
The integration of Large Language Models (LLMs) into industrial robotics offers new opportunities for natural-language interaction, explainability, and high-level decision support. However, their adoption in safety-critical and regulated manufacturing environments requires architectural mechanisms that ensure strict separation between reasoning and execution, standardised interfaces, and full traceability. This paper introduces the Industrial Model Context Protocol (I-MCP), a framework that formalises the interaction between LLM-based agents and intelligent industrial automation systems. I-MCP defines declarative context specifications, procedural agent responsibilities, and executable MCP tools as distinct components, enabling LLMs to orchestrate industrial tasks without direct access to physical systems. All operational capabilities are exposed through MCP-compliant services and recorded via structured, audit-ready logs. The framework is demonstrated through a case study on automated inspection and cleaning of composite tooling, in which an operator specifies tasks using natural language while a sensor-equipped robotic system performs perception, execution, and reporting via MCP-mediated services integrated with a ROS2-based control stack. The results illustrate how I-MCP supports explainability, traceability, and the safe integration of LLMs into industrial workflows, providing a scalable pathway for the deployment of next-generation LLM-enabled manufacturing systems in regulated production environments.
Manufacturing systems are progressively anticipated to evolve from automation-focused efficiency to human-centered intelligence, adaptability, and collaborative decision-making, in accordance with the Industry 5.0 vision. The amalgamation of Large Language Models (LLMs) with Extended Reality (XR) establishes a formidable framework for incorporating cognitive reasoning, contextual awareness, and intuitive interaction directly into industrial settings. Although recent research have shown the efficacy of LLM–XR systems in task execution and operational support, their comprehensive impact throughout the manufacturing lifecycle has not been extensively investigated. This analysis analyzes LLM–XR integration from a lifecycle viewpoint by combining industrial studies primarily published between 2015 and January 2026. It encompasses the entire manufacturing lifecycle, including applications in design, planning, assembly and disassembly, maintenance, training, quality inspection, and human–robot collaboration. These innovations collectively signify a shift from localized operational assistance to shared cognitive platforms that facilitate spatial-semantic reasoning, explicable decision-making, and human-AI collaboration. The research delineates essential technological facilitators that support the efficient implementation of LLM-XR, encompassing multimodal interaction, spatially grounded knowledge enhancement, reasoning-based intelligence, and ergonomics-conscious system design. Furthermore, ongoing issues of system reliability, grounding in domain knowledge, computational limitations, and user trust are rigorously analyzed. This review synthesizes LLM–XR integration across the manufacturing lifecycle and develops a generalized conceptual reference architecture that organizes recurring functional components, interaction mechanisms, information flows, and adaptive feedback processes across design, planning, execution, and quality-related activities.
Sustainable-resilient supplier selection is becoming increasingly challenging because decision data are often incomplete, vague, or conflicting. Many fuzzy multi-attribute decision making (MADM) methods still face two key limitations: (i) uncertainty may not be represented with sufficient flexibility, and (ii) they use either objective or subjective criteria weights, which can lead to inconsistent decisions. To overcome these issues, this study proposes a hybrid MADM decision-support framework developed in the p, q-quasirung orthopair fuzzy set (p, q-QOFS) environment. First, the essential criteria weights are objectively evaluated by Weights by ENvelope and SLOpe (WENSLO) method and subjectively by Fuzzy Weighted Zero-Inconsistency (FWZIC) within the p, q-QOF environment. Subsequently, the subjective and objective weights are integrated to obtain balanced final criteria weights for decision-making. Finally, the feasible suppliers are evaluated and ranked using a p, q-QOFS-based Mixed Aggregation by Comprehensive Normalization Technique (MACONT) method against sustainable and resilient criteria under uncertainty. The methodology is validated through a real-world case study of a food manufacturing company addressing a sustainable-resilient supplier selection problem with multiple feasible suppliers. The proposed hybrid MADM approach in the p, q-QOFS environment selects A1 as the most suitable sustainable-resilient supplier, underscoring the dominant role of social and resilience criteria and demonstrating stable ranking performance across different evaluation scenarios. Sensitivity analyses confirm that the solutions remain stable under different decision-making scenarios. Thus, the study presents a robust and practical decision-support model to support sustainability-and resilience-focused supplier selection in food supply chains, which are often highly disruption-sensitive. Although the case study is limited to one industry, the proposed framework can be extended to support related supply chain decisions such as order allocation and multi-period sourcing.