Industrial edge computing in time-sensitive, heterogeneous environments faces significant challenges in delivering deterministic, low-latency, and fair task offloading under mixed-criticality traffic. Existing solutions struggle with unpredictable delays, resource contention, and the joint optimization of offloading and scheduling. In this work, we propose a task queue mapping mechanism-based architecture that incorporates formal mixed-deterministic traffic modeling under time-sensitive networking, enabling unified computation and network resource management. The joint task offloading and traffic scheduling problem is rigorously formulated as a two-stage Markov decision process with embedded fairness constraints. To address the strong coupling and computational complexity, we develop a fairness-aware deterministic offloading and resource scheduling (FA-DORS) framework, a hierarchical multiagent deep deterministic policy gradient algorithm, enhanced by a contribution-based local reward mechanism to ensure adaptive resource allocation and fairness. Extensive experimental results demonstrate that in the proposed architecture, FA-DORS achieves lower task latency, higher completion rates, improved resource utilization, and enhanced transmission determinism while ensuring fairness in multitask offloading compared to baseline methods. In particular, FA-DORS achieves an 8.89%–39.15% reduction in average end-to-end delay and up to 39.73% reduction in average jitter under high load, compared with the baseline method.

Pose-guided human image generation aims to render a source image in a specific pose. Current methods predominantly employ 2D-based signals, which exhibit inherent information deficits, as pose conditions. This leads to difficulty in establishing precise source-target appearance-pose correspondence and further causing uncertainty in predicting self-occluded regions’ appearance. To address these issues, we propose a 3D Pose Conditional Diffusion model (3DPCD) that leverages a human parametric model to integrate comprehensive and adjustable 3D control into forward–backward diffusion steps. Specifically, we employ Fourier-transformed SMPL-X as the 3D pose representation to facilitate precise source-target correspondence by understanding the complete pose information. Building on this, we further propose a hierarchical appearance-pose alignment method, which aligns appearance with the complete pose information at both global and local levels. Moreover, motivated by the fact that human pose transformation is a progressive process in 3D space and our 3D pose representation is adjustable, we integrate progressively interpolated 3D control into a series of sampling steps. This effectively mitigates uncertainties in pixel transfer between poses. It should be noted that the proposed explicit pose-guided strategy also supports flexible adjustment of pose, shape, and viewpoint. Both quantitative and qualitative evaluations demonstrate that our 3DPCD outperforms state-of-the-art methods on the widely used DeepFashion InShop benchmark and our newly constructed PoseWeb-33 dataset, which features richer appearance variations and more diverse conditional poses.

With the growing incorporation of deep neural network (DNN) models into modern software systems, the prohibitive construction costs of DNN models have become a significant challenge in software development. To address this challenge, model reuse has been widely applied to reduce model training costs; however, indiscriminately reusing an entire model may incur significant inference overhead. Consequently, DNN modularization—borrowing the idea of modularization in software engineering—has increasingly gained attention, enabling module reuse by decomposing a DNN model into modules. In particular, the emerging modularizing-while-training (MwT) paradigm, which outperforms modularizing-after-training by incorporating modularization into the model’s training process, has been demonstrated as a more effective approach for DNN modularization. However, existing MwT approaches focus on small-scale convolutional neural network (CNN) models at the convolutional kernel level. They struggle to handle diverse DNNs and large-scale models, particularly Transformer-based models, which consistently achieve state-of-the-art results across various tasks. To address these limitations, we propose NeMo, a scalable and more generalizable MwT approach. NeMo operates at the neuron level—a fundamental component common to all DNNs—thereby ensuring applicability to Transformers and various DNN architectures. Moreover, we design a contrastive learning-based modular training method, equipped with an effective composite loss function, hence being scalable to large-scale models. Comprehensive experiments on two Transformer-based models and four CNN models across two widely used classification datasets demonstrate NeMo’s superiority over the state-of-the-art MwT method. Results show average performance gains of 1.72% in module classification accuracy and a 58.10% reduction in module size. Our findings demonstrate that NeMo exhibits efficacy across both CNN and large-scale Transformer-based models. Moreover, a case study based on open source projects demonstrates the potential benefits of NeMo in practical scenarios, offering a promising approach for achieving scalable and generalizable DNN modularization.

