This paper investigates intelligent configuration management for modular production systems by integrating operational semantics with knowledge graph representations. The proposed approach formalizes configuration knowledge and system behavior through operational semantics, enabling consistent reasoning across modular components. A knowledge graph is constructed to capture configuration rules, dependencies, and operational constraints, supporting dynamic reconfiguration and decision-making in complex manufacturing environments. Case studies demonstrate that the proposed framework improves configuration efficiency, adaptability, and semantic consistency in modular production systems.

The coexistence of manual and autonomous driving leads to nonstandard final parking states (FPS) of adjacent vehicles, which can hinder passenger door access if a standard FPS is enforced for the target vehicle. To enhance passenger comfort during boarding and alighting, this paper proposes an efficient bidirectional search hybrid A* (BHA*) method for parking path planning under adjacent vehicle deviation scenarios. A characterization variable for passenger comfort is constructed based on allowable and required door opening angles under adjacent vehicle constraints. An optimization model for the FPS is established with a comprehensive objective that balances passenger comfort for both target and adjacent vehicles while ensuring safe distances. A genetic algorithm is employed to determine the optimal FPS, and Voronoi potential fields combined with a bidirectional search strategy are used to generate feasible and comfortable parking paths. Experimental results demonstrate that the proposed BHA* method effectively improves parking feasibility and passenger comfort in complex parking environments.

Industry 4.0 demands intelligent and autonomous manufacturing systems that can adapt to dynamic and uncertain environments. A key enabler of such systems is self-learning, which supports effective decision-making under uncertainty. However, conventional approaches typically rely on large volumes of physical production data, limiting their applicability during early deployment phases. To address this challenge, this paper proposes a digital twin-based self-learning decision-making framework that enables virtual training and validation of decision models prior to physical deployment. The framework comprises four modular components: a digital twin for simulating the production environment, a learning module for generating and evaluating decisions, a real controller for executing validated actions, and the physical production system. A time-delayed deployment mechanism is introduced to ensure the safe transfer of learned behaviors from the virtual environment to real-world operation. Experimental results demonstrate that the proposed framework enhances decision quality, adaptability, and deployment efficiency for industrial robotic systems.

In automatic parking scenarios, obstacles on both sides of a target parking space can significantly degrade the feasibility and comfort of passenger boarding and alighting (CPBA) when a standard final parking state is maintained. Moreover, the presence of dynamic obstacle vehicles in parking lots poses substantial challenges to the adaptability of path-planning algorithms. To address these issues, this paper proposes a method for final parking state optimization and two-stage path planning in dynamic parking environments while explicitly considering CPBA. A CPBA characterization variable is first constructed based on the allowable opening angle of the vehicle door constrained by adjacent obstacles. To improve path quality, the basic motion curve of the vehicle is formulated to support queuing, parking, and transition maneuvers. A two-stage path planning framework is then developed to integrate parking lot system-level planning with vehicle-level trajectory generation, enabling safe, adaptive, and comfort-oriented automatic parking in dynamic scenarios. Experimental results demonstrate the effectiveness of the proposed approach in enhancing parking feasibility, adaptability, and passenger comfort.

This paper proposes an omni-scale spatio-temporal attention network for accurate impact localization in sandwich composite panels. The proposed network captures spatio-temporal features across multiple scales to effectively model the complex wave propagation patterns induced by impact events. By integrating attention mechanisms at different temporal and spatial resolutions, the method enhances sensitivity to impact-related signatures while suppressing noise and irrelevant variations. Experimental results demonstrate that the proposed approach achieves superior localization accuracy and robustness compared to existing methods, making it suitable for structural health monitoring of composite sandwich structures.

This paper presents LLM-MANUF, an integrated framework for fine-tuning large language models to support intelligent decision-making in manufacturing systems. The proposed framework adapts general-purpose LLMs to manufacturing-specific tasks through domain-aware data curation, task-oriented fine-tuning strategies, and structured knowledge integration. By aligning language model reasoning with manufacturing process semantics, LLM-MANUF enables effective decision support across planning, scheduling, and operational optimization scenarios. Experimental studies demonstrate that the proposed framework significantly improves decision accuracy, robustness, and adaptability in complex manufacturing environments.

