🏭 DCG Blast Furnace Slag — Real-Time Viscosity & RPM Controller

IIT (ISM) Dhanbad · Mineral & Metallurgical Engineering · Innovation 1

4 ML models · Bayesian-Optuna tuning · SHAP explainability · PCA anomaly detection · Qwen2.5-7B expert reports

🌡️ Slag Input Parameters

Temperature (°C)

Typical blast furnace tap: 1450–1550°C

1400 1600

Basicity CaO/SiO₂

Higher = CaO breaks silicate network → lower viscosity

0.8 1.4

Al₂O₃ (wt%)

Network former — raises viscosity (Xin et al. 2025)

8 18

MgO (wt%)

Network modifier — slightly reduces viscosity

4 12

Coke Rate (kg/t iron)

Proxy for furnace heat input

450 550

Tap Time (minutes)

Elapsed time since tap started — slag cools over time

0 90

Optical Basicity Λ (display only — auto-computed)

Duffy & Ingram (1976) — computed from mole fractions

0.55 0.75

Λ auto-computed from Basicity, Al₂O₃, MgO. Slider is read-only.

📊 ML Predictions

🔬 Data Quality Check (PCA Anomaly Detector)

🔵 Auto-Computed Optical Basicity Λ

🌲 Random Forest

⚡ XGBoost

🐱 CatBoost

🧠 Neural Network

⭐ Tuned Best Model (Primary Prediction)

💿 Disc RPM Recommendation

📈 Optimal Window Proximity

💬 Expert Explanation

⚡ Instant Analysis (SHAP attribution + rule-based)

🤖 Qwen2.5-7B Metallurgist Report (SHAP-enriched LLM prompt)

Optimal Viscosity Window: 0.055 – 0.080 Pa·s
< 0.055 → Reduce RPM (too fluid) | > 0.080 → Increase RPM (too viscous)
SHAP prompt methodology: Xin et al. 2025 (BO-CatBoost+SHAP), Chen et al. 2026 (optical basicity)

🏭 DCG Blast Furnace Slag — Real-Time Viscosity & RPM Controller

IIT (ISM) Dhanbad · Mineral & Metallurgical Engineering · Innovation 1

🌡️ Slag Input Parameters

📊 ML Predictions

💬 Expert Explanation

📌 Global SHAP Feature Importance (Tuned Best Model)