Application of Neural Network for Estimation of Pistachio Powder Sorption Isotherms

Transcription

1 American-Eurasian J. Agric. & Environ. Sci., (): -, ISSN - IDOSI Publications, DOI:./idosi.aejaes... Application of Neural Network for Estimation of Pistachio Powder Sorption Isotherms Hamid Tavakolipour and Mohsen Mokhtarian Department of Food Engng, Sabzevar Branch, Islamic Azad University, Sabzevar, Iran Abstract: Moisture sorption isotherms for pistachio powder were determined by gravimetric method at temperatures of,, and C. Maximum hysteresis level in sorption isotherms curve was observed at range of water activity.-.. On the other hand, hysteresis was diminished between sorption isotherm curves when temperature reached to C. Some mathematical models were tested to measure the amount of fitness of our experimental data including BET, GAB, Smith, Caurie, Halsey, Harkins and Jura, Freundlich, Kuhn, Oswin and Iglesias and Chirife. This mathematical analysis proved that Caurie model was the most appropriate one. Thermodynamic relationship of Clasius-Clapeyron was applied to determine isosteric heat of adsorption and desorption of pistachio powder moisture. As well, adsorption-desorption moisture content of pistachio powder were predicted by means of artificial neural network (ANN) approach. The results showed that, MLP network can able to predicted adsorption-desorption moisture content with R values. and., respectively. Comparison of ANN results with classical sorption isotherm models revealed that ANN modeling had greater accuracy to predict equilibrium moisture content of pistachio powder. Key words: Pistachio Modeling Sorption isotherm Neural network approach Isosteric heat of sorption INTRODUCTION stability of raw and processed agricultural products, the sorption isotherms can be used to obtain their optimum The pistachio (Pistacia vera L.) is cultivated in the residual moisture content and end point of a drying Middle East, United States and Mediterranean countries process, to design their suitable packaging systems and []. Pistachio is one of the most important Iranian storage conditions and to select proper ingredients for horticultural products with high export value for the preparing a formulated intermediate moisture food []. country and Persian cultivars recognized have different Maskan and Karatas () determined the sorption taste and flavour. According to Food and Agricultural characteristics of whole pistachio nuts at, and C Organization (FAO), Pistachio production in was and evaluated their data for fitting to some sorption reported MT, which Iran is dedicated of about equations []. They found some important parameters.% of the global production of pistachio in this year such as monolayer moisture content and heat effects on []. According to position of this product in the country water sorption. Additionally, the adsorption studies were economy, suitable preservation and storage conditions to done on few pistachio varieties and their products, e.g., prevent spoilage and damage it, is important. raw pistachio (Pistacia terebinthus L.) and its protein During various postharvest processes such as isolate by Ayrancy and Dalgic (), pistachio nut paste drying, storage and packaging, water adsorption and at different temperature values [, ]. However, very desorption processes play an important role on limited data are available on sorption isotherms and other deteriorative reactions (DRs) such as browning, lipid characteristics of different Persian pistachio varieties. oxidation and microbial growth of pistachio nuts. The Presently, neural networks enact an important key as influence of moisture on DR has been explained in terms a powerful analysis machine in predicting the process of a w and this approach is now well established in parameters. The scientists were used ANN in various field controlling reaction and predicting food stability. of food processing. For example, used an ANN approach Because a is the most important parameter for the as intelligent tools to prediction of food drying parameters w Corresponding Author: Hamid Tavakolipour, Department of Food Engng, Sabzevar Branch, Islamic Azad University, Sabzevar, Iran.

