Biaxial motion control systems are widely employed in manufacturing industries, where improving throughput and reducing machine costs are critical goals. To achieve these, lightweight materials are increasingly being used in structural components, though this often introduces higher flexibility in the machine links, leading to position-dependent precision loss in the end-effector. This article addresses the challenge of maintaining contouring accuracy in such systems by proposing a novel contouring error-bounded control algorithm for biaxial switched linear systems. The algorithm employs model predictive control (MPC) to enforce state, input, and contouring error constraints across different system modes, even when mode switching is not known in advance. While the exact switching signal is unknown, the controller assumes knowledge of the minimum dwell time the system remains in each mode. The proposed algorithm guarantees recursive feasibility and ensures closed-loop system stability. The effectiveness of the method is validated through a high-fidelity simulation of a dual-drive industrial laser machine, demonstrating that the contouring error is consistently maintained within the specified tolerance.

Deep learning-based wood defect detection systems suffer significant performance degradation when switching between species due to domain shift. Unsupervised domain adaptation (UDA) can mitigate this issue, but its effectiveness is limited when only scarce unlabeled target domain samples are available. In this article, we propose vision language model (VLM)-enhanced asymmetric low-rank adaptation (LoRA) diffusion (VALD), a generative augmentation framework that synthesizes target style defect samples to bridge the species gap. VALD employs a two-phase asymmetric LoRA fine-tuning strategy for stable diffusion. Specifically, the first phase trains a full LoRA with class-specific textual inversion tokens in the source domain; the second phase freezes the downprojection matrix and token embeddings, fine-tuning only the upprojection matrix in the target domain to adapt the visual style while preserving defect semantics.In addition, a large VLM is introduced to extract multidimensional visual background descriptions, which are incorporated into generation prompts to enhance sample diversity. Extensive experiments on 12 cross-species wood defect detection transfer tasks demonstrate that VALD consistently improves multiple UDA methods, achieving up to 90.55% accuracy and outperforming state-of-the-art diffusion-based augmentation approaches.

Series Snake robots possess strong adaptability for unstructured environments, but their trajectory tracking control is hindered by nonlinear dynamics and model uncertainties. This article proposes an optimal tracking control strategy based on an actor-critic reinforcement learning framework. The method integrates line-of-sight guidance with serpentine gait generation, while a neural network identification system approximates the solution of the Hamilton-Jacobi-Bellman equation for unknown dynamics. Actor and critic networks are employed to update control policies and cost functions online, reducing dependence on precise models. Rigorous theoretical analysis proves that position and velocity errors achieve semi-global uniform ultimate boundedness. Both simulations and prototype experiments were conducted based on a servo-driven yaw-pitch linkage alternating series snake robot. The results verify that the proposed method can achieve accurate trajectory tracking, rapid convergence, and stable joint control, demonstrating its effectiveness and superiority compared to existing methods.

In the wood industry, machine vision-based measurement systems are widely adopted for the real-time inspection of wood surface defects, effectively ensuring product quality, safety, and improving processing efficiency. However, existing defect inspection algorithms primarily prioritize accuracy on limited datasets, while ignoring the frequent variations in wood species and the instability of practical inspection environments, leading to severe performance decline when applied to unseen wood species. To address this limitation, an efficient wood defect segmentation method for seen and unseen wood species is developed, comprising a data acquisition platform with line scan cameras for high-resolution wood lath imaging, and a lightweight domain generalization-driven segmentation network (distribution consistency-guided network, DCGNet) that enhances generalization capacity without increasing the inference-time computational cost. Specifically, an adaptive species discriminator is adopted during the training process to encourage the extraction of texture-agnostic features, and a novel multi-class distribution discrepancy loss is designed to strengthen intra-class compactness and inter-class discriminability. Extensive cross-species experiments are conducted on four distinct wood species. The results demonstrate that the proposed method can effectively improve defect segmentation performance for both seen and unseen wood species, while achieving superior generalization accuracy across four domain settings.

