1 0 X ðr.35 0.9 1911.43 3.5 68. thermal properties estimated with this consideration presents higher values as compared with cases that include change phase and heating of vapor. r = 0. we consider available the transient temperature measurements sj0 (°C) taken at node central.99 1. r = 0.3 1905. or r¼R ð7Þ ð8Þ where k (W/m °C) is the thermal conductivity of the fruit. the numerical model must be able to solve the direct problem with values arbitrated to the magnitudes Table 1 Air drying conditions and parameters used in the experimental cases Cases Te hm  107 Def  1010 X0 Xe v hfg cv Time 1 2 3 4 5 6 29. For the inverse problem of interest here. calculated numerically by the implicit finite difference method and sj0 (°C) is the experimental temperature of the banana at thermocouple central.2 60. r¼0  oX  Def ¼ hm ðX R  X e Þ.5 121. ð12Þ n where T j0 (°C) is the temperature of the banana at node central. j is the time indicator. (1). the parameter a is regarded as unknown quantity.17 3.19 6. one has an inverse problem that allows one to determine the unknown magnitudes. (1)–(8).36 0.62 5. as well as all parameters appearing in the formulation.95 0. T(r.

zi(k + 1). (20) and (21). In each step. . . . i. Recombination is employed to generate a trial vector by replacing certain parameters of the target vector with the corresponding parameters of a randomly generated donor vector. The interest in evolutionary algorithms is increasing fast. random vector differentials to them. 1]. the direct problem must be solved several times. . ui2 ðk þ 1Þ. k is the generation. r2 and r3 are mutually different integers and also different from the running index.i + 1. fm > 0 is a real parameter called mutation factor. . . Step 4: Mutation operation (or differential operation): Mutation is an operation that adds a vector differential to a population vector of individuals according to the following equation. zi ðk þ 1Þ ¼ xi.r1 ðkÞ þ fm ½xi. The Differential Evolution method was used as the optimization technique and is described as follows.r2 ðkÞ  xi. and a trial vector. an index rnbr(i) 2 {1. . If the objective function of the trial vector is better than that of the target. . where / indicates the method for selecting the parent chromosome that will form the base of the mutated vector. and recently differential evolution. non-differentiable and non-convex functions by Storn (1997). . M is the individual’s index of population. . The different variants of DE are classified using the following notation: DE///b/d. 1997. For each vector. such as genetic algorithms. crossover and mutation procedures. 2. . the mutation factor (fm). b indicates the number of difference vectors used to perturb the base chromosome. The fundamental idea behind DE is a scheme whereby it generates the trial parameter vectors. It is necessary to specify that the vectors x. . . Thus. ui ðk þ 1Þ ¼ ½ui1 T ðk þ 1Þ. which are the difference between two randomly chosen parameter vectors. which involved the following steps: Step 1: Parameter setup: The user chooses the parameters of population size. and was successfully applied in the optimization of some wellknown non-linear. Since the procedures for the solution are usually iterative. whose simple yet powerful and straightforward features make it very attractive for numerical optimization. xin ðkÞT stands for the position of the i th individual of population of N real-valued ndimensional vectors. randomly selected with uniform distribution from the set {1. 2. 2. i  1. . xi ðkÞ ¼ ½xi1 ðkÞ. recombination is applied to the population. the DE mutates vectors by adding weighted. . M individuals (real-valued n-dimensional solution vectors) with random values generated according to a uniform probability distribution in the n dimensional problem space. There are a variety of evolutionary algorithms that have been proposed and studied.1. . . which controls the amplification of the difference between two individuals so as to avoid search stagnation and is usually taken from the range [0. mutation and selection to evolve from a randomly generated starting population to a final solution. is generated with . 