2 Am-Euras. J. Agric. & Environ. Sci., (): -, [], freezing and thawing times [], osmotic dehydration for three consecutive weighting at -day intervals. parameters such as solid gain (SG) and water loss (WL) Moisture content of pistachio samples was measured by [], anthocyanin concentration [], antioxidant activity vacuum oven worked at C and mbar for h []. [] and equilibrium moisture content []. For the adsorption process, pistachio samples were The aim of this study were: () to evaluation of placed in suspended dishes in jars containing silica gel. feasibility of neural network to predict absorption and Again, the same procedure was repeated to reach the desorption moisture content of pistachio powder in order equilibrium moisture content. Desorption isotherms were to momentarily monitoring of storage conditions, () to determined on samples hydrated in a glass jar over determine the sorption isotherms and hysteresis effects distilled water. At high a wvalues (a w>.), a small amount and () to evaluate the suitability of various mathematical of toluene was placed in a capillary tube fixed to the inner models for fitting the isotherms and determine the wall of different jars to prevent microbial spoilage of the isosteric heat of sorption. pistachio samples []. MATERIALS AND METHODS Empirical Modeling: Ten mathematical models were used for fitting of sorption isotherm curves of pistachio powder Material: Raw and dried Kerman variety whole pistachios at different temperature [, ]. The equations of these were purchased from a local market. They were size sorted models are shown in Table. and separated to split and nonsplit samples. Medium- The goodness of the fitting was evaluated by sized and splitted pistachio nuts with a moisture content calculating two statistical parameters namely coefficient of.% (d.b.) were selected for the tests. After harvesting, of determination (R ) and mean relative deviation modulus the pistachio nuts should be dried from a moisture P (defined in equation ). Each model with highest R content of about % to safe storage moisture of less value and lowest P(%) value is the best model for fitting than % wet basis (w.b.). Pistachio nuts dried to a range of sorption moisture isotherm data of pistachio powder of % (w.b.) are graded higher in organoleptic quality []. The P(%) parameter can be calculated as follows: indicators such as crispness and sweetness and lower in N X pi, X bitterness and rancidity than those dried in the range of ei, P(%) = N -% (w.b.) moisture content. Pistachio powder was X i= ei, () produced by weighing g of pistachio kernels crushed in a home mill (Black and Decker, London, U.K.) for s where X e,i is experimental moisture content, X p,i is until its average particle size reached microns. predicted moisture content and N is number of Various saturated salt solutions including LiCl, experimental data. CHCOOK, MgCl, KCO, Mg(NO), NaNO, NaCl and KCl were used to obtain constant relative humidity []. Design of Artificial Neural Network: An artificial neural network was composed of simple processing elements Determination of Equilibrium Moisture Content: called neurons that are connected to each other by Sorption isotherms of pistachio powder were determined weights. The neuron is grouped into distinct layers and by a static gravimetric method at,, and C. interconnected according to a given architecture. According to Tavakolipour and Kalbasi Ashtari (), eight saturated salt solutions were prepared to provide a Table : Mathematical models for fitting of sorption data to pistachio range of.,.,.,.,.,. and. a w values []. After transferring ml of each salt solution into separate glass jars, the jar grids were suspended. About g sample of pistachio powder was weighed separately and placed on grids in the jars, which were then tightly closed and kept in convective ovens at different temperatures for equilibration. The required time to reach equilibrium moisture content for sample of Models BET Kuhn Freundlich Equation X = A (a ) pistachio powder was close to day. Equilibration was achieved when the changes in moisture content (d.b.) did not exceed.% and it was less than. g/g dry solids powder [, ] X = X Ca /[( a )( + a (C-))] ï w w w GAB X = X CKa /[( Ka )( Ka + CKa )] ï w w w w Smith X = A B ln( a ) w Oswin X = C [a /( a )] w w n X = [A/ln(a )] + B w Halsey X =[ K/ln(a w)] /n w /B. Harkins and Jura X = [B/(ln(a w) A)] Iglesias and Chirife X = A + B[a w/(-a w)] Caurie X = exp(a + Ba w)

