A Computational Fluid Dynamics and Support Vector Regression Combined Method for Predicting the Flow Performance of the Five‐Strand Tundish

Abstract

To optimize the flow field of a multistrand tundish and improve the steel quality, a prediction model integrating the techniques of computational fluid dynamics (CFD) and support vector regression (SVR) is proposed. A total of 125 cases with different structure parameters, i.e., the diameter and inclination angle of the orifices, are designed. The flow field of liquid steel in the tundish is calculated by CFD, and a sample dataset is established. Then, the SVR, combining the genetic algorithm with K‐fold, is employed to train and predict the sample dataset. In the results, it is shown that the SVR model demonstrates good learning performance on the training dataset. The average absolute errors of predictions for dead volume fraction and consistency are 0.50% and 0.71%. In addition, the SVR model takes much less computational cost than the CFD method to predict the flow performance of tundish in a given baffle structure.