1989). the crossover rate (CR). . . DE uses a rather greedy and less stochastic approach to problem solving than do the classical evolutionary algorithms. Differential evolution (DE) is a population-based stochastic function minimizer (or maximizer) relating to evolutionary computation. The bin acronym indicates that the recombination is controlled by a series of independent binomial experiments. . then the target vector is replaced by the trial vector in the next generation.r3 ðkÞ. it is desirable to have a precise method for the solution of the direct problem that requires a relatively short computational time. evolution strategy. the boundary constraints of optimization variables. . ð13Þ where i = 1. xi2 ðkÞ. . . The basic idea is to maintain a population of candidate solutions that evolve under selective pressure that favors better solutions (Ba¨ck. Step 5: Recombination operation: Following the mutation operation. n} is randomly chosen using uniform distribution. zi2 ðkÞ. DE combines simple arithmetical operators with the operators of recombination. kmax. The DE algorithm was first introduced by Storn and Price (1995). 3. . . . DE differs from conventional genetic algorithms in its use of perturbing vectors. Optimization method Evolutionary algorithms are computer-based problemsolving systems of evolutionary computation field based on principles of evolution theory. . . They share a common conceptual base of simulating the evolution of individual structures using selection. Step 2: Initialization of the population: Set generation k = 0. and the stopping criterion of maximum number of iterations (generations). Step 3: Evaluation of the population: Evaluate the fitness (objective function) value of each individual. and d indicates the recombination mechanism used to create the offspring population. Evolutionary algorithms have been used in many problems. r1. T zin ðkÞ stands for the position of the i th individual of a mutant vector.. Goldberg. zi ðkÞ ¼ ½zi1 ðkÞ. These initial individual values are chosen at random from within userdefined bounds (boundary constraints). u and z used in DE represent the parameters Ai’s presented in the following section in Eqs. . The variant implemented in Matlab (MathWorks) was the DE/rand/1/bin. genetic programming. dealing with multidimensional and multimodal searches.to be determined. uin ðk þ 1Þ . N}. evolutionary programming. . due their robust and powerful adaptive search mechanisms.1. Fogel. . Initialize a population of i = 1. & Michalewicz.

The numerical simulations were performed for values of the heat transfer convective coefficient. body shape and temperature between the body surface and the fluid. 1]. 2003). and initial radius.97. h¼ ð15Þ In this case. (17)):   R ¼ 0:4721 þ 0:1819X e þ 0:1819ðX  X e Þ R0 .2 GHz Pentium IV processor with 2 GB of RAM.68Re0. The mean radius and the moisture content were fitting by a linear regression. of the target vector xi(k) is lower than that of the trial vector.8. if f denotes the objective function under minimization.39 m/s. v (m/s) is the air velocity of the drying. . The performance of a DE algorithm usually depends on three parameters: the population size N. 30 independent runs were made for the optimization method involving 30 different initial trial solutions. if randbðjÞ > CR or j 6¼ rnbrðiÞ: ð14Þ In the above equations. could be continuously recalculated according to the new average moisture content. the mutation factor fm. Re is the Reynolds number calculated by Re ¼ qvd . the shrinkage was treated like an elastic grid. hfg (J/kg) is the latent heat of vaporization of water. Numerically. xij ðkÞ. Using the Eq. Step 6: Selection operation: To decide whether or not the vector ui(k + 1) should be a member of the population comprising the next generation. The stopping criterion depends on the type of problem. hm (m/s) is the mass transfer convective coefficient. Thus this coefficient is enough sensitive the small changes in velocity. In foods with higher moisture content there are heat and mass transfer difficulting the experimental measure of heat transfer coefficient. 