The dexterous multi-fingered robotic hand， serving as a key end-effector， is pivotal for enabling robots to perform fine-grained grasping and compliant manipulation. Its advancement holds significant importance for promoting automation in manufacturing， enhancing the intelligence of service robots， and expanding applications in specialized environments. Focusing on humanoid multi-fingered dexterous hand technologies， this paper systematically reviews the current state-of-the-art and future trends. It begins by elucidating the fundamental concepts， system architecture， and typical characteristics of dexterous hands. This is followed by a comprehensive of research achievements from domestic and international teams and commercially available mainstream multi-fingered dexterous hand products， covering various degrees-of-freedom designs and their respective hardware and software implementations. Key technologies， including core hardware components， multi-modal sensory fusion， and control strategies， are critically analyzed. The paper subsequently summarizes practical applications across domains such as industrial assembly， daily life assistance， and operations in extreme environments. Current challenges， particularly in reliability， multi-modal coordination， generalization capability， human-robot safety， and integration and application， are identified. Finally， future research directions are prospected from multiple perspectives， including standard establishment， novel mechanical structures， advanced multi-modal perception and fusion， bionic evolution， and embodied intelligence， aiming to provide valuable insights for in-depth research and groundbreaking applications of dexterous hands.

To address the challenges of frequent bearing failures under complex working conditions， as well as the low real-time performance and strong dependence on manual feature extraction in traditional diagnostic methods， this paper proposes a bearing fault diagnosis method based on a deep learning model combining a Multi-Scale Convolutional Neural Network （MSCNN） and Long Short-Term Memory （LSTM）， and develops an intelligent bearing health management system. The system adopts an end-to-end diagnostic workflow， directly taking raw time-domain vibration signals as input. It extracts hierarchical local features across different frequency domains through MSCNN， and captures the temporal evolution of fault characteristics using LSTM， thereby achieving high-accuracy automated fault classification. To enhance the interpretability of diagnostic results and support intelligent maintenance decisions， the system integrates the Chinese large language model iFLYTEK Spark， which generates natural language diagnostic reports and maintenance suggestions through standardized prompts. The system is deployed on a domestically developed Phytium quad-core processor platform， ensuring full autonomy and reliability of both hardware and software components for industrial applications. Experimental results show that the proposed system achieves an average classification accuracy of 98.46% on the CWRU bearing dataset， and 96.73% on the AITHE bearing fault dataset， demonstrating strong robustness and cross-dataset generalization under complex and noisy conditions. With real-time visualization of diagnostic results and maintenance recommendations through a human-machine interface （HMI）， this system provides a reliable and intelligent solution for equipment health management and predictive maintenance.

To ensure the safety and stability of the power transmission systems and achieve comprehensive inspection and maintenance of newly constructed power transmission towers， this paper proposes a centrally symmetric quadrupedal humanoid climbing robot designed for existing foot pegs used by maintenance workers to climb towers. Each limb is configured with 3-1-2 arrangement. At the end of each limb， a large-tolerance， semi-enclosed hook-type gripping tool is designed specifically for the foot pegs. This tool features high tolerance， eliminating the need for precise end-effector positioning and enabling rapid engagement with the foot pegs. Humanoid climbing gait planning method is developed， facilitating the robot's full-range climbing of the power transmission tower by quickly hooking and gripping the foot pegs using the hook-type tool. Targeting a 40-meter-high self-supporting transmission tower， the robot's full-range climbing dynamics model and simulations were completed. Simulation results demonstrate that the proposed robot configuration can achieve humanoid full-range climbing of the tower， with a climbing time from the base to the top of less than 30 minutes， matching the efficiency of maintenance personnel. This provides a feasible solution for robotic maintenance applications in power transmission towers.