Inspection robots are increasingly deployed in substations for automated instrument monitoring. However, due to navigation errors and mechanical wear, robots fail to capture inspection images with the required quality and precision. To address this issue, we propose a reinforcement learning-based alignment control method. The proposed approach combines visual perception with an attention mechanism to accurately extract contour features of target instruments. The state representation is formed using the extracted features, raw images, and robot pose information. The alignment task is formulated as a Markov Decision Process (MDP), and a Dueling Deep Q-Network (Dueling DQN) is employed to learn a control policy that guides the robot to adjust its pose toward optimal alignment in real time. Experimental results demonstrate that the proposed algorithm significantly improves alignment accuracy and efficiency compared to conventional algorithm, and provides a promising solution for intelligent inspection in substations.

The substantial energy demands of buildings are increasingly supplied by renewable sources like photovoltaics. However, their intermittency necessitates the integration of stationary energy storage systems (ESS) within building energy management systems (BEMS) to stabilize power and coordinate multi-energy flows. The proliferation of electric vehicles (EVs) facilitates their integration with ESS, forming a combined battery system (CBS) that expands the arbitrage potential and flexibility of BEMS. To fully exploit the potential of CBS in optimizing BEMS operational costs, this paper proposes a deep reinforcement learning (DRL) real-time joint energy scheduling method based on heterogeneous battery systems. We first analyze the aging characteristics of different battery types within the CBS, and propose an innovative degradation assessment framework tailored to heterogeneous energy storage systems in vehicle-to-grid scenarios. This framework introduces a cycle degradation coefficient to provide real-time feedback on battery aging costs, making it suitable for DRL-driven scheduling. To achieve optimized collaborative scheduling of ESS and EVs, we propose an enhanced DRL algorithm incorporating double dueling and prioritized experience replay mechanisms. This algorithm addresses challenges such as complex state features, action coupling, and decreased learning efficiency in heterogeneous energy storage environments. It also prioritizes the travel demands of EV users to promote their participation. Experimental simulations from a real-world commercial building validate the effectiveness of the proposed approach, achieving a 43.39% reduction in system operating costs compared to the mixed-integer linear programming approach under equivalent conditions.

Accurate segmentation of shoe upper processing boundaries is crucial for automated trajectory generation and high-precision robotic control. However, developing a robust method is challenging due to the frequent style changes in High-Mix Low-Volume production. The reliance on large-scale annotated datasets renders traditional supervised methods impractical due to the prohibitive cost of annotation and retraining. To address these issues, a multimodal-based point cloud segmentation strategy is proposed for shoe upper processing boundaries. First, an unsupervised adaptive local spectral contrast filtering algorithm is designed to remove large amounts of background noise and isolate potential target regions by exploiting boundary color characteristics. Then, an unsupervised dynamic ellipsoidal neighborhood color-spatial region growing algorithm is developed based on geometric features of slender and closed boundary shapes to suppress interferences flanking the boundaries. Finally, a Siamese network is designed to perform few-shot matching against boundary templates exported from Shoemaster, effectively decoupling intrinsic boundary signals from complex extrinsic interferences to achieve precise segmentation. Experimental results demonstrate that the proposed method achieves a stable mean Intersection over Union (mIoU) of approximately 0.80. Compared to existing supervised and unsupervised baselines, this strategy exhibits superior generalization across diverse styles and effectively resolves the data dependency bottleneck.

The integration of photovoltaic (PV) systems, stationary energy storage systems (ESSs), and electric vehicles (EVs) alongside demand response (DR) programmes in industrial parks presents opportunities to reduce costs and improve renewable energy utilisation. Coordinating these resources is challenging because office and production zones have distinct operational objectives, and battery ageing costs are often ignored. This paper proposes a DR-based energy management framework that jointly optimises grid interaction costs, thermal comfort, EV departure state-of-charge requirements, carbon emissions, and battery ageing. We model heterogeneous load characteristics using a dynamic energy distribution ratio and incorporate dispatch-level ageing models for both ESS and EV batteries. The problem is formulated as a Markov decision process (MDP) and solved with a deep deterministic policy gradient (DDPG) algorithm. High-fidelity simulations using data from a practical industrial park in China show the framework maintains indoor comfort while significantly reducing total operating costs, yielding savings of 44.58% and 40.68% compared with a rule-based DR strategy and a conventional time-of-use arbitrage approach, respectively.