Source localization in graph information propagation is essential for mitigating network disruptions, including misinformation spread, cyber threats, and infrastructure failures. Existing deep generative approaches face significant challenges in real-world applications due to limited propagation data availability. We present SIDSL (Structure-prior Informed Diffusion model for Source Localization), a generative diffusion framework that leverages topology-aware priors to enable robust source localization with limited data. SIDSL addresses three key challenges: unknown propagation patterns through structure-based source estimations via graph label propagation, complex topology-propagation relationships via a propagation-enhanced conditional denoiser with GNN-parameterized label propagation module, and class imbalance through structure-prior biased diffusion initialization. By learning pattern-invariant features from synthetic data generated by established propagation models, SIDSL enables effective knowledge transfer to real-world scenarios. Experimental evaluation on four real-world datasets demonstrates superior performance with 7.5-13.3% F1 score improvements over baselines, including over 19% improvement in few-shot and 40% in zero-shot settings, validating the framework's effectiveness for practical source localization. Our code can be found here (https://github.com/tsinghua-fib-lab/SIDSL).

A visible light communication (VLC)-based sensing system is proposed for gait recognition and classification. Human-induced reflections are modeled using a time-varying channel representation, enabling the capture of gait dynamics without requiring wearable devices. A low-cost sensing module with embedded processing and multi-channel photodetectors is implemented. Filtered signals are transformed into spectral-spatial features and analyzed using deep learning models. Experiments involving ten participants across eight gait types demonstrate that contrastive and multi-scale models achieve over 98% accuracy, highlighting the potential of VLC-based sensing for unobtrusive, privacy-preserving, and real-time human gait recognition.

While significant progress has been achieved in single-person image generation, generating realistic images of multiple individuals with diverse poses and occlusions remains challenging. This complexity stems from intrinsic layout conflicts in the skeletal structures of multiple people, which hinder current models from accurately rendering fine details in interactive regions, leading to limb distortion and visual artifacts. In this work, we propose LAMP, a framework for pose-accurate and visually coherent multi-person image generation across various occlusion scenarios and interactions. Our key insight is to resolve spatial conflicts in overlapping skeletal regions by disentangling individual control conditions through hierarchical modeling. We introduce an occlusion-aware layered control model that leverages existing robust methods to process disentangled single-person skeletons, while designing a novel composition strategy that enables accurate perception of visibility relationships in occluded regions. Additionally, we design occlusion-aware prompt guidance to mitigate identity ambiguity among multiple individuals caused by occlusions. Both quantitative and qualitative results demonstrate LAMP’s superiority over state-of-the-art methods in generating high-quality multi-person images with diverse poses, complex occlusions, varied styles, and semantic-rich contexts.

