NEURAL NETWORKS WIN EXTRUSION PROCESS IDENTIFICATION AND CONTROL

Although neural networks have become one of the key research objects w ithin artifical intelligence, relatively little information is availab le on neural networks related to food process control. The interest in such areas as dynamic modelling of food processes as increased, not l east due to dramatic improvement and availability of the calculation m ethods and hardware. In the present case, flat bread extrusion was use d as an example food process. Dynamic changes of torque, specific mech anical energy (SME) and pressure were identified (modelled) and contro l ed using two independently taught feed-forward artificial neural net works (ANN). SME, torque and pressure are system parameters which can be controlled with process parameters, such as feed moisture, mass fee d rate and screw speed. Target parameters, such as product expansion i ndex, bulk density, etc. are normally difficult to measure on-line, bu t can be estimated as functions the system parameters. For the modelli ng of the whole flat bread extrusion cooking process a MIMO (multi inp ut and multi output) approach was necessary. The neural network topolo gy for the process model was 21-9-3 and for the controller 18-20-2. Th e process model was taught with 629 real data samples and the controll er with 115 synthetic samples created with the process model. When tes ting the MIMO controller, the SME and pressure set points were quite w ell reached. One of the clear advantages of neural networks in the con troller design is the ease of constructing a complex MIMO controller.