Artificial Intelligence echniques for Foo Drying echnology Koksal Erenturk Ataturk University College of Engineering Department of Electrical&Electronics Eng. Erzurum, urkey erenturk@yahoo.com Abstract: Applications of artificial intelligence techniques, such as artificial neural networks, fuzzy logic, genetic algorithms an neural-fuzzy systems, in engineering have gaine momentum in past ecae. Main applications of these techniques in engineering are estimation, optimization an control process. In this paper, some of the applications are stuie an both simulation an real-time experimental results are given. Artificial neural networks an genetic algorithms are very useful for estimation an optimization process for rying technologies. However, fuzzy logic is also capable of both classification an control of the rying process. Estimation, optimization an control applications of artificial intelligence methos are given in etail for ifferent types of foo rying applications. Echinacea angustifolia an carrot are selecte as application examples. A fuzzy logic base control approach is employe to control a convective type rier. Estimation an optimization applications of artificial neural networks an genetic algorithms are compare with non-linear regression analysis. In aition, fuzzy control is also compare with a classical control technique to conclue the robustness of the fuzzy control in terms of classical control. Accoring to the results, it is observe that artificial intelligence techniques have several avantages such as: ecreasing computation time, increasing stability an accuracy. Moreover these techniques coul be applicable for ifferent type processes with simple changes in configuration. Introuction Drying behavior of ifferent materials has been propose in the literature by various researchers on both theoretical an application grouns uring the past 60 years. here have been many stuies for moeling of rying behavior an etermining the rying kinetics of various vegetables an fruits such as onion (Sarsavaia, Sawhney, Pangavhane, & Singh, 1999, grape (Dincer, 1996, potato (Diamante & Munro, 1993, pistachio (Miilli, 2001, kiwifruits (Maskan, 2001, re pepper (Akpinar, Bicer, & Yiliz, 2003, rosehip (Erenturk, Gulaboglu, & Gultekin, 2004a an b an Echinacearoots(Erenturk, Erenturk, & abil,2004c. Dynamic moeling of the rying characteristics of agricultural proucts, using artificial intelligence methos incluing genetic algorithms an neural networks has gaine momentum, because learning ability of the neural network is suitable for ientifying plant an fruit responses, which are complex processes to which mathematical approaches are not easily applie. Stuies to ientify nonlinear an ifficult-to-efine system behavior with ai of neural networks were conucte on grain rying by Farkas, Reményi, & Biró (2000a an b an relea, Courtois, & rystram (1997. Kaminski, Strumillo, & omczak (1998 also use an artificial neural network for moeling of moisture content an quality inex for vitamin C in slice potatoes an green peas. On the other han, the genetic algorithm is one of the search methos an optimization techniques for an optimal value of a complex objective function by simulation of the biological evolutionary process base, as in genetics, on crossover an mutation. Morimoto, De Baeremaeker, & Hashimoto (1997a evelope an artificial neural network-genetic algorithm intelligence approach for optimal control of fruit-storage process. Morimoto, Purwanto, Suzuki, & Hashimoto (1997b use genetic algorithm for optimization of heat treatment for fruit uring storage. Hashimoto (1997 introuce applications of artificial neural networks an genetic algorithms to agriculturalsystems. Fuzzy set theory is a theory about vagueness an uncertainty. his theory provies an approximate, an yet effective, means of escribing the behavior of systems that are too complex or ill-efine to permit precise mathematical analysis. Fuzzy controllers were evelope to imitate the performance of human expert operators by encoing their knowlege in the form of linguistic rules. he fuzzy control is also nonlinear an aaptive in nature, which gives ita robustperformance uner parameter variations. Fuzzy control systems provie control through a set of membership functions quantifie from ambiguous terms in control rules. As fuzzy control can be implemente by a small number of rules, it has a short initial evelopment perio. he number of the rules is etermine by require accuracy. 375