To address the problems of scattered knowledge systems in ship assembly and ineffective mining and utilization of massive process data， this paper proposes an automatic knowledge graph construction technology for the shipbuilding domain based on large language models. This method uses large language models to convert unstructured and semi-structured ship data into structured data to build a ship process corpus. It models ship ontology knowledge structure with the assistance of large language models， designs an instruction prompting framework for ship assembly domain， and achieves efficient entity-relationship extraction， to complete the automatic construction of knowledge graphs. Additionally， the method uses triple sets constructed by general large language model instruction prompts as fine-tuning training sets to further fine-tune specialized small language models， ensuring the security of specific private ship data while reducing computational resources. Experimental results show that this method outperforms traditional baseline models in key metrics such as accuracy， providing a new technical approach for knowledge management and intelligent upgrading in the shipbuilding domain.

As an important component of intelligent manufacturing and intelligent operation and maintenance systems， industrial inspection robots are playing a key role in various complex industrial scenarios. With the continuous progress of deep learning， multi-sensor fusion， and autonomous navigation technologies， industrial inspection robots have significantly been improved in terms of accuracy， efficiency， and adaptability. This article systematically reviews the concept， key technologies， and typical applications of industrial inspection robots， and focuses on analyzing the research status of core technologies such as perception and recognition， autonomous positioning and navigation， advanced control， and intelligent decision-making. It also assesses the maturity and industrialization progress of current technologies by combining practical applications in fields such as power， workshops， and special environments. Despite significant achievements in this field， challenges still exist in perception accuracy， dynamic environment adaptability， and task execution intelligence. The development of key technologies is expected to continue in the directions of multi-source data fusion， autonomous learning， and collaborative operation. The article aims to provide a systematic reference and guidance for future research and industrial development of industrial inspection robot technology.

To address issues such as low detection accuracy， slow speed， missed and false detections， and large model parameter sizes in complex scenarios from a UAV perspective， this paper proposes an improved RBGE-YOLO algorithm model. Firstly， RFAConv is introduced in the backbone network to replace the original Conv， enhancing the model's ability to extract and fuse image features. Secondly， the neck network is reconstructed using BiFPN-GLSA to improve feature fusion and spatial feature utilization efficiency. Thirdly， a dual-layer small target detection structure is designed to strengthen the feature information of small targets. Finally， the Inner-EIoU loss function is utilized to address the limitations of IoU. Experiments on the VisDrone2019 dataset show that RBGE-YOLO improves Precision， Recall， mAP@0.5， and mAP@0.5：0.95 by 4.7%， 2%， 3.6%， and 2.5%， respectively， compared to the original YOLOv8s， while reducing the number of parameters by 16.4%. This achieves model lightweighting while significantly enhancing detection performance.

To address the inefficiency of traditional manual visual inventory counting and the high deployment costs of existing automated solutions in Automated Storage and Retrieval Systems （AS/RS）， this paper proposes an intra-warehouse visual inventory system based on modular visual devices. A retrofit-free stacker-accessible modular visual inventory device is designed. On the basis of a YOLOv8-powered visual inventory algorithm for multi-surface information fusion from a single view， the accurate counting of complex stack patterns （e.g.， non-full stacks and staggered stacks） is effectively solved by combining front pallet layer identification with top pallet layer counting. The system also features a non-intrusive integration architecture between the Warehouse Visual Stock System （WVSS） and the existing Warehouse Control System （WCS） via a database， enabling dynamic task scheduling and data closed-loop. Experimental results on four palletized cargo datasets demonstrate a stack quantity recognition accuracy of 96.3% with a processing time of 0.11 seconds per storage location. This solution provides a new engineering path for automated warehousing， characterized by high precision， low deployment cost， and minimal operational disruption.

