Towards a Digital Twin For Quarry Sites: From Requirements to Operational Components

Off-road quarry environments are complex systems where machines, materials, and humans interact under harsh, safety-critical, and resource-constrained conditions. While digital twins promise to transform these operations through virtual experimentation and optimization, practical adoption is limited by three key challenges: integrating models across multiple temporal and spatial scales, balancing computational efficiency with physical fidelity, and maintaining modular architectures that can evolve with changing site requirements. This licentiate thesis contributes toward addressing these challenges by developing a foundational framework for digital twin implementation in quarry operations and demonstrating its feasibility through two enabling components. First, through industry-embedded case studies combining semi-structured interviews, expert workshops, and site observations, the research maps simulation-optimization requirements across three operational levels and proposes a hierarchical modeling framework that defines interfaces between site-level planning, operational coordination, and machine dynamics. This framework establishes how information should flow between high-level production scheduling and low-level equipment control while maintaining computational tractability. To demonstrate technical feasibility within this framework, the thesis develops two machine-learning components at the dynamics level. A torque-prediction model uses expert-guided feature selection and Shapley Additive exPlanations (SHAP) analysis to achieve high-fidelity estimates with minimal sensor inputs, providing a template for interpretable surrogate modeling. A Long Short-Term Memory (LSTM) based world model enables efficient reinforcement learning for autonomous bucket filling, showing major improvements in both productivity and energy efficiency compared to baseline controllers in simulation environments. This research establishes the architectural foundation and demonstrates core technical capabilities necessary for quarry digital twins, while explicitly deferring full system integration, field validation, and cross-site deployment to future doctoral work. The contributions provide a structured approach to multi-level modeling for quarry digital twins, establishing methodological foundations for integrating site level planning, operational coordination, and machine dynamics models while demonstrating that machine learning can deliver computationally efficient surrogates suitable for real-time applications.