3 A multilayer perception (MLP) networks is one of the most popular and successful neural network architectures, suited to wide range of engineering application involved drying. Mathematically: Am-Euras. J. Agric. & Environ. Sci., (): -, n y = f( wx) + b j ij i j i= () where y is the net input of each neuron in the hidden and j output layers, x is input, n is number of inputs to the i neuron, w ij is the weight of the connection between neuron I and neuron j and b is the bias associated with j th j neuron. Each neuron consists of a transfer function that express of its internal activation level. Output from a neuron is determined by transforming its input using a Fig. : Schematic structure of ANN, x : air temperature, suitable transform function. x : water activity (a w) and y: equilibrium moisture In this work, Number of - hidden layers with - content (absorption/desorption) neurons per hidden layer, learning rate=., momentum coefficient=., activation functions of sigmoid logarithms where P ANN is predicted ANN output parameter, P exp is (Eq. ) and hyperbolic tangent (Eq. ) in each hidden and experimental data and N is number of observations []. output layer and training period (epoch)= were used in order to find the best configuration. RESULTS AND DISCUSSION logsig( z) = (,+) () The effect of temperature on the isotherm curves of + exp( z ) pistachio powder: Pistachio powder adsorption and desorption isotherm curves at different temperatures are z z e e tenh( z) = (-,+) () presented in Fig.. These curves are shown as a function z z e + e of temperature where temperature increased the equilibrium moisture content of pistachio powder In this network, air temperature and water activity (a w) decreased for both desorption and desorption. This were selected as input parameters and equilibrium behavior is consistent with thermodynamic laws and moisture content (absorption/desorption) was selected as showed the reverse effect of increasing of moisture output parameter (Fig. ). Data randomly divided into two content or temperature on sorption enthalpy. However, groups, % used for training and reminder % for some of sugars and low molecular weight food testing of network. Data modeling accomplished by using components are exceptions and with increasing SPSS statistical software version (). In order to temperature they will more hygroscopic []. The study of evaluation of network performance and selected the best the curves observed that for water activity in range of topology were used from two criteria such as determine of.-. corresponding to the capillary condensation coefficient (R ) and mean relative error (MRE). region, when adsorption and desorption temperature at minimum ( C) its hysteresis reached to the highest N value. With increasing temperature hysteresis was ( PANN, i Pexp, i) reduced and for water activity lower than. and higher i= R = N () than. its reached to the lowest value []. ( PANN, i PANN, i ) Empirical modeling: In this study the BET, GAB, i= Oswin, Smith, Halsey, Henderson, Kuhn, Freundlich, Harkins and Jura, Iglesias and Chirife and Caurie equations were used to fit the sorption data for pistachio N ( PANN, i Pexp, i ) MRE = () powder. Statistical and model parameters are shown in N P i= exp, i Table. The calculation of R and P values for the ten

4 Am-Euras. J. Agric. & Environ. Sci., (): -, Table : The model constants and statistical parameters for experimental adsorption/desorption data at different temperatures Temperature ( C) Models Constants Adsorption Desorption Adsorption Desorption Adsorption Desorption Adsorption Desorption BET Xï C P (%) R GAB Xï C K P (%) R Smith A B P (%) R Oswin C n P (%) R Kuhn B A P (%) R Halsey K n P (%) R Freundlich A B P (%) R Harkins and Jura A B P (%) R Iglesias and Chirife A B P (%) R Caurie A B P (%) R models showed that the highest R (average. for adsorption and. for desorption) and the lowest P values (average. for adsorption and. for desorption) were related to Caurie equation in the first order and Smith equation with R (average. for adsorption and. for desorption) and P values (average. for adsorption and. for desorption) in the second order for fitting sorption data in all storage temperatures. Isosteric Heat of Sorption: Isosteric heat of sorption or binding energy defined as the amount of energy required to remove water from the substrate in excess of the amount of energy required for free water vaporization.