In recent years， indigenously developed， aerospace-specific 3D structural design systems in China have undergone robust development， with notable achievements in the R&D of core components. However， with the widespread adoption of Large Language Models （LLMs）， establishing an effective interface between 3D structural design and AI-driven methodologies remains a central challenge. Furthermore， existing LLMs lack the capacity for precise reasoning over 3D geometry and complex physical fields， such as aerodynamics， which precludes their direct application in the intelligent design of aircraft structures. Among aerospace structural components， the aircraft wing is critical for generating lift. Its design process is highly complex， heavily reliant on expert experience， and tightly coupled with aerodynamic performance. Consequently， traditional design paradigms are characterized by lengthy iteration cycles and substantial costs. To address this challenge， this paper presents Airfoil-LLM， an intelligent design interface for the 3D modeling of aircraft wings， using the wing as a representative case study. Based on the Transformer architecture， this interface integrates natural language encoding with the decoding of CAD modeling sequences to enable intelligent and automated 3D wing generation. To support model training and validation， we have constructed a large-scale 3D wing design dataset. This comprehensive dataset comprises parameterized 3D CAD models， a wide spectrum of flight conditions from subsonic to supersonic regimes， key aerodynamic performance metrics， and multi-level textual descriptions. Experimental results demonstrate that Airfoil-LLM is capable of deeply comprehending textual descriptions ranging from simple geometric attributes to complex， coupled "geometry-performance" requirements. The system generates 3D models that align closely with the design targets in both geometric shape， achieving a maximum Intersection over Union （IoU） of 0.831， and aerodynamic performance.

The advancement of Industry 4.0 necessitates the deployment of intelligent， low-cost robotic systems on edge devices. However， the high computational complexity of Deep Reinforcement Learning （RL） algorithms presents a major obstacle to their implementation on resource-constrained platforms such as the Raspberry Pi. To overcome this challenge， this paper introduces a lightweight RL framework tailored for industrial robot sorting tasks. The core contributions are threefold： First， we propose a joint compression method combining Gradient Sensitivity-guided structured Pruning （GS-Pruning） with hierarchical quantization， which reduces model size by over 90% and achieves real-time inference below 35 ms on a Raspberry Pi while preserving policy accuracy. Second， we design a Dynamic Weight Adaptive Reward function （DWAR） that balances sorting efficiency， motion stability， and energy consumption， successfully suppressing robotic arm jitter and cutting average energy use by 18.1%. Third， we construct an end-to-end deployment system， RPi-EdgeRL， featuring a multi-threaded pipeline and a safety watchdog to guarantee stable and efficient autonomous operation. Experiments conducted on a FR3 collaborative robot validate our framework， achieving a 93.5% success rate in complex sorting tasks and confirming the feasibility and superiority of this low-cost， high-efficiency solution for real-world industrial applications.

Based on human-machine coupling modeling and biomechanical analysis， a shoulder-elbow rehabilitation assistive device featuring 5 degrees-of-freedom （DoF） rotational joints and 3-DoF sliding adjustments was developed. Motion capture experiments were conducted to obtain personalized scaled musculoskeletal models and reproduce upper limb rehabilitation movements through inverse kinematics. Utilizing Hill-type muscle models and the Computed Muscle Control （CMC） algorithm， the study analyzed muscle forces and energy consumption during rehabilitation training. Results demonstrated significant reductions in muscle forces for primary movers under assistive support： the long head of biceps brachii showed a 51.34% average force reduction， while the lateral head of triceps brachii exhibited 49.05% decrease. Energy consumption decreased by 30.74% and 36.56% in the long and short heads of biceps brachii respectively， with peak reductions exceeding 40%， indicating sustained unloading effects during elbow motion. Secondary muscles including the posterior deltoid and medial head of triceps brachii maintained moderate 10% reductions， balancing unloading requirements with joint stability to prevent over-intervention. The analysis confirms that the rehabilitation assistive device effectively reduces muscular burden and energy expenditure during training， mitigates muscle overload risks， and provides efficient assistance for patient rehabilitation.