To achieve high-quality upper spraying adhesive under multiple influencing factors, this paper proposes a multi-objective trajectory planning method for shoe upper spraying adhesive manipulators based on quintic B-splines and the Beluga Whale Optimization. Quintic B-spline interpolation models are employed for trajectory planning of the articulated space manipulator. This method ensures a continuous and stable spray path while maintaining jerk continuity, all while keeping velocity, acceleration, and jerk within the kinematic constraints. A time-energy-jerk optimization model is established, and the Beluga Whale Optimization is employed to address the optimal trajectory planning problem under time-energy-jerk constraints. The comprehensive objective of time-energy-jerk is divided into energy-jerk and execution time. These sub-objectives are then optimized sequentially. The trajectory simulation results demonstrate that the proposed method achieves efficient motion trajectories, satisfying the requirements of high precision, energy efficiency, and high performance.

In autonomous industrial inspection, identifying a specific device is a critical problem since appearances of devices are similar and they are densely deployed in the factory. The machine vision is not sufficient for the device identification due to navigation errors of inspection robots and similar appearances of devices. The devices in the Industrial Internet of Things (IIoT) generally support wireless communication, thus using wireless signals to improve the accuracy of device identification becomes a reasonable solution. However, it is a challenge to implement this idea since the wireless signals are unstable and poorly distinguishable in industrial fields with densely deployed devices. Motivated by the fact that the inspection robot can autonomously adjust its position and the strength of wireless signals is strongly correlated with distance, in this paper, we propose a proximity estimation with position adjustment based on signal-spatial relations (PEPA-S) algorithm for similar industrial devices identification. The signal-spatial relations include the signal relation and the spatial relation between industrial devices, which are adopted in the PEPA-S algorithm to improve its performance with the help of position adjustment. The experiments demonstrate that the PEPA-S can provide guaranteed estimation accuracy and improve the efficiency of similar industrial devices identification.

The growing penetration of electric vehicles (EVs) creates significant opportunities to enhance renewable energy utilization efficiency and improve the flexibility of smart grids through their integration with stationary energy storage systems (ESS). However, the heterogeneous degradation characteristics of ESS and EV batteries are often overlooked, resulting in inaccurate system cost estimation. In addition, the absence of effective incentive mechanisms that align the interests of multiple stakeholders significantly constrains EV participation in vehicle-to-grid (V2G) programs. To tackle these challenges, this paper proposes a differentiated degradation-aware and V2G-incentivized deep reinforcement learning (DRL) framework for real-time coordinated ESS–EV energy scheduling. Specifically, tailored degradation models are developed for ESS and EV batteries according to their distinct characteristics. These models are then embedded into the DRL decision-making process to enable real-time degradation awareness. To incentivize EV participation in V2G operations, a comprehensive incentive mechanism is designed by jointly considering travel demand guarantees, degradation compensation, and economic rewards. Finally, a DRL-based algorithm integrated with dynamic electricity price forecasting is proposed to maximize system operating profit. It adopts a dynamic ensemble strategy to adapt to time-varying price patterns and incorporates a prioritized experience replay mechanism to address the sparsity of coordinated ESS–EV strategies. Simulation results demonstrate that the proposed method improves system operating profit by 9.67%–30.55% compared with three baseline DRL approaches.

