Narjes Malekjani, Seid Mahdi Jafari, Mohammad …scientificfinding.gau.ac.ir/uploading/scientificfinding...Narjes Malekjani, Seid Mahdi Jafari, Mohammad Hashem Rahmati, Ebrahim Esmaeel

Narjes Malekjani*, Seid Mahdi Jafari, Mohammad Hashem Rahmati, Ebrahim Esmaeel Zadeh,and Hbibollah Mirzaee

Evaluation of Thin-Layer Drying Models andArtificial Neural Networks for Describing DryingKinetics of Canola Seed in a Heat Pump AssistedFluidized Bed Dryer

Abstract: In this study, drying characteristics of canola seedsweredeterminedusingheatedambient air at 40, 50and60°C,relativehumidity of 20, 40and60%andconstant velocity of 3m/s. To select a suitable drying curve, six thin-layer dryingmodels were fitted to experimental data. The models werecompared according to three statistical parameters: R2,reduced chi-square (χ2) and root mean square error. Usingsome experimental data, an Artificial neural network model,trained by Feed Forward Back-Propagation algorithm, wasdeveloped to predict moisture ratio values based on the threeinput variables. Different activation functions and severalruleswereusedtoassesspercentageerrorbetweenthedesiredandpredictedvalues.Accordingtotheresults, theapproxima-tion of diffusion drying model had better agreement with thedrying data. The artificial neural network model was able topredict the moisture ratio quite well with R2 of 0.9994. Thepredictedmean square error was obtained as 0.00012575.

Keywords: canola, thin-layer drying model, dryingkinetics, neural network

*Corresponding author: Narjes Malekjani, Student of Food Scienceand Technology, Department of Food Science and Technology,University of Tehran, Tehran, Iran, E-mail: [email protected] Jafari,Department of Food Science and Technology, GorganUniversity of Agricultural Science and Natural Resources, Gorgan, Iran,E-mail: [email protected] Hashem Rahmati, Department of Agricultural Machinery,Gorgan University of Agricultural Science and Natural Resources,Gorgan, Iran, E-mail: [email protected] Zadeh,Departmentof AgriculturalMachinery,GorganUniversity of Agricultural Science and Natural Resources, Gorgan, Iran,E-mail: [email protected] Mirzaee, Department of Food Science and Technology,Gorgan University of Agricultural Science and Natural Resources,Gorgan, Iran, E-mail: [email protected]

1 Introduction

Rapeseed (Brassica napus L.) is one of the world’s majorsources of edible vegetable oils [1]. Keeping grain cool

and dry is the key to successful canola storage. Canolaseed is usually harvested at high moisture content.Therefore, drying is an important post-harvest treatmentprior to optimal storage. Improper drying condition is oneof the causes of poor rapeseed quality [2]. Drying is aprocess comprising simultaneous heat and mass transferwithin the material and between the surface of the mate-rial and the surrounding media [3].

Simulationmodels are helpful in designing and improv-ing drying systems or for the control of the drying operation.Thin-layer dryers refer to the grain dryers in which all grainsare fully exposed to the drying air under constant dryingconditions, i.e. at constant air temperature and humidity.All commercial flow dryers are designed on thin-layer dryingprinciples. Thus, thin-layer drying simulation is the best cri-terion tomodel the food drying process [4].

Several thin-layer equations have been presented inthe literature. Although a considerable amount of datahas been reported in the literature regarding the thin-layer drying modeling of various agricultural products(fruits, crops and vegetables) like pistachio [5, 6], roughrice [7], green pepper, green bean and squash [8], grape[9, 10], onions [11], red pepper [12], garlic cloves [13] andear corn [14], little information is available on the thin-layer drying of canola seed. On the other hand, drying incontrolled drying air relative humidity and its influenceon the dried product characteristics and correspondingdrying models has not been studied thoroughly.

Sometimes, it is impossible to establish thin-layerequations due to the complexity of the phenomenon.Therefore, “black box” models based on input–outputpatterns like artificial neural networks (ANNs) are useful[15–18].

