Abstract The performance digital twin of production system is crucial and indispensable for the analysis and prediction of production performance parameters (e.g., bottlenecks, productivity) and the optimization of system configurations (e.g., order quantities, labor/machine numbers). The simulation models of these digital twins are typically constructed based on the discrete event modeling method, which can describe the logical and behavioral relationships among various objects within the system. Currently, a variety of commercial discrete-event simulation software tools (e.g., AnyLogic, FlexSim, Plant Simulation) are used to develop simulation models for the performance digital twins of production system. Generally, modeling and simulation is an expert job requiring substantial knowledge and experience. This report discusses data-driven simulation modeling and large language model (LLM)-assisted simulation modeling, and summarizes their respective advantages and disadvantages. With a focus on digital twin-oriented discrete event simulation modeling supported by large language models, this paper gives a comprehensive review of text2code and text2file model generation methods, presents text2module generation method for more intuitive and simple interaction with simulation software. Examples are given to show the process of LLM-assisted simulation modeling. The potential use of LLM in different phase of simulation study are also discussed.

Abstract Transparent, credible, traceable, and standardized carbon footprint (CF) data play a crucial role in low-carbon product design and manufacturing process optimization. However, due to the large number of components, diverse manufacturing processes, and various lifecyle scenarios of mechatronic products, it leads to challenges in CF data acquisition and modeling. These issues hinder the decision-making of low-carbon design and manufacturing across the entire product lifecycle. To this end, this research systematically investigates CF data acquisition, lifecycle modeling, optimization and decision-making. For the CF data, we review the existing data sources and acquisition methods, and proposed an efficient product CBOM (Carbon Bill of Materials) construction and acquisition approach by integrating enterprise information system. For the CF modeling, we analyze the pros and cons of hierarchical models and complex network models in product lifecycle carbon information expression, and propose a data-model co-driven framework for low-carbon product design. This framework could assists engineers in CF data acquisition, prediction, and traceability optimization during the design phase.

Abstract This presentation addresses the challenges of information silos and semantic inconsistency in alumina smart manufacturing by proposing a digital twin-driven information model system. We introduce a three-layer architecture integrating physical entities, mechanistic models, and data-driven models. Through self-developed tools, equipment and processes are standardized into semantic information models and dynamically instantiated as OPC UA services, creating a virtual twin that strictly maps the physical line. The system validates semantic interoperability, enables plug-and-play integration, and supports dynamic simulation through a virtual-physical closed loop, demonstrating enhanced process transparency and decision support. This work provides a technical pathway for building evolvable, knowledge-embedded digital twins in process industries.

Abstract Precise assembly of composite fuselages is critical for aircraft assembly to meet the ultra-high precision requirements. Due to dimensional variations, there is a gap when two fuselages assemble. In practice, actuators are required to adjust fuselage dimensions by applying forces to specific points on fuselage edge through pulling or pushing force actions. The positioning and force settings of these actuators significantly influence the efficiency of the shape adjustments. This talk introduces a reinforcement learning (RL) framework that enables sequential decision-making for actuator placement selection and optimal force computation. Specifically, our methodology innovatively formulates the actuator placement problem as a submodular problem, where the sub-modularity properties can be adopted to efficiently achieve near-optimal solutions. Then we employ the Dueling Double Deep Q-Learning (D3QN) algorithm to refine the decision-making capabilities of sequential actuator placements via our designed orthogonal space. The proposed methodology has been comprehensively evaluated through numerical studies and comparison studies, demonstrating its effectiveness and outstanding performance in enhancing assembly precision with limited actuator numbers.

Abstract With increasing safety requirements in elevator systems, traditional maintenance methods are insufficient for dynamic risk assessment of braking systems characterized by strong multiphysics coupling and transient behaviors. This study proposes a digital twin framework for elevator braking systems that integrates mechanism-based modeling, real-time sensing, and data-driven analysis to enable dynamic mapping between physical and virtual entities. A braking-unit-oriented model has been developed to analyze coupled thermo–mechanical performance during braking. Multi-source operational data are acquired in real time and fused with mechanism models and deep learning techniques to enhance prediction accuracy and response speed. This proposed framework supports risk evaluation under complex operating conditions and improves the reliability of performance monitoring. Overall, this approach offers theoretical insights and practical guidance for the intelligent operation and maintenance of elevator braking systems.

Abstract Aircraft assembly traditionally relies on rigid fixtures and excessive locators to ensure geometric accuracy, often causing forced contacts, deformation, and residual stress. To achieve collaborative control of geometric and mechanical states, sensors such as FBGs, force sensors, and displacement gauges are integrated with digital twin technologies for real-time perception and regulation.This work presents three representative applications: (1) deformation sensing of reconfigurable fixture locating arms using FBG-based strain measurements and shape reconstruction; (2) shape sensing of thin-walled aircraft panels through curvature-based surface reconstruction for deformation monitoring; and (3) online perception and collaborative control of fuselage posture and stress using hybrid stress modeling, kinematic calibration, and impedance control. Experimental results demonstrate accurate real-time sensing, visualization, and regulation of deformation, posture, and stress. The proposed framework establishes a synchronized digital twin of aircraft assembly and provides a practical reference for intelligent manufacturing applications requiring integrated geometric and mechanical state control.

Abstract This presentation introduces a multi-twin collaborative system for human-machine collaborative manufacturing, aiming to support the transition from conventional human-machine interaction to deeper human-machine integration. The central challenge addressed is the lack of dynamic interaction and collaborative decision-making mechanisms among heterogeneous digital twins in complex manufacturing scenarios. To overcome this limitation, the proposed architecture integrates four types of digital twins: the human digital twin, robot digital twin, equipment digital twin, and product digital twin. A collaborative application interaction center is further designed as the system hub to enable real-time data communication, intention understanding, collaborative scheduling, and closed-loop decision execution. This work provides a reusable technical pathway for digital-twin-enabled, human-centric intelligent manufacturing.