1] and CR is a crossover or recombination rate in the range [0. whose coefficients of correlation were higher than 0. R. Saravacos and Kostaropoulos (1995) studied the heat transfer coefficient observing changes in the heat transfer coefficient from 10 to 200 W/m2 °C when the velocity changes from 0. if randbðjÞ 6 CR or j ¼ rnbrðiÞ. v. Pr is the Prandtl number. When ones study the drying of products with bigger moisture content. To illustrate the effectiveness of the optimization procedure several simulations were performed. in the range between 15 W/m2 °C and 35 W/m2 °C. surface rugosity. X0 (kgw/kgdm) is the initial moisture content. the target vector is replaced by trial vector in the next generation. randb(j) is the jth evaluation of a uniform random number generation with [0. where Te (°C) is the equilibrium temperature. the food shrinkage too. q (kg/m3) l is the air density and l (Pa s) is the air viscosity (Kreith & Bohn. However. In the tests.2. Proceed to Step 3 until a stopping criterion is met. fluid properties. Otherwise. (16).uij ðk þ 1Þ ¼ ( zij ðk þ 1Þ. and the recombination rate CR. The program was run on a 3. R0. If the objective function. xi ðkÞ. then xi ðk þ 1Þ ¼ ui ðk þ 1Þ. the setup of DE used was the following: fm = 0. h. the objective function of each trial vector ui(k+1) is compared with that of its parent target vector xi(k). The values for heat transfer convective coefficient were obtained based on the Nusselt number. Thus. This means that the number of nodes in the radius was maintained constant and the radial subinterval size was changed at each time step. the target is allowed to advance to the next generation. In optimization tests. calculated from Eq. X . ð17Þ developed through an experimental test correlating the banana mean diameter to its moisture content was obtained in Queiroz and Nebra (2001).1 m/s to 5 m/s. The heat transfer coefficient is dependent of fluid velocity. Each optimization approach was implemented in environment computational Matlab (MathWorks). in the mathematical model is need to consider.3. The shrinkage is strong and fast at surface since X decrease quite fast main at the beginning of the drying. are in the range between 0. Thus the coefficients determinate using such hypothesis has more applicability. kNu .52Pr1/3. the population size M was 10 and the stopping criterion is when kmax is equals to 100 generations. otherwise: Def (m2/s) is the effective mass diffusivity. f.33 m/s and 0. d (m) is the diameter of the banana. if f ðuðk þ 1ÞÞ < f ðxi ðkÞÞ. d ð16Þ where k (kg/m3) is the air thermal conductivity. besides of heat and mass transfer. cv (J/kg K) is the specific heat of vapor of water and time is given in hours. specifically the vegetables and fruits. CR = 0. (17) the radius. Nu is the Nusselt number given by Nu = 0. those works present results for six experiments with different conditions of temperature and relative humidity to drying of banana. Step 7: Verification of stop criterion: Set the generation number for k = k + 1.97 + 0. while the values of air velocities. Xe (kgw/kgdm) is the equilibrium moisture content. as presented in Table 1. 3. The shrinkage phenomenon was included in this work where an empirical equation (Eq. Thermophysical properties The experimental results for temperature used in this study were obtained from Pe´rez (1998) and Queiroz and Nebra (2001). presented in Table 1. it is compared with the corresponding vector xi(k). usually kmax.

