The Industrial Internet integrates industrial systems with advanced Internet technologies to establish an intelligent implementation platform and diversified service ecosystem for cloud manufacturing. However, the extensive user base introduces substantial uncertainties in temporal, financial, and operational requirements of cloud manufacturing tasks. While existing studies propose various solutions, their reliance on subjective criteria for demand variation analysis leads to inadequate handling of fuzzy demands. A novel fuzzy demand-oriented optimization method is proposed for cloud manufacturing service composition, employing fuzzy sets and membership functions to establish an objective quantification framework for modeling uncertain demands. The approach formulates a multi-objective optimization model incorporating four key metrics: service cost, service time, service quality, and resource utilization rate, with fuzzy satisfaction functions constructing constraints containing random variables to ensure robust realization of fuzzy demands. An enhanced NSGA-III algorithm is developed featuring opposition-based learning mechanisms and GKM-based reference point generation to enhance population diversity and convergence efficiency. Validation through benchmark functions and practical cloud manufacturing scenarios confirms the method's effectiveness in addressing fuzzy demand challenges, with the co-evolution algorithm demonstrating superior convergence and diversity performance.

Natural and safe Human-to-Robot (H2R) object handover is a critical capability for effective Human–Robot Collaboration (HRC). However, learning a robust handover policy for this task is often hindered by the prohibitive cost of collecting physical robot demonstrations and the limitations of simplistic state representations that inadequately capture the complex dynamics of the interaction. To address these challenges, a two-stage learning framework is proposed that synthesizes substantially augmented, synthetically diverse handover demonstrations without requiring a physical robot and subsequently learns a handover policy from a rich 4D spatiotemporal flow. First, an offline, physical robot-free data-generation pipeline is introduced that produces augmented and diverse handover demonstrations, thereby eliminating the need for costly physical data collection. Second, a novel 4D spatiotemporal flow is defined as a comprehensive representation consisting of a skeletal kinematic flow that captures high-level motion dynamics and a geometric motion flow that characterizes fine-grained surface interactions. Finally, a diffusion-based policy conditioned on this spatiotemporal representation is developed to generate coherent and anticipatory robot actions. Extensive experiments demonstrate that the proposed method significantly outperforms state-of-the-art baselines in task success, efficiency, and motion quality, thereby paving the way for safer and more intuitive collaborative robots.

The manufacturing industry is undergoing a profound transformation toward smart, digital, and flexible production systems under the Industry 4.0 framework. Within this paradigm, Digital Twin (DT) serves as a key enabler, bridging physical and digital domains to simulate, analyse, and optimise manufacturing operations. Concurrently, robotic systems, enhanced by smart sensor perception, Industrial Internet of Things connectivity, and adaptive control mechanisms, are increasingly deployed to handle complex and dynamic tasks. However, the evolving demands of the modern manufacturing industry require a high degree of flexibility and responsiveness, necessitating more intelligent solutions. The Robot Digital Twin (RDT) has emerged as a transformative approach, facilitating dynamic adaptation and continuous operational improvement. This review offers a comprehensive examination of the literature on RDT in manufacturing from both technology and application perspectives, aiming to provide insight for researchers and practitioners in Industry 4.0. The paper introduces a four-layer RDT system architecture and summarises how Industry 4.0 technologies, e.g., the Industrial Internet of Things, Cloud/Edge Computing, 5 G, Virtual Reality, Modelling and Simulation, and Artificial Intelligence, converge and influence the RDT system based on this architecture. Furthermore, the review covers domain-specific and system-level applications, such as assembly, machining, grasping, material handling, human-robot interaction, predictive maintenance, and additive manufacturing systems, with an analysis of their development status. Finally, the trends, practical challenges, and future research directions for RDT systems in manufacturing are summarised at different levels.