Accurate prediction of reaction temperature in rotary kiln is essential to realize its advanced process control and operational optimizations. However, the complexity of the physical and chemical reactions in the rotary kiln makes it difficult for the traditional mechanism model to characterize the dynamic kiln process. In this study, a deep learning-based temperature prediction model is proposed to accurately track the temperature changes during the production process of rotary kiln. The proposed model integrates a temporal convolutional network (TCN) with a long short-term memory (LSTM) network. The former enables the proposed model to be aware of local context and can effectively compute local characteristics, whereas the latter with its time-memory capability can better capture the long-term dependencies of data to extract deep-level features. To enhance the prediction accuracy, the proposed model is further improved by introducing a novel feature fusion and hybrid pooling layer to merge the original input with TCN output features, which efficiently preserve the detailed information and significantly reduce model complexity. Attention mechanism is also incorporated after the LSTM network to concentrate on the key moment information and improve the model performance. Monitoring data from a zinc rotary kiln at a field site is used for model training and testing. Results demonstrate that the proposed model can achieve the best mean square error, 0.165, exhibiting a promising prediction accuracy of the rotary kiln’s temperature. It outperforms the state-of-the-art machine learning-based prediction models such as LSTM, Gated Recurrent Unit (GRU) and TCN.Note to Practitioners—This work considers high-accuracy prediction of the temperature changes during the production process of rotary kiln. It is critical to well predict kiln tail temperature by capturing long-term dependencies and extracting global and local features from time series data. Yet, this is an extremely challenging task because rotary kiln operation is a complex thermal process with multivariate, pure hysteresis, strong interference, nonlinearity and strong coupling characteristics. As a result, these issues inhibit the establishment of a model that can accurately characterize the production process of the rotary kiln and impedes the realization of automation in volatile kiln production. Current prediction methods usually adopt mechanism-based and data-driven, there are suffer from long computation time, unknown kinetic parameters and poor accuracy. Thus, they fail to effectively learn features and accurately establish model. This work proposes a temperature prediction dynamic model named enhanced TCN-LSTM, which combines a temporal convolutional network (TCN) with a long short-term memory (LSTM) network, incorporating a unique feature fusion technique along with a hybrid pooling layer and attention layer. This integration enables the model to extract both global and local features, thereby preserving detailed information efficiently while significantly reducing model complexity and enhance the stability and robustness. Experimental results based on the zinc rotary kiln dataset demonstrate that the proposed model exhibits a promising prediction accuracy of the rotary kiln’s temperature. In the future, it can be readily implemented and applied in many industrial areas such as systems modeling, time series forecasting, and predictive maintenance.

Deep-learning-based structured light 3D reconstruction technology (SL3D) provides excellent solutions for intelligent manufacturing. However, the scarcity of real-world datasets covering full-process data and diverse objects hampers the validation of new ideas. Moreover, limited research on dataset construction strategies, such as scene backgrounds and sample distribution, reduces network performance. We investigate the impact of background stability on foreground accuracy (BS-FA) and find a whiteboard background improved foreground prediction accuracy by up to 82% over a black background. Guided by BS-FA, we develop the SL3D-BF, a background-effective SL3D dataset for industrial use, featuring approximately 2,100 scenes with diverse objects like metal/plastic workpieces, plaster sculptures, and standard parts for precise evaluation. It uniquely includes shadow and foreground masks, absent in prior datasets, and offers full-process data from gratings to 3D point clouds, totaling 100,800 gratings. We also establish an initial benchmark for future research by conducting evaluation experiments with advanced methods. Furthermore, we investigate the relationship between the spatial frequency of sample occurrence and the model predictive ability to minimize the time and resource demands of dataset construction. Most importantly, SL3D-BF is also a valuable resource for tasks like depth estimation, defect detection, and semantic segmentation. The dataset is available at: https://github.com/LiYiMingM/Dataset_SL3D_BF.

Accurate estimation of burden surface depth plays a crucial role in constructing the temperature field and optimizing reaction control in volatile kilns. However, most image-based depth estimation techniques require high-quality input images and achieve limited accuracy, which restrict their applications in actual harsh working conditions such as high temperature, heavy dust and dense smoke. In this study, a deep learning-based monocular depth estimation model is proposed to measure the burden surface depth in the volatile kiln head zone. The proposed model integrates an encoder-decoder network with an attention module. The encoder-decoder network outputs a set of deep semantic features, while the attention module intelligently fuses multi-level features to predict a probability distribution over depth intervals for each pixel. A volatile kiln prototype is designed and constructed to generate image datasets of the kiln head zone which approximate real data collected from industrial production sites. Results demonstrate that the proposed model has a depth prediction error of RMSE = 11.008 mm for the burden surface region, outperforming state-of-the-art neural networks and the traditional depth-from-defocus method. Code and datasets are available athttps://github.com/LLLcong/Attention-MonoDepth.