After the invention of fuzzy logic by Zaeh, the fuzzy moeling an fuzzy ientification of systems has foun numerous practical applications in control, preiction an inference. In many cases, reucing to esign time an costs the fuzzy logic approach allows the esigner to hanle efficiently very complex close-loop control problems. Fuzzy control also supports nonlinear esign techniques that are now being exploite in motor an temperature control applications. A fuzzy logic base controller ajusts the system input to get a esire output by just looking at the output without any requirement mathematical moel of to be controlle system. For this reason fuzzy logic base controller systems iffer from classical control systems an it is possible to get esire control actions for complex, uncertain, an non-linear systems by using fuzzy logic controller (FLC withouttherequirement oftheir mathematical moelsan parameterestimation. In this stuy, applications of artificial intelligence techniques, such as artificial neural networks, genetic algorithms an fuzzy logic, for foo rying technologies are stuie. Estimation, optimization an control applications of artificial intelligence methos are given in etailfor ifferent types of foo rying applications. Echinacea angustifolia an carrot are selecte as application examples. A fuzzy logic base control approach is employe to control a convective type rier. Estimation an optimization applications of artificial neural networks an genetic algorithms are compare with non-linear regression analysis. In aition, fuzzy control is also compare with a classical control technique to conclue the robustness of the fuzzy control in terms of classicalcontrol. Mathematical Moel of Foo Drying Process he flow of moisture from the agricultural material to its surrounings can be consiere as analogous to the heat transfer from a boy immerse in col flui. Comparing the rying phenomenon with Newton s law of cooling, the rying rate willbe approximately proportional to the ifference in moisture content between the materialbeing rie an equilibrium moisturecontentattherying airstate. Hence: M Drying rate = t+ t t M t (1 Similarly,the moistureratiosof Echinacea an carrotare obtaine from: 376 MR M M e = (2 M 0 M e As propose by earlier authors an given in able 1, the rying curves obtaine were processe for rying rates tofin the mostsuitable moelamong thefour ifferentexpressions(akpinar etal,2003. Moel no: Moel name: Moelequation: 1 Newton MR = exp( kt 2 Page n MR = exp( kt 3 Moifie Page n MR = exp( (kt 4 Henerson an Pabis MR = a.exp( kt able 1: hinlayerrying curve moelsconsiere. he correlation coefficient (r was one of the primary criteria for selecting the best equation to efine the rying curves. In aition to r, the coefficient of etermination (r 2,reuce Chi-Square (χ 2,an sum of squares of the ifference between the ata an fitvalues (SSR were use to etermine the quality of the fit.he best results of the propose criteria were obtaine by using the moifie Page equation (Maamba et al.,1996; Panchariya etal.,2002 asshown in Eq.(3: n MR = exp( (kt (3 he epenence of the rying rate constant,k, an rying parameter, n, on the rying air variables was moele as an Arrhenius-type equation. his epenence of both constants on the variables can be expresse in thefollowing form: a a 1 a 2 3 k = a 0V exp( (4
b b 1 b2 3 n = b 0V exp( (5 Artificial intelligence techniques for foo rying process he selecte structure of the applie neural network, with its four inputs an single output,is shown in Figure 1. here is no feeback from the output to the inputs. Since the physical structure of a thin layer ryer consists of three main parts (the input variables, the rying be itself an the output variables a three layer feeforwar neural network was chosen for moeling purposes (Farkas et al.,2000a. In the hien layer, 30 hien neurons were use for Echinacea an 25 hien neurons were use for carrot. For training, the classical backpropagation algorithm was use (Farkas et al.,2000b for the both cases. In this stuy, a logarithmic sigmoi activationfunction was use. (a (b Figure 1: Neuralnetwork structurefora Echinacea an bcarrot. Higher r, r 2, χ 2 an SSR values were obtaine by using the neural network compare with that of moifie Page moel. he results have shown that the inicators for gooness of fitof the propose neural