290 3.2. Digital twins: making consequences testable A mining value chain digital twin provides the simulation capability required to evaluate design alternatives under realistic conditions. The digital twin replicates the physical system, including throughput rates, equipment availability, reliability profiles, failure modes, buffer logic, routing constraints and demand scenarios, and enables controlled testing of changes before capital is committed. For each design option, the model produces outputs including constraint location, throughput and value outcomes at different confidence levels (e.g. P10/P50/P90), downside risk exposure and the capital avoided or deferred under each alternative. This creates a safe environment to compare options under realistic operating variability, strengthening engineering judgement with system-level evidence. 3.3. AI-assisted scenario discovery The third layer integrates digital twin simulation with reinforcement learning for guided scenario discovery. The process follows a structured sequence: ● Inputs: Operational and design data define the system baseline and constraints. ● Digital twin simulation: The model replicates the system to enable controlled testing of changes. ● Scenarios tested: Design and process variations are explored across the solution space. ● AI-driven agent: Analyses results to identify variable combinations driving performance shifts. ● Reinforcement learning loop: Iteratively proposes targeted scenarios based on measured outcomes, directing exploration towards high-value regions of the solution space. ● Insights: Quantifies impact and reveals the key drivers of system performance. The AI layer can interrogate model results through natural-language queries, surface recurring bottlenecks and sensitivities, recommend new scenarios, explain trade-offs and produce executive-ready insights. This expands the range of scenarios that decision-makers can evaluate beyond what manual analysis would allow. 3.4. Data-driven Minimum Viable Solution (MVS) The outputs from the digital twin and AI layer feed into a data-driven Minimum Viable Solution (MVS) framework. While digital twins make uncertainty testable, MVS makes the capital response disciplined. The framework asks what outcome must be protected, which scope materially protects it, what risk remains if scope changes, and what the cost of additional confidence is. It then classifies each scope element into five decision categories: keep, remove, defer, resize or record. This ensures that capital decisions are traceable, evidence-based and aligned to system-level outcomes rather than individual functional requirements. 4. KEY RESULTS, OUTCOMES AND INSIGHTS 4.1. Case study: maintenance facility project The framework was applied to a maintenance facility project at a large open-cut mining operation. The facility was required to support the long-term operation of the mine’s heavy vehicle fleet. During definition, the capital estimate had increased materially from the original basis, raising the central question: is the proposed scope genuinely required to protect production and fleet availability? Uncertainty around the future fleet profile, maintenance demand and congestion risk was driving scope growth, compounded by limited ability to test end-to-end system behaviour with traditional tools. A system-level digital twin was built and used to test alternatives across bay configurations, maintenance scheduling, labour productivity assumptions and supporting infrastructure requirements. AI-assisted scenario discovery expanded testing beyond manually defined cases, with the analysis framed around identifying a
RkJQdWJsaXNoZXIy MTM0Mzk2