In Table 3. practically being constant in the next hours. As expected. Lima (1999) proposed the equation aðX  Þ ¼ A1 A A2 ð1  X  3 Þ þ A4 . (20). Krokida.9815 0. 2002) are simple to use however these equations are not always in agreement with experimental data. and to adjust the parameters by optimization approaches. Saad & Scott. Delaunay.6654 0.9741 0.9–1. (1)–(8) considers some strict assumptions (homogeneous material and infinite cylinder). Boillereaux. sj0  sj0 where sj0 (°C) is the mean experimental temperature of the banana at thermocouple central. Saravacos. the theoretical model for moisture content is the more complete in terms of incorporation of the physical phenomena. such coefficient was predicted using (Sweat.3674 0. more than 20. Table 2 Parameters of Eq.4994 0. Maroulis. 1995). the R2 values between reported and estimated curves. through of R2 values that the results predicted have a good agreement with experimental values. it can be observed that the thermal diffusivity must have a behavior of decreasing very accented in the first hours of drying.9937 0. 1995) as k ¼ 0:148 þ 0:493X . X  Þ ¼ A1  AX2 þ A3 þ A4 : T ð21Þ where apparent thermal diffusivity is dependent of dimensionless average moisture content in the section. (6).0985 0. The R2 value of 0. The functional forms of thermal properties are generally unknown. from the experimental data. The best fitness in the least square sense between the experimental temperatures and the temperatures obtained by mathematical model (Eqs. respectively. (20) and (21).4155 0. 2 P j s0  T j0 ðaÞ 2 R ¼ 1 P ð22Þ 2 .3658 0.4622 0. 4. such values are shown in Tables 2 and 3 and the values for objective function are shown in the last column in same tables.When the boundary condition. (21) is adequate adjusting the apparent thermal diffusivity with the hypothesis considered. several preliminary analyses were made. & Le Bail.9923 0. thus the number of parameters for adjust are 3 and 4. .0758 0. Results and discussion The proposed approach was analyzed for the case in which three parameters (see Eq. showing that the apparent thermal diffusivity obtained from the inverse method was fitted by function presented in Eq. Some authors have proposed replacing polynomials by piecewise linear functions of temperature.5000 0. Due to this fact it was proposed the use of two non-linear functions.4607 0. Fig. X  . 2001. The mathematical model described in Eqs.9969 0.5770 17.4799 0. The parameters were adjusted by inverse method using a Differential Evolution approach. Deviations between experimental and simulated temperatures were calculated using the multiple correlation coefficient (Pearson coefficient). . ð18Þ was proposed in Pe´rez (1998) where the apparent thermal diffusivity is dependent of temperature. 2 shows the moisture content predicted by numerical simulation for three cases of drying.. (21)) were treated as unknowns Ai for i = 1. & Panagiotou. T e 4.9971 0.0762 0. For example. 1997. Pe´rez. (20) Cases A1  1011 A2 A3 R2 f 1 2 3 4 5 6 6.7960 19.9875 11.0 is considered sufficient for that the apparent thermal diffusivity obtained in this work to be well adjusted with the experimental data. accordant Eqs. To characterize the apparent thermal diffusivity variable as a function of the moisture content. generally the number of parameters to adjust is relatively large. (1)–(8)) is shown in Tables 2 and 3. Eq. in successive trials as. A usual solution consists of representing these functions by empirical polynomials. ð19Þ where apparent thermal diffusivity is dependent of dimensionless moisture content average in the section. & Bransier. especially in the case of foods with multiple compositions. where the shrinkage effect of the material is included and Def and hm are considered. Havet. X e Analyzing the temperature variation rate with the time in the center of the banana. and dimensionless average temperature in the section.7950 11. is used is necessary evaluated the thermal conductivity of the fruit. where X is in wet basis. In Table 2 one observe. indicating that the function presented in Eq. 1986. Maroulis. 1998).0146 0. and the convergence of optimization approach is delicate (Jarny. (20)) and four parameters (see Eq. e X  ¼ XX0X .4806 0. Note that several authors have derived equations to predict thermophysical properties.4141 0. and time indicator j.0954 .3027 0. & Saravacos. e T  ¼ TT0T .3000 0. Semi-theoretical equations (Krokida. . and the proposed second function was. A preliminary choice of these functions could be an obstacle to a correct approximation of these thermal properties dependent of the temperature or moisture. aðT  .0064 0.3197 9. with orders sufficiently high to correctly represent the properties variations (Chourot. for apparent thermal diffusivity the following equation: aðT Þ ¼ A1 þ A2 T þ A3 T 2 þ A4 T 3 þ A5 T 4 þ A6 T 5 . Thus the first function proposed for apparent thermal diffusivity in this work was aðX  Þ ¼ A1  AX2 þ A3 ð20Þ . even if the parameters of these functions are adjustable. r = 0.