In the process industries, nonlinear and large time-delay systems pose significant challenges for efficient model predictive control (MPC). The advent of deep learning offers innovative techniques for precise modeling and control; however, deep neural architectures have limited application in control problems. This study introduces a deep neural networks-based model predictive control (DNNs-MPC) that can utilize various gradient-based neural network models as predictors, enhancing the predictive capabilities of MPC and improving performance for nonlinear systems with large time-delay. To achieve this, we first employ dilated convolution and recurrent neural networks to develop a dynamic system modeling predictor, effectively capturing the system's nonlinear characteristics. Concurrently, to address challenges associated with the objective function, we propose an optimization strategy that incorporates three objective functions and employs a multistage weight optimization method to improve control performance and ensure output stability. Furthermore, to derive the optimal control strategy, an adaptive gradient descent method is applied to accelerate the solution process and quickly obtain optimal control signals. Finally, the effectiveness of our method is validated through numerical simulations and a case study of an industrial rotary kiln, demonstrating significant improvements in control performance, system stability, and response accuracy.

In fringe projection profilometry systems, accurately reconstructing 3D objects with varying surface reflectivity requires high dynamic range (HDR) imaging. However, the limited dynamic range of single-exposure cameras poses challenges for capturing HDR fringe patterns efficiently. This paper introduces a deep learning-based HDR structured light 3D reconstruction pipeline, comprising an HDR Fringe Generation Module and a Phase Calculation Module. The HDR Fringe Generation Module employs an end-to-end network with attention guidance and feature distillation to reconstruct HDR fringe images from short- and long-exposure low dynamic range (LDR) inputs. The Phase Calculation Module processes the phase information from HDR fringes to enable 3D reconstruction. On a metallic HDR dataset, the method achieved a phase error of 0.105, comparable to the 4-exposure 6-step Phase Shifting Profilometry (PSP) method (0.069), with only 8.3% of the projection time. Experimental results demonstrate the robustness of our approach under diverse object geometries, exposure levels, and challenging global illumination environments. In quantitative measurements, our method achieved accuracies of sub-50 $\mu $ m on ceramic spheres, flat plates and metal step object. Ablation experiments confirmed that feature distillation and attention module effectively enhance the HDR Fringe Generation Module, producing high-quality HDR fringe patterns critical for reconstructing objects with HDR surface reflectivity. Furthermore, we constructed an HDR imaging metal dataset comprising 1,700 samples of machined metal parts with diverse shapes, sizes, and materials, making it a benchmark in the field of HDR structured light measurement. Our method offers a general HDR imaging-based structured light 3D reconstruction approach, integrating the two modules into an efficient, end-to-end solution for objects with HDR reflective surfaces.

As human-computer interaction (HCI) advances, gesture recognition has emerged as a transformative technology for human-computer interaction. Traditional methods, often camera or glove-based, are restricted by various environmental conditions and user-specific demands, highlighting the need for more universal, non-intrusive, and sustainable solutions. Addressing this, we present PowerGest, a self-powered gesture recognition system based on a solar cell array. This innovative system leverages the dual functionalities of solar cells: energy harvesting and gesture sensing, providing an alternative to conventional methods. It integrates a designed low-powered data acquisition chip with a wireless transmission module and a user-friendly interface. PowerGest employs a series of signal processing methods and utilizes several machine learning algorithms, achieving over 97% accuracy for both numeric input and activity control gesture recognition tasks. With its broad applications in robotic control, text input, and more, PowerGest contributes to a more sustainable and intuitive HCI experience. Project demo: https://drive.google.com/drive/folders/10KEul8PAfvTUomi0JvZ8u411JyScCXQI?usp=sharing.

With the rapid development of industrial network, Industrial Internet of Things (IIoT) has become an indispensable part of industrial network development. However, due to the vulnerability of Industrial Internet of Things to network intrusion attacks. Therefore, anomaly detection in IIoT is particularly important. In this paper, an effective intrusion detection approach ROSE-BOX (Random fOrest, SmotE, BO-Xgboost) is proposed to detect multi-class cyberattacks based on Random Forest, SMOTE and BO-XGBoost in IIoT. It is worth mentioning that BO-XGBoost is obtained by optimizing the parameters of XGBoost using Bayesian optimization. Finally, compared with other existing methods, the proposed approach has better detection performance, with an accuracy rate of over 99.85%.

