Modeling of apricot weight loss during drying with infrared dryer using genetic algorithm-artificial neural network optimization methods

One of the oldest methods for the preservation of food is drying, which consists in removing water from the product in order to provide microbiological safety and the most popular drying method includes convection (Hassan-Beygi 2009; Salehi et al. 2014). One of the ways to shorten the drying time is to supply heat by infrared radiation (IR). Application of infrared heating to food drying is recently of special interest because of the progress in radiator construction. Their efficiency is between 80% and 90%, the emitted radiation is in narrow wavelength range and they are miniaturized (Pan et al. 2008; Salehi et al. 2014). Modeling of agricultural products drying is one of the best ways to controlling the drying time and conditions. Artificial neural networks (ANNs) are information processing networks constituting a set of highly interconnected neurons arranged in multiple layers that can be trained to fit one or more dependent variables to any degree of accuracy using a set of independent variables as inputs. ANNs were successfully applied to problems from various areas including business, medical and industrial fields. Once the ANN is trained using experimental data, it can be used in a purely predictive mode to calculate the dependent variable(s) for any values of input variables. Process modeling is an area where ANNs of various configurations and structures have been considered as alternative modeling techniques, particularly in cases where reliable mechanistic models cannot be obtained. Genetic algorithm (GA) optimization technique can be used to overcome this inherent limitation of ANN. GA are search techniques for an optimal value, mimicking the mechanism of biological evolution. They have a high ability to find an optimal value (global optimal value or at least near global one) of a complex objective function, without falling into local optima. The mathematical chromosomes could be operated upon by quasi genetic operations of selection, crossover, and mutation. These three parameters are repeated until desired convergence on optimal or near-optimal of the solutions is achieved. Mutation is a genetic operator that alters one or more gene values in a chromosome from its initial state. Mutation enhances the GA ability by intermittently injecting a random point in order to better search the entire parameter space, which allows the GA to possibly escape from local optima. This can result in entirely new gene values being added to the gene pool. With these new gene values, the GA may be able to arrive at a better solution than was previously possible (Ramzi et al. 2015; Salehi and Razavi 2016). Fresh apricots were obtained from market. Slices of apricot with 5 mm thickness were prepared with the aid of a steel cutter and were immediately placed into the dryer. The apricot slices were dried in an infrared dryer. In this study, for drying and increasing the shelf life of apricot, infrared radiation method was used. The effect of infrared lamp power at three levels 150, 250 and 375 watts, the distance of lamp from sample at three levels 5, 7.5 and 10 cm and time of 160 minute on drying of apricot were examined. Modeling of process was done with the genetic algorithm–artificial neural network (GA-ANN) method with 3 inputs (lamp power, distance and time) and 1 output (weight loss). During drying, radiation properties of the material are changing due to decreasing water content. As a consequence, its reflectivity increases and the absorptivity decrease. Generally, solid materials absorb infrared radiation in a thin surface layer (Pan et al 2008). The effects of infrared power and distance on the moisture content of apricot slices are shown in Figures 3 and 5. As expected, the moisture content was decreased by increasing the power because of the
increased temperature and heat transfer gradient between the air and samples. The weight loss of apricot samples were 36.34, 52.00 and 87.90 min at 150, 250 and 375 W, respectively (32 min and 10 cm). In conclusion, experimental results showed that the infrared power has a significant effect on the evolution of moisture content. Hence, drying of thin layers seems to be more efficient at far-infrared radiation, while drying of thicker bodies should give better results at near-infrared radiation (Nowak and Lewicki, 2004). The weight loss reduced from 87.43 to 78.26 % when the distance was increased from 5 to 10 cm (150 W and 98 min).
GA-ANN model was developed for modeling of apricot weight loss during drying with infrared dryer. In this study, ANN with 1–30 neurons was trained using GA to find the optimal network configuration. It was found that ANN with 7 neurons in one hidden layer could predict apricot weight loss with high correlation coefficient and low NMSE (0.9987, 0.0026, respectively). The prediction efficiency of the GA-ANN model for unseen data is presented in Fig. 7. The calculated correlation coefficient value for estimation of apricot weight loss show high correlation between predicted and experimental values. Table 3 illustrates the weights and bias values of optimized network, which could be applied in a computer program for estimation of apricot weight loss during drying with infrared dryer. The results showed that an acceptable agreement between the predicted and experimental data can be achieved using GA–ANN model. The results of infrared drying of apricot showed that with increasing in lamp power and decreasing in sample distance from the heat source, the drying rate was increased. The GA-ANN modeling results showed a network with 7 neurons in 1 hidden layer with using sigmoid activation function can be predicting the weight loss in apricot drying by infrared method with correlation coefficient equal to 0.9987 and mean squared error equal to 1.9215. Sensitivity analysis results by optimum ANN showed the drying time was the most sensitive factor to control the weight loss of apricot slides.