Control of a Hydroforming Press with Bayesian Networks

The demands to automatic control for industrial plants are growing due to an increased complexity of the manufacturing processes. To face these challenges, intelligent control is getting more and more important. For example, neural networks and fuzzy logic are regularly used. The usage of Bayesian networks is seldom mentioned even if many training algorithms are available and Bayesian networks are also able to act under real-time conditions. That means that main preconditions for a self-adaptive controller are given. This paper explains how a Bayesian network is employed as a controller. The main idea is to use the desired value as if it were already observed and to use marginalization for the calculation of the input. This principle is successfully applied to the control of a hydroforming press. As a result the process characteristics in terms of an uniform blank draw-in and the preforming pressure are improved.