Calibration between robots and cameras is critical in automated robot vision systems. However, conventional manually conducted image-based calibration techniques are often limited by their accuracy sensitivity and poor adaptability to dynamic or unstructured environments. These approaches present challenges for ease of calibration and automatic deployment while being susceptible to rigid assumptions that degrade their performance. To close these limitations, this study proposes a data-efficient vision-driven approach for fast, accurate, and robust hand–eye camera calibration, and it aims to maximise the efficiency of robots in obtaining hand–eye calibration images without compromising accuracy. By analysing the previously captured images, the minimisation of the residual Jacobian matrix is utilised to predict the next optimal pose for robot calibration. A method to adjust the camera poses in dynamic environments is proposed to achieve efficient and robust hand–eye calibration. It requires fewer images, reduces dependence on manual expertise, and ensures repeatability. The proposed method is tested using experiments with actual industrial robots. The results demonstrate that our NBV strategy reduces rotational error by 8.8%, translational error by 26.4%, and the number of sampling frames by 25% compared to artificial sampling. The experimental results show that the average prediction time per frame is 3.26 seconds.

Digital Twin (DT) of a manufacturing system mainly involving materials and machines has been widely explored in the past decades to facilitate the mass customization of modern products. Recently, the new vision of Industry 5.0 has brought human operators back to the core part of work cells. To this end, designing human-centric DT systems is vital for an ergonomic and symbiotic working environment. However, one major challenge is the construction and utilization of high-fidelity digital human models. In the literature, preset universal human avatar models such as skeletons are mostly employed to represent the human operators, which overlooks the individual differences of physical traits. Besides, the fundamental utilization features such as motion tracking and procedure recognition still do not well address the practical issues such as occlusions and incomplete observations. To deal with the challenge, this paper proposes a systematic design framework to quickly and precisely build and utilize the human-centric DT systems. The mesh-based customized human operator models with rendered appearances are first generated within one minute from a short motion video. Then transformer-based deep learning networks are developed to realize the motion-related operator status synchronization in complex conditions. Extensive experiments on multiple real-world human–robot collaborative work cells show the superior performance of the proposed framework over the state-of-the-art.

The conventional automation approach has shown bottlenecks in the era of component assembly. What could be automated has been automated in some high tech industrial production, leaving manual work performed by humans. To achieve ergonomic working environments and better productivity, human–robot collabora tion has been adopted for this purpose through combining the strength, accuracy and repeatability of robots with adaptability, high-level cognition, and flexibility. A reliable human–robot collaborative setting should be supported by dynamically updated and precise models. For this purpose, the digital twin can realise the digital representation of physical collaborative settings through simulation modelling and data synchronisation but is limited by communication delay and constraints. This chapter will develop a digital twin-enabled approach to human–robot collaborative assembly. Within this approach, a sensor-driven 3D modelling of the physical devices of interest is developed to realise the physical-to-digital transformation of human–robot workcell, and a Wise-ShopFloor-based platform enabled by sensor data is used to develop a digital twin model of the physical human–robot workcell. Then, function blocks with embedded algorithms are used for assembly planning, decision making and robot control, and a time-ahead execution and planning approach is developed for reliable human–robot collaborative assembly. Finally, the performance of the developed system is demonstrated by a case study of a partial car engine assembly.

In the huge automotive service aftermarket, the efficient and timely supply of maintenance parts has always attracted continuous concern from automotive service enterprises and end users,which is limited by the regional collaborative efficiency and information interaction among many suppliers over the automotive parts supply networks. However, the supply networks consist of enterprises with different manufacturing capabilities, and are filled with multisource, massive, heterogeneous information that contains multiple entities and overlapping triplet relation, leading to difficulties in achieving uniform representation and adaptive understanding of information. Entity-relation extraction is essential for unified information representation.In this paper, we devise an entity relationship extraction(ERE) method based on heterogeneous graph neural networks and entity feature fusion, which treats entities and relation as nodes in a graph, and iteratively integrates node representation to identify the most suitable node features for ERE tasks. The method introduces an innovative mechanism: firstly, we extract the subject entities and fuse their features into node representations using an attention mechanism; and then, the relations and object entities are jointly extracted to achieving end-to-end triplet extraction. Experiments are conducted using parts supply chain data from partners. The results validates the effectiveness of the method and obtain outstanding performance in the automotive parts supply chain(APSC) networks.

