Integrating Modeling & Simulation and Artificial Intelligence for Training and Decision-Support in Complex Systems

Port terminals and industrial plants represent some of the most complex and safety-critical operational environments within modern logistics and maritime infrastructures. Operators must perform tasks in dynamic contexts characterized by high traffic density, hazardous materials, multimodal interactions between humans and machines, unpredictable environmental conditions, and fragmented information flows dispersed across digital and physical systems. These characteristics highlight the need for new methodological approaches capable of enhancing situational awareness, training effectiveness, and strategic decision support. This dissertation proposes an integrated simulation framework that unifies Modeling and Simulation (M&S), Extended Reality (XR), Artificial Intelligence (AI) and Data Analytics into a coherent architecture for advancing safety, resilience, and performance in port terminals. The work is grounded in the MS2G paradigm (Modeling, interoperable Simulation and Serious Games), enabling high-fidelity virtual environments, multi-layer modeling, and interoperable simulation components that represent physical processes, human behavior, cyber dynamics, and environmental conditions. At the core of the framework lies COYOTE, an innovative real-time simulator specifically designed to reproduce the operational reality of container terminal yards. The system integrates a physics-based 3D virtual environment developed in Unity, stochastic process modeling, discrete-event simulation structures, and agent-based representations of terminal vehicles, cranes, and human operators. Each entity interacts with the environment through dynamic state machines, navigation algorithms , obstacle-avoidance logic, and AI-driven behavioral models. A key strength of the simulator is the integration of Extended Reality. Through VR headsets and immersive audio-spatial rendering, operators visualize dynamic yard conditions, machine movements, dangerous-good leaks, limited-visibility scenarios (fog, night operations), and proximity alerts. The multisensory experience improves attention, risk perception, and decision-making under stress, allowing safe reproduction of scenarios that would be too dangerous or costly to train in real life. The research introduces also a comprehensive risk-analysis and behavioral-assessment module, which computes real risk exposure (RE), perceived risk exposure (pRE), and operational performance indicators based on mission duration, correctness of inspections, number and severity of incidents, and proximity to hazards. These Measures of Merit (MoMs) provide quantitative evaluation of operator behavior, supporting both training and scientific investigation of human factors in ports. A central methodological component of the dissertation is the experimental validation campaign conducted with professional operators from PSA Genova Pra’, one of Italy’s largest and most technologically advanced container terminals. More than 350 simulation trials were executed under varying levels of traffic density, weather conditions, scenario complexity, and augmented-reality activation. A second comparative dataset was collected from university engineering students to evaluate differences in learning curves, risk perception, and task accuracy between experienced personnel and non-experts. The results demonstrate: • a statistically significant improvement in task accuracy and inspection correctness across repeated missions; • a decrease in total and average risk exposure, indicating better navigation, safer positioning relative to moving vehicles, and more cautious behavior; • faster mission execution times without increasing risk, demonstrating development of competence and procedural fluency; • higher risk-awareness gains when augmented-reality indicators were activated; • strong correlation between pRE and RE, allowing calibration of new cognitive-behavioral models for operator perception. These findings provide robust empirical validation of COYOTE’s effectiveness as both a training system and a research platform for analyzing human behavior in hazardous port environments. Building upon this foundation, the dissertation introduces an extended Digital Twin architecture, integrating simulation models, data streams, and AI-based reasoning to support decision-makers. The architecture allows reproduction of port-wide operational states, assessment of accident evolution, evaluation of emergency responses, and exploration of future strategic scenarios, including hybrid cyber-physical threats. Intelligent agents simulate different behaviors, equipment failures, and cascading effects across operational layers, replicating the complexity of modern multi-domain risks. Overall, this thesis advances the state of the art in simulation-based training, safety engineering, and digital-twin–enabled decision support for port systems. By merging real-time physics modeling, AI-driven agents, immersive XR, and a validated experimental methodology involving professional operators, the research provides a novel scientific and technological foundation for improving resilience, situational awareness, and operational excellence in critical maritime infrastructures.