To address the issues of suboptimal performance， and high rates of missed detection and false detection in steel surface defect detection technology in industrial production environments， an improved YOLO11 algorithm called GCI-YOLO11 has been proposed. Firstly， in the feature extraction part， the GC-C3k2 module based on the GCNet attention mechanism was designed to enhance the algorithm’s capability to extract contextual feature information from images. Secondly， the CARAFE upsampling algorithm was introduced in the neck part to enable the algorithm to aggregate contextual information within a large receptive field， reducing the loss of feature information during the upsampling process. Finally， Inner-CIoU was used to replace CIoU for loss function optimization， and auxiliary regression box was introduced to improve detection accuracy and model generalization capability. Experimental results show that， GCI-YOLO11 achieved improvements of 2.9% and 2.3% in mAP₅₀ and mAP_50-95 on the NEU-DET dataset， and 1.6% and 0.3% in mAP₅₀ and mAP_50-95 on the GC10-DET dataset， showing better detection performance.

In order to reduce the monitoring and operating pressure of operators during deep peak shaving， an intelligent desulfurization control system is proposed to address the problems of poor measurement accuracy and large inertia and delay in the controlled objects that prevented long-term stable automatic operations. By using BP neural network， a mapping relationship is constructed between signals such as flue gas flow rate， SO₂ concentration in the raw flue gas and slurry pH to achieve soft measurement of slurry pH value； Replacing conventional PID with variable structure predictive control and combining it with more accurate and reasonable feedforward signals ensures the control effect of the desulfurization system under rapidly changing load and coal quality conditions. Later， utilizing the unit ICS system， the desulfurization intelligent control system is successfully applied to a 650 MW unit. The operation results show that after the system is put into operation， the SO₂ concentration at the outlet is stably controlled within 25 mg/m³， and the deviation between the pH value of the slurry and the set value is kept within 0.2， and there are no significant fluctuations during the variable load and pH meter flushing process. The desulfurization is automatically put into operation for a long time， effectively reducing the operating pressure of the operators.

Industrial robots， with their high flexibility and large working range， have gradually become another important processing equipment besides CNC machine tools in national strategic fields such as aerospace and maritime industry in China. The dynamic characteristics of the robot end are dominated by its joints. To improve the dynamic performance of the robot， it is necessary to start with its weak joints， make improvements and innovations on the basis of the existing series form， and explore new high-stiffness driving methods and robot configurations. A 2-RPR robot for milling large propellers is proposed， which includes a six-axis robot main body and a double electric cylinder branch chain. The translation of the electric cylinder drives the rotation of the robot joint， thereby improving the stiffness of the whole robot. In order to meet the processing space requirements of large propellers， the length parameters of each link of the robot are optimized based on the genetic algorithm to realize the optimization of the working space of the whole robot， so that the working space meets the processing range of a single blade and has the maximum utilization rate.

Electric vehicle charging and battery swapping stations face multiple operational challenges including low service efficiency， poor economic benefits， and weak grid interaction capability. A V2G coordinated scheduling model based on hierarchical deep reinforcement learning is proposed which effectively reduces decision complexity through a collaborative architecture of strategic and tactical layers， and significantly enhances system responsiveness. Empirical research demonstrates that the model exhibits substantial practical value in actual charging station environments， primarily reflected in reasonable growth of operational revenue， optimized energy utilization efficiency， significant improvement in service quality， and effective reduction in user waiting times. Compared with traditional scheduling methods， the SAC algorithm adopted in this study shows stronger adaptability and stability when facing complex decision environments， effectively responding to uncertainties such as traffic flow fluctuations and electricity price changes. The research results provide an implementable intelligent scheduling solution for electric vehicle charging and battery swapping stations， offering valuable technical reference for addressing actual operational issues in the industry， and contributing positively to the sustainable development of the electric vehicle industry.

