With the rapid advancement of generative artificial intelligence, embodied intelligence is increasingly being integrated into robotic tasks within smart manufacturing. Industrial operations typically feature long-horizon sequences and demand high precision. However, although embodied models, particularly Vision-Language-Action (VLA) models, have shown significant promise in autonomous control with superior generalization capabilities, their inherent black-box nature lacks the transparency and trustworthiness essential for manufacturing tasks. To address this challenge, we propose the Industrial Embodied Intelligence Trustworthy Interpretable Agent (IEI-TIA), a high-level trustworthy supervisory agent designed for industrial sorting and palletizing tasks. Rather than directly generating low-level control commands, IEI-TIA monitors robotic execution, diagnoses possible failure types at key operational stages, and provides corrective next-step instruction via invoked skills, thereby improving the reliability and interpretability of long-horizon and repetitive robotic tasks. An alignment-enhanced parameter-efficient finetuning method is proposed utilizing paired vision-language data to enable robust trustworthiness identification across various possible robotic failure types in different subtasks in sorting and palletizing. Additionally, we construct a specialized dataset and establish benchmarks for industrial embodied intelligence behavior diagnosis and instructions. Experimental results demonstrate that the fine-tuned trustworthy agent achieves 93.7% diagnosis accuracy for various task failures, representing a substantial improvement of approximately 38% over baseline general-purpose large models. By integrating the agent’s step-by-step operational instructions, the success rate of robotic long-horizon and repetitive tasks is improved by 14.0%.

To meet higher requirements of flexible manufacturing, smart manufacturing is developing to intelligently deal with changing demands in product customization with better generalization and adaptation. For instance, robotic systems are anticipated to realize embodied and spatial intelligence in manufacturing, to intelligently generalize in handling diverse objects in changing environments. However, the insufficiency of 3D scene data significantly hinders embodied and spatial intelligence learning. Therefore, based on digital twin, the digital genealogy is proposed to generate more diverse synthetic data, rather than synchronizing the same scene with physical world by digital twin. Also, the digital genealogy focuses on the whole evolution process from industrial parts to products in manufacturing, rather than narrowly focusing on current state in digital twin. In digital genealogy, DG-DNA for various industrial parts, similar with biology, is proposed to constrain reliable generation results of parts. To generate digital genealogy scenes with diverse industrial parts, parts matching and generation methods are both adopted with constraints of DG-DNA. Specifically, an artificial intelligence generative algorithm, named DGIP-Gen, is proposed to generate target industrial part given specific DG-DNA. The experimental results have demonstrated the generated parts are diverse and meet the specific constraints of different DG-DNA requirements, to support embodied and spatial intelligence learning.

The task planning and motion planning problems in multi task and multi robot collaboration are usually considered as independent parts to be solved. However, in long-horizon, multi-step collaborative scenarios, this independent solution approach is difficult to effectively handle the coupling problem between task planning and motion planning, such as the inability to simultaneously consider robot motion collisions during task allocation. To address these limitations, we develop a unified hierarchical reinforcement learning framework that enables agents to learn effective policies in multi task and multi robot collaborative motion planning, supplemented by two techniques: 1) using a shared graph attention network and distributed police networks to assign target tasks to each robot at a high level, and 2) training decentralized motion planning policies for each robot at a low level to control the workspace state and target end actuator posture. The high-level and low-level policies are trained in parallel and stabilized through expert dataset guidance. To verify the effectiveness of the method, we conduct experiments on a general object placement platform and an engine hydraulic column assembly platform. The results indicate that this method can more efficiently complete multi task and multi robot collaborative work.

In industrial applications, large models have exhibited superior generalization capabilities that are unattainable with smaller models. However, when faced with edge scenarios and highly diverse industrial samples, their deployment remains challenging due to high computational costs and unreliable output. To address these challenges, we propose CoLLM, a fuzzy large-small model collaborative framework, which dynamically selects between small and large models based on the characteristics exhibited by the samples. Specifically, this approach estimates uncertainty from input samples to guide model selection: low-uncertainty samples are processed by the small model for efficiency, while high-uncertainty or complex samples are routed to the large model for improved accuracy. It first constructs a fuzzy decision-making agent based on the fuzzy neural network (FNN) to assess sample complexity and determine the appropriate model for inference. Furthermore, a self-reflection mechanism is proposed to refine the large model's output, reducing the risk of unreliable output. Experimental results in industrial time series datasets demonstrate that our framework improves the computational efficiency of large models up to 14.54x while maintaining or improving prediction accuracy.

3-D modeling technologies play a crucial role in modern manufacturing. 3-D models are often exported as boundary representations for compatibility and data protection, which remove the modeling history and limit editability. To restore modeling sequences from such models, researchers employ neural networks to infer the possible modeling steps. This approach needs a large amount of labeled sequence data, and annotating such data is time-consuming. To address this issue, we propose a self-supervised pretraining method that generates modeling sequences directly from boundary representation models. Training data are first generated using a heuristic modeling sequences generation algorithm. Before training, each B-rep model is preprocessed into a zone graph representation. We then introduce the modeling operation evaluation network, which extracts features and scores each candidate operation to sequentially reconstruct the model. By selecting the most suitable operation at each step, the network progressively reconstructs the modeling sequence. This approach effectively reconstructs modeling sequences, restores the editability of B-rep models, and significantly reduces the reliance on labeled data.

Accurate and fast industrial time-series prediction is essential for safe and reliable operation of industrial equipment. Recent deep learning methods enable extracting complex temporal patterns by utilizing large-scale parameters and multiresolution feature extraction. However, they cause substantial computational complexity and limit their application at the edge. In this article, we design an adaptive temporal-channel multiresolution network (AMR-Net) that dynamically adjusts time-series resolution to avoid redundant computation. The motivation is that low-resolution feature representations are sufficient for predicting “easy” samples, whereas “hard” samples require high-resolution features to capture fine-grained information. For the AMR-Net, time series are initially input through a temporal-channel resolution decomposition (TCRD) module, which efficiently extracts low-resolution representations. Samples exhibiting high prediction confidence are expedited through early exit mechanisms, avoiding further processing. Meanwhile, high-resolution subnetworks capture the fine-grained information to discern the “hard” samples. Experiments on CMAPSS and N-CMAPSS datasets demonstrate that AMR-Net can improve computational speed by 15x while maintaining high accuracy.

With the continuous expansion of industrial systems, multisource and heterogeneous industrial data have increased rapidly, making the construction of a structured industrial knowledge system a core requirement in the industrial domain. Industrial knowledge graph (IKG) serves as a key approach for knowledge structuring and relation modeling and has become an indispensable foundation for industrial tasks. However, existing IKG construction methods still face core challenges such as data heterogeneity, complex semantic understanding, frequent knowledge changes, and limited automation. Inspired by the construction of IKG by industry experts, we propose CoMA-IKG, an large language model (LLM)-driven collaborative multiagent framework for automated construction of IKG. In the industrial data processing stage, an LLM-driven adaptive chunking agent is developed to achieve semantically complete and self-adjusting segmentation. In the triple extraction stage, a cluster of LLM-driven agents for progressive triple reasoning extraction and mechanism-aware logical discrimination is constructed to enable accurate industrial triple extraction under stepwise reasoning and industrial mechanism constraints. In the IKG evolution stage, an LLM-driven co-evolution agent is developed to generate evolution commands automatically based on the structural state of the IKG and real-time industrial data changes, enabling autonomous updating and continuous evolution of the IKG. Experimental results show that CoMA-IKG significantly outperforms existing automated knowledge graph construction methods in terms of relation mining, logical reasoning, and dynamic evolution of the IKG.