ANNs are mathematical tools whose functioning isinspired by that of the human brain [19, 20]. They are apromising tool for simulating variables of processesbecause of their simplicity. They have the ability tolearn the complex relationships without a priori knowl-edge of model structure [21–24].

doi 10.1515/ijfe-2012-0136 International Journal of Food Engineering 2013; 9(4): 375–384

Brought to you by | New York UniversityAuthenticated | 128.122.65.156

Download Date | 4/12/14 6:07 AM

Prediction of heat and mass transfer in the dryingprocess of mango and cassava was achieved using neuralnetworks [25]. ANNs were also used for rapid determina-tion of the drying kinetics (k and n) used in Page equationin drying potato slices by Islam et al. [26]. Cubillos andReyes [27] also used ANN approach for modeling thedrying of carrots.

The objective of this study was to investigate the adapt-ability of a thin-layer dryingmodel to canola seed, by choos-ing the best among six thin-layer drying equations andcompare neural networks with the mathematical models forthe prediction of thin-layer drying of canola seed.

2 Materials and methods

2.1 The experimental facility

The fluidized bed drying chamber received wet solids anddischarged dried products. Processed air was dehumidi-fied by the evaporator and then was heated to a highertemperature by the heater. The air was then arrived intothe fluidized bed chamber from the bottom. Moisture wastransferred from the solids into the air when two phases

were in contact. The schematic of the designed dryerapparatus and components of the dehumidifying systemis shown in Figure 1.

The dehumidifying system consisted of an evapora-tor, a compressor, a condenser and an expansion valveand a damper. The working fluid (R22 refrigerant) at lowpressure is vaporized in the evaporator by receiving heatfrom the surroundings. That causes temperature of the airat the surrounding to decrease. If the temperature of airwas below the dewpoint, condensation occurs; the air iscooled and dehumidified. Thus, evaporator acted as acondenser or dehumidifier for the air.

The working fluid then goes to compressor, wherethe enthalpy of the working fluid raises and discharges itas superheated vapor at high pressure. The superheatedvapor then goes to the condenser, where heat is removedfrom the working fluid and transferred to the process airaround the condenser. As a result, the temperature of theair increases. Finally, the working fluid is throttled usingan expansion valve to the low pressure line and entersthe evaporator to complete the cycle.

To control the humidity level at a specified point, adamper was installed in the drying apparatus. Whenthe humidity of the air reaches the determined level,the sensor switches the damper on and produces vapor.

9

4

3

21

DETAIL ASCALE 1 : 5

DETAIL CSCALE 1 : 10 5

6

7

8

Figure 1 Schematic of dryer apparatus. (1) Feeding perforated screen, (2) main fluidization chamber, (3) supplementary fluidizationchamber, (4) centrifugal blowing fan, (5) demoisturizing compressor, (6) condenser, (7) evaporator, (8) air heater and (9) moisturizing unit

376 N. Malekjani et al.: Drying Characteristic of Canola Seed



When the relative humidity of drying air increases upto the desired level, the humidifier would be switchedoff.

2.2 Material preparation and dryingconditions

Samples of freshly harvested canola seed (RGS003 vari-ety) were cleaned, the oven drying technique was used tomeasure the initial moisture content of the grains accord-ing to AOAC method 934.06 [28]. After determining theinitial moisture content, sufficient amount of water wasadded to the seeds in order to increase the moisturecontent to a desired level, then the samples were sealedin double layers of polythene bags and stored in a refrig-erator at 5°C [29].

Before starting the experiments, the samples weretaken from the refrigerator and kept at room temperaturefor about 24 h, so that the samples had reached thermalequilibrium [30].

The experiments for the thin-layer drying were com-pletely randomized. Different combinations of three dry-ing air temperature levels (40, 50 and 60°C) and threerelative humidity levels (20, 40 and 60%), for a total ofnine experiments, were each performed in triplicate.These temperatures combined with conditioned relativehumidity were selected to study drying of canola seed inlow temperature levels that has almost no effects oncanola seed chemical properties. The experiments wereconducted, after the system had reached a steady state.The time required to stabilize the whole system wasabout 2 h.

