Tundish Open Eye (TOE) formation is a phenomenon that occurs due to the sweeping off of the overlying slag layer by the strong recirculatory flows generated by the upwelling argon gas-liquid steel plume[1-4]. The exposed layer of liquid steel readily reacts with ambient air to form harmful inclusions that may directly pass on to the molds, thereby degrading the quality of semi-finished products such as billets, blooms or slabs. The inclusions can also get attached on the inner walls of the Submerged Entry Nozzles (SENs) and clog them, resulting into replacing them prematurely. Apart from SEN life reduction, the overall productivity of the continuous casting process also suffers. Since these issues have a direct impact on steel quality, productivity and thus, profitability of steel making process, an efficient and practically implementable solution is the need of the hour.
Previous research on TOEs has focused on physical, mathematical and mechanistic modeling. Although these models can fairly reproduce the phenomenon of TOE and provide us a fundamental understanding of the problem, they are unable to perform on-line and dynamic predictions. In the recent past, machine learning has seen great success due to its versatility and strength in making predictive models. These models can perform predictions in real-time, once they are trained to do so.
Artificial neural networks (ANN) based models have flexibility in data input and variable relationships makes it an ideal model for many different applications. There are just a handful of cases, where machine learning algorithms based on neural network, have been utilized. The present work demonstrates the viability of using machine learning techniques (ANN) for predicting open eye formation during the continuous casting operation.