The distributed flexible job shop scheduling problem (DFJSP) has emerged as a critical challenge in the field of scheduling optimization due to its intricate resource allocation and the demand for production–logistics collaboration across multiple factories. However, most existing studies related to DFJSP only focus on the production and transportation process of jobs within a single factory, while neglecting the cross-factory logistics and the heterogeneous characteristics of transportation resources. Therefore, this article first investigates the distributed flexible job shop scheduling problem with heterogeneous transportation (DFJSPHT) resource constraints and proposes an end-to-end deep reinforcement learning (DRL) scheduling method to minimize the makespan. An innovative heterogeneous disjunctive graph model is constructed to uniformly represent the states of factories, machines, operations, and transportation resources in DFJSPHT, and the scheduling process is modeled as a Markov decision process (MDP). Next, a resource release strategy is developed to enhance the efficiency of transportation resources. To enhance the feature expression ability of the model, a graph neural network (GNN) is employed to capture the problem characteristics, and the policy network is trained using the proximal policy optimization. Comparative experiments are conducted on synthetic and benchmark instances demonstrate that the proposed method outperforms the classical priority scheduling rules and two popular DRL-based scheduling methods in solving DFJSPHT, with performance improvements exceeding 10% in most instances.

Grasping tasks are crucial in industrial manufacturing, where precise and efficient object grasping ensures smooth assembly processes and stable operations. Robotic arms, deployed in industrial environments, require timely and accurate computations to perform these tasks. This paper introduces BGRN, an RGB-D fusion-based binary grasp prediction network designed for lightweight grasp pose prediction in such settings. We propose a binary grasp pose prediction framework that significantly reduces memory usage by quantizing both weights and activations to 1 bit. Additionally, an interaction fusion module improves the integration of RGB and depth images, while an information recovery connection helps mitigate feature loss caused by binarization. Experimental results show that BGRN achieves competitive accuracy and notable reductions in memory usage and computational load compared to full-precision models.

Accurate demand forecasting for products in the parts supply chain system (PSCS) is critical for enterprises to optimize production and inventory operations. To address the challenges of inadequate modeling of complex inter-product relationships and the limited accuracy of existing forecasting methods, this paper proposes a demand forecasting method based on a Graph Convolutional and LSTM Network with Embedded Multi-Head Self-Attention (GL-MHSA). The method first extracts hybrid distance features, including Euclidean and pattern distances, from product sales data in the PSCS to mine product associations and construct graph-structured relational data. A Graph Convolutional Network (GCN) is then used to capture structural association features among products, while an LSTM network models the temporal dependencies in the demand sequences. The extracted features are fused through a Multi-Head Self-Attention (MHSA) mechanism to obtain a comprehensive feature representation. This representation is concatenated with other auxiliary features to form the final input for demand prediction. Experimental results on an automotive PSCS dataset show that the proposed GL-MHSA model achieves more accurate modeling of product associations and significantly improves demand forecasting performance compared to existing approaches.

In the context of Industrial Internet of Things (IIoT), time-series data is essential for maintenance and operational efficiency. However, challenges in IIoT data transmission, such as network instability, and in data annotation, like the high costs, lead to a shortage of high-quality labeled data, hindering system performance and industrial intelligence. Although traditional methods, such as signal imputation and denoising, have been employed, generative artificial intelligence (GAI) offers new possibilities for the generation of industrial time-series data. To address these challenges, we propose a novel score-based diffusion architecture specifically designed for industrial data generation. The score-based approach effectively leverages the gradients of the data distribution, offering a more structured and stable generative process compared to generative adversarial networks (GANs). Furthermore, our model incorporates predictive-corrective (PC) samplers with Langevin dynamics annealing to further optimize the generation process. Experimental results on turbofan engine datasets demonstrate that our model overcomes the inherent training instabilities of GANs, providing a more reliable and effective method for synthesizing high-fidelity industrial time-series data.

The Industrial Internet of Things provides an opportunity for flexible and collaborative manufacturing, but introduces more risk and more communication overhead from the Internet to the industrial field. To avoid attacks from unreliable service providers and requesters, Industrial Demilitarized Zone (IDMZ) is introduced in conjunction with firewalls to provide new communication modes between edge servers and industrial devices. As the number of tasks being offloaded to the edge side increases, optimal task offloading to balance the risk and the communication overhead with limited demilitarized buffer size becomes a challenge. Therefore, this paper establishes a mathematical model for secure task offloading in the Industrial Internet-of-Things considering dense communication with different communication modes. Then, a Parallel Gbest-centric differential evolution (P-G-DE) is designed to solve this task offloading problem with a heuristic-embedded initialization strategy, a modified Gbest-centric differential evolutionary operator and a circular-rotated parallelization scheme. The experimental results verify that the proposed method is capable of providing a high-quality solution with a lower risk and a shorter execution time in seconds, compared to six state-of-the-art evolutionary algorithms.

Recently, foundation models (such as ChatGPT) have emerged with powerful learning, understanding, and generalization abilities, showcasing tremendous potential to revolutionarily promote modern industry. Despite significant advancements in various fields, existing general foundation models face challenges in industry when dealing with the data of specialized modalities, the tasks of varying-scenario with multiple processes, and the requirements of trustworthy output, which makes industrial foundation model (IFM) a necessity. This article proposes a system architecture of termed IFMsys, including model training, model adaptation, and model application. Specifically, in model training, a base model is constructed by pretraining on multimodal industrial data and fine-tuning with fundamental industrial mechanisms. In model adaptation, the base model is developed into a series of task-oriented and domain-specific IFMs through fine-tuning with representative tasks and domain knowledge. In model application, an industrial agent-centric collaboration system and a comprehensive application framework of IFM are proposed to enhance the industrial product lifecycle applications. In addition, a prototype system of the IFM, namely, MetaIndux, is delivered, with application examples presented in typical industrial tasks. Finally, future research directions and open issues of IFM are prospected. We hope this article will inspire the advancements in the theories, technologies, and applications in this emerging research field of IFM.

Industrial time series prediction is highly important for the predictive maintenance of Industrial Internet of Things devices. Deep learning methods have demonstrated state-of-the-art (SOTA) performance in the field of time series prediction. However, time series data from complex industrial scenarios often contain substantial uncertainty. This makes it difficult for deterministic deep learning models to achieve accurate predictions. Moreover, existing static methods often fail to meet the real-time requirements of industrial environments. To address the challenges, this study introduces fuzzy learning into deep learning models to overcome the drawbacks of fixed model representations. Therefore, we propose a fuzzy dynamic transformer (FDformer) that can adaptively adjust network depth according to the complexity of individual samples. Subsequently, we design a fuzzy feature extraction mechanism to capture feature information within the fuzzy membership degree, enabling the feature-level fusion of the fuzzy representation with the dynamic depth representation. Finally, we propose a training method for dynamically allocating loss weights, emphasizing the contribution of various samples to different exits, thereby improving the performance of time-series dynamic networks. Experiments on multiple datasets indicate that FDformer achieves minimal computational costs and excellent prediction accuracy across multiple datasets, outperforming SOTA algorithms.

Industrial time-series prediction is crucial for Industrial Internet of Things. Due to the complexity and variation of modern industry, knowledge transfer for varying data has been an attractive research area. However, conventional methods may overlook the intradomain distribution and the mutual information, leading to incorrect semantic alignment and loss of prediction-relevant information. To address these issues, a contrastive learning-based domain adaptation method, contrastive temporal prediction adaptation, for industrial time-series cross-domain prediction is proposed. It leverages a contrastive domain generalization and a contrastive self-supervised alignment method to obtain stable representations and capture the relationship between the data distribution and labels, to bring samples with similar labels closer in the feature space. Besides, an instancewise adversarial discrimination is developed to leverage the data distribution to mitigates interference from irrelevant information. The performance of our method is verified through experiments on CMAPSS dataset. The results demonstrate that our method outperforms existing methods.

Industrial Internet of Things (IIoT) faces significant security challenges such as data privacy and vulnerabilities. Unsupervised anomaly detection aims to identify abnormal patterns by monitoring multivariate time series data of IIoT without anomaly annotation. Previous deep-learning-based methods have high-computation cost, which hinders their deployment in edge devices. In this article, we propose a dynamic anomaly detection network (DADN), which introduces a dynamic anomaly detection mechanism to enable efficient inference. Specifically, a bilateral early-exit mechanism is designed so that each sample can dynamically exit at a certain layer during the forward process to support the anomaly judgement, and the layer where sample exits is adaptively determined at the inference stage. Experimental results show that DADN significantly reduces computational costs and enhances F1 scores in industrial anomaly-detection benchmarks, as shown by a 58.22% decrease in GFLOPS on the SWAT dataset, outperforming previous representative method (anomaly transformer).