Experimental validation for the inverse method using the parameters of the Table 3 (central temperature. (20) using the parameters of the Table 2 for all cases.4442 0. for fourth case were 2.0032 0.2465 9.3111 0. The predicted temperatures are in excellent agreement with experimental data obtained in Pe´rez (1998). It is important to observe that in practice. Fig. so. through of the values of multiple correlation coefficients shows no significant differences between reported and estimated curves (see Table 2). in r = 0).1072 1.3452 0.4040 7.9977 0.7  107 (m2/s).5350 12.49  1010 (m2/s) and 1. 3.1858 0. The inclusion of shrinkage and convective effect lends more credibility to the apparent thermal diffusivity obtained and presented in Tables 2 and 3.2354 0.1880 18.9998 0. Predicted temperatures were calculated with parameters estimated under the inverse method using the Differential Evolution.9999 0. Statistical analysis Fig.3011 0. The minimum and maximum values for apparent thermal diffusivity obtained in this work using Eq. (21) Cases A1  1011 A2 A3 A4  1010 R2 f 1 2 3 4 5 6 5.44  109 (m2/s) and maximum equals 2. In this figures the shrinkage and convective effect at banana surface are included in the mathematical model.0106 Fig.4758 0. in r = 0). Statistical analysis shows good agreement between reported and estimated curves.9993 0.0312 0.3015 0. Drying kinetic curves modelled considering shrinkage and convective condition at the surface.0084 0.0298 20. Experimental validation for the inverse method using the parameters of the Table 2 (central temperature.0000 7.3805 0. Fig.1297 0.3460 0. 4. bananas shrink by about 43 ± 47% their original diameter during drying in accordance with Queiroz and Nebra (2001).9999 0.Table 3 Parameters of Eq. This fact reveals the importance of including this phenomenon in the theoretical model. with values lower than the obtained by Lima (1999) with minimum equals 1. (20). .0074 0.9924 10.3120 0. 2.4741 0. this is a complete model due the incorporation of physical phenomena in the banana’s drying. respectively. for example.4458 0. Considering all cases such differences occur due Fig.3790 0. compared with the experimental data for all cases. 3 shows predicted and experimental temperatures at the thermal center (r = 0) predicted by Eq. 4 shows the temperature profiles at the thermal center (r = 0) of infinite cylinder predicted by Eq. (21) using the parameters of the Table 3.1858 0.9974 0.0128 0.88  107 (m2/s).

8203 18.9915 0.5443 19.4923 18.9260 17.9608 0.9754 0. drying temperature.9349 0.4067 18. Dimensionless average moisture content  ln(a) in whole drying process for the case 4 using Eq.9678 20.9359 0.9255 0. content until 5 min of drying process for three cases of drying using Eq.4369 19. In accordance with Szczesniak (1983).5080 17.9110 19. (20) and (21) to evaluate the apparent thermal diffusivity.8801 0.9640 0.9833 0.8595 19. 2  107] (m2/s).3644 18.3405 18. 6 is illustrated the behavior of natural logarithm of thermal diffusivity in function of dimensionless average moisture Table 4 Natural logarithm of thermal diffusivity as a function of dimensionless average moisture content using Eq. The results of thermal diffusivity obtained for other five cases using the same equation are similar.9793 0. Thus in Table 4 is presented the apparent thermal diffusivity with smaller time step.9873 0.3375 19.9292 0.1198 19.7061 19.8652 0.8727 0.9209 0. In Fig. Obviously. (20) Case 1 Case 2 Case 3 X ln(a) X ln(a) X ln(a) 1.6456 19.0334 18.5107 19.7628 19.9108 0.8008 19. body shape and methodology.9567 0. Results of natural logarithm of thermal diffusivity obtained with Eq. 2.7383 19. this affirmation is dependent of moisture content of the food during the process that was submitted. The numerical simulations were performed using Eqs. 5 for whole process of drying.8249 18.9447 0.9819 0. respectively. i.9496 0. (20) for fourth case of drying are shown in Fig.5805 18.9895 0. In this context.0022 1.9350 0.0180 20. from zero until five minutes of drying process for Eq.7569 18.1665 19.6212 18.2599 19.5718 18.8532 1.e.4120 19.the different mathematical model adopted in each work.6037 19.0043 18.9592 20.9827 19.9763 0. (20). however such differences are acceptable.9497 0.9636 18.5219 19.5807 19. Dimensionless average moisture content  ln(a) until five minutes of drying process using Eq.2562 19.9139 19. 