شبيه سازي فرآيند خشك شدن شلتوك با استفاده از شبكه هاي عصبي مصنوعي

چكيده انگليسي :

Simulation of Rough Rice Drying using Artificial Neural Networks Mojtaba Tohidi Mojtaba thd@yahoo com March 16 2010 Department of Agricultural Machinery Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Morteza sadeghi sadeghimor@cc iut ac ir Abstract Rice Oriza sativa L as an important world food is one of the most valuable crops with a long history of cultivation Drying operation is used to prevent deterioration of such a strategic product Since the quality of product is directly affected by the drying operation controlling of this process is vitally important Common methods to study the affecting factors on the drying process of agricultural products are statistical and mathematical models which mostly encounter several simplifications influencing the accuracy of the model These methods hand in several differential and or algebraically equations which must be solved and interpreted However if the problem involves several input and output variables as in drying process these methods are very complex to use Nowadays artificial intelligence technology as a success of rapid computer technology introduces artificial neural networks ANN for solving models for systems and processes problems An ANN is a set of computational elements which are connected to each other like biological neurons and is able to communicate information without any previous knowledge of their relations In this study the ANN and multivariate regression methods were used to predict some affecting parameters on rough rice drying operation in a deep bed mode The input parameters for these models were inlet air velocity 0 5 0 8 and 1 1 m s 1 inlet air temperature 40 50 60 70 and 80 C and inlet air relative humidity 40 50 60 and 70 In addition to drying kinetic three dependent variables including product output rate as the dryer capacity index evaporation rate as the drying kinetic quality and kernel cracking percentage as the dried product quality were investigated To create neural networks training test and evaluation patterns drying experiments were performed under different conditions by using a laboratory dryer Then the results of the experiments were used in the design of neural networks About 70 percent of the experiments were used for training the network and the rest for its test and evaluation To predict the dependent parameters three well known networks namely multi layer perceptron MLP generalized feed forward GFF and modular neural network MNN were examined Drying kinetic was predicted using a network with four mentioned inputs Other three parameters were firstly predicted with three separate ANNs and then a single ANN was designed to predict all these three parameters simultaneously Sensitivity analysis process gives valuable information about the model sensitivity to the input variables sensitivity analysis was done according to Hill s method Results showed that the separate networks were more accurate than the latter network The best results for drying kinetics was obtained by the GFF network with topology of 4 15 1 learning algorithm of Levenberg Marquardt LM and active function of hyperbolic tangent with normal mean square error NMSE value of 0 000792 and correlation coefficient of 0 996 The most accurate network for predicting product output rate evaporation rate and kernel cracking percentage was GFF network with topology of 3 4 4 1 and NMSE of 0 001325 topology of 3 7 1 and NMSE of 0 001554 and topology of 3 11 1 and NMSE of 0 001 251 respectively The sensitivity analysis results revealed that the inlet air temperature and relative humidity had the maximum and minimum effect on all output variables respectively However three input parameters directly affect the output variables sensitivity coefficient 0 1 Comparison of results showed the ANN method could predict the dependent variables more accurate than the fitted regression relation Therefore it can be concluded that ANN is a simple and fast

توحيدي، مجتبي

شبيه سازي فرآيند خشك شدن شلتوك با استفاده از شبكه هاي عصبي مصنوعي

سينتيك خشك شدن , فراسنجه هاي فرآيند و محصول