Abstract In recent years, the rapid development of large-scale satellite constellations has challenged the mass production capabilities of satellite manufacturers, especially in assembly. To this end, satellite manufacturers are working to construct the satellite mass assembly shop-floor to enable mass assembly. However, there is still a lack of a modularized manufacturing system oriented to flexible production for satellite mass assembly, as well as production control methods and logistics scheduling methods. This presentation first provides an overview of the current state of the satellite assembly and analyzes its needs. Then, the theory of digital twin-based satellite mass assembly is introduced. Based on the theory, the key theologies of digital twin-based satellite mass assembly are presented, including digital twin enhanced satellite mass assembly process management and control method, digital twin-based satellite mass assembly shop-floor logistics distribution method. Finally, a case study in a real satellite mass assembly shop-floor are introduced, and the system application verifies the feasibility and effectiveness of the proposed theory and technologies.

Abstract Digital twins have made significant progress and are now widely applied across various fields. By enabling real-time data interaction, digital twins can replicate physical environments in a safe and controlled virtual space. As research into digital twins continues to expand, a range of technologies and tools are being employed in their development. However, there is still a lack of standardized procedures or methodologies to guide the creation of digital models, the integration of these models with physical entities, and other key steps in the development process. This report analyzes the need for standards in digital twin development and introduces existing standardization systems in this domain.

Abstract Fault diagnosis and Remaining Useful Life (RUL) prediction of rolling bearings, the key rotating components of complex equipment, are important means to ensure normal operation of equipment, prevent unplanned downtime, and improve the maintenance efficiency of the equipment. The fusion of data-driven models and physics models is gradually becoming the mainstream trend for solving the problem of equipment fault diagnosis and prediction. Currently, various fault diagnosis and prediction methods based on models and data-driven methods have been proposed and successfully applied in different scenarios. However, due to the multifactorial and nonlinear characteristics of bearing performance degradation, it is difficult to establish an accurate physics model to describe the evolution mechanism of bearing performance degradation. Additionally, the effectiveness of pure data-driven methods relies on the quantity and quality of training samples, neglecting fault and failure mechanisms, and the interpretability of analysis results is poor. There is an urgent need to study the problem of rolling bearing fault diagnosis and RUL prediction under the conditions of incomplete data and lack of fusion with mechanisms to overcome the limitations of a single method, and to develop a high-precision multi-type Data Model Fusion (DMF) method.

Abstract To be confirmed

Abstract In recent years, as a promising way to realize smart manufacturing, digital twin shop-floor has attracted more and more attentions. Frontier researches have preliminarily shown that the interaction, which is the core feature of digital twin, is beneficial for dynamic analysis, real-time production management and remote shop-floor control. However, current research pays scant attention to the seamless control under uncertain conditions, which could lead to outdated or ineffective control because of the interaction delay and uncertainty. To address this problem, this paper firstly proposed a digital twin shop-floor construction framework towards seamless control under uncertain conditions. Moreover, connotations of seamless and resilient control are also introduced. Then, the Lego-style modeling and configuring method of digital twin shop-floor are discussed, to provide the basis for digital twin shop-floor development and seamless control. Predictive interaction mechanism for resilient control is further explained in detail. Finally, a digital twin shop-floor for chemical fiber production is chosen as the case to validate the effectiveness and feasibility of proposed framework and method.

Abstract The approach proposes a self-attention-enhanced Async-MAPPO method to tackle unstable cross-domain communication and delays in digital-physical fusion networks. The approach integrates traffic shaping, network twinning, and traffic scheduling, incorporating a frame preemption mechanism to prioritize critical data. A digital twin provides a simulated environment for algorithm training, while a multi-agent Actor-Critic structure with a multi-objective reward function optimizes latency and synchronization. Experiments on a heterogeneous platform confirm superior performance under dynamic congestion compared to baseline methods.

Abstract Currently, digital testing technology has emerged as a vital approach for aerospace equipment, with accurate modeling serving as the foundation for ensuring the credibility and reliability of its outcomes. Modeling methods range from conventional techniques, such as CFD and MATLAB simulations, to deep learning-based techniques. While the former depends on mathematical principles for characterization and the latter utilizes neural networks, research on these approaches often remains segregated, lacking sufficient exploration of the model fusion. To address this gap, this paper proposes a multi-dimensional digital-physical fusion modeling method for aerospace equipment testing. Consequently, with a fixed-wing UAV as the research subject, system model based on SysML, geometry model and mechanism model for flight simulation of UAV are constructed separately. Then, to enhance the fidelity of the mechanism model, a deep learning-based lift coefficient prediction model, was developed utilizing physical wind tunnel testing data of airfoils, which was then integrated with the mechanism model to establish a digital-physical fusion model. Finally, the mechanism model, data model, and fusion model are validated, demonstrating the effectiveness and adaptability of the proposed method.