5 is possible see that the most important change in diffusivity is made during intensive drying at the initial phase of the process.8504 0.9326 0.9030 0. Conclusions Fig.2385 19.9028 0.9073 0. for the three cases of drying.8877 0. 5.9267 0. there are differences with other works of the literature due the different values of moisture content.9957 0. are used the values presented in Table 4.8578 0. and 3.9707 0.8663 19.9603 0.8984 16.9938 0.0666 19. it is possible to observe that the thermal diffusivity is decreasing with the decreasing moisture content also for the same drying interval (5 min) the least moisture content is obtained for case 3 because in this drying configuration is used the biggest drying temperature as presented in Table 1. As expected.9461 0.8357 15.9600 0.0000 0.9531 0.8161 19.9715 0.9398 0.9653 0.9163 0. The proposed optimization procedure using Differential Evolution method was successfully applied to the determination of apparent thermal diffusivity dependent of moisture content and/or temperature average in the section radial of the banana during the drying process.9117 0.0428 19. (20) using the cases 1.0000 0. In Fig.2653 18. (20).7615 19.1579 19. 5.9878 0..6590 17. Analyzing the results presented in Table 4.6594 19.3516 19.3432 19.0000 0.8953 0.8085 19.9427 0. (20).9302 0.9700 0. A statistical analysis shows no significant differences between the predicted and experimental profiles of temperature at the thermal center .9796 0.9151 19.9519 0.9677 0.6715 19.43 (dry basis) as shown in Table 1. 6.5997 19.8431 0. decreasing its value with the increasing of the temperature.7119 19. the thermal diffusivity is decreasing with the decreasing moisture content.5165 17.9187 0.9393 0. the values of the thermal diffusivity are in the range between [1  107.4806 19.9540 19.0660 Fig.4347 19.9433 0. Such equations are validated for several values of temperature in the range between approximately 17 °C and 65 °C while the moisture content is in the range of 0.6739 18.9548 0.01 and 3.

& Samarasinghe. Lima.. J. (2002). USA. Chourot. A. Chicago. & Panagiotou.). F. Lima.D. Switzerland.... H. & Le Bail.. Food Technology. Havet. pp. N. Fogel. (2002). 231–245. P. Krokida. Kacimov. 669–675. 78(2). T. SP.. Sablani. However it is clear that diffusivity depends on several parameters. The proposed procedure can be extended to the determination of other thermophysical properties in different processes like thermal conductivity. 127–132. D. A. & Jarny. International Journal of Food Properties.. D. (1996). UNICAMP. N.. (1999).09  1010 (m2/s). The simultaneous determination of thermal conductivity and heat capacity within an orthotropic medium by using conjugate gradient algorithm. Sa˜o Jose´ dos Campos. 55(1). Thermal properties estimation during thawing via real-time neural network learning. The thermal properties of liquid foodsreview. Transactions of ASAE. 89–94. D. (2007). In Proceedings of the winter meeting of the American society of agricultural engineers. 281–304. Diffusion phenomenon in prolate spheroidal solids. (2004). S. M. J. (2004). & Silva Neto. J. The shrinkage and convective effects assumptions in the mathematical model provides greater reliability on the calculated thermal diffusivity. R. for the granted financial support.. L. (2006). Transport properties in processing of fruits and vegetables. R. J. Handbook of evolutionary computation. heating and/or freezing. because of its accuracy and short time for parameters estimation.. Jarny. 