The flow structure in the mold of a continuous steel caster has a significant impact on the quality of the final product. Conventional sensors used in industry are limited to measuring single variables such as the mold level. These measurements give very indirect information about the flow structure. For this reason, designing control loops to optimize the flow is a huge challenge. A solution for this is to apply non-invasive sensors such as tomographic sensors that are able to visualize the flow structure in the opaque liquid metal and obtain information about the flow structure in the mold. In this paper, ultrasound Doppler velocimetry (UDV) is used to obtain key features of the flow. The preprocessing of the UDV data and feature extraction techniques are described in detail. The extracted flow features are used as the basis for real time feedback control. The model predictive control (MPC) technique is applied, and the results show that the controller is able to achieve optimum flow structures in the mold. The two main actuators that are used by the controller are the electromagnetic brake and the stopper rod. The experiments included in this study were obtained from a laboratory model of a continuous caster located at the Helmholtz-Zentrum Dresden Rossendorf (HZDR).
Keywords: industrial control; industrial process tomography; model predictive control; ultrasound doppler velocimetry.