345 ADAPTIVE CONTEXT-AWARE MULTI-EXPERT SYSTEM FOR A MINE SAFETY DIGITAL TWIN BASED ON COLLISION ALERT SEVERITY ANALYSIS *A. Gutarra Sanchez1, W. Calla1, A. Olarte1, A. Franca2, H. Goncalves2, R. Martins2 1Smart Centre, Hexagon Mining, Peru, (*Presenting author: anthony.gutarra@hexagon.com) 2Smart Centre, Hexagon Mining, Brazil ABSTRACT Collision Avoidance Systems (CAS) are widely used in large-scale mining operations to support real-time interaction awareness between mobile assets. These systems generate proximity alerts that reflect the interaction dynamics of mining environments. Interpreting these alerts within the operational context requires additional analytical support, as interaction risk may be prioritized differently across mining operations depending on site-specific operational conditions and safety practices. This work presents an adaptive, context-aware multi-expert framework for contextual interpretation and prioritization of CAS alerts together with spatial risk analysis. The methodology integrates automatic mine segmentation, a fuzzy multiexpert system for contextual risk estimation, and spatial aggregation techniques to identify operational risk hotspots. The approach relies exclusively on standard CAS data, enabling deployment without additional sensors or infrastructure. The framework was implemented within the Hexagon Mining CAS environment and evaluated using real operational data from large-scale open-pit iron and copper mining operations in Brazil and Peru, comprising several thousand interaction events across multiple equipment combinations. Results show that, even across a large volume of interactions, the expert system identified a small subset of high-severity, operationally relevant events based on its contextual risk interpretation. Validation against a CAS safety specialist yielded a sensitivity above 95% and a precision around 90%. The spatial aggregation stage identified recurring high-risk patterns, particularly at mixed-traffic intersections involving haul trucks and light vehicles. A representative scenario illustrates a common interaction pattern consistently observed across sites. The analysis supported practical safety recommendations for traffic control and intersection management. Overall, the framework provides an additional analytical layer that translates CAS interaction alerts into contextual safety indicators aligned with the operational risk perception of each mining site, forming the basis for a mine safety digital twin. KEYWORDS Collision Avoidance System, Mining Safety, CAS Alert Prioritization, Fuzzy Expert System, Safety Digital Twin, AI-Driven Decision Making, Smart Mining Operations 1. CONTEXT AND PROBLEM STATEMENT Collision Avoidance Systems (CAS) are widely deployed in large-scale mining operations to support real-time collision prevention between mobile assets. These systems typically rely on GNSS-based positioning to determine equipment location, speed, and heading; vehicle-to-vehicle communications (commonly via RF) to detect nearby assets; and predictive logic to assess potential conflicts and generate alerts (Ruff, 2007; Hexagon Mining, 2020). Industry vehicle-interaction frameworks also emphasize that these controls operate within complex, highly dynamic conditions where human factors, traffic density, and operational context strongly influence collision risk (EMESRT, 2024). While CAS technologies provide reliable real-time proximity awareness between mobile assets, interpreting interaction alerts within the operational context remains challenging in complex mining environments (Ruff, 2007; Hrica et al., 2022). In medium- to large-scale open-pit operations, these alerts are
RkJQdWJsaXNoZXIy MTM0Mzk2