The demand for accurate and real-time indoor localization and human sensing is rising with the development of smart environments, with applications in security and smart homes. Effective systems enhance safety, energy efficiency, and user experience by leveraging existing infrastructure, reducing deployment costs, and integrating seamlessly. Traditional methods rely on dedicated hardware, while communication or lighting infrastructure can provide dual-purpose solutions. This research focuses on Visible Light Communication (VLC) technology, which uses visible light for data transmission. Our VLC system utilizes existing lighting infrastructure to transmit data, providing localization and human sensing functionalities. The system design strategically incorporates specific VLC transmitters and receivers to enhance sensing performance. The collected data is processed using advanced algorithms and machine learning models, ensuring robust, real-time, and cost-effective localization with an accuracy of 96.6% and human activity recognition with an accuracy of 98.3%. This multi-functionality system demonstrates the potential for VLC in healthcare monitoring and home automation applications.

In response to the challenges presented by conventional material and texture recognition methods, our research introduces a system using visible light communication (VLC) technology. This approach provides a non-contact, non-destructive, dual-functional solution, overcoming the limitations of cost, safety, and environmental adaptability associated with traditional methods. Through a comprehensive design integrating hardware and software, our system utilizes VLC for precise and efficient recognition. Extensive testing confirms its effectiveness, achieving 97.7% accuracy in material identification and 93.8% in texture detection. This study highlights VLC's potential in enhancing automated recognition systems across various applications.

SolarSense is a self-powered sensing system for gesture recognition using solar cell arrays, thereby offering a sustainable approach to human-computer interaction (HCI) within industrial settings. The system effectively employed the sensing and energy harvesting capabilities of solar cells, achieving over 97.0% accuracy in recognizing diverse gestures. The design incorporates a low-power wireless data acquisition chip, a signal processing framework, and a user interface to realize robotic control and text input applications. SolarSense enhances HCI with its eco-friendly and user-centric approach, which is suitable for Internet of things (IoT) scenarios. Demo: https://youtu.be/RmPolChw_c4.

The intelligent goal of process manufacturing is to achieve high efficiency and greening of the entire production. Whereas the information system it used is functionally independent, resulting to knowledge gaps between each level. Decision-making still requires lots of knowledge workers making manually. The industrial metaverse is a necessary means to bridge the knowledge gaps by sharing and collaborative decision-making. Considering the safety and stability requirements of the process manufacturing, this article conducts a thorough survey on the process manufacturing intelligence empowered by industrial metaverse. First, it analyzes the current status and challenges of process manufacturing intelligence, and then summarizes the latest developments about key enabling technologies of industrial metaverse, such as interconnection technologies, artificial intelligence, cloud-edge computing, digital twin (DT), immersive interaction, and blockchain technology. On this basis, taking into account the characteristics of process manufacturing, a construction approach and architecture for the process industrial metaverse is proposed: a virtual-real fused industrial metaverse construction method that combines DTs with physical avatar, which can effectively ensure the safety of metaverse's application in industrial scenarios. Finally, we conducted preliminary exploration and research, to prove the feasibility of proposed method.

Zinc rotary kiln is an important equipment in the nonferrous metallurgical industry. Due to unclear internal working conditions, its operation based on experience is random. Digital twin (DT) with virtual–real integration and synchronization ability is a necessary method to realize real-time and accurate monitoring for key variables, while its nowadays practical application face challenge about accuracy and consistency. Therefore, this article proposes a DT driven soft sensing method for key variables in rotary kiln. First, a thermodynamics and chemical reactions coupling model is built after analyzing the mechanism in kiln. Second, key parameters of DT were identified through analysis of limited and multisource data to ensure its consistency. Finally, in DT deployment stage, to realize the DT soft-sensing real-timely, a model reduction and DT distributed computing method was proposed to improve simulation timeliness. Practical application proved that soft sensing method based on DT can accurately and effectively obtain real-time monitoring results of key variables in the rotary kiln.

