Table 1 - Sample input data generated from simulator applications. Input Data Sample-1 Sample-2 Sample-3 Sample-4 Sample-5 PLS_pH 1.09 1.36 1.25 1.03 1.52 Cu_in_PLS_gpl 1.63 1.81 1.77 1.76 1.42 Fe_in_PLS_gpl 17.03 11.10 12.8 14.97 15.02 Feed_Flow_lpm 17776 16132 16695 19502 19485 Extractant_Conc_pct 7.3 11.7 7.2 18.8 19.5 Organic_Flow_lpm 19597 24785 13778 23516 15021 H2SO4_flow_lpm 290.73 205.99 229.70 184.64 159.06 ORP_mV 385.49 486.60 418.16 424.40 435.80 5.1 Accuracy of the Hybrid Model The hybrid model combines first‑principles solvent‑extraction chemistry with a machine‑learned residual layer, enabling predictions that are both physically consistent and operationally realistic. Model performance was evaluated by comparing hybrid predictions against plant‑measured KPIs across a wide range of operating conditions, including variations in ore grade, feed acidity, O/A ratio, extractant strength, Fe levels in PLS, and hydrodynamic stability. The key KPIs—Copper Throughput (t/day), SX Extraction Efficiency (%), and Iron Concentration in the Rich Electrolyte (g/L)—were used to quantify predictive accuracy. Standard evaluation metrics (MAE, RMSE, R²) consistently showed that the hybrid formulation outperformed both standalone first principle -based and purely data-driven models. As illustrated in Fig. 5–7, hybrid model predictions for copper throughput and SX efficiency closely matched plant behavior, exhibiting minimal residuals and maintaining accuracy across varying head grades. The hybrid formulation also captured nonlinear interactions involving PLS pH, Fe loading, and organic strength— phenomena that equilibrium-only first principle models only partially represent—leading to more reliable estimation of extraction performance. Notably, forecasts of iron concentration in the rich electrolyte (Fig. 7) improved substantially due to the residual learner’s ability to correct for unmodeled operational realities such as entrainment, crud formation, and subtle hydrodynamic effects. Collectively, these results demonstrate that the hybrid twin delivers high‑fidelity, plant‑realistic KPI predictions while preserving chemical and mass‑balance feasibility, making it well‑suited for real‑time optimization, recommendation generation, and operator decision support in industrial SX/EW circuits. Figure 5 - Copper Recovery - Actual Plant Vs Hybrid model prediction Figure 6 - SX Efficiency - Actual Plant Vs Hybrid model prediction Figure 7 - Iron in Rich Electrolyte - Actual Plant Vs Hybrid model prediction
RkJQdWJsaXNoZXIy MTM0Mzk2