Lime Saturation Factor as a Determinant of Free Lime Content in Portland Cement Clinker: A Machine Learning-Based Preventive Quality Control Study

Free lime (FCaO) is the most critical and rapidly available quality indicator of Portland cement clinker, and its control depends on the composition of the raw meal entering the kiln. This study investigates the relationship between the Lime Saturation Factor (LSF) of raw meal as the upstream compositional parameter and clinker FCaO as the downstream quality outcome. A total of 960 valid daily production records collected from Kiln Line V of PT Semen Tonasa, Indonesia (January 2023–December 2025), integrating Distributed Control System (DCS) and laboratory quality-control data, were analyzed using descriptive statistics, Pearson correlation, independent-samples t-tests, quartile-based risk stratification, and supervised machine learning. XGBoost was compared with Linear Regression, Random Forest, and Artificial Neural Network models using 25 process parameters. The mean FCaO was 1.2552% (SD = 0.224), with 12.4% of observations classified as defective (FCaO > 1.5%). LSF showed a significant positive correlation with FCaO (r = +0.295, p < 0.001), and defective batches exhibited significantly higher LSF values than normal batches. Defect frequency increased from 7.5% in the lowest LSF quartile to 22.9% in the highest. XGBoost achieved the best predictive performance (R² = 0.517, MAE = 0.105%, RMSE = 0.144%, MAPE = 8.96%), confirming LSF as an important upstream predictor. The findings demonstrate that raw meal LSF is an interpretable and actionable parameter for preventive clinker quality control and early-warning monitoring before kiln entry.