During the machining process of parts, choosing appropriate machining strategies optimizes production costs effectively. However, when the amount of data is limited, existing neural networks often struggle to fit the data accurately. Meanwhile, existing neural networks suffer from information dilution and lack effective mechanisms for direct information transfer between nonadjacent surfaces. This article proposes a method to extract General Machining Unit data. This data improves few-shot training performance. We investigate the distribution and information flow within General Machining Units and design a new way of data augmentation. In addition, to address the information dilution, a novel wormhole mechanism is proposed to aggregate information that spans the topological connections. In the backbone, we propose the Brep-WH layer that integrates wormhole mechanisms and attention pool layers. Both the Brep-WH network and the General Machining Unit data successfully improve the accuracy of the milling strategy dataset and the Fusion 360 Gallery segmentation dataset.

The breakthroughs of next-generation information technologies have accelerated the advancement of industrial cyber-physical intelligence (ICPI), particularly in system intelligence and applications. However, this progress has also brought challenges in ensuring operational reliability and system intelligence. Anomaly detection, a critical component of fault-tolerant and intelligent ICPI, is usually addressed by treating it as a one-class classification and location problem. While autoencoder frameworks have shown promise in addressing this challenge, most existing methods usual struggle with precise anomaly identification or require resource-intensive region-based training. Furthermore, the dynamic nature of anomalies and the scarcity of labeled training data complicate the development and evaluation of anomaly detection models. In this article, an innovative tensor-based knowledge distillation approach (TKDA) is introduced, which integrates a pretrained teacher network, a tensor-decomposed student network, and a denoising module into a unified framework. Anomalies are identified and localized by analyzing differences in intermediate activation values between teacher and student networks during data processing. Extensive experiments demonstrate that TKDA addresses the limitations of low accuracy in anomaly location and inefficiency in computational processes, achieving significant improvements across diverse datasets, including F-MNIST, MNIST, CIFAR-10, MVTecAD, Retinal-OCT, and two medical datasets.

Over the past few years, multiview attributed graph clustering has achieved promising performance via various data augmentation strategies. However, we observe that the aggregation of node information in multilayer graph autoencoder (GAE) is prone to deviation, especially when edges or node attributes are randomly perturbed. To this end, we innovatively propose a tensor-representation-based multiview attributed graph clustering framework with smooth structure (MV_AGC) to avoid the bias caused by random view construction. Specifically, we first design a novel tensor-product-based high-order graph attention network (GAT) with structural constraints to realize efficient attribute fusion and semantic consistency encoding. By imposing attribute augmentation mechanisms and smooth constraints (SCs) on the proposed high-order graph attention autoencoder simultaneously, MV_AGC effectively eliminates the instability of reconstructed graph structures and learns a more compact node representation during training. In addition, we also theoretically analyze the stronger generality and expressiveness of the proposed tensor-product-based attention mechanism over the classical GAT and establish an intuitive connection between them. Furthermore, to address the performance degradation caused by clustering distribution updating, we further develop a simple yet effective clustering objective function-guided self-optimizing module for the final clustering performance improvement. Experimental results on the six benchmark datasets have demonstrated that our proposed method can achieve state-of-the-art clustering performance.

Complex industrial equipment plays a critical role in specific tasks within industrial edge scenarios. Predicting their health status accurately is essential to ensuring safety and reliability in the production process. However, real-world industrial edge scenarios often have limited resources and stringent real-time requirements, making it difficult to deploy high-precision deep learning models directly at the edge. To address this issue, this article proposes an efficient adaptive-gating binary neural network (ABNN). First, a trend-aware encoder (TAE) is proposed to optimize the binarization process of the input layer. Next, a learnable precision indicator (LPI) is proposed to adjust the inference precision level. Finally, an adaptive-gating convolution is proposed to improve the representational capabilities while maintaining the fitting ability without significantly increasing the computational cost. Additionally, a field-programmable gate array (FPGA) hardware accelerator is designed for the proposed network. ABNN achieves approximately a 7% improvement in accuracy and a 45% gain in efficiency compared to the baseline model.

Recently, deep learning-based models such as transformer have achieved significant performance for industrial remaining useful life (RUL) prediction due to their strong representation ability. In many industrial practices, RUL prediction algorithms are deployed on edge devices for real-time response. However, the high computational cost of deep learning models makes it difficult to meet the requirements of edge intelligence. In this article, a lightweight group transformer with multihierarchy time-series reduction (GT-MRNet) is proposed to alleviate this problem. Different from most existing RUL methods computing all time series, GT-MRNet can adaptively select necessary time steps to compute the RUL. First, a lightweight group transformer is constructed to extract features by employing group linear transformation with significantly fewer parameters. Then, a time-series reduction strategy is proposed to adaptively filter out unimportant time steps at each layer. Finally, a multihierarchy learning mechanism is developed to further stabilize the performance of time-series reduction. Extensive experimental results on the real-world condition datasets demonstrate that the proposed method can significantly reduce up to 74.7% parameters and 91.8% computation cost without sacrificing accuracy.

Implementing advanced AI techniques in industrial manufacturing requires large volumes of annotated sensor data. Unfortunately, collecting such data is often impractical due to extreme environments and the manual burden of expert annotation. Recent advancements in artificial intelligence generated content (AIGC) have inspired the exploration of industrial time-series generation to mitigate data shortages. However, existing AIGC models encounter difficulties in generating industrial time series due to their complex temporal dynamics, multichannel intercolumn correlations, and diverse frequency characteristics. To address these challenges, we propose MetaIndux-TS, a frequency-informed AIGC foundation model based on diffusion model frameworks. This model is designed to generate industrial time-series data under a variety of working conditions, across different types of equipment, and with variable lengths. Specifically, MetaIndux-TS integrates dual-frequency cross-attention networks, transforming time series into the frequency domain to model multivariate dependencies and capture intricate temporal details. In addition, the contrastive synthesis layer is constructed to generate high-fidelity time series by comparing periodic and long-term trends with initial noisy sequences. Comprehensive experiments show that MetaIndux-TS outperforms state-of-the-art models (SSSD, Dit, and TabDDPM), achieving a 57.5% improvement in fidelity and 20.4% in predictive score. MetaIndux-TS exhibits zero-shot generation capabilities for samples under unseen conditions, offering the potential to address data collection challenges in extreme environments. Codes are available at: https://github.com/Dolphin-wang/MetaIndux.

With the remarkable success of generative models like ChatGPT, artificial intelligence generated content (AIGC) is undergoing explosive development. Not limited to text and images, generative models can generate industrial time series data, addressing challenges, such as the difficulty of data collection and data annotation. Due to their outstanding generation ability, they have been widely used in Internet of Things, metaverse, and CPSS to enhance the efficiency of industrial production. In this article, we present a comprehensive overview of generative models for industrial time series from deep-generative models (DGMs) to large-generative models (LGMs). First, a DGM-based AIGC framework is proposed for industrial time series generation. Within this framework, we survey advanced industrial DGMs and present a multiperspective categorization. Then, we systematically propose the roadmap to construct industrial LGMs from four aspects: large-scale industrial dataset, LGMs architecture for complex industrial characteristics, self-supervised training for industrial time series, and fine-tuning of industrial downstream tasks. Furthermore, we introduce an evaluation benchmark that systematically assesses fidelity, diversity, and utility. We include a case study on aircraft engine maintenance, demonstrating the application of DGMs in industrial predictive maintenance. Finally, we conclude the challenges and future directions to enable the development of generative models in industry.

Digital twin has made significant achievements in applications across various fields by bridging the gap between virtual and physical world. However, the modeling of digital twin faces challenges of labor-intensive and time-consuming manual modeling or coding. With the emergence of artificial intelligence foundation models, the integration of large language models (LLMs) and digital twin may contribute to enhancing modeling efficiency significantly. Therefore, in this paper, a framework of digital twin modeling code generation is proposed based on LLMs. Given instructions in natural language prompts, modeling code in digital twin software will be generated from LLMs. Firstly, a tree diagram-based code generation method is proposed to generate main section of modeling code. Additionally, a randomized code completion method is proposed to complete missing parts which are unspecified or unclear in prompts. Finally, a case study is given in NVIDIA Omniverse for a digital twin construction based on framework proposed in this paper.

