Enhancing the stationary state prediction in Model Predictive Control systems to avoid Dross defect in heavy-section foundries

A Model Predictive Control (MPC) is a system designed to control a production plant. These systems are composed by several phases, being one of the most important ones the phase for the prediction of the plant situation in a given time. In a previous work, we presented a machine-learning approach for this prediction phase that replaced the need of developing a single mathematical function with a more generic classification approach. However, standalone classifiers had some drawbacks like to select the most adequate classification models for the learning data and task. In this paper we extend our previous work with a general method to foresee Dross defects building a meta-classification system through the combination of different methods and removing the need of selecting the best algorithm for each objective or dataset.