Mining Process Optimization Platform
Published:
An end-to-end machine learning platform for optimizing mineral processing operations — SAG milling, flotation, and thickening — deployed across multiple mining divisions and processing plants.
Business impact
This platform delivered measurable operational gains at industrial scale: throughput uplift exceeding +100 TPH in SAG milling — translating to millions in additional annual revenue — alongside measurable copper recovery improvements across flotation circuits. Optimization recommendations run on a 4-hourly production cadence, embedded directly into the operational workflow across multiple mining divisions.
| Metric | Result |
|---|---|
| Throughput | +100 TPH in SAG milling |
| Recovery | Measurable improvement across flotation circuits |
| Optimization cycle | Hourly tracking, 4-hour recommendation cadence |
| Deployment | Multiple mining divisions |
Strategic context
In large-scale copper mining, a 1% recovery improvement or a 100 TPH throughput gain translates to tens of millions USD annually. This platform moved operational decisions from experience-based to data-driven, embedding analytics into the daily operational cadence. The multi-division deployment required balancing standardized methodology with division-specific calibration — a technical and organizational challenge that defined the platform’s architecture.
Key Performance Indicators — Process impact
The platform embeds analytics into the operational cadence, shifting decision-making from experience-based to data-driven with measurable production impact.
| KPI | Baseline | With system | Impact |
|---|---|---|---|
| Decision basis | Operator experience, shift-to-shift variability | Data-driven recommendations every 4 hours | Consistent, auditable operational decisions |
| Optimization cycle | Manual setpoint adjustments | Automated scenario simulation with confidence intervals | Faster response to ore variability |
| Value realization | Unknown improvement potential | +100 TPH throughput, measurable recovery gains | Quantifiable annual production value |
| Multi-division scalability | Site-specific solutions | Configurable shared platform across divisions | Reduced per-site implementation cost |
The challenge
Large-scale mining operations involve complex, interconnected processes where small improvements in throughput or recovery translate into significant economic impact. SAG mills, flotation banks, and thickeners each have dozens of controllable variables and hundreds of sensor readings, creating a high-dimensional optimization problem that evolves continuously with ore characteristics.
System architecture
The platform follows a modular pipeline architecture built on Kedro for reproducibility and MLOps best practices:
- Data ingestion: Streaming and batch pipelines pulling from SCADA systems, lab analyses, and operational databases via Azure Data Factory
- Feature engineering: Domain-informed features including rolling statistics (mean, variance, percentiles over configurable time windows), lag variables capturing process inertia, ore property indicators derived from assay data, and operational regime detection via hidden Markov models or change-point algorithms
- Model training: Ensemble of XGBoost, gradient-boosted trees, and neural networks trained on historical operational windows
- Recommendation engine: Scenario simulation generating actionable setpoint recommendations with confidence intervals
- Operational dashboard: Real-time monitoring of adherence to recommendations, KPI tracking, and expert feedback loops
Results
- Throughput uplift scenarios exceeding +100 TPH in SAG milling operations
- Measurable improvements in copper recovery across flotation circuits
- Hourly tracking cycles with 4-hourly optimization recommendations in production
- Multi-division deployment with division-specific parameter tuning
Technology stack
- Orchestration: Kedro pipelines with Azure Databricks execution
- Compute: PySpark for distributed processing, Delta Lake for data storage
- ML: XGBoost, scikit-learn, with MLflow experiment tracking
- Infrastructure: Docker containers on Azure Container Registry, CI/CD via Azure Pipelines
- Visualization: Power BI dashboards with Streamlit prototypes for model exploration
This project is part of proprietary corporate work. Source code is not publicly available.