5 Am-Euras. J. Agric. & Environ. Sci., (): -, Adsorption Desorption C.... Water activity (aw) Adsorption Desorption C.... Water activity (a w) Adsorption Desorption C.... Water activity (a w) Adsorption Desorption.... Water activity (aw) Fig. : The absorption and desorption isotherms curves of pistachio powder at different temperatures C Furthermore, the isosteric heat of sorption shown in Fig. is a function of moisture content and it varied inversely with the amount of water vapor absorbed by the solid. The higher moisture content, the lesser energy was required to remove water molecules from the pistachio powder, as observed in Fig.. At moisture content more than %, the binding energy in pistachio powder decreases to zero, because water characteristics of pistachio approaches to free water. Also, results obtained Fig. : Isosteric heat of absorption and desorption of from binding energy of moisture adsorption and pistachio powder at different equilibrium moisture desorption showed that binding energy for adsorption contents isotherm is larger than desorption isotherm and causes some structural changes during drying process. When It's a valuable tool for understanding the water sorption moisture content in pistachio powder reduced from mechanism. Generally, sorption phenomena can be % to % (dry basis), the binding energy increased from explained by the differential form of the Clausius-. to. kj/mol. Clapeyron equation: Artificial Neural Network Modeling: In order to d( lmaw) Q = s () possibility of predicting of adsorption and desorption d(/ T) R moisture content of Persian pistachio powder, artificial where Q s is isosteric heat and R is universal gas constant. neural network model was used. For this aim, a Some selected values of isosteric heat of sorption were combination of layers and neurons with different estimated from the slope of ln a w versus /T plots at activation functions were used for modeling perceptron various moisture content values. The slope of the line neural network. Neural network include one and two (-Q s/r) decreases to zero as moisture content in hidden layers, to neuron were selected randomly and agricultural commodity increases which is an indicative of network power was estimated to predict adsorption and reduced water interactions, i.e., less binding energy []. desorption moisture content of Persian pistachio powder.

6 Am-Euras. J. Agric. & Environ. Sci., (): -, Table : ANN results to predict equilibrium moisture content (EMC) of pistachio powders with different activation functions at various temperatures and a w Number of neurons Activation Statistics No. of hidden layers function EMC parameters One hidden layers Logsig Adsorption MRE R Desorption MRE R Tanh Adsorption MRE R Desorption MRE R Two hidden layers Logsig Adsorption MRE R Desorption MRE R Tanh Adsorption MRE R Desorption MRE R Predicted AMC (% d.b) (a) Adsorbtion moisture content (AMC) Linear (Adsorbtion moisture content (AMC)) R =. Experimental AMC (% d.b) Predicted DMC (% d.b) (b) Desorbtion moisture content (DMC) Linear (Desorbtion moisture content (DMC)) R =. Experimental DMC (% d.b) Fig. : Predicted and experimental values of MLP network to predict EMC of pistachio powder, (a) Absorption moisture content with logsig activation function and (b) Desorption moisture content with tanh activation function Suitable learning epoch was selected according to and two hidden layers has been shown, topology of -- previous study in field of ANN. The results showed that, - (i.e. network with inputs, neurons in the first and the best learning epoch for logarithm sigmoid (logsig) and second hidden layer and output) with logsig had the tangent hyperbolic (tanh) activation function was best result to predict absorption moisture content. On the obtained equal []. The best learning epoch in each other hand, result of perceptron neural network with activation function was chosen base of achieving the same activation functions (logsig and tanh) with one and lowest mean relative error. two hidden layers indicated that, neural network with The obtained result of MLP network with logsig and structure of -- (i.e. network with inputs, one hidden tanh activation function and different configurations are layer content six neurons with two output) and tanh shown in Table. Investigation of obtained result for activation function had goodness result to predict MLP with logsig and tanh activation function with one desorption moisture content of Persian pistachio powder.