The Industrial Internet of Things (IIoT) amalgamates cutting-edge information technologies, including artificial intelligence, big data, and cloud computing, to establish a sophisticated platform for intelligent predictive maintenance of complex industrial equipment. While numerous predictive maintenance methodologies have been proposed, much of the existing research predominantly emphasizes predictive techniques, with limited attention devoted to developing a comprehensive predictive maintenance framework. To bridge this scholarly gap, this paper proposes a novel cloud-edge adaptive framework for equipment predictive maintenance in IIoT. Positioned across the cloud, edge, and equipment planes of the IIoT infrastructure, this framework adeptly addresses challenges such as highly generalized collaborative modeling, scenario-specific modeling, and continuous dynamic evolution of equipment predictive maintenance in the Industrial Internet. Consequently, this framework offers a methodical and holistic solution to predictive maintenance for industrial equipment.

Edge computing exploits node devices situated in close proximity to terminals to deliver distributed computing services directly to users, with FPGAs serving as the predominant platform. With the amplification in volume and intricacy of deep learning models, effectively deploying these models on FPGA devices introduces substantial challenges. To counter this predicament, this paper introduces an FPGA-based hardware acceleration framework for reconfigurable edge devices tailored for Convolutional Neural Networks (CNNs). Initially, a method for dynamically quantizing network models is devised to markedly curtail model memory consumption. Subsequently, a meticulously engineered hardware acceleration unit is formulated to attain augmented computing parallelism via meticulous temporal redesign. Finally, model deployment on FPGA devices for inference verification is showcased utilizing fully connected and convolutional neural networks as exemplars. On the MNIST dataset, the FPGA inference unit attains remarkable accuracy and computational efficiency in comparison to CPUs and GPUs

Investigating the methods of handling real-world objects with robotic arms is a crucial aspect of artificial intelligence research. By inputting images from cameras or other devices, neural networks extract suitable end-effector poses to achieve precise object manipulation. Current methods face challenges of inadequate regression precision and excessive classification, which negatively impact network performance. Thus, we propose a new method based on the local grasp unit. This approach transforms grasp prediction into evaluating whether the features of local grasp units are suitable for grasping. Stable local grasp units are selected by introducing a binary grasp pose discriminator with local point clouds as inputs. In addition, to tackle the issue of distant regions in point clouds of grasp pose discriminator, we propose the innovative Attention-Bridging Layer. The Attention-Bridging Layer selects representative points from different regions and enables information exchange between them, facilitating information flow between distant areas and enhancing the network’s performance.

With the development of modern information technology and 5G, cloud-edge intelligent collaboration can make full use of the powerful computing power of cloud computing and the real-time response capability of edge computing to improve the overall efficiency of industrial intelligent systems. Thus, it shows strong application potential in digital twins, foundation models, and meta-universes. However, most of the existing researches focus on resource collaboration, data collaboration and application collaboration in cloud-edge computing framework, and there are many shortcomings in the research of intelligent collaboration framework. Therefore, we first propose a cloud-edge intelligent collaboration framework, which consists of three parts: terminal layer, edge artificial intelligence (AI) layer and cloud AI layer, including two core processes: cloud-edge intelligent training and cloud-edge intelligent inference. Then, we systematically analyze the key technologies of cloud-edge intelligent collaboration, including model segmentation, model early exit and foundation models. Finally, we put forward the application of cloud-edge intelligent collaboration in digital twin and AI Generated Content (AIGC), and through the analysis of typical application scenarios.

Internet-of-Things (IoT), is the comprehensive interconnection systems of computational, networking and physical devices with the important goal of providing proactive and personalized services efficiently. The foundation of such services is big data integration and processing among various devices, which brings important challenges including data fusion, transferring and sharing of computational results. On the other hand, decentralized blockchain platforms provide novel technologies for reliable IoT data integration and processing. In addition, to facilitate decentralization and distribution of IoT big data, tensor-train (TT), as a tensor decomposition method, can play a vital role. Therefore, in this paper, a tensor-train-based permissioned-private (P2) blockchain is proposed to realize the organization, integration, sharing and applications of IoT data for intelligent IoT services. To demonstrate the performance of the proposed method, case studies with IoT data are carried out on permissioned-private chain platform to measure its performance.

Unsupervised domain adaption (UDA), which transfers knowledge from a labeled source domain to an unlabeled target domain, has attracted tremendous attention in many machine learning applications. Recently, there have been attempts to apply domain adaption for sensor time series data, such as human activity recognition and gesture recognition. However, existing methods suffer from some drawbacks that hinder further performance improvement. They often require access to source data or source models during training, which is unavailable in some fields because of privacy protection and storage limit. Typically, the source domains may only provide an application programming interface (API) for the target domain to call. On the other hand, current UDA methods have not considered the temporal consistency and low-signal-to-noise ratio (SNR) of sensor time series. To address the challenges, this article presents a black-box domain adaption framework for sensor time series data (B2TSDA). First, we propose a single/multi-source teacher-student learning framework to distill the knowledge from the source domains to a customized target model. Then we design a new temporal consistency loss by combining an adaptive mask method and dynamic threshold method to maintain consistent temporal information and balance the learning difficulties of different classes. For the multisource black-box domain adaption, we further propose a Shapley-enhanced method to determine the contribution of each source domain. Experimental results on both single-source and multisource domain adaption show that our framework has superior performance compared to other black-box UDA methods.

Smart manufacturing has been transforming toward industrial digitalization integrated with various advanced technologies. Metaverse has been evolving as a next-generation paradigm of a digital space extended and augmented by reality. In the metaverse, users are interconnected for various virtual activities. In consideration of advanced possibilities that may be brought by the metaverse, it is envisioned that industrial metaverse should be integrated into smart manufacturing to upgrade industry for more visible, intelligent and efficient production in the future. Therefore, a conceptual model, named IMverse Model, and novel characteristics of the industrial metaverse for smart manufacturing are proposed in this article. Besides, an industrial metaverse architecture, named IMverse Architecture, is proposed involving several key enabling technologies. Typical innovative applications of the industrial metaverse throughout the whole product life cycle for smart manufacturing are presented with insights. Nonetheless, in prospect of future, the industrial metaverse still faces limitations and is far from implementation. Thus, challenges and open issues of the industrial metaverse for smart manufacturing are discussed, then outlook is provided for further research and application.

Industrial multivariate time series (MTS) is a critical view of the industrial field for people to understand the state of machines. However, due to data collection difficulty and privacy concerns, available data for building industrial intelligence and industrial large models is far from sufficient. Therefore, industrial time series data generation is of great importance. Existing research usually applies generative adversarial networks (GANs) to generate MTS. However, GANs suffer from the unstable training process due to the joint training of the generator and discriminator. This article proposes a temporal-augmented conditional adaptive diffusion model, termed Diff-MTS, for MTS generation. It aims to better handle the complex temporal dependencies and dynamics of MTS data. Specifically, a conditional adaptive maximum-mean discrepancy (Ada-MMD) method has been proposed for the controlled generation of MTS, which does not require a classifier to control the generation. It improves the condition consistency of the diffusion model. Moreover, a temporal decomposition reconstruction UNet (TDR-UNet) is established to capture complex temporal patterns and further improve the quality of the synthetic time series. Comprehensive experiments on the C-MAPSS and FEMTO datasets demonstrate that the proposed Diff-MTS performs substantially better in terms of diversity, fidelity, and utility compared with the GAN-based methods. These results show that Diff-MTS facilitates the generation of industrial data, contributing to intelligent maintenance and the construction of industrial large models.