network moel are better than the values obtaine by the moifie Page moel. hese results are shown in able 2. herefore, the propose neural network moel was selecte to represent the thin layer rying behavior of E. angustifolia because of the higher values of r an r 2, an the lower values of χ 2 an SSR than that by the moifie Page moel. It can be clearly seen from able 2 that the accuracy of the neural network moel provie a better fit an better results. he performance of the neural network moel for E. angustifolia is illustrate in Fig.2a, 2b an 2c for ifferent rying air temperatures, rying air flow rates an root sizes. Detaile information forthiscase coul befoun in(erenturk, Erenturk, & abil,2004c Moel name Moelconstants Correlation coefficient(r Coefficientof etermination (r 2 Newton k=0.004 0.9862 0.9726 1.27E-3 0.348 Page k=0.014 n=0.790 0.9938 0.9876 3.78E-4 0.089 Moifie Page k=0.004 n=0.790 0.9965 0.9930 3.29E-4 0.089 Henerson &Pabis k=0.004 a=0.915 0.9896 0.9793 9.64E-4 0.262 NNE - - 0.9994 0.9989 3.96E-05 0.0109 able 2: Resultsofstatisticalanalyses on the moeling of moisturecontentsan rying time. χ 2 SSR 377
(a (b (c Figure 2: he performance of the neuralnetwork moelfore. angustifoliaforarootsizes brying airflow an cifferentrying airtemperatures. Similiar to the previous case, the rying rate k an the rying parameter n of the moifie Page moel forcarrot were bestescribe by Arrhenius-type moelan shown below: k = 42.66V n = 5.48V 0.3123 0.0846 0.8437 0.1066 2386.6 exp( exp( 452.5 (r=0.987 (r=0.954 Above expressions can be use to estimate the moisture content of carrot at any instant uring rying, because the regression coefficient,r,is foun with acceptable accuracy. he accuracy of the establishe moel was evaluate by comparing the compute moisture ratio uner any particular rying conitions with the observe moistureratio. During both regression routines an etermination of the epenence of the rying rate constant, k, an rying parameter, n, on the rying air variables, the propose GA approach in (Erenturk, Erenturk, 2007 was employe for allexperimentalruns. After the step by step proceure escribe in (Erenturk, Erenturk, 2007, thefollowing relationship between rying variablesan rying characteristics was obtaine. k = 26.64V n = 5.29V 0.4199 0.0856 0.8362 0.1023 2223.1 exp( exp( 443.4 (r=0.996 (r=0.962 Regaring above r, r 2, χ 2 an SSR values etermine by using regression analysis an the mathematical moel optimize by applying GA were liste in able 3. he accuracy of the mathematical moel optimize by using GA was observe moresatisfactorythan thatofregression analysis. Moel name Correlation coefficient(r Coefficientofetermination (r 2 χ 2 SSR Newton 0.9964 0.9928 2.36E-3 2.553 Page 0.9938 0.9876 2.62E-3 2.825 Moifie Page 0.9991 0.9981 2.45E-3 2.698 Henerson &Pabis 0.9976 0.9917 2.42E-3 2.611 M oifie Page before optimization 0.9985 0.9971 2.52E-3 2.725 able 3: Statisticalresultsofthe mathematical moelsoptimize by using GA Another AI technique suitable for rying process is fuzzy logic. Fuzzy controllers were evelope to imitate the performance of human expert operators by encoing their knowlege in the form of linguistic rules. Since the fuzzy control is also nonlinear an aaptive in nature, these properties give FC a robust performance uner parameter variations. Fuzzy control systems provie control through a set of membership functions quantifie from ambiguous terms in control rules. As fuzzy control can be implemente by a small number of rules, ithas a short initial evelopment perio. he number of the rules is etermine by require accuracy. A fuzzy logic base controller ajusts the system input to get a esire output by just looking at the output without any requirement mathematical moel of to be controlle system. For this reason fuzzy logic base controller 378
