350 capabilities. The proposed algorithms were implemented in Python and deployed within the CAS server as part of an automated pipeline. This setup enabled continuous ingestion of alert data, real-time contextual severity estimation of interactions, and integration with the safety monitoring environment. 4. RESULTS AND DISCUSSION 4.1. Intelligent Mine Segmentation (Stage 1) The proposed segmentation approach was applied to telemetry data from mobile mining equipment to automatically derive functional zones within large-scale open-pit mining operations, including iron mining sites in Brazil and copper mining operations in Peru. The system identified zones corresponding to typical operational contexts, such as haulage routes, loading, dumping, and other activity areas. The segmentation was performed over a spatial discretization of the mine, where each spatial unit was assigned to a functional zone according to its dominant characteristics. A representative subset of the segmentation results was reviewed with safety and operations specialists, confirming the coherence of the identified zones and yielding an average spatial classification accuracy above 90% across the analyzed operations. Misclassifications were mainly observed in transition areas between zones, where functional boundaries are inherently diffuse. This behavior is consistent with data-driven spatial segmentation and highlights the need for a subsequent refinement step. Due to confidentiality constraints, detailed segmentation maps cannot be presented; however, validation with domain specialists confirmed the consistency of the identified zones across different operational contexts. 4.2. Intelligent CAS Alert Prioritization Using a Fuzzy Expert System (Stage 2) The fuzzy expert system for alert prioritization, described in Stage 2, was evaluated over a representative three-month operational period using aggregated data from Collision Avoidance System (CAS) logs from large-scale open-pit iron and copper mining operations in Brazil and Peru. The dataset included several thousand CAS alerts corresponding to equipment interactions across different operational zones, including Haul Truck (HT)–Light Vehicle (LV) interactions. According to the system design, collision risk severity is estimated on a continuous scale from 0 to 100%. For reporting purposes, the severity output was grouped into three aggregated levels: Low, Medium, and High. Alerts classified within the High category were considered priority events requiring operational attention. The results are summarized in the histogram shown in Figure 4. Across the evaluated dataset, approximately 2–3% of interactions were categorized within the High severity level according to the contextual risk interpretation produced by the expert system, while the remaining interactions were distributed across the Medium and Low levels. Figure 4 – Severity distribution of CAS alerts after fuzzy expert system prioritization. This distribution illustrates how the expert system translates CAS interaction alerts into contextual risk indicators, enabling operations to highlight interactions that may require closer review according to sitespecific risk interpretation. This behavior aligns with the operational role of CAS, which continuously
RkJQdWJsaXNoZXIy MTM0Mzk2