Figure 2 – General Scheme of the Extract, Transform, and Load (ETL) Process. Source: Microsoft (2025). Data modeling was performed using a star schema dimensional approach, an architecture that organizes information by separating quantitative facts (such as laboratory results and timestamps) from descriptive dimensions (e.g., barrel type, material, or location). This structure optimizes query performance and facilitates the analysis of geochemical time series, enabling efficient comparisons across materials and complex scenarios. Based on this framework, analytical and predictive models were developed to estimate consumption rates and depletion times of sulfides and acid-neutralizing agents. These models were evaluated through cross-validation techniques and comparisons among different forecasting approaches to ensure the reliability of results within a real operational environment. Finally, the deployment phase integrated the models and automated workflows into an operational platform, enabling analytics within a real mining environment and facilitating interactive visualization of results to support decision-making. In this sense, the deployment phase goes beyond conceptualization, bringing the power of artificial intelligence directly into daily operations. The adoption of these models in real-world environments allows mining operations to function more sustainably and efficiently, supporting data-driven decision-making, regulatory compliance, and the minimization of environmental risks. Figure 3 illustrates the implementation of the temporal evolution of sulfide consumption and the behavior of neutralizing agents used in different treatments—such as calcium carbonate (CaCO₃) in the form of solid limestone, calcium oxide (CaO) applied as lime slurry, and combined systems using both agents—along with a comparison of three predictive methods for estimating their depletion under a mining scenario.
RkJQdWJsaXNoZXIy MTM0Mzk2