Hybrid artificial artificial neural network for prediction and control of process variables in food extrusion

Two hybrid Artificial Neural Network (ANN) chains have been implemented and tested for configuration and control of a food extrusion process. The first architecture (Configuration Chain) predicts the process variables screw speed N and water content w starting from product characteristics. It calculates also energy and flow variables as Specific Mechanical Energy, shaft torque, shear stress, pressure and temperature variations. The second architecture (Control Chain) is a closed loop ANN-chain in which a differential neurocontroller regulates the extrusion process by calculating adjustments of N and w. Both chains are feedforward multilayer ANNs trained using the Back Propagation Algorithm. The ANNs in each chain are built using the programme MemBrain and are triggered and automated by an AngelScript code. A total of 42 patterns have been used (24 for training and 18 for verification). For both ANN-chains the quality of each ANN is presented as well as proof of concept of the whole chain. Industrial relevance: The present study is intended to offer a method to efficiently optimise in an innovative way the process of extrusion in the food industy. In particular, the research aims to develop two Artificial Neural Network chains for (1) prediction of optimal process variables (screw speed and water content are considered) starting from desired product characteristics and (2) prevent process instabilities on the predicted variables by using a neurocontroller. The advantages for the industry are the opportunity to select the best machine variables for each reasonable desired product and the possibility to stabilise and regulate in a connectionistic way a multivariable system like the extrusion process which is difficult to be regulated in a deterministic manner. (C) 2013 Elsevier Ltd. All rights reserved.