Modeling moisture absorption kinetics of barley grain using viscoelastic model and neural networks

ماشین های کشاورزی

Article 5, Volume 5, Issue 2 - Serial Number 10, 2015, Pages 270-280   

Document Type: Research Article

DOI: 10.22067/jam.v5i2.28305

Authors

M. Kamali; S. J. Razavi^* ; M. Sadeghi; S. M. Shafaei

Biosystems Engineering Department, College of Agriculture, Isfahan University of Technology, Isfahan, Iran

Abstract

Introduction: Barley is one of the most important grains with high digestible starch making it a main source of energy in human nutrition as well as in livestock rations formulation and feeding. Starch is the main part of barley grain and it has an inverse relation with its protein. It has a digestible foodstuff of 80 to 84 percent of its dry matter content. Barley as livestock foodstuff should be processed and it is done in several ways. A customary method for processing barley in dairy farms is its size reduction by milling (Hunt, 1996). An alternative method of barley processing is steam rolling. However, because of the high cost of steam generators a method of soaking with heating has been considered as an alternative method for steam rolling (Yang et al., 2000). The rate of moisture absorption by grains during the soaking process varies considerably and depends on the size of the grain, water temperature and the length of soaking. High temperature water soaking is an ordinary way to reduce the time duration for reaching a high rate of moisture absorption during the soaking process (Kashaninejad et al., 2009). Various studies have shown that these models have adequate accuracy in analyzing drying and moisture absorption processes for most agricultural products (Abu-Ghannam and McKenna, 1997). Some researchers have modeled beans moisture absorption behavior using 14 mathematical models and found that the Weibull model had the most conformity with variations in experimental data (Shafaei and Masoumi, 2014c). Observations made by researchers indicate that the moisture absorption process in various materials encompasses a primary phase with a fast rate and a second phase with a lower rate. The second phase in moisture absorption is called the relaxation phase. The main problem with all the mathematical and experimental models is the lack of the model’s ability to evaluate the rate of moisture absorption in the secondary phase. Artificial Neural Network (ANN) as an important artificial intelligent method comparable to human brain capabilities is applied to train and store data in the form of weighted networks (Dayhoff, 1990). This method has superiority to many ordinary statistical and model making methods. In comparison to linear regression models, ANN does not require placing estimated values around mean values and for this reason it retains actual variations in the data being analyzed. Prediction by using trained ANN enables the researchers to decrease or increase input and output variables.Therefore, it is possible to produce a multivariate model with an output even more than the objectives deemed necessary (Heristev, 1998). The goal of this research was to predict instant moisture content of three barley varieties (Reyhan3, Fajr and MB862) during the soaking process under three temperature levels (10, 20 and 45 ◦C) using two conventional ANN methods of multilayer perceptron (MLP) and radial basis function (RBF) in comparison with viscoelastic mathematical model and reporting the results.
Materials and method: Barley varieties were collected from the Isfahan Province Agriculture Organization grain depository and were cleaned and the debris were separated before the experiments. The selected grains were sorted to three groups of small, medium and large grains sizes. To exclude the effect of grain size during moisture absorption, the medium size grains were used. The moisture content of the grains was determined based on the ASAE S352.2 DEC97 (ASAE, 1999) which were %8.23, %8.62 and %8.89 on a dry basis for Reyhan3, Fajr and MB862, respectively with no significant difference at %5 probability level (p>0.05). Experiments were conducted under three temperatures (10, 20 and 45 ◦C) in the refrigerator, at room temperature and in the oven, respectively for each variety. In each experiment, 10 medium size grains were selected randomly and weighed with an AND laboratory scale model Gf-400 (made in Japan) and placed in foam containers having 200 mg of distilled water. Grains were weighed after a predetermined period of elapsed time (5, 10, 15, 30, 60, 120, etc. minutes). The experiments were conducted with three replications and moisture absorption rates were determined by the equations presented by McWatters et al., 2002. The experiments were conducted on a time table based on which the time for the dissolving of grains was reached. In this case, the moisture content of the grains reaches the saturation point. According to equations presented by Peleg, as water density increases as much as 0.01 gram due to grains dissolving in water, the saturation point has been reached (Peleg, 1988). For this reason, distilled water density was measured and controlled before and after each experiment by a pycnometer. Neural network was designed according to the two methods of multi-layer perceptron (MLP) and radial basis function (RBF) with three neuron layers. The first layer, i.e. input layer, is independent variables of temperature and time.The second layer, i.e. hidden layers, is the networks hidden layer and the third layer, i.e. output layer, is the dependent variable of moisture content which was selected. In each case, the nonlinear reduced gradient, combined gradient and BFGS algorithm, and Trigonometric, Logarithmic, Gaussian, and Logical functions were used to train, test and evaluate the network. To evaluate the predicting viscoelastic model and the network, we used statistical indices maximum value of coefficient of determination (R2) and minimum value of mean square error (RMSE).
Results and Discussion: Moisture absorption curves showed that as the temperature increases, moisture absorption rate increases as well. Higher equilibrium moisture levels are obtained in water with higher temperatures. This phenomenon is the result of increased moisture diffusion in grains due to higher temperature levels. Higher water temperatures causes grain internal material which is mainly starch to gelatinize and, thus, the internal tissues resistance to moisture absorption reduces (Ranjbari et al., 2011). The moisture absorption rate increases as immersion temperature and gelatinization temperature reach closer to each other.
Conclusions: The results of this study showed that although viscoelastic mathematical model has an adequate accuracy for instant prediction of barley grain moisture content, it has a lower accuracy compared to intelligent models. On the other hand, among the two neural network methods, MLP method has a higher accuracy in predicting moisture content compared to RBF method. MLP obtained the best results for three varieties of barley because of back- propagation learning algorithm with BFGS algorithm and 2-4-1 network structure. According to the prediction of the best neural network which was selected, three-dimensional graphs of moisture content based on temperature and time variables, showed that with an increase in temperature and duration of immersion, moisture absorption increases for three varieties of barley.

