An Economical Method for Artificial Neural Network Process Modeling by the Model-Modifier Approach

Abstract

In this paper we present our model-modifier approach as an economical method for the development of accurate manufacturing equipment models. The model modifier method leverages knowledge from one ANN model to another of a similar type, thus reducing the development effort required as compared to starting from scratch. The economy afforded by this knowledge-sharing technique was evaluated on a Chemical Vapor Deposition (CVD) reactor. The results show that the model-modifier approach is a valid method for transferring knowledge between similar ANN models and that significant savings in training data accrue from this approach. In our case, a highly accurate ANN model was developed with a mere one-fifth of the data that would have been required without this approach. Further, we have also shown that an ANN model developed by the model-modifier approach can be easily and reliably utilized for process optimization.