KEYWORDS Solvent Extraction, Electrowinning, Copper, Digital Twin, Operator Assistant, AI at the Edge, Physical AI 1. INTRODUCTION Copper solvent extraction–electrowinning (SX/EW) is a mature hydrometallurgical route that selectively transfers Cu²⁺ from pregnant leach solution (PLS) into an oxime-based organic phase and recovers metal via electrowinning. The underlying cation-exchange equilibria link copper loading with proton release—two H⁺ per mole of Cu—so extraction inherently alters the acidity profile and, in turn, selectivity and kinetics [1–3,7]. Distribution-ratio formulations that depend on extractant concentration and pH, consistent with chelation stoichiometry and slope analysis, are widely used to characterize stagewise behavior [2,3]. Industrial design balances mass transfer, organic-to-aqueous ratios (O/A), and acid management across mixer–settler trains and stripping contacts, guided by McCabe–Thiele analysis and equilibrium isotherms [1,4,7]. Accurate representation of acidity via [H⁺], pH, and H₂SO₄ equivalents is central to prediction and control [5,6]. Against this backdrop, SX/EW players increasingly face declining ore grades, ageing equipment, and variability in PLS composition particularly elevated Fe which depresses copper extraction efficiency (~80–85%), degrade organic performance, and lower electrowinning current efficiency (~60–75%), resulting in throughput losses (≈5–10%), higher specific energy use, and more lower-grade cathodes. Assay latency and fragmented data integration exacerbate the response lag. To address these gaps, this paper investigates an operator-assistant approach that combines first principle-based modeling with data-driven inference, executes at the edge, and delivers interpretable, ranked recommendations under human-in-the-loop supervision. This paper articulates an architecture that ingests planning targets, laboratory assays, field sensors, and historian data into a unified Operator Assistant comprising a Digital Twin (hybrid, first principle - first with data-driven residuals) and an Optimizer for prescriptive guidance, connected via open industrial protocols (e.g., OPC UA, MQTT). The contributions of this work are: (i) a researchgrade formulation of a first principle -based, stagewise SX model that captures Cu/Fe partitioning and acid balance with explicit dependence on [H⁺] and [HR]; (ii) the formulation of a hybrid Digital Twin that couples this first principle model with machine learning residuals and an embedded Optimizer for actionable decision support (iii) an edge-oriented integration of plant data streams and open protocols to support real-time state estimation, what-if analysis, and ranked recommendations, learning residuals and an embedded Optimizer for actionable decision support 2. ARCHITECTURE (CONTROL LAYERS) The architecture illustrated in Figure 1 represents a structured approach to integrating heterogeneous data streams and advanced computational modules for improved decision-making in solvent extraction operations. At the foundation of this architecture lies a multi-layered data acquisition system that consolidates information from four distinct sources. Planning systems provide high-level production targets, operational schedules, and constraint sets, which serve as boundary conditions for optimization. Laboratory Analysis contributes periodic assay data, such
RkJQdWJsaXNoZXIy MTM0Mzk2