Modern industrial processes generate many interassociated variables, which are more likely to implicit associations knowledge for describing irregular changes at different times to accurately describe behavior changes. Motivated by this issue, a novel knowledge-data-based synchronization states analysis method is proposed in this article for process monitoring. Its advantage mainly refers to integrating physical–chemical mechanism knowledge to handle the representation of associated relationships between numerous monitor variables. Furthermore, this method utilizes the trend distributions of variable changes to observe the differences between operation states and their parents online, which can maintain the simple, practical, and efficient advantage of data-driven process monitoring. Specifically, global process monitoring can be achieved by the synchronization status exceeding its corresponding threshold ($\chi ^{2}$distribution). At the same time, the local cause of backtracking can also be identified by whether the weighting of eigenvector components of each variable exceeds their corresponding thresholds ($\chi ^{2}$distribution). This novel proposed process monitoring method is applied to one practical hydrometallurgical zinc purification process consisting of copper and cobalt removal processes. The application's comparable performance shows the applicability and effectiveness of this proposed method.

In order to address the simulation requirements arising from limited computational resources, non-reusable complex models, and low-latency demands for edge devices within the realm of the Industrial Internet of Things (IIoT), we have successfully defined and developed a Lightweight Simulation System(LSS) specifically designed for industrial applications. This system draws inspiration from the architectural principles of edge and fog computing. LSS leverages an innovative architectural paradigm known as Function as a Service (FaaS), which demonstrates an even finer granularity compared to the traditional microservices architecture. Within the framework of LSS, we have discretely compartmentalized distinct operational elements, such as data acquisition and storage, model computation, and model visualization, into Docker microservices. This strategic shift in design effectively alleviates the need for developers to be overly concerned with the intricacies of server management and domain-specific industrial knowledge. Consequently, this enables them to shift their focus towards determining "what needs to be accomplished," thereby transforming the granularity of work from being server-centric to a more task-centric perspective. In comparison to simulation platforms within other frameworks, LSS distinguishes itself through its remarkable attributes, which include a lightweight structure, minimal coupling, and superior real-time performance. These unique qualities substantially enhance the collaborative, shared, and reusable capabilities of simulation systems across various dimensions. To substantiate the viability of LSS, this paper provides empirical evidence by applying it to Electrolysis and partial cascade algorithms in the context of hydrometallurgy. The results underscore the fact that the LSS system significantly improves modeling efficiency, fosters the reusability of models and algorithms, and elevates the level of industrial automation across diverse processes.

Concrete cracks are one of the most apparent indicators for possible structural deterioration and need to be periodically inspected. However, for current image-based automated crack inspection techniques, accurate and detailed crack quantification and assessment remain a challenging task. Most of these techniques require high-quality input images, which may be difficult to ensure in practice. Besides, simply merging crack detections from multiple images to generate a large crack map may result in an inaccurate outcome for crack severity assessment. In this article, a novel crack quantification framework is proposed to identify complete crack geometric properties utilizing a set of unordered inspection images. To realize this, cracks in images are detected by an instance segmentation convolutional neural network. Subsequently, the crack segmentations from multiple separate images are systematically aggregated through voxel-based reconstruction and Bayesian data fusion. This framework outputs a crack model that can retrieve accurate geometric properties of each crack segment by recognizing the crack's inherent branching patterns. The capability and performance of the proposed crack quantification framework are validated on cracked concrete specimens in a laboratory setting. Also, a field test on a cracked concrete wall was carried out using images captured by a UAV to demonstrate the efficacy of the proposed framework in practical conditions.

In classification, a large number of features often make it difficult to select appropriate classification features. In such situations, feature selection or dimensionality reduction methods play an important role in classification. ReliefF algorithm is one of the most successful filtering feature selection methods. In this paper, some shortcomings of the ReliefF algorithm are improved, on the problem of poor stability of neighbor samples selection, proposing the method of using the average value of multiple random selection to improve the anti-volatility of the algorithm. And redundant analysis is added to the ReliefF algorithm to eliminate the redundant features. The experimental results show that the improved ReliefF algorithm can effectively establish the classification feature sets, achieve the better classification accuracy.