Even though Additive Manufacturing (AM) has become a critical enabler of manufacturing in various industries, its full potential in terms of process quality and productivity has not been achieved yet. The recent developments in Artificial Intelligence (AI) can help toward this goal, especially through Human–AI Collaboration (HAIC). However, existing approaches are focused on certain aspects of the problem, without comprehensively tackling the issue. This study proposes a holistic and AM-specific HAIC framework that combines the different components of human expertise, explainable AI, simulation-based forecasting, and variable-based process control into an integrated decision-making structure. The key findings include the identification of the most important variables that should be utilized, including their classification through the input of experts in terms of importance (utilizing the presented M-S metric), controllability, and the most suitable agent (human, AI, both) to effectively control each variable. Finally, the concept of the framework for effective HAIC in AM is analyzed, including the operational sequence of sensing, AI analysis, human evaluation, decision implementation, and feedback loops. Two complementary case studies are presented; the first provides a conceptual example, and the second one develops a quantitative scenario that allows the comparison of three decision pathways—AI-only, Human-only, and HAIC.

As Industry 5.0 emerges as a human-centric evolution of industrial systems, this study investigates the effectiveness of training interventions in companies aimed at supporting the transition to Industry 5.0, emphasizing human-centric and resilient skill development. Drawing from multiple case studies involving engineers and operators, the research applies both meta-analysis and meta-regression to assess the added value of experiential learning approaches such as Teaching and Learning Factories. In addition, a novel methodology combining quantitative analyses with qualitative interpretation of emerging competences is presented. Principal Component Analysis and classification frameworks are employed to identify and organize key competence clusters along technological, organizational, and social dimensions. Special attention is given to the emergence of human-centered competences such as decision empowerment, which are shown to complement traditional operational capabilities. The findings confirm that experiential training interventions enhance both self-efficacy and adaptive operational readiness, while the use of fusion techniques enables the generalization of results across heterogeneous corporate settings. This work contributes to ongoing discourse on Industry 5.0 readiness by linking training design to strategic company incentives and highlights the role of structured evaluation in informing future policy and implementation pathways.

Digital twins (DTs) consist of various technologies and therefore require a wide range of data. However, many businesses often face challenges in providing sufficient data due to technical limitations or business constraints. This can result in inadequate data for training or calibrating the models used within a digital twin. This paper aims to explore how knowledge can be generated from federated digital twins—an approach that lies between digital twin networks and collaborative manufacturing—and how this can be used to enhance understanding for both AI systems and humans. Inspired by the concept of federated machine learning, where data and algorithms are shared across different stakeholders, this idea involves different companies collaborating through their respective DTs, a situation which can be referred to as federated twinning. As a result, the models within these DTs can be enriched with more-detailed information, leading to the creation of verified, high-fidelity models. Human involvement is also emphasized, particularly in the transfer of knowledge. This can be applied to the modeling process itself, which is the primary focus here, or to any control design aspect. Specifically, the paradigm of thermal process modeling is used to illustrate how federated digital twins can help refine underlying models. Two sequential cases are considered: the first one is used to study the type of knowledge that is required from modeling and federation; while the second one investigates the creation of a more suitable form of modeling.

Fault-tolerant control systems deployed in manufacturing and process industries must maintain output regulation under actuator degradation and parametric drift while producing operator-readable intervention records. This paper proposes a three-layer cognitive supervisory control framework: a continuously synchronized nominal plant model serves as a lightweight digital twin, driving residual-based anomaly detection that responds to model-mismatch faults before tracking error accumulates; an event-triggered supervisory layer applies composite anomaly scoring, risk estimation, and dwell-time-constrained mode switching over a bank of pre-verified stable controllers; and an optional bounded LLM reasoning agent forming Layer 3 of the cognitive architecture produces real-time natural-language audit records at each supervisory event. The framework is functionally non-intrusive—the closed-loop response is numerically identical to standalone LQI in three of four benchmark scenarios and on an unstable plant variant, with zero supervisory interventions. Under abrupt 75% actuator loss, a single targeted intervention achieves lower overshoot and better tracking than all baselines, including MPC whose feasibility-degradation under reduced actuator authority is confirmed empirically. On near-undamped plants representative of compressor and pipeline dynamics, integral absolute error improves significantly relative to standalone LQI; on a distributed thermal process representative of laser welding and induction heating, IAE improves substantially under abrupt heater power loss. Parametric sensitivity analysis across fault severity, detection threshold, onset timing, measurement noise, score weights, and dwell-time confirms robustness of the detection architecture. The LLM layer’s contribution is regulatory rather than performative: its value is the causally connected audit record generated at the moment of each decision, not IAE improvement—a form of real-time explainability that after-the-fact attribution methods such as SHAP and LIME cannot provide by construction.