Intelligent recognition of substation building electrical wiring diagrams is a key step to ensure the correctness and safety of substation design and to improve the efficiency of engineering operations. However, existing methods struggle with severe text-graphic interference, extreme scale variations of symbols, and their inability to explic itly model complex, imbalanced topological connections. To solve these challenges, this paper proposes a novel framework for electrical wiring drawing recognition, formulated as a topology-aware graph learning problem. First, we propose a CAD-image fusion strategy that leverages deterministic vector topology to guide visual pars ing, effectively suppressing text-graphic interference to extract high-fidelity physical wiring structures. Second, by formalizing heterogeneous devices and their physical connections into a formal graph data structure, we develop a GraphSAGE-based classifier capable of decoding the complex, multi-hop configurations of complete electrical assemblies. Last, to handle unique geometric variations and structural sparsity, domain-tailored optimizations are introduced, including Dynamic Snake Convolution (DSConv) for capturing elongated busbars and a Gated-Attention mechanism to regulate extreme topological imbalance during graph aggregation. Experimental results on real-world 110 kV substation diagrams demonstrate the effectiveness of our framework, achieving 96.4% mAP for element detection, 98.95% edge precision for topology extraction, and 95.47% accuracy for assembly classification.

Wood surface defect detection is critical for quality control and production efficiency in the wood industry. While current deep learning-based methods have yielded promising results, their reliance on full supervision demands extensive and high-quality annotations, particularly for pixel-level segmentation tasks. To address this limitation, this study proposes a two-stage unsupervised method for wood defect segmentation. Firstly, a reconstructive network is designed to learn the feature distribution of normal patterns. By combining anomaly generation and mapping-reconstruction strategy, it can achieve anomaly simulation without labels. Additionally, a pixel correlation module is introduced to enhance global feature representation and improve the reconstruction fidelity. Then, a discriminative network is constructed to locate anomalous regions by analyzing the residuals between the reconstruction results and the anomalous input. Furthermore, an efficient feature fusion module is introduced to refine detailed information in the discriminative network. Experiments conducted on wood defect dataset demonstrate the effectiveness of the proposed method.

Intelligent recognition of substation electrical wiring diagrams is key to ensuring design correctness, safety, and operational efficiency. However, current methods struggle with stability and generalization due to drawing diversity and complexity. To solve this, we propose a recognition method based on CAD data, leveraging object detection and vector data parsing. First, we use YOLOv8 to capture basic electrical elements. Second, leveraging the parsable nature of CAD data, we design a hierarchical extraction strategy that uses element detection results to accurately extract wiring structures for topological analysis. Experimental results on real-world diagrams show 94.1% mAP for element detection and high-fidelity wiring extraction.

Wood surface defect segmentation is extremely critical for defect refinement and quality control of wooden products. However, it is a challenging task to develop an efficient method with current algorithms due to the complicated characteristics of wood defects with obscure boundary, intraclass difference and interclass similarity. To address these issues, a lightweight network via multi-dimension boundary perception and guidance is proposed for precise segmentation of wood defects. At first, based on the Segformer, a boundary prediction branch is added to enrich detailed boundary information in the encoder, and supervised by the Gaussian signal and cosine similarity, to balance the effect of the boundary gradient information. Then, a double-flow enhancing module is designed to integrate the adjacent level features, by embedding two enhancing paths, to adaptively generate discriminative information of the defects. Finally, a binary segmentation head following the predicted map is introduced to strengthen the penalty for the false prediction results of the boundary. Experimental results demonstrate the proposed method outperforms the state-of-the-arts on our wood surface defect dataset, as well as on three public datasets.

Machine vision-based meter reading technologies have been developed to monitor the status of devices in industrial sites. However, the integrity of the pointer can be destroyed by the occlusion on the meter, such as dirt or rain. In this case, the pointer detection becomes a challenging task in the meter reading. To overcome this challenge, in this paper, the pointer generative adversarial network (Pointer-GAN) is proposed for pointer mask generation. Specifically, an occlusion simulation method is developed in the data preprocessing to provide sufficient images of occluded meters in the training phase, and the dilated convolution is adopted in the residual block to strengthen the correlation among pointer features in the long-range. The next, the multi-scale attention mechanism is designed for preventing the pointer feature in the low-level from being affected by the noise. Finally, the dense dilated convolution block is utilized to integrate the pointer feature in the low- and high-level for the pointer mask generation. The experiments demonstrate that the Pointer-GAN can generate the pointer mask with higher accuracy for the meters under occlusion compared to the other methods, thereby improving success rates of reading meters in different occlusion scenarios.

