📦 Resource pdf

Metallurgical Response Curve Calibration Guide

The Metallurgical Response Curve (MRC) Calibration Guide is a standardized methodology for quantifying and modeling the relationship between feed ore characteristics (e.g., grade, mineralogy, hardness) and downstream metallurgical performance (e.g., recovery, concentrate grade, energy consumption). It enables predictive process optimization by empirically calibrating dynamic response curves using plant data, laboratory testwork, and statistical regression. The guide provides protocols for data validation, curve fitting, uncertainty quantification, and integration into digital twin and real-time optimization frameworks.

📖 Overview

Metallurgical Response Curve Calibration is foundational to modern mine–mill–concentrator integration, bridging geological variability with process engineering outcomes. At its core, the MRC represents a multivariate, non-linear transfer function that maps key feed attributes—such as modal mineral abundance (e.g., liberated chalcopyrite %), textural parameters (grain size distribution, locking), and chemical assays (Cu, S, Fe, As)—to metallurgical KPIs like flotation recovery, leach extraction rate, or grinding specific energy. Calibration requires harmonized datasets spanning drill-core geochemistry, QEMSCAN/MLA mineralogical imaging, comminution tests (Bond Work Index, JK Drop Weight), and synchronized plant operational logs, all time-aligned and spatially referenced to mining blocks or drawpoints. The calibration process follows a rigorous workflow: (1) data curation and outlier detection using multivariate statistical control charts; (2) dimensionality reduction (e.g., PCA or PLS regression) to identify dominant drivers; (3) piecewise or spline-based curve fitting (often with Bayesian regularization to manage sparse or noisy data); and (4) cross-validation via temporal holdout or geostatistical block simulation. Critically, the guide emphasizes uncertainty propagation—assigning confidence intervals to predicted responses using bootstrapping or Gaussian process regression—to support risk-informed production scheduling and blending decisions. Practically, calibrated MRCs are embedded in mine planning software (e.g., Deswik, Vulcan, MineSuite) and advanced process control (APC) systems to enable 'what-if' scenario analysis, real-time feed-forward control, and dynamic circuit reconfiguration. For example, detecting a shift toward more silicate-rich ore may trigger automatic adjustments to reagent dosages, pulp pH, or cyclone cut point—before metallurgical performance degrades. The guide also prescribes governance protocols: version control of curves, periodic re-calibration triggers (e.g., after major equipment change or new orebody zone entry), and audit trails linking curves to underlying assay and testwork certificates.

📑 Key Components

1 Feed Characterization Database
2 Response Curve Fitting Engine
3 Uncertainty Quantification Framework

🎯 Applications

  • Block-model-based recovery forecasting
  • Real-time flotation circuit optimization
  • Mine-to-mill reconciliation and bottleneck diagnosis

📐 Key Formulas

Multivariate Recovery Response Function

R_i = f(\text{Cu}_{\text{lib}}, \text{S}_{\text{tot}}, D_{80}, \text{HGI}, \text{pH}, \text{[collector]}) + \varepsilon_i

Predicts metal recovery (R_i) for ore sample i as a function of liberation-corrected grade, total sulfur, grind size (D80), Hardgrove Grindability Index (HGI), pH, and collector concentration; ε_i is the residual error term

Calibrated Specific Energy Model

E_{sp} = \alpha \cdot \left(\frac{\text{Wi}}{10}\right) \cdot \left(\frac{1}{\sqrt{P_{80}}} - \frac{1}{\sqrt{F_{80}}}\right) \cdot \left[1 + \beta \cdot \left(\frac{\text{SiO}_2}{\text{Fe}_3\text{O}_4}\right)\right]

Modified Bond equation incorporating mineralogical ratio correction factor to predict specific grinding energy (E_sp) in kWh/t

Bayesian Prediction Interval

R_i^{\text{pred}} \in \left[\hat{R}_i - t_{\nu,1-\alpha/2} \cdot \sigma_{\text{pred}},\; \hat{R}_i + t_{\nu,1-\alpha/2} \cdot \sigma_{\text{pred}}\right]

95% prediction interval for recovery estimate, where σ_pred combines model parameter uncertainty and residual variance

🔗 Related Concepts

Mineral Liberation Analysis (MLA) Geometallurgy Digital Twin for Mineral Processing

📚 References

#geometallurgy #process optimization #mining digital twin