University of Tehran International Journal of Mining and Geo-Engineering 2345-6930 59 2 2025 06 01 A conscious lab-based approach for modelling and investigating the impact of influential parameters on feed rate and differential pressure in cement vertical roller mills 129 135 102423 10.22059/ijmge.2025.395567.595253 EN Rasoul Fatahi School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran Hadi Abdollahi School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran Mohammad Noaparast School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran Mahdi Hadizadeh School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran Journal Article 2025 05 19 Vertical Roller Mills (VRMs) are widely used in energy-intensive industries like cement, steel, and chemicals due to their efficiency in grinding, drying, and material transport. However, two critical aspects remain underexplored: the correlation between operational variables and differential pressure (dp) and the influence of key parameters, such as feed rate, on mill performance. To address these gaps, this study utilized advanced machine learning methods, including Random Forest (RF), LightGBM, and Shapley Additive Explanations (SHAP), integrated within a Conscious Lab-based (CL) framework. The study focused on modelling feed rate as a manipulated and dp as a controlled variable, with SHAP employed to analyze variable interactions. Findings identified operational factors such as working pressure, dp, counter pressure, and mill fan speed as significant determinants of feed rate setpoints. Working pressure emerged as the most influential variable impacting both dp and feed rate, establishing its critical role in stabilizing operations and regulating performance. Key variables, such as working pressure, mill fan speed, and feed rate, were also identified as primary contributors to dp, reflecting the principles of the CL framework for dynamic control. Validation tests revealed LightGBM as the best-performing algorithm, achieving the highest R² values (0.98 and 0.97) and lowest RMSE (1.34 and 0.16) for feed rate and dp prediction, respectively, making it the optimal model for predicting feed rate and dp. This study highlighted the potential of combining machine learning with the CL framework to accurately model complex relationships among variables, optimize VRM operations, and advance sustainable energy-efficient practices in the cement industry. https://ijmge.ut.ac.ir/article_102423_cd1904982254bb80f0fbd22ba62dcd4f.pdf