Integrating Neural Network and Symbolic Inference for Predictions in Food Extrusion Process

Abstract

Predicting process outputs in a food extrusion process is a difficult task due to multiple variables and their highly nonlinear relationship. Experimental data have been collected by earlier researchers to fit statistical models to identify process conditions that result in the “best” output. A neural network is developed and trained with experimental data to capture the process knowledge, and map the relationship between process variables and process output. An expert system is developed that uses the neural network as an inference engine component to make exact predictions. It also has a knowledge base that contains a set of symbolic rules. This allows the system to provide a more comprehensible form of prediction that helps engineers gain a better understanding of the problem dynamics.