Most thin-layer models do not include air velocity asan independent variable [31]. The air velocity was fixed at3 m/s in order to have a steady fluidization of the seedsample, because the terminal velocity of canola seed wasdetermined about 3 m/s before the drying experiments.The air velocity was checked with an anemometer (TESTO405-V1, Germany) with a resolution of 0.01 m/s. A

thermocouple temperature sensor with a resolution of0.1°C was used for measurement of drying temperature(ELREHA GMBH, Germany). Each experimental replicateused a sample of about 100 g of canola seed.

During the drying process, the samples were periodi-cally weighed using an electronic balance (Satorius,Germany) to determine weight loss, from which the dry-ing curves were obtained.

Weighing of the sample to record the change inweight was carried out every 10 min. The change inmoisture content, relative to drying time, can be calcu-lated based on the weight change. When the samplesreached the predetermined moisture content, the experi-ment was stopped. The final desired moisture contentwas 8% (dry basis) as being ideal for safe canola seedstorage [32, 33]. The moisture content at any time indrying experiment was calculated by [34]:

Mf ¼ W0 �Wf

W0� 100 ð1Þ

2.3 Mathematical modeling of drying curves

Drying curves were fitted with six thin-layer drying mod-els, namely, the Page, the two-term exponential,Approximation of diffusion, the Verma et al., the Wangand Singh and the modified Handerson and Pabis model(Table 1). The moisture ratio of canola seed during dryingexperiments was calculated using the following equation:

MR ¼ M �Me

M0 �Með2Þ

where MR, M, M0 and Me are the moisture ratio, moisturecontent at any time, initial moisture content and equili-brium moisture content, respectively.

The coefficient of determination (R2) was one of theprimary criteria for selecting the best model to describethin-layer drying curves of canola seeds. However, thereare some statistical test methods such as the reduced chi-

Table 1 Mathematical models applied to drying curves

Model no. Model name Model References

1 Page MR ¼ expð�ktnÞ Page [35]2 Two-term exponential MR ¼ a expð�ktÞ þ 1� að Þ expð�katÞ Yaldiz et al. [10]3 Wang and Singh MR ¼ M0 þ at þ bt2 Ozdemir and Devres [36]4 Approximation of diffusion MR ¼ a expð�ktÞ þ 1� að Þ expð�kbtÞ Yaldiz and Ertekin [8]5 Verma et al. MR ¼ a expð�ktÞ þ 1� að Þ expð�gtÞ Verma et al. [37]6 Modified Henderson and Pabis MR ¼ a expð�ktÞ þ b expð�gtÞ þ c expð�htÞ Karathanos [38]

N. Malekjani et al.: Drying Characteristic of Canola Seed 377



square (χ2) and root mean square error (RMSE) to evalu-ate the goodness of fit of the models. Lower the χ2 andRMSE values and higher R2 values, better the fitness.These parameters can be calculated by eqs (3)–(5)[39, 40]:

R2 ¼ 1�PN

i¼1 ðMRpre;i �MRexp;iÞ2PNi¼1 ðMRpre;i �MRpre;iÞ2

ð3Þ

χ2 ¼Pn

i¼1 ðMexp;i �Mpre;iÞ2N � n

ð4Þ

RMSE ¼ 1N

XnI¼1

ðMpre;i �Mexp;iÞ2" #1=2

ð5Þ

where MRexp,i is the experimental moisture ratio at obser-vation i, MRpre,i is the predicted moisture ratio at thisobservation, N is number of observations and n is num-ber of constants.

2.4 Neural network design

To obtain the best prediction by the network, severalarchitectures were evaluated and trained using theexperimental data. The Feed Forward Back-Propagationalgorithm was utilized in training of all ANN models.

This algorithm uses the supervised training techni-que where the network weights and biases are initializedrandomly at the beginning of the training phase.