1–6.. Strohalm. Theoretical and experimental analysis of the drying kinetics of bananas. B. optimization and machine learning. Mikki. International Journal of Heat and Mass Transfer. Determination of thermal diffusivity of foods using 1D Fourier cylindrical solution.. Y. & Campo. (1999). A. 4(2). SP. S. M. (1983a). Araga˜o. References Ayhan. Bristol. G. USA.. Case study: Drying of banana. Faistel. F. G. UNICAMP. Boillereaux. K. Luikov. T. & Zitny´. Philadelphia/NY.. M. G.. In Proceedings of the third international conference on inverse problems in engineering. & Scott. Mendoncßa. Study of the temperature behavior of bananas during the drying process. M.. B. R. Flow geometry optimization of channels with baffles using neural networks and second law of thermodynamics. & Bohn. 1–19. Estimation of temperature dependent thermal properties of basic food solutions during freezing. A.. 665–679. An inverse analysis for polymers thermal properties estimation.. A.. Cadet. In Proceedings of the eighth international heat transfer conference. Heat and mass transfer in capillary-porous bodies. (1995). 261–266. A. Z. Campinas.. M. Y. 1–8). Landfeld. Carciofi.. Delaunay. T.. (2003). thesis. Oxford: Pergamon Press. Ba¨ck. & Kostaropoulos. & Tandiroglu. Campos Velho. Journal of Food Engineering. K. The results obtained in this work validate the proposed method as a tool to the determination of apparent thermal diffusivity in the drying temperature range. (1975). A. J..47  1011 (m2/s) using the Eq. (1999). Choi. San Francisco. D. B. Journal of Food Engineering. B.88  107 (m2/s) while the lower value was 9. Z.). Determination of thermal diffusivity of mortadella using actual cooking process data. 17–23. Journal of Food Engineering. Brazil (in Portuguese). Oxford: Institute of Physics Publishing/Oxford University Press. & da Silva. Goldberg.. Thermal properties of raw. and specific heat in drying. WA. In IMACS congress on scientific computation. J. A. M. & Nebra. (1997). Acknowledgements The second and last authors would like to express theirs thanks to CNPq (Conselho Nacional de Desenvolvimento Cientı´fico e Tecnolo´gico – Brazil). 35(9). S. (2001). P. wetting. 19–24). (1983b). Karlik... 5(1). M. S. Saravacos.. inverse problems in engineering: Theory and practice.. Developments in Chemical Engineering and Mineral Processing.. International Journal of Food Properties. MA: Addison-Wesley. Khachf. Paper no.. (1994). A neural network approach in a backward heat conduction problem. & Banerjee. M. 305–311. Computational Mechanics. Chakraverty. A.. F.. G. K. E. G. L. D. Princı´pios de Transfereˆncia de Calor (1st ed. Oxford: Clarendon Press. 67(3). S. B. Thermal conductivity prediction for foodstuffs: Effect of moisture content and temperature. & Bransier.. H. (2003). 801–804. Numerical modelling of high pressure thawing: Application to water thawing. Luz. I. Queiroz. P. R. 179–200. Saravacos. SP. D.. B. Genetic algorithms in search. A. A. R. D. & Saravacos.using Eq. Mujumdar. G. Baı¨ri.. Master thesis. CA. Z.. our main objective was to validate the inverse method using Differential Evolution optimization approach coupled with a simple one-dimensional heat transfer model besides to obtain a value acceptable for thermal diffusivity. applied mathematics. Perret. (1966). S. E. (2001). (1995). L. V. 139–143. Reading. . F. M. cooling.. Modelling heat and mass transfer in drying of biological materials: A simplified approach to materials with small dimensions. M. Bailleut.. M. D. (1998). 10(3/4). 48(3). Maroulis. & Le Bail. J. A. Port Ludlow. 