Integrating large language models (LLMs) into healthcare diagnosis demands systematic frameworks capable of handling complex medical scenarios while preserving domain-specific expertise. Here we propose KG4Diagnosis — a hierarchical multi-agent framework that integrates LLMs with automated medical knowledge graph construction, covering 362 common diseases across specialties. The system mirrors real-world clinical workflows via a two-tier architecture: a general-practitioner agent conducts initial assessment and triage, then delegates to specialized agents for in-depth diagnosis in specific domains. Our core innovation lies in an end-to-end knowledge graph generation pipeline, which combines semantic-driven entity and relation extraction, multi-dimensional decision-relationship reconstruction from unstructured medical texts, and human-guided reasoning for knowledge expansion. KG4Diagnosis establishes an extensible, modular foundation for accurate and scalable AI-powered medical diagnosis systems.

This paper investigates the use of large language models to advance capability matching in manufacturing system reconfiguration. The proposed approach leverages LLMs to interpret heterogeneous capability descriptions, align system requirements with available resources, and support intelligent reconfiguration decisions. By integrating language-based reasoning with manufacturing knowledge representations, the framework enhances flexibility, scalability, and robustness in complex manufacturing environments. Experimental studies demonstrate the effectiveness of the proposed method in improving reconfiguration accuracy and decision efficiency.

This paper presents a comprehensive framework for manufacturing system reconfiguration and optimisation that integrates digital twin technologies with modular artificial intelligence. The proposed approach combines virtual system representations with intelligent decision-making modules to enable adaptive reconfiguration, performance optimisation, and scenario analysis in complex manufacturing environments. By leveraging digital twins for real-time system monitoring and modular AI components for reasoning and optimisation, the framework enhances flexibility, scalability, and responsiveness in manufacturing systems. Case studies demonstrate the effectiveness of the proposed framework in supporting intelligent reconfiguration and optimisation across diverse production scenarios.

This paper presents an automatic defect detection approach for wire bonding in microwave components using multi-stage hybrid methods based on deep learning. The proposed framework combines multiple processing stages to progressively enhance defect-related features while suppressing noise and irrelevant variations. By integrating deep learning models with hybrid detection strategies, the method improves robustness and accuracy in identifying wire bonding defects under complex inspection conditions. Experimental results demonstrate the effectiveness of the proposed approach for quality assurance in microwave component manufacturing.

This paper investigates human-centered artificial intelligence technologies for human–robot interaction in social settings. The proposed approaches emphasize user-centric design principles to enhance interaction naturalness, adaptability, and social acceptance of robotic systems. By integrating perception, learning, and interaction strategies, the study demonstrates how AI-driven methods can improve engagement and usability in socially interactive environments. Experimental evaluations highlight the potential of human-centered AI to support more effective and socially aware human–robot interaction.

This paper explores efficient decision-making approaches for small and medium-sized enterprises (SMEs) by leveraging knowledge graphs implemented with Neo4j and AI-based vision technologies. The proposed framework integrates structured relational knowledge with visual perception to support data-driven operational decisions in industrial environments. By combining graph-based reasoning with AI vision, the approach enhances situational awareness, scalability, and decision accuracy for SMEs adopting low-cost digital automation solutions. Experimental results presented at LoDiSA 2023 demonstrate the feasibility and effectiveness of the proposed method.

This paper proposes a runtime-condition modelling approach for proactive intelligent products by integrating knowledge graphs with embedding techniques. The proposed model captures dynamic operational conditions and contextual knowledge to enable intelligent products to anticipate system states and support proactive decision-making. By combining symbolic knowledge representations with continuous embedding spaces, the framework enhances adaptability, scalability, and reasoning capability in intelligent product systems. Experimental results demonstrate the effectiveness of the proposed approach in improving responsiveness and intelligence in knowledge-driven product lifecycle management.

This paper presents a PLC orchestration automation framework to enhance human–machine integration in adaptive manufacturing systems. The proposed approach coordinates programmable logic controllers and higher-level control logic to enable flexible task allocation, adaptive system behavior, and improved collaboration between human operators and automated components. By integrating orchestration mechanisms with manufacturing execution and control layers, the framework supports dynamic adaptation to changing production requirements. Experimental results demonstrate that the proposed method improves system responsiveness, robustness, and overall operational efficiency in adaptive manufacturing environments.