The error minimization process is achieved using agradient descent rule. There were three inputs and one

output parameters in the experimental tests. The threeinput variables are time (in min), temperature (°C) andrelative humidity (%). The output for evaluating dryerperformance is MR. Therefore, the input layer consistedof three neurons, and the output layer had one neuron(Figure 2).

Figure 2 shows the configuration of multilayer neuralnetwork for predicting MR (moisture ratio). Tangent (Tan)function suited best for the output layer. This arrange-ment of functions in function approximation problems ormodeling is common and yields better results. However,many other networks with several functions and topolo-gies were examined. The training [(mean square error(MSE)] was chosen to be 0.00001 for all ANNs. The com-plexity and size of the network was also important, so thesmaller ANNs had the priority to be selected. Differenttraining algorithms were also tested, and finally,Levenberg–Marquardt (trainlm) was selected. The com-puter program MATLAB 7.9, neural network toolbox, wasused for ANN design.

3 Results and discussion

3.1 Drying curves

Figure 3 shows final drying time versus temperature atconstant drying air relative humidity. It is obvious thatat high-temperature levels, the difference between totaldrying times is lower than that at low temperatures.Total drying time at 60% drying air relative humidity

Hidden layerInput

Time

Temperature

Relative humidity

Output

MR

Figure 2 Configuration of neural network for predicting MR (moisture ratio)




reduced about 22.57% compared with experiments per-formed at 20% RH at constant air temperature, it canbe discussed on the basis of the fact that a drying airwith low relative humidity has a greater potential todiffuse in the particles and remove the moisture fromthem. Whereas increasing air temperature from 40 to60°C resulted in shortening the drying times about54%. This is because increasing the temperature leadsto an increase in drying rate because of the higher bedtemperature of the particles; this increases the intra-particle moisture diffusion leading to a higher dryingrate. The increased transport properties of the fluidswith the increase in temperature have been wellknown, and the experimental data are in agreementwith the basic concepts of the mass transfer. A higherbed temperature increases the moisture diffusion rate,resulting in an increased drying rate. These results arein agreement with earlier researches [41–43].

In the other words, the effect of air temperaturecompared with air relative humidity is significant intotal drying time of canola seed. The drying rate washigh at the beginning of drying and became low as themoisture content decreased. Figure 4 presents the experi-mental data (moisture ratio vs time) obtained for air attemperatures ranging from 40, 50 and 60°C at differentrelative humidities. In all three figures, the best-fittingcurve is also included (as will be discussed later).Moreover, it is proven that no constant drying rate periodis significant in canola drying, and drying of canola seedoccurred in the falling rate phase.

3.2 Analysis of data

The moisture ratio, MR, of each experimental series wascalculated. Non-linear regression analysis using MATLABcomputer program was made for fitting the measureddata (moisture ratio and time) into six thin-layer dryingmodels, displayed in Table 1. Tables 2–4 show the fittingresults (R2, RMSE and χ2) for the models listed in Table 1,using the experimental data, with the best-fitting modelin bold type. The best model describing the thin-layerdrying kinetics was selected with the highest R2 averagevalues, and the lowest RMSE and χ2 average values. Bycomparing R2, RMSE and χ2 average values, it is clear thatthe approximation of diffusion model satisfactorily fitsexperimental data on drying kinetics observed in thefalling rate period and has better agreement with theexperiments. The approximation of diffusion model con-stants are reported in Table 5. In all the experiments, thevalues of R2 and RMSE were 0.999997–0.999999 and0.005534–0.026604, respectively.

Chen and Morey [44] and Xanthopoulos et al. [45]stated that data points in a plot of the residuals versuspredicted values should tend to fall in a horizontalband centered on zero line, displaying no systematictendencies toward a clear pattern. Therefore, plottingthe experimental data against the predicted values ofdimensionless moisture ratio, Figure 5, there were nosystematic patterns. The proximity of residualsaround zero line shows the sufficiency of the derivedmodel.

