Nonlinear model of a rice drying process using neural networks

Background: The production quality of rice is highly depended on the drying process as drying weakens the rice kernel. A look at the production process of rice in the industry was taken. The drying of rice influences the storage capacity of the grain, the energy consumption, the final mass of the grain and the percentage of whole grains at the end of the process. Objective: The main objective was to analyse the drying of rice by making an artificial neural network to model and simulate it. Methods: The modeling of a rice drying process using neural networks was presented. These models are suitable to be used in combination with model-based control strategies in order to improve the drying process. The implementation, preprocessing and data retrieval for the design of an artificial neural system was analyzed. Controlling the drying factors is of major importance. Feedforward and dynamic neural networks were compared based on their performance. Results: It was concluded that when some part of the dataset is given as training, even with one dataset, a back-propagation network simulates very well the other parts of the drying curve. It can be said that the approximations done by the networks to obtain a nonlinear model of the rice drying process are quiet good. Conclusions: Firstly, because of the too little data available for training, the networks were not as good as expected. More data is needed to realy have a powerfull network capable of approximated very well the drying curve. Secondly, a backpropagation network can be a good solution for modelling and for use in a controller if more data is available, in contrast a linear network gave bad results. thirdly, a network with little number of layers is the best option. A perfect mapping from the input to the output is impossible due the differences in each test and the imperfect sensors.