Automated Cross Channel Temperature Predictions for the PFR Lime Kiln Operating Support

The Parallel Flow Regenerative (PFR) lime kiln process is challenging with respect to the energy efficiency, product quality and production stops, due to the inability of the human operators to accurately predict the evolution of the process. Monitoring and controlling of such processes encounter several issues, related to the high mass and heat inertia of the process, data quality, production stops, operator’s experience, as well as unknown exogenous factors (e.g., quality of the fuel, and raw material properties). Hence, an automated control/optimization mechanism for properly configuring the process is not straightforward. In this paper, we present a selection of mechanisms for data preprocessing together with domain specific feature analysis that allow for capturing the short-term changes of the critical parameters of the process. Through these mechanisms, automated predictive modeling can be performed that can be used by the kiln operator or a predictive-based controller to modify fuel feed strategies to meet energy efficiency and product quality requirements. We validate the proposed data-based preprocessing and modeling approaches through experiments in real-world data sources.