Simulation of turbulent flow and particle transport in the continuous casting of steel

The quality of continuous cast steel is greatly affected by fluid flow in the mold region, especially involving transient phenomena. Mathematical models are being applied to investigate many different aspects of these phenomena, but their accuracy must be validated before they can be applied with confidence. As part of a long-term effort to develop and apply comprehensive models of the continuous casting process, this work evaluates the relative accuracy of models of three different fluid flow phenomena in continuous casting through comparison with measurements. Firstly, transient flow simulations of velocities in the mold region are compared with digital particle image velocimetry (PIV) measurements in a single phase water model. Large-eddy simulations (LES) are found to reasonably match the flow measurements, including transient flow variations, except at long time scales, which could not be modeled owing to the excessive computation costs. The standard K-E model produced very good agreement with time-averaged velocities for relatively little computation time, although it is inaccurate at predicting the transient variations. Secondly, particle trajectory calculations are compared with water model measurements to study the distribution and flotation removal of inclusion particles. The LES model was able to match the measurements both qualitatively and quantitatively. Thirdly, steady, multiphase flow computations are compared with flow patterns observed in both a water model and an operating steel caster with argon gas injection. For the same conditions, the water model and steel caster produced very different flow behavior. The computational model was able to match the measured flow patterns in both cases. This work suggests that computational flow modeling has the potential to match real processes as well or better than water models, especially when complex related phenomena such as particle motion and multiphase flow are involved. Much work is still needed to further improve the models and to apply them in parametric studies.