This paper proposes a modular artificial intelligence framework integrated with the Asset Administration Shell (AAS) to streamline testing processes in manufacturing services. The approach combines modular AI components with standardized asset representations to support interoperability, automation, and scalability in testing workflows. By embedding intelligence within the AAS structure, the framework enables efficient orchestration of testing activities, improved data consistency, and enhanced decision support. Experimental studies demonstrate the effectiveness of the proposed solution in reducing testing effort and improving service efficiency in manufacturing environments.

This paper presents a comprehensive review and expert evaluation of agent-based manufacturing systems. The study analyzes the evolution of agent-based approaches in manufacturing, covering system architectures, coordination mechanisms, decision-making strategies, and industrial applications. By synthesizing insights from both academic literature and expert assessments, the paper identifies key challenges, performance considerations, and future research directions for deploying agent-based manufacturing systems in complex industrial environments.

This paper proposes a maturity model to characterize and assess the autonomy of manufacturing systems. The model defines structured autonomy levels by considering system intelligence, decision-making capability, adaptability, and human–machine collaboration. By providing a systematic framework for evaluating autonomy progression, the proposed maturity model supports benchmarking, strategic planning, and the design of next-generation autonomous manufacturing systems. Case studies and expert evaluations demonstrate the applicability and usefulness of the model in industrial contexts.

This paper proposes a variational Bayesian learning framework with a reliable likelihood approximation for accurate process quality evaluation. The proposed method addresses uncertainty modelling and inference challenges in complex industrial processes by introducing a robust approximation strategy for likelihood estimation. By integrating variational inference with reliable probabilistic modelling, the approach improves prediction accuracy and uncertainty quantification in process quality assessment. Experimental results on industrial datasets demonstrate the effectiveness of the proposed method in enhancing evaluation reliability and decision support for industrial informatics applications.

This paper proposes a multi-phase scheduling method for reconfigurable flexible job-shops with multi-machine cooperation based on a scout and mutation-based Aquila optimiser. The proposed approach decomposes the scheduling problem into multiple coordinated phases to address reconfigurability, machine cooperation, and operational constraints simultaneously. By integrating an enhanced Aquila optimisation strategy with scouting and mutation mechanisms, the method improves solution quality, convergence speed, and robustness. Experimental results demonstrate the effectiveness of the proposed scheduling framework in reducing makespan and improving system performance in complex reconfigurable manufacturing environments.

This paper proposes a coupling optimisation method that jointly considers production scheduling and computation offloading for intelligent workshops with a cloud–edge–terminal architecture. The proposed approach integrates manufacturing task scheduling with computational resource allocation to minimise production makespan and computation latency simultaneously. By modelling the interactions between physical production processes and distributed computing resources, the method provides coordinated optimisation strategies for intelligent manufacturing systems. Experimental results demonstrate that the proposed method effectively improves production efficiency while reducing computation delay in cloud–edge–terminal environments.

This paper presents a framework for manufacturing system reconfiguration that integrates artificial intelligence techniques with digital twin technology. The proposed framework combines intelligent decision-making with virtual system representations to support adaptive reconfiguration, performance evaluation, and what-if analysis in complex manufacturing environments. By leveraging AI-driven reasoning and digital twins for real-time system monitoring, the approach enhances flexibility, scalability, and responsiveness in manufacturing system reconfiguration. Results presented at FAIM 2022 demonstrate the effectiveness of the proposed framework in supporting intelligent and adaptive manufacturing systems.

This paper proposes a global interactive attention-based lightweight denoising network for accurately locating internal defects in CFRP laminates. The proposed method integrates global contextual feature extraction with interactive attention mechanisms to enhance defect-related representations while maintaining a lightweight network structure. By effectively suppressing noise and highlighting defect features, the network achieves improved localization accuracy under limited computational resources. Experimental results demonstrate that the proposed approach outperforms existing methods in defect detection performance, making it suitable for real-time and industrial nondestructive testing applications.