2030 40 50 60 70

30

40

50

60

70

80

90

100

Tim

e (m

in)

Temperature (°C)

20% RH

40% RH

60% RH

Figure 3 Drying air temperature versus final drying time at different air relative humidities for drying of canola seed




00 20 40 60 80 100 120

0.2

0.4

0.6

0.8

1

1.2M

R

Drying time (min)

40°C – 20%

50°C – 20%

60°C – 20%

40°C – 40%

50°C – 40%

60°C – 40%

40°C – 60%

50°C – 60%

60°C – 60%

Figure 4 Plots of canola moisture ratio (M/M0) versus time at different temperatures and relative humidities

Table 2 Coefficient of determination for RH ¼ 20%, varying T

Model Temperature

40°C 50°C 60°C

R2 RMSE χ2 R2 RMSE χ2 R2 RMSE χ2

Page 0.999967 0.047865 0.000417 0.99998 0.032965 0.000198 0.999984 0.024527 0.000109Two-term exponential 0.999948 0.058304 0.000618 0.999872 0.075947 0.001049 0.999766 0.08237 0.001234Wang and Singh 0.999089 0.272495 0.013501 0.998338 0.337558 0.020717 0.997067 0.399386 0.029002Approximation of diffusion 0.999988 0.022972 0.000158 0.999997 0.010303 3.18e–05 0.999997 0.008658 2.25e–05Verma et al. 0.995945 0.402904 0.0487 0.99876 0.197952 0.011755 0.999829 0.06318 0.001198Modified Henderson and Pabis 0.999988 0.016242 0.000226 0.999997 0.007569 4.91e–05 0.999994 0.008286 5.89e–05


Model Temperature

40°C 50°C 60°C


Page 0.999982 0.038213 0.000265 0.999989 0.032965 0.000198 0.999994 0.014809 3.99e–05Two-term exponential 0.999984 0.024527 0.000239 0.999926 0.058466 0.000622 0.999849 0.069192 0.00087Wang and Singh 0.999391 0.237671 0.01027 0.998419 0.323023 0.018972 0.997416 0.382101 0.026546Approximation of diffusion 0.999987 0.026604 0.000212 0.999996 0.011583 4.03e–05 0.999999 0.005534 9.19e–06Verma et al. 0.996451 0.409237 0.050243 0.997535 0.272682 0.022307 0.999349 0.125227 0.004705Modified Henderson and Pabis 0.99998 0.022991 0.000453 0.999996 0.008191 5.75e–05 0.999999 0.003913 1.31e–05





Model Temperature

40°C 50°C 60°C


Page 0.999987 0.035262 0.000226 0.999987 0.027693 0.000139 0.999966 0.044129 0.000354Two-term exponential 0.999991 0.028537 0.000148 0.999967 0.043222 0.00034 0.99985 0.085528 0.00133Wang and Singh 0.999581 0.207056 0.007795 0.998881 0.288095 0.015091 0.998458 0.33434 0.020324Approximation of diffusion 0.999991 0.023053 0.000159 0.999996 0.011833 4.2e–05 0.999986 0.02272 0.000155Verma et al. 0.994576 0.534991 0.085865 0.99656 0.346788 0.036079 0.999606 0.117659 0.004153Modified Henderson and Pabis 0.99999 0.017313 0.000257 0.999996 0.008808 6.65e–05 0.999986 0.016339 0.000229

Table 5 Regression analysis coefficients related to the approximation of diffusion model for each experiment

Temperature (°C) Relative humidity (%) Coefficients of drying models

k a b

40 20 0.769306 0.041212 0.26756240 0.141324 0.207636 0.27190460 0.192271 0.155532 0.174299

50 20 0.365744 0.896550 0.38229740 0.335218 0.104682 0.34011260 0.297409 0.114499 0.282693

60 20 0.280030 0.124267 0.50408540 0.219952 0.156923 0.49673460 0.334231 0.092549 0.389822

1

0.8

0.6

0.4

Y = 1.0028x – 0.0021R2 = 0.9995

0.2

00 0.2 0.4 0.6

MRexp

MR

pre

0.8 1 1.2

1.2

Figure 5 Plot of predicted values of MR versus the measured values for approximation of diffusion model