Industrial Internet of Things enables various manufacturing processes executed in distributed production lines and flexible workshops. With different cloud–edge–device collaboration ways, interconnected manufacturing tasks and computational tasks are cooperatively completed in manufacturing cells, cloud resources, and edge resources. Large-scale decision variables and complex precedence constraints make the scheduling problem intractable. To this end, this article proposed a discretized soft actor–critic configured differential evolution algorithm to find a stable solution for the cloud–edge–device collaborative task-scheduling problem. A mathematical model is established to describe the relationship between different tasks, the variables, the main constraints in collaboration, and the scheduling targets. A decentralized partially observable Markov decision process is modeled with five neural networks and three discretized loss functions to formulate the discretized soft actor–critic policy efficiently and enable it to find the best differential evolution configurations for different scheduling cases. Experimental analysis of four cloud–edge–device scheduling instances indicates that the proposed method trained in one case is adaptable to the other three cases. In the four cases, the proposed method reduces the total objective by 30.82% and 44.35% at most compared to five deep-reinforcement-learning-based differential evolution algorithms and seven typical evolutionary algorithms, respectively.

Prediction of industrial time-series is crucial for various Industrial Internet of Things applications. Despite the high accuracy of deep learning methods for time-series prediction, the significant memory requirements of deep learning models pose a challenge for the limited computational resources of industrial edge devices. To address this issue, this work proposes BTFormer, which achieves a high compression rate while maintaining competitive performance. First, a binary adaptive attention module is proposed to mitigate the loss of attention information caused by binarization. Second, a trend information soft-link is proposed to propagate trend information between layers and improve the representation ability of the model. Finally, a distribution-guided distillation strategy is proposed to optimize the training process. The experiments demonstrate that BTFormer effectively reduces model memory usage by 31.0 times and improves computational efficiency by 32.8 times while maintaining competitive performance.

Intelligent fault diagnosis models have demonstrated a superior performance in industrial prognostics health management scenarios. However, these models may struggle to generalize in complicated industrial environments, when encountering new working conditions and handling low-resource and heterogeneous data. To cope with the aforementioned issues, we focus on constructing a universal training framework with domain generalization technique that will encourage fault diagnosis model to generalize well in unseen working conditions. Firstly, a model-agnostic meta-learning based training framework called Meta-GENE is proposed for homogeneous and heterogeneous domain generalization. Secondly, a gradient aligning algorithm is introduced in meta-learning framework to learn domain-invariant strategy for robust prediction in unseen working conditions. Thirdly, a semantic matching technique is proposed for utilizing heterogeneous data to alleviate low-resource problem. Our method has yielded excellent performance on the PHM09 fault diagnosis dataset and achieved superior results on a set of generalization tasks across various working conditions.

The intelligent manufacturing system is a complex, large-scale, interconnected system composed of many intelligent agents, and there may be physical or information space couplings between the agents. A distributed monitoring system and optimization control method are proposed to ensure the system completes its tasks safely and efficiently. The distributed monitoring system based on the average consensus algorithm is equivalent to the centralized design method, in which the submonitoring system only requires local and neighbor subsystem information. The advantage of this design is that it uses local and interactive information to achieve global diagnosis. In addition, sending data from all subsystems to a central computing node is challenging to implement in large-scale manufacturing systems. Based on the centralized plug-and-play (PnP) optimization control method, an average consensus algorithm distributed manufacturing system PnP optimization control method is proposed. Its advantage is that it uses local information and interactive information to achieve global control optimization. On this basis, an integrated architecture for distributed fault detection and optimization control is developed. The simulation results verify the feasibility and effectiveness of proposed method.

Time-series prediction plays a crucial role in the Industrial Internet of Things (IIoT) to enable intelligent process control, analysis, and management, such as complex equipment maintenance, product quality management, and dynamic process monitoring. Traditional methods face challenges in obtaining latent insights due to the growing complexity of IIoT. Recently, the latest development of deep learning provides innovative solutions for IIoT time-series prediction. In this survey, we analyze the existing deep learning-based time-series prediction methods and present the main challenges of time-series prediction in IIoT. Furthermore, we propose a framework of state-of-the-art solutions to overcome the challenges of time-series prediction in IIoT and summarize its application in practical scenarios, such as predictive maintenance, product quality prediction, and supply chain management. Finally, we conclude with comments on possible future directions for the development of time-series prediction to enable extensible knowledge mining for complex tasks in IIoT.

Time series extrinsic regression (TSER) aims at predicting numeric values based on the knowledge of the entire time series. The key to solving the TSER problem is to extract and use the most representative and contributed information from raw time series. To build a regression model that focuses on those information suitable for the extrinsic regression characteristic, there are two major issues to be addressed. That is, how to quantify the contributions of those information extracted from raw time series and then how to focus the attention of the regression model on those critical information to improve the model's regression performance. In this article, a multitask learning framework called temporal-frequency auxiliary task (TFAT) is designed to solve the mentioned problems. To explore the integral information from the time and frequency domains, we decompose the raw time series into multiscale subseries in various frequencies via a deep wavelet decomposition network. To address the first problem, the transformer encoder with the multihead self-attention mechanism is integrated in our TFAT framework to quantify the contribution of temporal-frequency information. To address the second problem, an auxiliary task in a manner of self-supervised learning is proposed to reconstruct the critical temporal-frequency features so as to focusing the regression model's attention on those essential information for facilitating TSER performance. We estimated three kinds of attention distribution on those temporal-frequency features to perform auxiliary task. To evaluate the performances of our method under various application scenarios, the experiments are carried out on the 12 datasets of the TSER problem. Also, ablation studies are used to examine the effectiveness of our method.

Representation learning-based remaining useful life (RUL) prediction plays a crucial role in improving the security and reducing the maintenance cost of complex systems. Despite the superior performance, the high computational cost of deep networks hinders deploying the models on low-compute platforms. A significant reason for the high cost is the computation of representing long sequences. In contrast to most RUL prediction methods that learn features of the same sequence length, we consider that each time series has its characteristics and the sequence length should be adjusted adaptively. Our motivation is that an "easy" sample with representative characteristics can be correctly predicted even when short feature representation is provided, while "hard" samples need complete feature representation. Therefore, we focus on sequence length and propose a dynamic length transformer (DLformer) that can adaptively learn sequence representation of different lengths. Then, a feature reuse mechanism is developed to utilize previously learned features to reduce redundant computation. Finally, in order to achieve dynamic feature representation, a particular confidence strategy is designed to calculate the confidence level for the prediction results. Regarding interpretability, the dynamic architecture can help human understand which part of the model is activated. Experiments on multiple datasets show that DLformer can increase up to 90% inference speed, with less than 5% degradation in model accuracy.

Industrial Internet of Things is moving toward an intelligent level with large-scale collaborative cloud and edge resources, making it possible for online supervision, fast analysis, and precise control for many manufacturing job shops. However, online processing of large-scale industrial computation brings huge communication overhead and energy consumption among cloud, edge, and end devices. To improve the performance of the cloud–edge collaboration, this article establishes a practical model of task scheduling considering two kinds of cloud–edge collaborative modes. We propose a parallel group-merge evolutionary algorithm to assign thousands of tasks in seconds. The algorithm separates tasks into weakly correlated groups and applies modified evolutionary operators to find a subsolution for each group. Then, the subsolutions are merged to form a complete solution for fine-tuning based on the cross-use of heuristics. Experimental results show that the proposed method could assign thousands of tasks to cloud servers and edge servers in seconds, reduce the overall task computing time by 36.97%, and save the overall energy by 23.71% at most.

