Optimizing screw profiles for twin-screw food extrusion processing through genetic algorithms and neural networks

Abstract

Abstract Screw profile design is crucial to the amount of shear and mechanical energy that is imparted on the material being extruded. A genetic algorithm model, in combination with a neural network fitness function, was developed to predict screw profile design. Model was then used to predict the necessary screw profiles along with the necessary process conditions needed for different target products. Predicted screw profiles and extrusion conditions produced expected values of pressure, motor torque, specific mechanical energy (SME), expansion ratio (ER), water absorption (WAI), and water solubility (WSI). Neural network models displayed high R2 values (>0.979) for the process responses of pressure, motor torque, and SME, and slightly lower R2 values for product responses of ER (0.935) WSI (0.900), and WAI (0.847). This demonstrates the possibility for quick predictions of optimum screw profile designs for achieving desired characteristics in the final extrudates.