A decision support system using neural networks in a glass furnace process

A decision support system using artificial neural networks is implemented with real world data of a glass furnace process at Samsung. It provides the functions such as process model identification, set-point control and interpreting input factors. Since a glass furnace process is highly complex, a traditional attempt to develop a model from first principles often proves to be a difficult and costly procedure. However, the decision support system using artificial neural networks does not require a priori knowledge of a glass furnace process and proves to be useful in identifying the model directly by input/output data collected from the plant. This paper shows the method of finding the partial derivative value at some point from trained weights, the conversion method of a 3-layered perceptron network into a 2-layered one, and the interpretation method of neural networks solutions.