Abstract In manufacturing multi-physics products such as Pilot Solenoid Valves, stable yield rates have been achieved, yet assembly quality mechanisms remain unclear due to multi-source disturbances, coupled errors, and latent variations. Early detection and tracing of errors remain difficult, limiting further improvements. This paper proposes a Digital Physics Factor Fusion (DPFF)-driven assembly error testing method, integrating error modeling, early detection, and causal tracing. A DPFF framework is constructed to fuse physical and digital domains, and a symbolic directed graph (PSV-SDG) models causal relations and error propagation in the assembly process. An enhanced Principal Component Analysis (DPCA) with anomaly screening improves weak anomaly detection, while a reverse inference strategy with dynamic division and Bayesian path optimization enables efficient error tracing. Validation on a PSV production line reveals latent aging and sealing deviations, raising batch yield from 91% to 94%. Results demonstrate DPFF’s strong error perception and closed-loop control, with broad applicability to complex assemblies.

Abstract CNC machine tools constitute the core equipment of the manufacturing industry. While most CNC machine tools are capable of achieving micron-level accuracy at the time of delivery, the production accuracy of CNC machine tools decreases too much after long-term service, thereby constraining the advancement of high-quality manufacturing. To address the challenge in the performance analysis of CNC machine tools, this presentation proposes a data model fusion approach for service performance analysis within a digital twin framework. Specifically, this presentation illustrated a theoretical framework of data model fusion, and a method of the service performance modeling of digital twin machine tools. As a case study, a data model fusion method for motorized spindle thermal error prediction is introduced.

Abstract As a way of achieving the integration of cyber-physical systems, the digital twin shop-floor (DTS) provides extensive data about shop-floor operations through dynamic interaction. Operational data collected during the manufacturing process is characterized by its multi-dimensional, heterogeneous, and time-series nature. To fully leverage this information and facilitate informed decision-making, it is essential to develop an effective method for representing the data in a unified and integrated manner. This report proposes a knowledge graph-based operational data representation method for the DTS. The knowledge graph captures the entities and relationships related to the DTS, providing semantic integration of production line and process information. The resulting data representation facilitates DTS operational data analysis and cognitive decision-making promoting smart production during the manufacturing process. Additionally, the report focuses on 3C intelligent manufacturing and explores issues related to data fusion and production control in 3C DTS.

Abstract This study develops a digital-twin-driven feature fusion framework to predict surface roughness in aerospace curved component milling. It integrates heterogeneous data via 2D CNNs (vibration time–frequency features), Bi-LSTMs (cutting force sequences), and embedding layers (static parameters). A heterogeneous graph models all feature interactions, with graph attention dynamically weighting multi-factor dependencies. This data-physics fusion paradigm enhances prediction accuracy beyond traditional methods, supporting intelligent manufacturing of high-performance aerostructures.

Abstract Using transfer learning to predict the quality of different types of industrial products is a direction worth studying. Traditional machine learning models often struggle to generalize across product types due to differences in features and production processes. By pretraining a model on a large dataset from a primary product type, transfer learning allows the model to be fine-tuned with smaller datasets from other product types, leveraging shared knowledge and reducing the need for extensive data collection. The method explores transfer learning strategies in processing quality prediction of drying process in process industry and demonstrates its effectiveness in improving quality prediction accuracy, even with limited data from product categories. This approach enhances efficiency and reduces costs in industrial quality control.

Abstract Immersive teleoperation has become a critical approach for enabling embodied intelligent robots to perform complex tasks in dynamic and uncertain environments. The research presents a digital-twin-enhanced immersive teleoperation platform that integrates a master terminal, a physical slave robot, and a digital twin embodiment. The platform applies haptic and multimodal perceptual interaction, combining force, tactile, and visual feedback to enhance operator immersion, transparency, and control accuracy. The digital twin serves as a real-time virtual proxy, enabling predictive modeling, state monitoring, and adaptive optimization of robot behaviors, thereby improving reliability and safety in teleoperation tasks. Experimental validation in representative scenarios demonstrates that the proposed system significantly enhances task performance, reduces operational uncertainty, and supports scalable human–machine collaboration. The framework provides a promising foundation for advancing intelligent teleoperation in manufacturing, healthcare, rescue, and service robotics.

Abstract Configurable modeling and simulation of Production Logistics Digital Twin Systems (PLDTs) represents a promising research direction. However, when academic researchers and industrial practitioners engage in developing and simulating such PLDTs, they often encounter challenges including the complexity of constructing geometric models, intricate operational and collaboration mechanisms, and prolonged adaptation cycles for scheduling algorithms. We proposes a configurable modeling and simulation approach for production logistics digital twins, utilizing a "Part Tree-Mechanism Component Tree" structure to model complex workshop equipment, enables dimension parameter configurability and multi-degree-of-freedom animation configurability, facilitates rapid and flexible configuration of mechanism Model. An "Equipment-Controller-Statistician" architecture is adopted to represent process mechanisms, scheduling algorithms, and production capacity KPIs of production logistics workshops, forming an integrated simulation framework.