Facing the critical strategic demand for enhancing the resilience and security of industrial and supply chains at the national level， the paper focuses on digital twin technology as a key enabler for driving the digital and intelligent transformation of warehousing and logistics systems. A three-stage evolutionary trajectory is systematically outlined， progressing from static modeling to dynamic synchronization and ultimately to intelligent decision-making. In view of the structural challenges in traditional warehousing and logistics systems such as data silos， lagging equipment maintenance， rigid processes， insufficient flexibility， and lack of holistic optimization， this study constructs a systematic empowerment pathway encompassing five dimensions： “omni-domain data integration—intelligent operations and maintenance reconstruction—process simulation optimization—flexible collaborative scheduling—global decision simulation.” The construction of a unified data foundation enables standardized access and high-quality governance of multi-source heterogeneous data； Deployment of a predictive maintenance platform significantly enhances equipment reliability and system continuity； Application of virtual simulation and dynamic optimization technologies achieves intelligent restructuring of warehouse operations and efficiency multiplication； The construction of an elastic resource scheduling mechanism enhances the adaptability of the system to dynamic demands； By building a full-chain simulation decision-making sandbox， the system is boosted from empirically driven local decisions to data- and model-driven global autonomous optimization. In the practical application within the chemical fiber industry， this technology system has increased production efficiency by 5%， reduced operational costs by 15%， improved equipment utilization by 10%， and shortened fault-handling time by 30%. Looking ahead， digital twin technology will evolve toward “holistic coordination and intelligent symbiosis，” providing critical support for constructing autonomous as well as controllable modern warehousing and logistics systems and cultivating new productive forces.

In order to adapt to the characteristics of complex electronic equipment， such as multi-variety， variable batch， multi-level blind matching and vertical interconnection， a six-degree of freedom heterogeneous assembly robot arm has been independently developed and applied to the assembly line of complex electronic equipment. Firstly， the structure composition， configuration advantages and problems in practical application of the heterogeneous six-axis manipulator are introduced. Secondly， the forward and inverse kinematics algorithm of the heterogeneous six-axis manipulator is established by using D-H parameter method， and the kinematics model of the heterogeneous six-axis manipulator is constructed. Then the calibration algorithm， trajectory planning algorithm， collision control algorithm and other methods of the heterogeneous six-axis manipulator are studied. Finally， the field calibration experiment and MATLAB simulation analysis are used to verify the motion planning method， which proves the rationality and practicability of the relevant methods.

To address the challenges faced by mobile robots in overcoming obstacles in unstructured environments such as agricultural inspections and disaster rescue， this study proposes a design scheme for a four-wheeled mobile chassis that integrates a rocker-steering suspension with a crank-slider mechanism. First， kinematic and dynamic models of the walking mechanism were established to analyze the influence of key configuration parameters （e.g.， support wheel center distance， hinge distance） on terrain adaptability and load platform posture. A multi-objective optimization method was employed to determine the optimal parameter combination （L_F =200 mm，k₁=0.9）. Second， a three-dimensional virtual prototype was developed by integrating a crank-slider mechanism and symmetric frame design. Dynamic simulations conducted on the RecurDyn platform validated the chassis performance in traversing 18 mm speed bumps and 20 mm semi-cylindrical obstacles， showing pitch angle fluctuations within ±3°and peak torque demand ≤15 N·m. Finally， prototype tests demonstrated that the chassis can stably cross 90 mm speed bump-type obstacles under a 75 kg load， with a linear motion speed of 1.8 m/s and a path deviation of less than 20 mm/5 m. The results indicate that this design significantly enhances the terrain adaptability of mobile robots in unstructured environments， providing a reliable mobile platform for agricultural inspection， logistics， disaster rescue， and similar scenarios.

In response to the challenges posed by complex microhole backgrounds in frames， a multitude of small and medium-sized defective targets， and the high degree of shape randomness encountered in mobile phone visual inspections， we have developed an enhanced YOLOv8-burr model based on YOLOv8 model. This model incorporates a lightweight global attention transformation module， which leverages packet convolution， within the network neck region. It also integrates a multi-scale feature extraction module into the backbone and employs a polarization self-attention mechanism along with a CARAFE operator in the network sampling stage. These innovations enable the model to harness global feature information and multi-layer channel details for more precise detection of small target defects. The experimental results show that the improved model has a size of 14.4M and can achieve 92.1% accuracy of microhole defect recognition， and in the category of "burr" defects that are difficult to identify， the accuracy has been improved by 10.4% compared with the original model before improved， which meets the identification accuracy requirements of the robot for the identification of microhole machining defects in the mobile phone frame.