7 Am-Euras. J. Agric. & Environ. Sci., (): -, This network was able to predict absorption and function with neuron in first hidden layer had best result desorption moisture content with relative error values to predict desorption moisture content of pistachio. and., respectively. Likewise, R values for powder (R =.). Comparison of ANN results with absorption and desorption moisture content were classical models revealed that, ANN modeling had greater obtained. and., respectively. The similar results accuracy to predict adsorption-desorption moisture in field of sorption isotherm were reported by another content of pistachio powder. scientist in cases of raisin []. Fig. shows, model sensitivity diagram of predicted REFERENCES values by MLP network vs. experimental values for the best configurations (i.e. structure of --- with logsig. Tavakolipour, H. and M. Mokhtarian,. Neural activation function and structure of -- with tanh Network Approaches for Prediction of Pistachio activation function for predict absorption and desorption Drying Kinetics. International Journal of Food moisture content, respectively). The result indicated that, Engineering, : -. data were randomly located around the regression line.. FAO,. FAOSTAT information bank. Food and This could be a reason for carefully evaluating the neural Agriculture Organization of UN. networks to predict absorption and desorption moisture. Tavakolipour, H. and A. Kalbasi Ashtari,. content of Persian pistachio powder. Estimation of Moisture Sorption Isotherms in Kerman Pistachio Nuts. International Journal of Food CONCLUSIONS Engineering, : -.. Maskan, M. and S. Karatas,. Sorption The moisture sorption isotherms and isosteric heat Characteristics of Whole Pistachio Nuts of sorption are important in optimization of storage (Pistachio vera L.). Journal of Drying Technology, conditions and packaging of pistachio powder. Static : -. gravimetric method was used for determined the curves of. Ayrancy, E. and C. Dalgic,. Moisture Sorption pistachio powder isotherm curves at four storage Isotherms of Pistacia terebinthus L. and its Protein temperatures of,, and C and then they were Isolate. Lebensm. Wiss.u. Technology, : -. fitted by ten famous EMC mathematical models Among. Hayoglu, I. and O.F. Gamli,. Water Sorption different suggestive models for fitness of sorption data, Isotherms of Pistachio Nut Paste. International Caurie and Smith models were the best sorption models, Journal of Food Science and Technology, because of minimum deviations between theoretical and : -. actual data. Furthermore, the hysteresis was the highest. Goni, S.M., S. Oddone, J.A. Segura, R.H. Mascheroni value, when a w was in range of.-. and it reduced and V.O. Salvadori,. Prediction of Foods when it was less or more than. and., respectively. Freezing and Thawing Times: Artificial Neural On the other hand, the values of binding energy for Networks and Genetic Algorithm Approach. Journal adsorption and desorption isotherms were equal to of Food Engineering, : -.. kj/mol for pistachio powder. At moisture content more. Lertworasirikul, S. and S. Saetan,. Artificial than %, the binding energy in pistachio powder Neural Network Modeling of Mass Transfer During decreases to zero, because water characteristics of Osmotic Dehydration of kaffir Lime Peel. Journal of pistachio approaches to free water. Also, results obtained Food Engineering, : -. from binding energy of moisture adsorption and. Fernandes, A.M., P. Oliveira, J.P. Moura, desorption showed that binding energy for adsorption A.A. Oliveira, V. Falco, M.J. Correia and P. Meloisotherm is larger than desorption isotherm and causes Pinto,. Determination of Anthocyanin some structural changes during drying process. Concentration in Whole Grape Skins Using As well, in this study was evaluated possibility of Hyperspectral Imaging and Adaptive Boosting ANN to on-line monitoring of equilibrium moisture Neural Networks. Journal of Food Engineering, behavior of pistachio powder during drying and storage. : -. The results demonstrated that, neural network with logsig. Cimpoiu, C., V.M. Cristea, A. Hosu, M. Sandru and threshold function could be predicted adsorption L. Seserman,. Antioxidant Activity Prediction moisture content with nodes per first and second and Classification of Some Teas Using Artificial hidden layer with R value equal.. On the order hand, Neural Networks. Journal of Food Chmistry, the results showed that, this network with tanh activation : -.

8 Am-Euras. J. Agric. & Environ. Sci., (): -,. Amiri Chayjan, R. and M. Esna-Ashari,.. Noshad, M., M. Mohebbi, F. Shahidi and Comparison between Artificial Neural Networks and S.A. Mortazavi,. Effect of Osmosis and Mathematical Models for Estimating Equilibrium Ultrasound Pretreatment on the Moisture Adsorption Moisture Content in Raisin. Journal of Agriculture Isotherms of Quince. Journal of Food and Engineering International: CIGR J., (). Bioproducts Processing, : -.. AOAC,. Official Methods of Analysis of the. Shivhare, U.S., S. Arora, J. Ahmed and th Association of Official Analytical Chemists, Ed, G.S.V. Raghavan,. Moisture Adsorption Association of Official Analytical Chemists, Isotherms for Mushroom. Lebensm.-Wiss. u. Washington, DC, USA. Technol., : -.