The Direct Sheet Plant located at Tata Steel IJmuiden combines continuous casting of steel slabs from liquid steel and rolling of steel sheets from steel slabs into one process. At the caster of the Direct Sheet Plant, steel slabs are casted with one third of the thickness compared with conventional casters. Therefore its casting speed has to be increased to stay gainful. But this leads to a drawback of this technique; the surface level of liquid steel in the mould becomes more sensitive to fluctuations. The PID mould level controller, delivered at commissioning performed unsatisfactory and the choice was made to design a model based mould level controller. The structure of the used model of the casting process is based on physical laws. Its parameters are estimated from experiments on the on scale water model of the caster.
In this work, system identification techniques are used to improve the mathematical model of the casting process. For this, again the water model of the casting process is used. Both the casting process and its water model are open loop unstable. Therefore closed loop system identification methods are used. With in mind a future implementation of a system identification experiment on the DSP caster, the experiment on the water model is designed using the least costly identification approach.
The admissible uncertainty the identified models should have are taken into the experiment design. Using a LMI-based optimization, the costs of the identification experiments are minimized while the uncertainty of the identified model is kept below the admissible uncertainty. To do the calculations, an initial estimate of the system and the relation which describe the variance of the estimate as a function of excitation power are used.
Results show that the least costly identification approach proved to work. The uncertainty of the identified models from least costly experiments is approximately the same as it was designed for. The identified models are also a good (improved) representation of the water models dynamics. These identified models of the water model dynamics can be used as a representation of the DSP caster dynamics. Or using a similar approach as in this work, a least costly identification experiment can be design at the DSP. In this way an online (while operating and producing) system identification experiments can be designed at the DSP.