3.3 Results of artificial neural networkmodeling

An ANN was developed based on the experimentalwork. Results showed that Feed Forward Back-Propagation training algorithm was well suited for pre-diction of moisture ratio based on different times, rela-tive humidities and temperature levels. Performancediagram of the ANN model is shown in Figure 6.Predicted versus experimental values for the studiedparameters are indicated in Figure 7. This figure

shows the results of analysis for moisture ratio. Ascan be seen, all the investigated prediction modelssimulate the experiments satisfactorily. The developednetwork had a good generalization in predicting thequality of the canola seed from the drying process.Thus, this network model could be used to determinethe moisture ratio of canola seed under the dynamicdrying system. ANN predictions for MR yielded coeffi-cient of determination (R2) of 0.9994 and MSE value of0.00012575. Summary of the various ANN networks hasbeen presented in Table 6.

100

10–1

10–2

10–4

10–3

10–5

0 5 10 15 20 25

Train

Validation

Test

Epoch

MS

E

Figure 6 Training error curve

1

0.8

0.6

0.4

0.2

0

1.2

0 0.2 0.4 0.6

Measured (MR)

Pre

dic

ted

(M

R)

0.8 1 1.2

Y = 0.999x + 0.001R2 = 0.9994

Figure 7 Correlation between the experimental data and the predicted values of the ANN model for moisture ratio (%)




4 Conclusions

Thin-layer drying experiments were conducted to deter-mine the thin-layer drying characteristics of canolaseed. Six thin-layer drying models were evaluated fortheir suitability. The approximation of diffusion modelwas the best model for describing the thin-layer dryingcharacteristics of canola seed. The analysis alsoshowed that the drying of canola seed occurred duringthe falling rate period and that no constant rate periodwas observed. This indicated that a diffusion phenom-enon is the governing physical mechanism of moisturemovement in the canola seed. The moisture diffusionmodel was determined. The final selected model, 3-10-2(3 neurons in input layer, 10 neurons in hidden layer 1and 2 neurons in output layer), successfully learned therelationship between input and output parameters. TheANN results are quite satisfactory in a way that R2

values are close to one, while MSEs were found to bevery low. Analysis of the experimental data by the ANNrevealed that there is a good correlation between theANN-predicted results and the experimental data.Generally speaking, ANN proved to be a reliable alter-native for canola seed thin-layer drying prediction dueto generality and simplicity.

Notations

a, b, c, g, h, n dimensionless drying constants in dryingmodels

Deff effective moisture diffusivity (m2/s)k drying constant in drying modelsM moisture content (% dry basis) (g water/g dry

solids)MR moisture ratioMSE mean square errorn number of constantsN total number of observationsr particle radiusR2 coefficient of determinationRH relative humidity in drying chamberRMSE root mean square errorT temperature (°C)t time (min)W weight (g)χ2 reduced chi-square

Subscripts0 initiale equilibriumexp experimentalf final

pre predictedw water

Table 6 Summary of the various ANN networks

Activation function Neurone in hidden layer 1 Neurone in hidden layer 2 MSE Epoch R2

TANSIG 5 – 0.00021076 22 0.99868TANSIG 10 – 0.00012575 23 0.9994TANSIG 15 – 0.0675940 7 0.62209TANSIG 20 – 0.04252 1 0.73264TANSIG 25 – 0.00013931 11 0.99914TANSIG 30 – 0.00066779 19 0.99842TANSIG 40 – 0.00021505 12 0.99924TANSIG 50 – 0.098054 0 0.4204TANSIG 60 – 0.061714 0 0.70734TANSIG 70 – 0.0058948 13 0.96663TANSIG 80 – 0.11153 4 0.37248TANSIG 90 – 0.016316 13 0.96196TANSIG 100 – 0.0025453 25 0.98876TANSIG 10 10 0.00037601 17 0.99901TANSIG 10 20 0.00067978 18 0.99741TANSIG 10 30 0.037463 18 0.76560TANSIG 10 40 0.0018337 18 0.99407TANSIG 20 10 0.0019113 24 0.98219TANSIG 20 20 0.10557 5 0.18731TANSIG 20 30 0.066668 16 0.59806TANSIG 20 40 0.0026562 29 0.99113




References

1. Donald BE, Bassin P. Rapeseed and crambe: alternative cropswith potential industrial uses. Bulletin 656. AgriculturalExperiment Station, Kansas State University, Manhattan,Walter R. Woods, 36 p, 1991. ISSN 0097-0484.

