Günter Sörgel

Neural Network Control for Steel Rolling Mills

Worldwide, steel and aluminum production and manufacturing is still one of the major basic industries with a huge amount of material and energy consumption. Hence, optimization of the various process control schemes which are involved can lead to significant savings. Artificial Neural Networks are a new information processing technique which provides a novel approach to process control problems and promises major improvements. Therefore, Siemens together with FORWISS has been studying and developing neural control schemes for a number of different process control problems which occur at hot line rolling mills (Lindhoff et al., 1994). In this paper we give a brief survey of the different control aspects which were tackled with this new approach and comment on their current status.

Microstructure monitor and process simulation

Abstract The knowledge of microstructure is essential to compute mechanical properties of hot rolled steel. Depending on process data and the chemical composition of the steel, a physical model is used to calculate microstructure parameters such as grain size, recrystallization and the γ-α-transformation. In addition to the physical model an artificial neural network is employed to calculate mechanical properties. New results from hot rolling mills are presented. The on-line application of the monitor system allows a quick release of strips after rolling and the saving of costs for quality control measurements. Process parameters could be adjusted to achieve desired mechanical properties of the product. Process data can either be obtained on-line from the process itself or can be simulated off-line, e.g. by Siemens’ HYBRid EXpert System HYBREX. Its central component is a high-performance flowsheet simulator, which enables complex plants to be modeled, simulated and evaluated. The architecture and capability of HYBREX will be presented as well as the newly achieved possibilities by integrating the microstructure model to the simulation tool.