Evaluation of free lime content in cement kilns by physics-guided support vector regression using factory energy management system

The cement industry is one of the most carbon-intensive industrial sectors owing to its high energy demand. In clinker production, free lime (F-CaO) is a critical quality indicator that reflects the adequacy of heat supply during high-temperature processing; excessive F-CaO indicates insufficient thermal energy and leads to deterioration of clinker quality and energy efficiency. Accurate prediction of F-CaO is therefore important for effective quality control and energy-efficient plant operation. Previous data-driven studies have selected input variables mainly from statistical correlations, often neglecting the underlying heat-transfer and reaction mechanisms of the pyro-process. To address this limitation, the energy flows and airflow distribution in the pre-heater, kiln, and cooler were quantified using a factory energy management system (FEMS) combined with a heat and mass balance model. Based on this analysis, physically meaningful input variables governing F-CaO formation were derived and incorporated into a support vector regression (SVR) model. Under an identical chronological train/test split and hyperparameter optimization protocol, the physics-guided inputs showed improved prediction performance compared with conventional correlation-based inputs, achieving MSE = 0.03, TIC = 0.07, and R = 0.73 on the hold-out test set. These results indicate that deriving input variables from physical principles can improve F-CaO prediction consistency within the studied operating window, providing a rational framework for physics-guided soft sensor development.