under constraints. A complementary rule-based layer encodes established SX operating heuristics (e.g., prioritizing pH correction during elevated Fe loading), ensuring the optimization results remain interpretable and consistent with standard operational practice. 4.4 Recommendation Workflow The plant input data is validated by the optimizer, after which the setpoint generator produces new candidate setpoints for the relevant control variables. These setpoints passed to the hybrid digital twin, which evaluates the expected KPIs—such as copper recovery, SX efficiency, and iron concentration in the rich electrolyte. Both the generated setpoints and their corresponding KPI predictions are submitted to the constraint validator. If the constraints are satisfied, the setpoints are scored and ranked based on KPI performance. After all candidate setpoints are evaluated, the highest-ranked option is selected and delivered to the operator as the recommended operating setpoint. Figure 40 - Operator Assistant workflow 4.5 Output to Operations The Operator Assistant provides operators with a concise set of recommendations that include expected extraction‑efficiency changes, projected copper‑throughput gains (t/day), impacts on Fe behavior and electrolyte quality, and short‑term predictive trends. This information is delivered via edge‑deployed dashboards with low latency, enabling timely operational adjustments during variable ore and PLS conditions. It provides operators with clear, data‑backed recommendations and links real‑time plant conditions to safe, optimized actions, improving recovery and stability while keeping humans in control. 5. EXPERIMENTS This section evaluates the performance of the hybrid digital-twin framework and the Operator Assistant across three dimensions: (i) prediction accuracy of key performance indicators (KPIs), (ii) accuracy and reliability of recommended control-variable setpoints, (iii) operational benefits delivered through the end-to-end recommendation workflow. All experiments were conducted using historical plant simulated data (Table 5.1) combined with offline validation scenarios curated in collaboration with process engineers.
RkJQdWJsaXNoZXIy MTM0Mzk2