2. Sadowska J, Fornal J, Ostaszyk A, Szmatowicz B. Drying condi-tions and processability of dried rapeseed. J Sci Food Agric1996;72:257–62.

3. Akpinar EK, Sarsilmaz C, Yildiz C. Mathematical modeling of athin layer drying of apricots in a solar energized rotary dryer.Int J Energy Res 2004;28:739–52.

4. ChakravertyA,SinghRP.Post-harvest technologyof cereals,pulsesand oilseeds. NewDelhi, India: Oxford and IBH Publishing, 1988.

5. Midilli A. Determination of pistachio drying behavior and condi-tions in a solar drying system. Int J Energy Res 2001;25:715–25.

6. Midilli A, Kucuk H. Mathematical modeling of thin layer dryingof pistachio by using solar energy. Energy Convers Manage2003;44:1111–22.

7. Basunia MA, Abe T. Thin-Layer solar drying characteristics ofrough rice under natural convection. J food Eng 2001;47:295–301.

8. Yaldiz O, Ertekin C. Thin layer solar drying of some vegetables.Dry Technol 2001;19:583–96.

9. Dincer I. Sun drying of sultana grapes. Dry Technol1996;14:1827–38.

10. Yaldız O, Ertekin C, Uzun HI. Mathematical modelling of thinlayer solar drying of sultana grapes. Energy 2001;26:457–9.

11. Singh H, Sodhi NS. Dehydration kinetics of onions. J Food SciTechnol 2000;37:520–2.

12. Doymaz I, Pala M. The effects of dipping pretreatment on air-drying rates of seedless grapes. J Food Eng 2002;52:413–27.

13. Sharma GP, Prasad S, Datta AK. Drying kinetics of garlic clovesunder convective drying conditions. J Food Sci Technol2003;40:45–51.

14. Friant NR, Marks BP, Bakker-Arkema FW. Drying rate of corn. TASAE 2004;47:1605–10.

15. Wu D, Zijiang Y, Liang L. Using DEA-neural network approachto evaluate branch efficiency of a large Canadian bank. ExpertSyst Appl 2006;31:108–15.

16. Fujiwara T. In: Bulsari AB, editor. Neural networks for chemicalengineers. Amsterdam: Elsevier-Publishers, 1995:283–95.

17. Haykin S. Neural networks, a comprehensive foundation. NewYork: Macmillan College Publishing Company, 1994.

18. Tai-Yue W, Shih-Chien C. Forecasting innovation performancevia neural networks – a case of Taiwanese manufacturingindustry. Technovation 2006;26:635–43.

19. Grossberg S. Studies of the mind and brain. Drodrecht,Holland: Reidel Press, 1982.

20. Kohonen T. Self-organization and associative memory, 2nd ed.Springer Series in Information Sciences, Vol. 8. Berlin:Springer Verlag, 1987.

21. Hill T, Marquez L, Oconnor M, Remus W. Artificial neural net-work models for forecasting and decision making. Int JForecasting 1994;10:5.

22. Savkovic-Stevenovic J. Neural networks for process analysisand optimization: modelling and applications. Comput ChemEng 1994;18:p 1149.

23. Shene C, Andrews B, Asenjo AJ. Optimization of Bacillus sub-tilis for the fed-batch fermentations for the maximization of the

synthesis of a recombinant b –1, 3-glucanase, CAB, Japan, Vol.7, 1998:219.

24. Shene C, Andrews B, Asenjo AJ. Fed-batch optimizations ofbacillus subtilis fed-batch ToC46 (pPFF1) for the synthesis ofa recombinant b –1, 3-glucanase: experimental study andmodelling. Enzyme Microb Tech 1999;24:247.

