Enhanced prediction of carbon monoxide in cement rotary kilns: a machine learning approach versus response surface methodology

This research employs machine learning (ML) techniques and Response Surface Methodology (RSM) to predict and analyze carbon monoxide (CO) concentration in the combustion process of a rotary kiln at the Khash Cement plant. Accurate modeling of the relationships between process variables is critical for controlling pollutant emissions and enhancing operational stability in energy-intensive industries. To this end, a dataset comprising 8,600 real-time operational records was collected from the plant’s PLC control system. After preprocessing, statistical analysis, and outlier removal, the data were used to train and evaluate the predictive models. CO concentration was designated as the dependent variable, while main burner fuel, calciner fuel, kiln feed rate, Cyclone 4 temperature, Cyclone 3 temperature, preheater fan outlet temperature, and kiln rotational speed were considered independent variables. Three machine learning algorithms—Random Forest Regressor (RFR), AdaBoost, and XGBoost—were implemented in Python to predict CO levels and evaluated using standard statistical metrics, including R², RMSE, MAE, and MAPE. For reference, the process was also modeled using the conventional Response Surface Methodology (RSM), which serves as an industry-standard statistical benchmark . The results indicate that the RFR model achieved the highest predictive accuracy (R² = 0.89), followed by XGBoost (0.78), while the RSM benchmark yielded an R² of 0.69, and AdaBoost achieved an R² of 0.57. Furthermore, an explainability analysis using SHAP (SHapley Additive exPlanations) identified Cyclone 4 temperature and calciner fuel as the most influential parameters affecting CO variations. The findings demonstrate that combining real-world industrial data with advanced machine learning models and established statistical techniques can provide reliable and interpretable decision-support tools for monitoring, emission control, and operational optimization in cement rotary kilns.