Aiming at the critical problem that real-time industrial defect detection systems are difficult to balance detection speed， accuracy and computational resource constraints in edge computing environments， a fast lightweight industrial defect detection architecture based on an efficient hybrid state space model is proposed. The architecture designs a C2f_EfficientViM_CGLU fast feature extraction module that deeply integrates the global sequence modelling capability of the visual state space model with the efficient local feature enhancement mechanism of convolutional gated linear units， achieving fast and efficient extraction of complex defect features. The HSM-SSD （Hidden State Mixer based State Space Duality） efficient state space modeling mechanism is introduced to process long sequence dependencies with O（n） linear complexity， significantly improving the fast recognition capability for irregularly shaped and sparsely distributed defects. A Slimneck fast lightweight feature fusion network is constructed through GSConv （Ghost Shuffle Convolution） sparse convolution and VoV-GSCSP （Variance of Variance Ghost Shuffle Cross Stage Partial） efficient feature fusion strategies， achieving significant improvements in inference speed and extreme model compression while ensuring detection accuracy. Comparative experimental results on NEU-DET and APDDD standard datasets show that the proposed network architecture achieves mAP50 of 92.13% on NEU-DET dataset， improving 9.77 percentage points compared to the baseline model YOLOv8n， with only 2.9 M parameters and 7.7 GFLOPs computational complexity， reducing parameters by more than 93% compared to the traditional Faster-RCNN method. The mAP50 on APDDD dataset reaches 89.68%， validating the good generalization performance and fast detection capability of the method. This study provides a theoretical foundation and an efficient and feasible fast detection technical solution for real-time quality control in Industry 4.0 intelligent manufacturing environments.

With the rapid development of computer vision and artificial intelligence， object detection technology has been widely applied in the field of fire protection. A smart fire extinguisher dial recognition device based on the improved YOLOv5 algorithm has been designed and implemented to address the low efficiency and misjudgment of manual recognition of fire extinguisher dials in traditional fire equipment maintenance and repair processes. This device integrates the Raspberry Pi 4B computing core and a high-resolution camera， combined with a closed enclosure， a light shielding side panel， and an adaptive fill light system， effectively suppressing external interferences and adapting to different fire extinguishers of varying capacities. At the algorithm level， the CBAM attention module and BiFPN feature fusion network are introduced to optimize the feature extraction ability and multi-scale object detection performance of the YOLOv5m model. At the same time， an image segmentation strategy is proposed to enhance the recognition accuracy of the model for small targets in complex backgrounds. The improved model has higher accuracy and precision. Onsite tests have shown that the device has a 100% accuracy rate in recognizing the pressure state of 3 kg， 4 kg， 5 kg， and 8 kg fire extinguishers in complex industrial environments. After integration with the intelligent IoT system for vehicle-mounted fire-fighting equipment diagnosis， the detection data of the device can be uploaded in real time to the cloud platform for further data analysis and status evaluation of the system. In the maintenance and repair process of fire extinguishers， it demonstrates high work efficiency and automation level， providing strong technical support for the maintenance and repair of fire-fighting equipment.

Spiral baffle is an important element in shell and tube heat exchangers. Effectively obtaining the spatial position information of the curved spiral baffle surface is the key to solving the tube hole machining of spiral baffle. Aiming at the problems of weak texture， multiple occlusions， reflective surface， and difficulty in meeting industrial requirements on accuracy and time for 3D reconstruction， this paper proposes a matching algorithm that combines feature detectors with dense vision and a variable field-of-view attention module， to obtain more feature points in weekly textured and reflective areas； The cascaded CasMVSNet architecture for dense point cloud reconstruction for spiral baffles is used to reduce surface voids and shorten reconstruction time. The experimental results on the self collected dataset show that the proposed 3D reconstruction technique yields a perfect spiral baffle surface with almost no voids or noise in weak texture areas. Compared to other algorithms， the reconstruction accuracy has improved by 3.96%， the cumulative error curve area of pose estimation has increased by 6.48%， and the reconstruction time has been reduced to within 180 seconds. This proves the reliability and effectiveness of the reconstruction technique.