25. Hernandez-Perez JA, Garca-Alvarado MA, Trystram G, Heyd B.Neural networks for the heat and mass transfer predictionduring drying of cassava and mango. IFSET 2004;5:57–64.

26. Islam Md, Sablani SS, Mujumdar AS. An artificial neural net-work model for prediction of drying rates. Dry Technol2003;21:1867–84.

27. Cubillos F, Reyes A. Drying of carrots in a fluidized bed. II.Design of a model based on a modular neural networkapproach. Dry Technol 2003;21:1185–96.

28. AOAC. Official method of analysis, association of official ana-lytical chemists (no. 934.06). Arlington, VA: AOAC, 1990.

29. Cassells JA, Caddick LP, Green JR, Reuss R. Isotherms forAustralian canola varieties. In: Wright EJ, Webb MC, HighleyE, editors. Proceedings of the Australian postharvest technicalconference, 25–27 June 2003, Canberra, 2003.

30. ANSI/ASAE S448.1. Thin-Layer drying of agricultural crops.ASAE standards 51st edition, 2004:598–600.

31. Sinicio R, Muir WE, Jayas DS, Cenkowski S. Thin-layer dryingand wetting of wheat. Postharvest Biol Technol 1994;5:261–75.

32. Crisp J, Woods JL. The drying properties of rapeseed. J AgricEng Res 1994;57:89–97.

33. Pathak PK, Agrawal YC, Singh BP. Thin-layer drying model forrapeseed. T ASAE 1991;34:2505–8.

34. LoCurto GJ, Zakirov V, Bucklin RA, Hanes DM, Teixeira AA,Walton OR, et al. Soybean friction properties. AnnualInternational Meeting, ASAE, Paper No. 97–4108, 1997.

35. Page GE. Factors influencing the maximum of air drying shelledcorn in thin layer. Indiana: Purdue University, 1949.

36. Ozdemir M, Devres YO. The thin layer drying characteristics ofhazelnuts during roasting. J Food Eng 1999;42:225–33.

37. Verma LR, Bucklin RA, Endan JB, Wratten FT. Effects of drying airparameters on rice dryingmodels. T ASAE 1985;28:296–301.

38. Karathanos VT. Determination of water content of dried fruitsby drying kinetics. J Food Eng 1999;39:337–44.

39. Demir V, Gunhan T, Yagcioglu AK, Degirmencioglu A. Mathematicalmodelling and thedetermination of somequality parameters of air-dried bay leaves. Biosyst Eng 2004;88:325–35.

40. Togrul IT, Pehlivan D. Modelling of drying kinetics of singleapricot. J Food Eng 2003;58:23–32.

41. KudrasT,EfremovGI.Aquasi-stationaryapproachtodryingkineticsin fluidized particulatematerials. Dry Technol 2003;21:1077–90.

42. Syahrul S, Hamdullahpur F, Dicer I. Exergy analysis of fluidizedbed drying of moist particles. EXERGY 2002;2:87–98.

43. Topuz A, Gur M, Gul MZ. An experimental and numerical studyof fluidized bed drying of hazelnut. Appl Therm Eng2004;24:1534–47.

44. Chen CC, Morey RV. Comparison of four EMC/ERH equations. TASAE 1989;32:983–90.

45. Xanthopoulos G, Oikonomou N, Lambrinos G. Applicability of asingle-layer drying model to predict the drying rate of wholefigs. J Food Eng 2007;81:553–9.




Documents

Narjes Malekjani*, Seid Mahdi Jafari, Mohammad …scientificfinding.gau.ac.ir/uploading/scientificfinding...Narjes Malekjani*, Seid Mahdi Jafari, Mohammad Hashem Rahmati, Ebrahim Esmaeel

Narjes Malekjani, Seid Mahdi Jafari, Mohammad …scientificfinding.gau.ac.ir/uploading/scientificfinding...Narjes Malekjani, Seid Mahdi Jafari, Mohammad Hashem Rahmati, Ebrahim Esmaeel