Development of a Steel Temperature Prediction Model in a Steel Ladle and Tundish in a Casting and Rolling Complex

The paper presents a prediction model for steel temperature in a specified time interval during its overheating over a liquidus point in the tundish of a continuous casting machine (CCM). This depends on the last temperature measurement in a ladle prior to casting with consideration to the ladle’s history and treatment during secondary metallurgical processes in an casting and rolling complex (CRC). The prediction of steel superheat temperature at CRC is based on two modeling stages: the first stage integrates two approaches—a machine-learning algorithm and a probabilistic-graphical model (PGM and Bayesian networks); and the second stage uses a method for evaluating probability distributions. The first approach gives a point estimate of the intermediate and final temperatures of specific heat. The PGM-approach is useful for scenarios with uncertain input data, which is often the case for metallurgical enterprises. The accuracy level in predicting the temperature of the metal in the tundish at the first stage reached 6.0°C. At the second stage, the model integration into the process control system improved prediction accuracy, as well as provided technological parameter control in real time. The model is used as an advisor (master) for the technical team of the ladle furnace and vacuum degasser.