Abstract A reliable human-robot workcell relies on accurate and nearly real-time updated models, especially in a constrained yet dynamic environment. This paper investigates digital twin-driven human-robot collaborative assembly enabled by function blocks. Leveraging sensor data, digital models are developed to precisely mimic physical human-robot collaborative settings supported by a digital-twin architecture. An advance-execution twin system based on the current status through real-time condition monitoring performs assembly planning and adaptive robot control using a network of function blocks. An augmented reality-based interaction method using HoloLens further facilitates human-centric assembly. An engine-assembly case study is performed to validate the effectiveness of the system.

Abstract Digital transformation is not merely the deployment of technologies; it is a shift in how complex systems are designed, interconnected, and evolved. Within this paradigm, digital twins have emerged as a key engineering construct, which simultaneously functioning as system components and as systems in their own right, capable of interacting, integrating, and co-evolving across digital and physical spaces. Digital twins were initially studied in manufacturing environments, where systems are relatively structured and bounded. In recent years, their research scope has expanded toward Logistics and Supply Chain Management (LSCM), a domain that is inherently systemic, distributed, and integration intensive. These characteristics imply that digital twins cannot be understood only as replicas of isolated assets, but rather as interacting entities embedded in networks of operations and decision processes. At the same time, both business research and digital twin development increasingly emphasize a service perspective. However, when moving from manufacturing to LSCM, several fundamental questions remain insufficiently structured: What types of services are actually required across supply chain contexts? Where in LSCM are these services applied? And what capabilities and enabling technologies are necessary to support them?

Abstract Urban air quality management suffers from a "Qualitative-Quantitative Divide", where rich sensor data remains disconnected from scientific literature and computational simulations. This fragmentation creates an epistemological impasse in which planners cannot feasibly test the thousands of potential intervention combinations described in policy reports using resource-intensive, high-fidelity models. To bridge this gap, this presentation presents a knowledge-infused Agentic AI framework designed to orchestrate a "Continuous Evidence Cycle" within the urban digital twin ecosystem. The framework utilises a three-pillar architecture: an ontology-driven knowledge graph that acts as a machine-readable plausibility filter for policy hypotheses; a reduced-form simulation pipeline capable of performing ex-ante causal testing in under 60 seconds; and a Retrieval-Augmented Generation Trust Anchor that grounds real-time sensor anomalies in a localised qualitative context. By transitioning from static decision-support tools to autonomous, reasoning-capable agents, this work provides a technical blueprint for proactive urban governance that integrates empirical observation with computational projection to protect public health.

Abstract Optimizing the forging process, crucial in advanced manufacturing, requires precise control of the forging speed to enhance quality and efficiency beyond the limits of constant-speed strategies. However, existing modeling methods struggle to capture the spatiotemporal evolution of high-dimensional physical fields, while conventional control approaches are either too slow or rely heavily on process repeatability, limiting their applicability under dynamic forging conditions. Addressing the challenges characterized by high dimensionality, nonlinearity, and the fact that its internal state is unmeasurable in aluminum alloy forging, this paper proposes a "slow-then-fast" variable-speed strategy. We leverage offline reinforcement learning (Offline RL) and a deep operator network (DeepONet) to build a surrogate model for fast multi-physics prediction. This model generates an optimal, constraint-satisfying speed profile in a high-dimensional action space, presenting a novel solution for high-performance forging.

Abstract With growing interest from academia and industry in sustainable investment, ESG disclosure in the apparel industry has emerged as a strategic imperative that balances economic development and environmental protection, yet the data authenticity of such disclosures remains a critical and unresolved challenge. In this study, we propose FSSL-DA to address the challenge. We first propose an ESG reporting system (BI-ESG) integrating blockchain and IoT to achieve automated data collection while ensuring data authenticity, consistency, and transparency. Second, within the BI-ESG system, spatial-temporal clustering based on Federated Semi Supervised Learning (FSSL) is adopted in the event modeling methodology. Third, we conduct FSSL-based data authenticity analytics with the spatial-temporal analytics integrated into the algorithms to generate labelled data for federated training. Additionally, a real-world case study is conducted in a multi-plant apparel enterprise, combining 12 months of operational data with synthetically injected anomalies to generate 216,000 minute-level samples. Experimental evaluation identified the optimal configuration which achieved 97.1% accuracy and 97.0% F1-score in detecting inauthentic ESG records. Sensitivity analyses confirm robustness across client numbers, data windows, and under differential privacy. This work demonstrates a scalable, privacy-preserving solution for ESG data verification, offering practical benefits for sustainable supply chains.

Abstract The energy consumption characteristics of mechanical machining system are complex, and equipment conditions exhibit strong time-varying properties, which makes it difficult to monitor the energy efficiency in real time. A digital twin-enabled energy efficiency monitoring method for mechanical machining system is proposed to solve this problem. The multi-source energy consumption characteristics of machining system are analyzed, and the energy efficiency monitoring index of machining system is constructed. A multi-source heterogeneous data acquisition method for machining system is presented, and a real-time machining workshop mapping model based on event and data-driven is established. The collision detection and Boolean operation are used to dynamically judge the running state of the machine tool, and the energy efficiency monitoring method of the machining system based on digital twin is proposed.

Abstract The proposed address resolution protocol (CPI ARP) is designed for the cyber–physical Internet (CPI) environment, integrating physical and digital logistics networks. By extending traditional ARP mechanisms, CPI-ARP addresses the practical requirement for coordinated asset tracking and state awareness of vehicles and physical shipment units (PSUs) within global logistics systems. The protocol enables coordinated interaction between digital and physical operations by linking logical network addresses (PIP) with physical addresses (PMAC) and physical locations. Key challenges related to synchronizing network and physical address systems, including scalability, mobility, and real-time data synchronization, are addressed in this protocol. Performance evaluations through simulation experiments under various network conditions demonstrate CPI ARP’s ability to reduce latency, improve packet delivery, and support logistics system responsiveness in intracity and cross border contexts. These findings pave the way for further development and standardization of CPI networks, contributing to the optimization of global logistics operations.

Abstract Indoor positioning is a key enabler of spatial–temporal digital twin, providing continuous physical-to-digital synchronization in complex indoor environments. Although Bluetooth Angle-of-Arrival (AoA) positioning is attractive for large-scale deployment, its integration into digital twin systems is hindered by limitations in positioning accuracy, deployment cost, and system latency. This paper proposes a digital twin–driven Bluetooth AoA indoor positioning system, in which indoor localization functions as a core spatial–temporal sensing layer of the digital twin. A LiDAR-IMU-guided data collection strategy is employed only during system initialization to automatically generate high-precision ground truth for model training. A Transformer-based network learns robust mappings from raw Bluetooth signals to spatial angles, which are fused across multiple anchors via Least Squares triangulation to estimate positions. The system is designed for deployment on edge devices, enabling real-time synchronization between the physical environment and its digital twin. Experimental results indicate sub-meter localization accuracy and end-to-end latency below 50ms on edge devices, validating the system as a practical spatial–temporal perception backbone for indoor digital twin applications.