Case Studies on Modeling Manufacturing Processes Using Artificial Neural Networks

Abstract

The modeling capability of an artificial neural network is studied through three different manufacturing processes. The first case study is a linear separable pattern classification problem in manufacturing process diagnosis. The performance between the neural network and the probability voting classifier is compared. The second case study uses a design of experiment to study an SMC compression molding process. Modeling and predicting performances between a regression model and a neural network model are compared in linear as well as nonlinear cases. The third case study investigates correlation models between the operating conditions and product quality defects of an automotive painting process. Results from a neural network model are compared with those of a probability voting classifier. An ad hoc modification named focused learning paradigm on the back-propagation algorithm is also introduced to speed up network learning.