The Challenge
Mineral processing operations are dynamic and influenced by shifting feed quality, evolving circuit configurations, and equipment health variability. Traditional control method soften rely on fixed logic, operator intuition, or spreadsheet-based models—making real-time optimisation difficult and reactive.
To modernise decision-making, the Processing Strategy Adviser (PSA) initiative was launched. At its core is a digital process twin—a high-fidelity simulation model that replicates plant behaviour and enables advanced scenario analysis, predictive optimisation, and recommendation generation.
The PSA combines first-principles process modelling with data-driven insights through a hybrid modelling framework, supporting flexible, plant-specific configurations and integration with real-time control systems.
Key technical challenges included achieving convergence speed, maintaining accuracy across diverse conditions, and integrating seamlessly with Advanced Process Control (APC) and Planning and Work Optimisation (PWO) systems.
The Solution
Opti-Num developed and prototyped the PSA using a modular, Python-based Process Modelling Framework (PMF) capable of simulating full circuit behaviour and recommending operational strategies.
- Model Development and Conversion - Legacy Excel models were reengineered and integrated into scalable Python modules, covering key plant areas such as comminution, flotation (cell-by-cell), classification, and reagent addition.
- Hybrid Model Architecture - White-box (physics-based) process logic was augmented with black-box (machine learning) parameter estimation techniques to improve prediction quality where direct modelling was impractical or insufficient.
- Scenario Simulation and Optimisation - The PSA allows users to simulate “what-if” scenarios, quantify the impact of input changes, and receive real-time recommended control setpoints aligned with operational KPIs such as recovery, grade, and energy efficiency based on current feed and processing conditions.
- Real-Time Readiness and Integration - The model architecture was designed to support feedback loops with APC and PWO systems, providing operator-in-the-loop or fully automated optimisation support. Model outputs include directional predictions, confidence bounds, and constraint-aware setpoint recommendations.
The Outcome
The initiative laid a solid technical and strategic foundation for plant-wide optimisation through digital twinning:
- Strategic Planning - Engineers can simulate future feed conditions and test processing responses in advance, reducing guesswork and improving recovery targets.
- Improved Control Insight – The model offers deeper understanding of circuit interactions, helping identify bottlenecks and refine control strategies.
- Optimisation at Scale – PSA recommendations extend beyond typical APC logic, factoring in downstream implications, feed variability, and mineralogical influences.
- Digital Foundation – The platform supports future extensions into closed-loop control, autonomous execution, and continuous model calibration.
- Cross-Site Scalability – Modular architecture supports replication across plants with different circuit layouts and control philosophies.
.png)
.png)
.png)
