73C-28 |
Dynamic modelling of drum drying using neural network-diferential hybrid model. |
G. C. RODRIGUEZ1, S. Estrada, M. A. Garcia, and M. A. Salgado. (1) Food Engineering Dept., Instituto Tecnológico de Veracruz, PO Box 1420, Veracruz , Ver., 91860, Mexico Control of the drying process for food products is a very complex task due to the variability of the raw materials and the common disturbances of the system. Development of new control algorithms can help to keep a constant end quality of food products. However, the development of these algorithms requires a mathematical representation of process dynamics that takes non-linearities into account. The aim of this work was to identify a dynamic model for a drum dryer using neural networks and a system of linear first order differential equations. The model obtained was a combination of neural network and a system of differential equations. The network identified (two hidden layer with three neurons in each layer) considered as input variables: drum dryer speed (Vrc) and steam pressure (Pv); and as output variables the gain (K) and time constants (t), which are the dynamic parameters of the studied system. The identified architecture was coupled a system of differential equations considering variations in the initial moisture (Xo). The efficiency of this hybrid model and a linear differential model were compared with experimental data obtained for milk drying. The hybrid model represented adequately the evolution of final temperature of the dry product with a smaller error than that of the differential model. The hybrid model was coupled to a classic PI control, and different control actions as time and perturbations in Xo were tested. The simulation results were validated with experimental data. The results obtained showed that a PI can absorb strong variations in Xo and short action times (5 y 10 s) and still keep the stability of the system. The simulations results showed that a hybrid model was able to represent suitably the non–lineal behavior of a drum dryer and can be used for testing different control strategies.
Session 73C, Food Engineering: Transport Processes and Kinetics
|