The product quality has become increasingly important for the modern manufacturing processes. Due to the measurement delay, data-driven soft sensor models are usually built for the quality prediction in advance. While most prior works develop the customized model for a specific scenario, some recent works explore the adaptive mechanisms for the model to tolerate the online changes. However, they either tackle the operational variations due to changing product specifications for market demands, or deal with the latent variations due to process uncertainties such as sensor degradation. To improve the generalization towards diverse processes with both variations, a novel slow-fast dual-branch method inspired by the complementary learning systems in neuroscience is proposed for the first time. The slow branch is composed of an enhanced multi-layer perceptron with attention-based embedding fusion and memory aware synapses to grasp and consolidate the long-term global knowledge under non-independent and identically distributed data samples. The fast branch contains a modified broad learning system with maximum correntropy criterion and adaptive sample weights to rapidly track the short-term time-varying patterns. The two branches are integrated via feature sharing and refined gradient boosting to mimic the interactions between neocortex and hippocampus of brain. Extensive experiments on three real-world manufacturing processes from distinct industries show the superior performance of proposed method over 15 state-of-the-art methods.

Due to the wide application of industrial robots (IRs) in the manufacturing industry and their significant energy consumption (EC), predicting EC under different trajectories and working conditions has attracted increasing attention. Data-driven modeling methods have proven to be a viable approach for revealing the quantitative relationship between IR operating parameters and EC. However, in manufacturing systems, the coexistence of numerous heterogeneous IRs necessitates a substantial amount of data with power labels and sufficient hardware computing resources to model the operational EC of each robot type. Motivated by these requirements, this paper proposes a transfer learning based method for modeling the operational EC of IRs. Based on an analysis of the temporal causal relationship between model input variables and operational EC, a time series information feature extraction method and an industrial robot operational energy consumption prediction network (ROEPN) are proposed, which combines layer normalization (LN), long short-term memory neural network (LSTM) and masked multi-head attention mechanism (MHA). Moreover, a rigorous pre-training-fine-tuning transfer learning scheme is designed and implemented on the target domain data, effectively achieving the transfer of ROEPN from the source domain to the target domain. Experiments were conducted on the HSR-JR612 and HSR-JR603, and the results demonstrate that the proposed EC model transfer method can predict EC for different IRs, trajectories and loads, with the mean absolute percentage error (MAPE) being less than 2.69% in the case of small samples.

Traditional scheduling less account of human-related dynamic events: worker skill degradation and worker mandatory rest. However, in actual production, workers experience fatigue accumulation that decreases work efficiency, thereby decreasing the precision of jobs, increasing rework rates, and even elevating processing risks. It conflicts with the idea of industrial resilience and human well-being for Industry 5.0. Therefore, a humancentric dynamic distributed flexible job shop scheduling problem (HDDFJSP) has been researched in this paper. Firstly, a multi-objective mathematical model of HDDFJSP is proposed to minimize makespan, worker fatigue, and scheduling deviation. Secondly, a Q-learning improved differential evolution (QLIDE) is designed to solve the HDDFJSP. In the QLIDE, a new four-layer encoding method and two initialization strategies are proposed to generate a high-quality initial population and a novel mutation strategy and two auxiliary mutation methods are designed to enhance the algorithm's exploitation capabilities. Furthermore, three neighborhood search strategies are introduced and combined with mutation operations as part of the Q-learning action phase to improve population convergence and diversity. Thirdly comparative test with four other well-known algorithms has been conducted and the results demonstrate the significant superiority of the QLIDE. Finally, the QLIDE is applied to solve a real case of a labor intensive hydraulic cylinder manufacturing enterprise. The results indicate that considering rescheduling can effectively help production managers to handle dynamic event of humans during the intelligent manufacturing systems.