163–167. Ecological Modelling. parboiled and debranned parboiled wheat and wheat bulgur. The thermal properties of tomato juice concentrates. A. Choi. A. The higher value obtained in this work to apparent thermal diffusivity was approximately 1. direction of the fibres and in particular of its temperature.. 47(2). J. M. The determination of thermophysical properties from an inverse method is an attractive technique both from the experimental and methodological point of view. (2002). Heat and mass transfer model including shrinkage applied to ellipsoidal product: Case study for drying of bananas. M. J. Non-iterative estimation of heat transfer coefficients using artificial neural network models. Celso Filho. among other the composition of the product. Brazil: Thomson Pioneira (in Portuguese). USA. Campinas. S. (2005). & Okos. J. E. Kreith. Kulasari. Z. Kuba´sek. (20).. B. (1989). A. Oliveira.. (2002). Carvalho. and simulation (pp. A. D. In Proceedings of the IV Brazilian conference on neural networks (pp. (1997). Journal of Food Engineering. Pe´rez. Transient conduction in spherical fruits: Method to estimate the thermal conductivity and volumetric thermal capacity. 99–105. Brazil (in Portuguese). The values provided in this study were obtained without a examination of all these characteristics... Thermal diffusivity estimation of the olive oil during its high-pressure treatment. R. & Michalewicz. J.. H. Journal of Food Engineering. Y. SP. Houska.. 86. (20) or Eq. Krokida. B. Identification of nonlinear thermal properties by an output least square method. Issamoto. Kamara´d. (21) were obtained the values 2. Saad.. Rheological properties of fluid fruit and vegetable puree products: compilation of literature data. J. 28(1). & Laurindo. G. J. (21). 57(1). Journal of Food Engineering. Energy Conversion and Management.. Journal of Food Engineering. D. M. II. Y. The mathematics of diffusion (2nd ed.. R. 33(2). 26(1). & Silva. Queiroz. 83-6516. 74(3). A. E. D. & Garcı´a de Maria. Lausanne. B. Brazil. Maroulis. 34(1). 286–291. Shyamal. R. Ph. L. M. B. V. Journal of Food Engineering. 49. & Nebra.. already using the Eq. D..93  107 (m2/s) and 1. 63–75. Laraqi. S. E. (1986). A. Crank.. & Okos. Boillereaux.

Storn.Simpson. Engineering Computation. An evolutionary algorithm to estimate unknown heat flux in a one-dimensional inverse heat conduction problem. . 1–41). 64(1). (1997). In M.). H. In Proceedings of the fifth international conference on inverse problems in engineering: Theory and Practice. R. Differential evolution – A simple and efficient heuristic for global optimization over continuous spaces... (2004). 38–56. Journal of Food Engineering. Engineering properties of foods. C. Rizvi (Eds. E. R. & Price. Oxford: Oxford University Press. A. (1995). & Ashlock. UK. (2005). 89–96. 10(5). 13(5). M. Genetic algorithm behaviour in the solution of an inverse thermal field problem. R. R. Cambridge. Rao & S. L. Thermal properties of foods. S. J. Szczesniak. (1986). Storn. Berkeley. New York: Marcel Dekker.. 709–711. Rao. D. Bagley (Eds. V. (1983). Physical properties of foods: what they are and their relation to other food properties. M. A. Applied Engineering in Agriculture. Numerical solution of partial differential equations (3rd ed. Bryden. V. USA. & E. D. A. Westport: AVI. K. Wood.. Journal of Global Optimization. Suram. In M. Sweat. (1985). G. B. J. A. Engineering properties of foods. NY: Marcel Dekker Inc. E. S. Selected thermophysical properties of rice bran. K. (1995).. R. S. International Computer Science Institute.) Physical properties of foods (pp. Smith. Tao.). Differential evolution: a simple and efficient adaptive scheme for global optimization over continuous spaces. 341–359. Peleg. & Corte´s. An inverse method to estimate thermophysical properties of foods at freezing temperatures: apparent volumetric specific heat. 11(4). Sweat. (1996).. & Liuzzo. Technical Report TR-95-012. (1994).