Keywords

Immersion; Moisture content; Multi-layer perceptron; Radial basis function; Training algorithm

References

1. Abu-Ghannam, N., and B. Mckenna. 1997. Hydration kinetics of kidney beans (Phaseolus vulgaris L.). Journal of Food Science 62 (3): 520-523.

2. ASAE. 1999. ASAE Standards, 46th ed. 1999. S352.2: Moisture measurement - unground grain and seeds. St. Joseph, Mich.: ASAE.

3. Dayhoff, J. E. 1990. Neural Network Principles. Prentice-Hall International, U.S.A.

4. Fletcher, R. 1987. Practical methods of optimization (2nd ed). New York: Wiley.

5. Heristev, R. M. 1998. The ANN Book. GNU Public License. USA.

6. Hunt, C. W. 1996. Factors affecting the feeding quality of barley for ruminants. Animal Feed Science and Technology 62 (1): 37-48.

7. Kashaninejad, M., A. M. Dehghani, and M. Khashiri. 2009. Modeling of wheat soaking using two artificial neural networks (MLP and RBF). Journal of Food Engineering 91 (4): 602-607.

8. Khanna, T. 1990. Foundation of neural networks. Addison-Wesley Publishing Company, U.S.A.

9. Khazaei, J., and S. Daneshmandi. 2007. Modeling of thin-layer drying kinetics of sesame seeds: mathematical and neural networks modeling. International Agrophysics 21 (4): 335-348.

10. Khazaei, J., and N. Mohammadi. 2009. Effect of temperature on hydration kinetics of sesame seeds (sesamum indicum L.). Journal of Food Engineering 91 (4): 542-552.

11. Maskan, M. 2002. Effect of processing on hydration kinetics of three wheat products of the same variety. Journal of Food Engineering 52 (4): 337-341.

12. McWatters, K. H., M. S. Chinnan, R. D. Phillips, L. R. Beuchat, L. B. Reid, and R. M. Mensa-Wilmot. 2002. Functional, nutritional, mycological and akara-making properties of stored cowpea meal. Journal of Food Science 67 (6): 2229-2234.

13. Mohsenin, N. N. 1986. Physical properties of plant and animal materials. 2nd ed. Gordon Breach Science, Publ; New York.

14. Peleg, M. 1988. An empirical model for the description of moisture sorption curves. Journal of Food Science 53 (4): 1216-1217.

15. Ranjbari, A., M. Kashaninejad, M. Alami, and M. Khomeiri. 2011. Effect of ultrasound pretreatment on water absorption characteristic of pea during steeping process. Electronic Journal of Food Processing and Preservation 2 (1): 91-105. (In Farsi).

16. Sayar, S., M. Turhan, and S. Gunasekaran. 2001. Analysis of chickpea soaking by simultaneous water transfer and water-starch reaction. Journal of Food Engineering 50 (2): 91-98.

17. Shafaei, S. M., and A. A. Masoumi. 2014a. Estimating moisture absorption kinetics of beans during soaking using mathematical models. Agricultural Engineering International: CIGR Journal 16 (3): 230-237.

18. Shafaei, S. M., and A. A. Masoumi. 2014b. Studying and modeling of hydration kinetics in chickpea seeds (Cicer arietinum L.). Agricultural Communications 2 (2): 15-21.

19. Shafaei, S. M., and A. A. Masoumi. 2014c. Use of viscoelastic model to describe water absorption of beans during soaking. Acta Horticulture (ISHS) 1054 (1): 191-196.

20. Shafaei, S. M., A. A. Masoumi, and H. Roshan. 2014. Analysis of water absorption of bean and chickpea during soaking using Peleg model. Journal of the Saudi Society of Agricultural Sciences. in Press.

21. Sopade, P. A., and J. A. Obekpa. 1990. Modeling water absorption in soybean, cowpea and peanuts at three temperatures using Peleg’s equation. Journal of Food Science 55 (4): 1084-1087.

22. Toma, M., M. Vinatoru, L. Paniwnyk, and T. J. On. 2001. Investigation of the effects of ultrasound on vegetal tissues during solvent extraction. Ultrasonics Sonochemistry 8 (2): 137-142.

23. Turhan, M., S. Sayar, and S. Gunasekaran. 2002. Application of Peleg model to study water absorption in chickpea during soaking. Journal of Food Engineering 53 (2): 153-159.

24. Yang, W. Z., K. A. Beauchemin, and L. M. Roda. 2000. Effect of barley grain processing on extent of digestion and milk production of lactating cows. Journal of Dairy Science 83 (3): 554-568.