The efficient pointer meter reading methods have been proposed based on machine vision to replace time-consuming manual inspections for the industrial monitoring. However, the interference factors, such as rain or dirt, can occlude meter, which poses obstacles in the recognition and labeling of pointer and scales. To solve these problems, we propose a multi-task network with pointer and main scale detection (PMSD-Net) for the occluded meter reading with synthetic data generation technology. Specifically, dense parallel dilated convolution block is proposed for correlating the pointer and main scale features with large receptive field. Multi-scale feature fusion is designed to purify noisy features for the detailed information extraction. The relation reconstruction mechanism is designed to reconstruct the feature relation under severe occlusion. Moreover, the keypoint detection branch is designed to detect meter center and pointer tip according to the segmented pointer, which can identify changeable position of the segmented pointer tip to determine the pointer orientation. Finally, the synthetic data generation technology is developed to generate massive labeled data with simulated interference factors in the meter for the training, which enhances the generalization ability of PMSD-Net in various occlusion scenes. Experimental results indicate that PMSD-Net can segment more accurate regions of pointer and main scale and detect the changeable position of pointer tip for occluded meters, thereby improving the accuracy in reading occluded meters.

Gears are the core components of transmission systems, and their health status is critical to the safety and stability of the entire system. In order to efficiently identify the typical fault types such as missing teeth and broken teeth in gears, this paper collects a rich sample under complex backgrounds from different shooting angles and lighting conditions. Then a hierarchical approach is used to describe gear faults on the image. The gear samples are first segmented for image extraction and then finely labeled for gear fault regions. In addition, imbalanced datasets are produced to simulate the environment with fewer fault samples in the actual industrial process. Finally, a semi-supervised learning framework is trained based on the above method and applied in actual environment. The experimental results show that the model performs well in gear target detection and fault diagnosis, demonstrating the effectiveness of the proposed method.

Building electrical load forecasting, as a necessary foundation for building energy management, is of great significance for building energy efficiency and sustainable urban development. However, the accuracy of forecasting can hardly be guaranteed due to the stochastic nature of occupant behavior. To overcome this challenge, this paper proposes a data fusion-based building electrical load forecasting method with occupancy data obtained by wireless sensing technology. Firstly, a wireless sensing scheme is developed, which utilizes pre-existing wireless devices within the building energy management system (BEMS), offering a cost-effective means of obtaining occupancy information without violating occupant privacy. Moreover, to estimate the pattern of occupant behavior in the entire building, an improved stacked sparse auto-encoder (ISSAE) model is developed, which involves unsupervised feature fusion from information sources of varying significance. Finally, to cope with the time-varying and strongly fluctuating building load, a multi-source data fusion forecasting model based on the ensemble deep random vector functional link (edRVFL) is proposed. This model integrates the contributions of the latest accuracy and diversity through the ranking-based dynamic integration strategy. The effectiveness of the proposed method is validated in a commercial building. The experimental results demonstrate that, compared with the load forecasting scheme without occupancy information, the proposed method can improve the forecasting accuracy on RMSE and MAPE by 13.21% and 14.97%, respectively, while cost-effectiveness and privacy are ensured.

As a core component of electric vehicles, lithium-ion batteries continue to face the challenge of balancing charging speed with battery health maintenance, which directly impacts driving range and charging efficiency. Existing fast charging solutions either rely heavily on precise battery models or lack sufficient understanding of the underlying battery mechanisms. To address these challenges, this study proposes a reinforcement learning-based fast charging strategy that incorporates the battery's state of health. First, a high-fidelity electrochemical model considering battery aging is introduced in this study. Then, a fast charging optimization problem with SOH constraints is formulated. Subsequently, a reinforcement learning algorithm is designed to autonomously search for the optimal charging strategy. Finally, experimental results demonstrate that the proposed charging strategy achieves a 44% reduction in charging time compared to the conventional constant current-constant voltage (CC-CV) method, while achieving a comparable capacity loss over 100 cycles.