346 generated continuously as part of normal operational monitoring under highly dynamic traffic conditions. Given this operational variability, safety and dispatch teams often benefit from additional analytical tools that help interpret CAS alerts according to the specific operational context in which they occur. In current industry practice, post-operational analysis of CAS alerts is often performed using simple rule-based or univariate criteria, such as alert duration, minimum distance, or aggressor speed; for example, events exceeding a fixed time threshold or involving equipment traveling above a predefined speed may be treated as higher risk (Ruff, 2007; EMESRT, 2023). While useful for basic filtering, such criteria may not fully capture the multidimensional nature of collision risk in dynamic mining environments. Integrating contextual operational variables such as traffic conditions, spatial characteristics, and equipment combinations can enhance the interpretation of interaction alerts and support safety teams in identifying patterns that may require operational attention (EMESRT, 2024; Hrica et al., 2022). Consequently, many operations continue to rely on reactive safety practices, where attention is focused on isolated alerts rather than on systemic risk patterns. Dispatch and safety teams are often required to review large numbers of interaction events with limited contextual information, reducing the effectiveness of preventive interventions. This situation highlights the need for methodologies capable of translating CAS interaction alerts into structured, context-aware risk indicators that integrate multiple operational variables, adapt to different mine zones, and provide interpretable outputs for preventive decision-making. In response to this gap, this work proposes an adaptive, context-aware multi-expert framework designed to support contextual interpretation and prioritization of CAS interaction alerts and identify spatial risk patterns using only standard CAS telemetry. By integrating intelligent mine segmentation, fuzzy expert reasoning, and spatial aggregation models, the approach aims to translate collision alerts into actionable safety information and support proactive, data-driven safety management in real mining operations. 2. OBJECTIVES AND SCOPE This work addresses three main topic areas of the World Mining Congress: AI and Data-Driven Decision Making, Health, Safety and Well-being, and Smart Operations and Systems Integration. The study focuses on advanced analytics and expert-driven reasoning to support collision-risk management in largescale open-pit mining operations. The objective of the paper is to develop and validate a practical, contextaware methodology that supports the contextual interpretation and prioritization of CAS interaction alerts. The proposed framework combines automatic spatial segmentation, fuzzy expert systems, and graph-based aggregation to translate CAS interaction alerts into structured safety indicators using only standard CAS telemetry, without requiring additional sensors or infrastructure. The methodology was implemented within the CAS server environment and evaluated using operational data from large-scale open-pit iron and copper mining operations in Brazil and Peru. The analysis covers the full pipeline from interaction-level risk estimation to spatial hotspot identification across different operational zones. Without modifying the CAS alerting logic, the proposed system introduces an additional analytical layer that supports the interpretation of interaction risk according to site-specific operational conditions and safety practices, enabling each mine to prioritize alerts in alignment with its own operational risk perception. 3. METHODOLOGY The proposed framework is designed to interpret and contextualize alerts generated by a Collision Avoidance System through the estimation of collision risk associated with each interaction. It adopts a hybrid data-driven and expert-driven paradigm based exclusively on CAS data, including mobile equipment positioning telemetry and collision alert logs. From this single source, complementary artificial intelligence (AI) techniques are integrated: unsupervised clustering to construct spatial context, fuzzy expert systems for collision risk estimation, and graph-based models for severity aggregation and risk propagation. The
RkJQdWJsaXNoZXIy MTM0Mzk2