Aiming at the challenges such as insufficient feature extraction capacity， and low accuracy in the fault diagnosis for asynchronous motors， an asynchronous motor fault diagnosis model based on convolutional neural network （CNN）-Transformer-BiLSTM is proposed. By linking large-scale convolutional kernels with the BiLSTM module to extract the temporal features of fault vibration signals， the model uses a gate mechanism and bidirectional temporal learning mechanism to effectively learn the feature relationships between fault signals at multiple moments. By connecting small-sized convolutional kernels with the Transformer module， the model further increases its ability to extract temporal features. The model uses a multi-head attention mechanism to achieve parallel and efficient processing of feature sequences and outputs a diagnosis result through a Softmax classification. Under 10 dB noise interference， the proposed model was compared with the MSCNN-LSTM-Attention， MSCNN， BiLSTM， and 1DCNN models. The comparison test results show that this model can effectively extract fault features， and the fault diagnosis accuracy rates were increased by 4.3%， 9.7%， 17.2%， and 18.5% respectively， demonstrating that this model has a higher fault accuracy under noise interference.

To solve the problem of accidents caused by damaged crane hooks in current industrial environment， and the low efficiency of crane loading and unloading， an improved YOLOv8n crane hook identification algorithm is proposed. Firstly， AKConv module is introduced to replace the Conv module in the backbone network. This module gives arbitrary parameters and shapes to the convolution kernel， providing rich choices between the convolution cores. Secondly， the ADown downsampling module is embedded in the backbone network， reducing the loss of feature information during the downsampling process. Finally， a CAFMAttention convolution attention fusion module is introduced to enhance the global and local feature extraction of hook recognition. Based on the experimental results， the improved YOLOv8n algorithm increases the precision， recall and mAP50 indicators by 4.6 %、4.2 % and 3.3 % respectively. The improved algorithm enables real time detection of hook positions， assisting operators in timely adjustment and decision-making， avoiding collisions or accidents， thereby improving safety in industrial environments. In addition， automatic hook recognition facilitates faster hook location identification while enabling precise cargo loading and unloading operations， consequently boosting work efficiency.

High-precision localization of unmanned vehicles in complex environments still faces challenges such as insufficient hardware-software coordination and the lack of standardized integration platforms， which restrict the construction of end-to-end data links and cyber-physical closed-loop interactions. To address these issues， this paper proposes a digital-twin-based localization platform architecture for unmanned vehicles. On the hardware side， a multi-source sensing and communication system is developed on the STM32 platform by integrating Global Navigation Satellite System （GNSS）， Ultra-Wideband （UWB）， Inertial Measurement Unit （IMU）， Bluetooth/Wi-Fi， and Narrowband Internet of Things （NB-IoT） modules. These modules are interconnected via high-speed buses and dedicated interfaces， forming the foundation for heterogeneous data acquisition and transmission. On the software side， the platform leverages the industrial Internet identification and resolution system together with digital twin technology to enable bidirectional real-time interaction and closed-loop control between the physical unmanned vehicle and its virtual counterpart. Core functionalities include data management， spatiotemporal localization， path planning， and geofencing. Experimental results demonstrate that the proposed platform ensures ranging accuracy and stability while enhancing system interactivity and integration， thereby providing a comprehensive hardware-software solution for high-precision unmanned vehicle localization.