Industrial Cyber Physical Systems can use data and information gained from across a variety of different environments to enable robots that are reconfigurable. Custom grasping is a basic operation a robot must be able to carry out for a given task, i.e., finding the best grasping point for emergent behaviors. However, environmental disturbance and limited data degrade the precision and speed of many tailored machine learning models on robot grasping detection. This paper proposes a region-based method to enable fast custom grasping through fewer RGB-D data. The grasping detection problem is simplified as a two-stage prediction problem. At the first stage, a robust grasp candidate generation strategy is proposed based on the Sobel operator. At the second stage, a region-based predictor is designed to locate the best grasping point-pair for an emergent task. The predictor is trained by a modified consistency based self-training method to realize semi-supervised learning. Experimental results show that the success rate of custom grasping of new emergent object can be increased by 3.4% on average using the proposed method. By introducing data augmentation strategies in training, the success rate is further increased by 9.2% on average. A robot is able to grasp new object with 91.5% success rate using less than 100 training samples. The number of training samples required for the proposed method is less than to 1% of which for the previous works. Note to Practitioners—This research was motivated by the problem of robot reconfigurability for various industrial automation processes and focuses mainly on the recognition of grasping point-pair of emergent object for different task. Existing approaches on robotic grasping detection are tailored to a given object and require expensive training with large amount of labeled data. This paper presents a region-based few shot learning approach that enables the robot to detect the best grasping point-pair autonomously and quickly. We show how to generate candidate point-pairs with image distortion and background disturbance. We then demonstrate how the best grasping point-pair can be located with much less training cost. Experiments suggest that this approach is feasible in robot automation for handling a class of objects. In future research, we will construct behavior learning module to enable evolving cyber-physical robotic system for more purposes.

For prognostics and health management of industrial systems, machine remaining useful life (RUL) prediction is an essential task. While deep learning-based methods have achieved great successes in RUL prediction tasks, large-scale neural networks are still difficult to deploy on edge devices owing to the constraints of memory capacity and computing power. In this article, we propose a lightweight and adaptive knowledge distillation (KD) framework to alleviate this problem. First, multiple teacher models are compressed into a student model through KD to improve the industrial prediction accuracy. Second, a dynamic exiting method is studied to enable an adaptive inference on the distilled student model. Finally, we develop a reparameterization scheme to further lessen the student network. Experiments on two turbofan engine degradation datasets and a bearing degradation dataset demonstrate that our method significantly outperforms the state-of-the-art KD methods and enables the distilled model with an adaptive inference ability.

The structural information detection of road conditions, which is adopted for improving driving comfort, patrol inspection, road maintenance, and accident rescue. In order to improve the trustworthiness of road condition detection, a real-time artificial intelligence road detection system based on binocular vision sensors is investigated in this article. The system is deployed on the low-power edge computing platform, which can upload the processing results to the cloud through the Internet-of-Things devices. The authors use binocular disparity information and image-based lightweight deep segmentation network to enhance the detection robustness and accuracy in the industrial Internet-of-Things application scenarios. Considering the small training dataset, a special data labeling regularization and training strategy have also been proposed for training this network. In addition, we employ multiframes feature matching and measurement data filtering to enhance the measurement accuracy. The experimental results demonstrate that our monocular–binocular fusion framework is robust and efficient.

Health indicator prediction, such as remaining useful life prediction and product quality prediction, is an important aspect of industrial intelligence. It is essential to process the massive multichannel industrial time series collected from the Industrial Internet of Things for the industrial health indicator prediction. At present, there are still three issues that need to be considered for industrial health indicator prediction. First, it is difficult to directly connect the distant positions in the industrial time series to extract the temporal relations, which decreases the efficiency of extracting the potential long-distance temporal relations and training networks. Second, it should be fully considered that data from different channels have different contributions. Equally dealing with the contributions of each channel will weaken the representational ability of prediction networks. Third, the loss function deals with early predictions and delay predictions equally, which will lead to high risks caused by delay predictions. In this article, for these issues, a novel multichannel temporal attention-based network (MCTAN) is proposed for industrial health indicator prediction, which can weigh contributions of different channels through the channel attention while avoiding the loss of the temporal information and directly connect each time series position to the local fields of the sequence through the multi-head local attention mechanism to efficiently extract potential long-distance temporal relations. Then, a weighted mean square error loss function differently dealing with early predictions and delay predictions by setting dynamic weights is presented to reduce delay predictions. Next, to deal with the above-mentioned issues systematically, a framework combining data preprocessing and MCTAN collaboratively is introduced to predict industrial health indicators through multichannel time series. Finally, the experiments are carried out on the commercial modular aero-propulsion system simulation dataset to measure the performances, including the accuracy of industrial health indicator predictions and the inference speed.

Cyber-physical-social systems (CPSS), an emerging cross-disciplinary research area, combines cyber-physical systems (CPS) with social networking for the purpose of providing personalized services for humans. CPSS big data, recording various aspects of human lives, should be processed to mine valuable information for CPSS services. To efficiently deal with CPSS big data, artificial intelligence (AI), an increasingly important technology, is used for CPSS data processing and analysis. Meanwhile, the rapid development of edge devices with fast processors and large memories allows local edge computing to be a powerful real-time complement to global cloud computing. Therefore, to facilitate the processing and analysis of CPSS big data from the perspective of multi-attributes, a cloud-edge-aided quantized tensor-train distributed long short-term memory (QTT-DLSTM) method is presented in this article. First, a tensor is used to represent the multi-attributes CPSS big data, which will be decomposed into the QTT form to facilitate distributed training and computing. Second, a distributed cloud-edge computing model is used to systematically process the CPSS data, including global large-scale data processing in the cloud, and local small-scale data processed at the edge. Third, a distributed computing strategy is used to improve the efficiency of training via partitioning the weight matrix and large amounts of input data in the QTT form. Finally, the performance of the proposed QTT-DLSTM method is evaluated using experiments on a public discrete manufacturing process dataset, the Li-ion battery dataset, and a public social dataset.

How to evaluate the performance variations of large-scale cloud data centres is challenging due to diverse nature of cloud platforms. Classic methods such as profiling-based evaluating methods tend to only provide global statistics for a system compared with cloud tracing based approaches. However, existing tracing based research lacks a systematic comparative multiview analysis from architecure-view to job-view and task-view, etc.to evaluate cloud performance variations, together with a detailed case study. We introduce MuCoTrAna, a multiview comparative workload traces analysis approach to evaluate the performance variations of large-scale cloud data centres which assists the cloud platform performance managers and big trace analysts. The efficiency of the proposed approach is demonstrated via case studies in Alibaba 2018 trace and Google trace. The multifaceted analysis results of traces reveals the qualitative insights, performance bottlenecks, inferences and adequate suggestions from global view, machine view, job-task view, etc.

A key step in remanufacturing is disassembly of the “core” or the returned product to be remanufactured. Disassembly sequence planning is challenging due to uncertainties in the conditions of the cores. Rust, corrosion, deformation, and missing parts may require disassembly plans to be changed and adapted frequently. Conventional industrial automation that usually serves in repetitive and structured activities may fail when it is applied to disassembly. This research investigates the flexible sequencing of robotic disassembly in the presence of failed automation operations and develops online recovery by incorporating backup actions. It starts with modeling the time and success rate of a backup action. The expected disassembly time and completion rate of a disassembly plan are deduced according to the failure probability of both the operations and their backup actions. A biobjective optimization model for robotic disassembly sequence planning is established using a dual-selection multiobjective evolutionary algorithm. Two solution selection criteria are combined to produce potential offspring candidates in each evolutionary generation. Experimental results show that the backup actions allow efficient recovery from automation and can potentially improve the robustness of robotic disassembly.Note to Practitioners—This research was motivated by the development of automated disassembly techniques. Industrial automation techniques usually use predetermined operation motions. Robotic disassembly using such an approach may fail due to uncertainties in the condition of the products (e.g., positioning and geometry). This article introduces backup actions for disassembly sequence planning and describes the logic and reasoning of their implementation. Our proposed method can theoretically increase the completion rate of automated robotic disassembly. Experimental studies suggested that backup actions are efficient in providing a reliable disassembly sequence and, thus, can improve the robustness of robotic disassembly. In future research, we will implement typical backup actions and establish an automated disassembly process with a replanning module.

Industrial Internet of Things data incorporate the fundamental elements of industrial processes, providing novel paradigms of predictive maintenance for complex industrial equipment. Remaining useful life prediction is critical in the predictive maintenance task of product lifecycle management, which has attracted increasing research attention. However, most existing prediction methods cannot effectively extract complex multiscale temporal patterns and cannot meet the real-time requirements of industrial sites. To address these issues, we propose a$T^{2}$-Tensor-aided multiscale transformer for accurate and effective prediction in this article. We defined the$T^{2}$-tensor to represent the multiscale temporal pattern by reconstructing the time series. Besides, a high-order transformer for multiscale feature extraction is proposed. Particularly, the multiscale characteristics can be captured through intertoken and intratoken. In addition, a transformer parameter lightweighting method with tensor ring decomposition is developed. Experiments demonstrate the accuracy and efficiency of the proposed method.