systems iffer from classical control systems an it is possible to get esire control actions for complex, uncertain, an non-linear systems by using fuzzy logic controller (FLC without the requirement of their mathematical moels an parameter estimation. For this purpose, a fuzzy logic base control approach is employe to control a convective type rier.simulation result is illustrate in Fig. 3. Fuzzy control (FC is also compare with a classical control technique to conclue the robustness of the fuzzy control in terms of classical control. Comparison results are given in able 4. Accoring to the results, it is observe that artificial intelligence techniques have several avantages such as: ecreasing computation time, increasing stability an accuracy. Figure 3: Fuzzy logic base temperature controlofaconvective rier. Controllertype Overshoot(% Risetime (h Steay state error( o C Fuzzy control - 0.28 1.23 PID 1.22 0.61 3.52 able 4: Performance evaluation of controllers. Conclusions In this stuy, applications of artificial intelligence techniques for foo rying processes are presente. In orer to estimate the rying behavior of ifferent type foos, a fee-forwar artificial neural network (ANN structure is esigne an applie to Echinacea an carrot.it is observe that ANN base estimation is more accurate than that of nonlinear regression analysis. In aition, for more complex operation, such as Arrheniustype moeling, GA base optimization technique is applie an more reliable results are observe. Fuzzy logic base control of a convective rier is also given an compare with a classical PID-type control technique to conclue the robustness of the fuzzy control in terms of classical control. Accoring to the results,itis observe that artificial intelligence techniques have several avantages such as: ecreasing computation time, increasing stability an accuracy. Moreover these techniques coul be applicable for ifferent type processes with simple changesinconfiguration. 379
References Akpinar, E. K., Bicer, Y., & Yiliz, C. (2003. hin layer rying of re pepper. Journal of Foo Engineering, 59, 99-104. Diamante, L. M., & Munro, P. A. (1993. Mathematical moelling of the thin layer solar rying of sweet potato slices. Solar Energy, 51, 271 276. Dincer, I. (1996. Sun rying of sultana grapes. Drying echnology, 14, 1827 1838. Erenturk, K., Erenturk, S., & abil, L. G. (2004c. A comparative stuy for the estimation of ynamical rying behavior of Echinacea angustifolia: regression analysis an neural network. Computers an Electronics in Agriculture, 45, 71 90. Erenturk, S., & Erenturk, K. (2007. Comparison of Genetic Algorithm an Neural Network Approaches for the Drying Process of Carrot. Journal of Foo Engineering, 78 (3, 905-912. Erenturk, S., Gulaboglu, M. S., & Gultekin, S. (2004a. he hin-layer Drying Characteristics of Rosehip. Biosystems Engineering, 89, 159-166. Erenturk, S., Gulaboglu, M. S., & Gultekin, S. (2004b. he effects of cutting an rying meium on the vitamin C content of rosehip uring rying. Journal of Foo Engineering, 68 (4, 513-518. Farkas, I., Reményi, P., & Biró, A. (2000a. A neural network topology for moelling grain rying. Computers an Electronics in Agriculture, 26, 147 158. Farkas, I., Reményi, P., & Biró, A. (2000b. Moelling aspects of grain rying with a neural network. Computers an Electronics in Agriculture, 29, 99 113. Hashimoto, Y. (1997. Applications of artificial neural networks an genetic algorithms to agricultural systems. Computers an Electronics in Agriculture, 18, 71-72. Kaminski, W., Strumillo, P., omczak, E. (1998. Neurocomputing approaches to moelling of rying process ynamics. Drying echnology, 16, 967 992. Maamba, P. S., Driscoll, R. H., & Buckle, K. A. (1996. he thin layer rying characteristic of garlic slices. Journal of Foo Engineering, 29, 75-97. Maskan, M. (2001. Drying,shrinkage an rehyration characteristic of kiwifruits uring hot air an microwave rying. Journal of Foo Engineering, 48, 177 182. Miilli, A. (2001. Determination of pistachio rying behavior an conitions in a solar rying system. International Journal of Energy Research, 25, 715 725. Morimoto,., De Baeremaeker, J., & Hashimoto, Y. (1997a. An intelligent approach for optimal control of fruit-storage process using neural networks an genetic algorithms. Computers an Electronics in Agriculture, 18, 205-224. Morimoto,., Purwanto, W., Suzuki, J., & Hashimoto, Y. (1997b. Optimization of heat treatment for fruit uring storage using neural networks an genetic algorithms. Computers an Electronics in Agriculture, 19, 87-101. Panchariya, P. C., Popovic, D., & Sharma, A. L. (2002. hin-layer moelling of black tea rying process. Journal of Foo Engineering, 52, 349-357. Sarsavaia, P. N., Sawhney, R. L., Pangavhane, D. R., & Singh, S. P. (1999. Drying behaviour of brine onion slices. Journal of Foo Engineering, 40, 219-226. relea, I. C., Courtois, F., & rystram, G. (1997. Dynamic moels for rying an wet-milling quality egraation of corn using neural networks. Drying echnology, 15, 1095 1102. 380