Process optimization and control are largely motivated by economic incentives. This is especially the case when large volumes are involved in production. Small improvements in the process can result in large gains. Also, in the competitive market situation, continuous improvement is necessary to be able to maintain and improve the market position.
How can a complex process be optimized? It is necessary to understand the functioning of the process before one can proceed to optimize it. Modeling a given process aims at understanding the functioning the process and the relationships between the process variables. How should the system be optimized? How can one avoid causing negative side effects in one part of the process (and eventually global loss) while optimizing another, in other words, how to avoid suboptimizing? These are some of the questions one is faced with when trying to optimize a process.
In this study, methods are presented with which one can learn about the process characteristics with the aid of a model created by a neural network from process measurements. Data of the incoming raw material, process parameter settings and end product characteristics from individual products are used for building a non-linear regression model of the measurement data.
In particular, an artificial neural network called the Self-Organizing Map is used. Neural networks have been quite promising in complex application areas where traditional methods have failed. Due to their inherently non-linear nature, they can handle much more complex situations than the traditional methods.
The methods presented in this work give a good starting point to a process modeling and optimization effort.