In cloud manufacturing systems, fault diagnosis is essential for ensuring stable manufacturing processes. The most crucial performance indicators of fault diagnosis models are generalization and accuracy. An urgent problem is the lack and imbalance of fault data. To address this issue, in this article, most of existing approaches demand the label of faults asa prioriknowledge and require extensive target fault data. These approaches may also ignore the heterogeneity of various equipment. We propose a cloud-edge collaborative method for adaptive fault diagnosis with label sampling space enlarging, named label-split multiple-inputs convolutional neural network, in cloud manufacturing. First, a multiattribute cooperative representation-based fault label sampling space enlarging approach is proposed to extend the variety of diagnosable faults. Besides, a multi-input multi-output data augmentation method with label-coupling weighted sampling is developed. In addition, a cloud-edge collaborative adaptation approach for fault diagnosis for scene-specific equipment in cloud manufacturing system is proposed. Experiments demonstrate the effectiveness and accuracy of our method.

Soft sensor, as an important paradigm for industrial intelligence, is widely used in industrial production to achieve efficient monitoring and prediction of production status including product quality. Data-driven soft sensor methods have attracted attention, which still have challenges because of complex industrial data with diverse characteristics, nonlinear relationships, and massive unlabeled samples. In this article, a data-driven self-supervised long short-term memory–deep factorization machine (LSTM-DeepFM) model is proposed for industrial soft sensor, in which a framework mainly including pretraining and finetuning stages is proposed to explore diverse industrial data characteristics. In the pretraining stage, an LSTM-autoencoder is first unsupervised pretrained. Then, based on two self-supervised mask strategies, LSTM-deep can explore the interdependencies between features as well as the dynamic fluctuation in time series. In the finetuning stage, relying on pretrained representation, the temporal, high-dimensional, and low-dimensional features can be extracted from the LSTM, deep, and FM components, respectively. Finally, experiments on the real-world mining dataset demonstrate that the proposed method achieves state of the art comparing with stacked autoencoder-based models, variational autoencoder-based models, semisupervised parallel DeepFM, etc.

In the modern supply chain system, large-scale transportation tasks require the collaborative work of multiple vehicles to be completed on time. Over the past few decades, multi-vehicle route planning was mainly implemented by heuristic algorithms. However, these algorithms face the dilemma of long computation time. In recent years, some machine learning-based methods are also proposed for vehicle route planning, but the existing algorithms can hardly solve multi-vehicle time-sensitive problems. To overcome this problem, we propose a novel multi-agent reinforcement learning model, which optimizes the route length and the vehicle’s arrival time simultaneously. The model is based on the encoder-decoder framework. The encoder mines the relationship between the customer nodes in the problem, and the decoder generates the route of each vehicle iteratively. Specially, we design multiple route recorders to extract the route history information of vehicles and realize the communication between them. In the inferring phase, the model could immediately generate routes for all vehicles in a new instance. To further improve the performance of the model, we devise a multi-sampling strategy and obtain the balance boundary between computation time and performance improvement. In addition, we propose a simulation-based vehicle configuration method to select the optimal number of vehicles in real applications. For validation, we conduct a series of experiments on problems with different customer amounts and various vehicle numbers. The results show that the proposed model outperforms other typical algorithms in both performance and calculation time.

Industrial Internet of Things (IIoT), as an important industrial branch of the Internet of Things (IoT), has an essential purpose to improve intelligent industrial production. For this purpose, IIoT big data should be efficiently processed to mine valuable information. In handing the IIoT big data, cloud-edge computing is getting more attention to reduce the interaction latency to meet the real-time requirement, especially in the field of prognostic and health management (PHM). It is expected that artificial intelligence (AI) technologies will significantly change the manner of processing IIoT big data. Therefore, new methods about PHM, combining cloud-edge computing with AI technologies, are required to process the IIoT big data for intelligent industrial manufacturing. As an essential element of PHM, predicting the remaining useful life (RUL) of industrial equipment plays an increasingly crucial role, especially for industrial intelligence. However, traditional methods pay much attention on prediction accuracy and neglect the influence of computing time. In this article, by combining cloud-edge computing with AI technology, a new data-driven method, namely, cloud-edge-based lightweight temporal convolutional networks (LTCNs), for RUL prediction is proposed. First, to meet the real-time requirement, a cloud-edge computing and AI-based framework for RUL prediction is presented. Second, a new model structure named LTCN is proposed and applied in the framework. Real-time prediction results will be obtained in the edge plane and higher accuracy prediction results will be obtained through historical information in the cloud plane. Third, an incremental learning approach based on updating partial parameters of LTCN is discussed to improve the accuracy of prediction models with newly collected data. Experiments show that our method can improve the prediction accuracy and reduce the computational time of RUL.

Integration of each aspect of the manufacturing process with the new generation of information technology such as the Internet of Things, big data, and cloud computing makes industrial manufacturing systems more flexible and intelligent. Industrial big data, recording all aspects of the industrial production process, contain the key value for industrial intelligence. For industrial manufacturing, an essential and widely used electronic device is the lithium-ion battery (LIB). However, accurately predicting the remaining useful life (RUL) of LIB is urgently needed to reduce unexpected maintenance and avoid accidents. Due to insufficient amount of degradation data, the prediction accuracy of data-driven methods is greatly limited. Besides, mathematical models established by model-driven methods to represent degradation process are unstable because of external factors like temperature. To solve this problem, a new LIB RUL prediction method based on improved convolution neural network (CNN) and long short-term memory (LSTM), namely Auto-CNN-LSTM, is proposed in this article. This method is developed based on deep CNN and LSTM to mine deeper information in finite data. In this method, an autoencoder is utilized to augment the dimensions of data for more effective training of CNN and LSTM. In order to obtain continuous and stable output, a filter to smooth the predicted value is used. Comparing with other commonly used methods, experiments on a real-world dataset demonstrate the effectiveness of the proposed method.

In the modern industry, the information has been sufficiently shared among the production equipment, intelligent subsystems, and mobile devices via advanced network technology. For this purpose, many challenges on plant-wide performance evaluation such as product quality prediction have been received considerable attention in complex industrial Internet of Things systems. In this article, an efficient and effective soft sensor based on the semisupervised parallel deepFM model is proposed for the product quality prediction. First, a label broadcasting method is presented to augment labeled samples from unlabeled samples. Then, a data binning method is introduced to discretize process variables for an unbiased estimation. Based on the modified deepFM model, quality information can be separately extracted from different components of the model while high- and low-dimensional features can be obtained. Manifold regularization is embedded into the back propagation algorithm, in which unlabeled samples issue can be further resolved. Experiments on a real-world dataset demonstrate the effectiveness and performance of the proposed methods.

Industrial Internet-of-Things (IIoT) has revolutionized almost every aspect of industrial manufacturing through industrial intelligence by incorporating production equipment, mobile terminals, and smart devices with wireless or wired networks. However, industrial visual information, such as images, videos, graphs, and texts, generated and collected from the industrial processes, contains various kinds of hidden value for industrial intelligence. Therefore, for the trend of providing ubiquitous industrial intelligence, new paradigms of perception and processing technologies of visual information such as recognition methods are required. However, industrial visual information is heterogeneous and complex with multiattributes, which presents significant challenges on visual information perception and processing technologies such as multiattributes recognition method. In this article, to provide industrial intelligence, a tensor-based visual feature recognition method is used to recognize the object from the perspective of multiattributes with the combination of attributes. To demonstrate its practical implementation, a case study about the industrial intelligence on the faulty location and diameter of bearings in the IIoT is described. Also, experiments on object recognition are carried out on the public image set COIL-100 to demonstrate the performance of the proposed method.

