Interpretable modeling for material transportation process in an industrial cement vertical roller mill

Area of Science:

• Industrial Engineering
• Materials Science
• Artificial Intelligence

Background:

• Vertical roller mills (VRMs) are crucial for efficient cement finish grinding.
• Quantifying the interplay between ventilation, fan operation, and pneumatic transport in VRMs using plant data is challenging.

Purpose of the Study:

• To develop and apply a data-driven, explainable machine learning framework (Conscious Lab - CL) for modeling material transport in an industrial VRM circuit.
• To enhance understanding of ventilation's role in pneumatic transport and circuit stability.

Main Methods:

• Utilized 1050 operating records from an industrial VRM at Tehran Cement Plant.
• Trained and evaluated three ensemble algorithms (CatBoost, XGBoost, Random Forest) using hold-out splits and nested cross-validation.
• Employed SHAP analysis for model interpretability and identification of dominant drivers.

Main Results:

• CatBoost demonstrated superior predictive accuracy for mill fan speed and power.
• SHAP analysis identified key variables like differential pressure, grinding pressures, water injection, and feed rate influencing transport.
• Nonlinear interactions among these variables were revealed.

Conclusions:

• The Conscious Lab (CL) framework offers an explainable, operator-centric tool for VRM process diagnosis and control.
• Ventilation parameters significantly govern pneumatic transport and overall circuit stability in VRMs.
• The findings support the development of intelligent control strategies for energy-efficient cement grinding.