Employing Information Hidden in Industrial Process Data

The paper refers to an intermediate stage of the running international project ProDaCTool which aims to develop the decision-support tool for operators of complex industrial processes. It should help to guarantee stable highest quality of product by optimal settings of machinery. The main idea of the project is to extract valuable information from a huge amounts of process data and to store it in the form of multi-dimensional mixtures of probability density functions. Resulting mixtures will be marked as \good" or \bad" according to selected quality criteria. Then, the system will advise the operator to optimize trajectory of the working point relative to distance from \good" parts of the data space. Solution of the whole task has been based on Bayesian approach. A cold rolling mill was chosen for experiments and nal implementation of the advisory system.