The industrial Internet of Things (IIoT) is growing quickly due to increasing deployment and integration of smart sensors, instruments, and devices, and software using wired or wireless networks. Through this integrated hardware-software approach, industrial practices will improve significantly, resulting in industrial intelligence for more efficient manufacturing. To realize such industrial intelligence, significant developments in IIoT big data processing and analysis are required to uncover and use hidden essential and valuable information of the production process. But large-scale, streaming, multiattribute IIoT data from production processes are noisy and have redundancies. Therefore, a suitable data processing technique such as tensor-train that can handle these IIoT data is needed. However, existing tensor-train decomposition methods are inefficient and cannot meet the processing demands of the large-scale IIoT big data. In this article, we propose an advanced (improved and highly efficient) distributed tensor-train (ADTT) decomposition method with its incremental computational method for processing IIoT big data. Finally, experiments are carried out on a typical and publicly available IIoT dataset - the bearing test data to verify and measure the performances of the proposed ADTT method.

Product quality prediction, as an important issue of industrial intelligence, is a typical task of industrial process analysis, in which product quality will be evaluated and improved as feedback for industrial process adjustment. Data-driven methods, with predictive model to analyze various industrial data, have been received considerable attention in recent years. However, to get an accurate prediction, it is an essential issue to extract quality features from industrial data, including several variables generated from supply chain and time-variant machining process. In this article, a data-driven method based on wide-deep-sequence (WDS) model is proposed to provide a reliable quality prediction for industrial process with different types of industrial data. To process industrial data of high redundancy, in this article, data reduction is first conducted on different variables by different techniques. Also, an improved wide-deep (WD) model is proposed to extract quality features from key time-invariant variables. Meanwhile, an long short-term memory (LSTM)-based sequence model is presented for exploring quality information from time-domain features. Under the joint training strategy, these models will be combined and optimized by a designed penalty mechanism for unreliable predictions, especially on reduction of defective products. Finally, experiments on a real-world manufacturing process data set are carried out to present the effectiveness of the proposed method in product quality prediction.

With the rapid development of 3D printing technology and the continuous breakthrough of new material technologies, 3D printing has received more and more attention in the fields of industry, medical, sports, and education. In the cloud manufacturing environment, how to efficiently manage 3D printing models is a critical issue that needs to be solved urgently. The paper begins with a study of 3D printing model aggregation and retrieval mechanism based on this problem. Firstly, we devised a set of 3D printing model management framework with high scalability. Under this framework, a meta-model library and feature library are established to realize the aggregation of 3D printing models. Then, we developed a sketch-based 3D model retrieval method, which can help platform users to create and retrieve personalized 3D printing models easily. This study provides new ideas and a reference pattern for the model design and retrieval methods of existing commercial 3D printing platforms.

In service-oriented manufacturing modes, service scheduling is important for providing just-in-time delivery of manufacturing services to customers. In this paper, the mathematical model of the dynamic cloud manufacturing scheduling problem is constructed, and a scheduling method based on dynamic data-driven simulation is proposed to improve the scheduling performance. The framework, scheduling rules and simulation strategies are discussed in detail. Specifically, three single scheduling rules and three combined scheduling rules are designed based on service time, logistics time, and subtask queue status of candidate services. The real-time information of tasks and services is involved in the simulation strategy to select better scheduling rules. The simulation models of proposed scheduling strategies are constructed and simulated in the Simio software, which is connected to a real-time information database. The proposed method is tested through a case study of numerical control machining in cloud manufacturing, and the results show that the proposed method is promising.

People face data-rich manufacturing environments in Industry 4.0. As an important technology for explaining and understanding complex data, visual analytics has been increasingly introduced into industrial data analysis scenarios. With the durability test of automotive starters as background, this study proposes a visual analysis approach for understanding large-scale and long-term durability test data. Guided by detailed scenario and requirement analyses, we first propose a migration-adapted clustering algorithm that utilizes a segmentation strategy and a group of matching-updating operations to achieve an efficient and accurate clustering analysis of the data for starting mode identification and abnormal test detection. We then design and implement a visual analysis system that provides a set of user-friendly visual designs and lightweight interactions to help people gain data insights into the test process overview, test data patterns, and durability performance dynamics. Finally, we conduct a quantitative algorithm evaluation, case study, and user interview by using real-world starter durability test datasets. The results demonstrate the effectiveness of the approach and its possible inspiration for the durability test data analysis of other similar industrial products.

With the conception and application of cloud manufacturing getting into manufacturing enterprises, more and more enterprises begin to establish the cloud system. How to collect the knowledge generated in these systems and pro-vide it to the enterprise manufacturing cloud for use is now an important issue. This paper presents an architecture of knowledge cloud based on manufacturing big data, according to the characteristics of manufacturing big data, building a knowledge database by ontology, encapsulation the knowledge service, and finally providing the provide the support for the enterprise manufacturing cloud.

Efficient service scheduling in the dynamic cloud manufacturing environment is very important for optimal supply-demand matching and timely delivery of products. This paper presents a systematic and deep analysis of the scheduling process and problem characteristics for dynamic service scheduling in cloud manufacturing. A simulation model is established for the dynamic service scheduling process of cloud manufacturing from aspects of demanders, tasks, services and path models. Both the attributes and activities of different models are studied, and a systematic scheduling simulation model is established. This work is probably valuable for the future simulation-based dynamic scheduling research in cloud manufacturing.

The problem of task allocation and service selection in the complex dynamic cloud manufacturing (CMfg) environment is complex for different types of manufacturing resources. With the rapid development of 3D printing technology, the supply-demand matching problem of 3D printing tasks and services in CMfg needs to be modelled specifically. In this paper, the service attributes of 3D printing services are analyzed, including model size, printing material, printing preciseness, cost, time and logistics. The service transaction model of 3D printing services is built. To reduce delivery time of tasks from service suppliers to service demanders, a 3D printing service matching and selection method (MST) is proposed to generate the optimal solutions. Experimental results show that the average task completion time with MST is less than that of the typical method when the amounts of tasks change. Besides, MST can balance the task assignment among different service providers.

Infovis (Information visualization) task taxonomy plays an essential role in guiding Infovis design and implementation. Infovis users with various roles in Infovis usually have different requirements for Infovis task modeling. Actually, Infovis research need a consistent taxonomy covering the tasks at different levels, or an interaction model that can facilitate Infovis system development with formal descriptions. But in fact, finding such a unified model is challenging. In this paper we propose a multilevel interaction model (MIM) for hierarchical tasks in Infovis systems. In MIM we define goal model, behavior model, and operation model that can model multilevel tasks in Infovis. In addition, we establish mapping models among MIM components, which can support Infovis systems design, development, application, and evaluation. Finally, we present a domain-specific Infovis application modeled by MIM. Application examples shows that MIM can effectively model multilevel tasks in Infovis and has potential to provide a framework enabling rapid prototyping of Infovis systems.

Cloud Manufacturing provides a new concept and model for manufacturing informatization, which has become a hot research topic in modern networked manufacturing area. Cloud manufacturing systems generate huge amounts of sensor data from distributed manufacturing devices, and how to manage them efficiently becomes one of the main challenges of Cloud Manufacturing. Various software frameworks have been developed in Cloud Computing to handle massive data management, and the Hadoop framework has proved an effective solution. In this paper we introduce Hadoop to deal with the sensor data management in Cloud manufacturing systems. First we analyze the requirement of massive sensor data management in Cloud Manufacturing and the defects of traditional RDBMS (Relational Database Management System), and then present a framework supporting parallel storage and processing of massive sensor data in Cloud manufacturing systems based on Hadoop. This framework can provide a promising solution for massive sensor data management in Cloud Manufacturing.

Optimal deployment of simulation virtual machines is an important issue in Cloud Simulation. Challenges involve resource cost prediction for simulation tasks as well as host physical machine selection for simulation virtual machines. In this paper we propose a novel approach using knowledge curves (i.e., curves as knowledge base) to solve this problem. First we present a resource cost estimation algorithm using empirical load curves synthesis, and then discuss a deployment target host selection algorithm by curves matching. This approach can provide a promising solution for intelligent deployment of virtual machines in Cloud Simulation. In addition, the proposed approach will be increasingly precise and effective as curve knowledge base increases.