Tool wear prediction is of great significance for reducing costs， improving efficiency and ensuring machining quality. To address the challenges such as difficulties in extracting features related to tool wear information， low utilization rate of the extracted features， and low prediction precision and accuracy， under the circumstances of complex environmental noise and a low signal-to-noise ratio， a Multi-scale Sample Reconstruction （MSR） method for vibration signals was first proposed to mitigate the impact of noise on the prediction effect of subsequent models. Subsequently， an improved model was put forward， which was based on the integrated model of the Residual Network （ResNet） and the Bidirectional Long Short-Term Memory （BiLSTM） network. In this improved model， the Criss Cross Attention （CCA） mechanism was integrated into each residual layer， and a Stacked Bidirectional Long Short-Term Memory Network （SBILSTM） was adopted. By comparing this improved model with the ResNet-BiLSTM model as well as traditional deep learning models， the results demonstrated that this method significantly enhanced the prediction precision and accuracy of tool wear.

To improve the object grasping success rate and pose prediction accuracy of robots through visual information in unstructured scenarios， a grasp detection model combining convolution and self-attention， i.e.， the Unified transformer grasp network （UFGNet）， is proposed. UFGNet uses an encoder-decoder structure. In the encoding phase， a layered Transformer module is adopted. The multi-head relationship aggregator in this module integrates the advantages of convolution and self-attention， and effectively extracts rich multi-scale features and global information at the shallow layer and deep layer respectively. The residual structure in the decoding stage enables the model to learn more complex feature representations， and pays more attention to the grasping area through the Shuffle Attention module， thereby improves the grasping detection performance. To verify the performance of UFGNet， the Cornell data set and Jacquard data set were used to train and test UFGNet， and the PyBullet simulation platform was built for capturing experiments. The results show that UFGNet has achieved 98.4% and 94.9% accuracy on the Cornell dataset and Jacquard dataset respectively， demonstrating competitive performance. In the simulation experiment， the average grasping success rate is 95.8%， and the experimental results verify the accuracy and robustness of UFGNet in grasping objects.

In order to solve the problem that multiple surface processing quality evaluation indexes need to be monitored online at the same time in the actual manufacturing process，a surface roughness recognition and dimensional accuracy prediction method based on RPM and MTL-GAMDenseNet-CA network is proposed. Firstly，the two-dimensional RPM diagram is obtained by two-dimensional conversion of the multi-channel vibration signal by RPM，and then the multi-channel RPM diagram is decomposed and reconstructed by using the two-dimensional discrete wavelet transform. Secondly，the GAM attention mechanism is introduced into the DenseNet model to form a new GAMDenseNet network as the encoder of the model，and then the CA attention mechanism is integrated into the decoder structure，and the MTL-GAMDenseNet-CA multi-task learning network model is constructed by using the hard-parameter sharing network architecture. The gradient normalization algorithm （GradNorm） adaptively adjusts the weight ratio of the two task loss functions of surface roughness recognition and dimensional accuracy prediction，and optimizes the model. Finally，the proposed method is verified by comparative experiments and ablation experiments. The experimental results show that the accuracy of the proposed method reaches 99.88% in the surface roughness recognition task. The mean absolute error （MAE） and root mean square error （RMSE） of the dimensional accuracy prediction task reaches 0.0177 and 0.0221， respectively. This shows that the proposed method can effectively realize the online monitoring of multiple evaluation indexes of surface processing quality.

In order to improve the cyber security communication capability of electric energy meter verification and calibration networks， this study proposes a cyber security communication technology based on an improved SDN controller and a combined encryption algorithm. A network-intrusion-detection model constructed upon CNN （convolutional neural networks） and LSTM （long short-term memory networks） identifies network intrusion flows， while a latency-aware traffic-scheduling algorithm blocks malicious network attacks， and realizes load balancing during electric energy meter verification and calibration； Combining Logistic chaotic mapping with RC6 algorithm to encrypt the communication information of the electric energy meter verification and calibration network， ultimately achieving the security of the electric energy meter verification and calibration network communication. Experimental results have shown that the average information entropy value of the electric energy meter verification and calibration network communication under this method is 7.992， which is 0.057 higher than the improved RC4 algorithm and 0.074 higher than the chaotic S-box encryption algorithm. The electric energy meter verification and calibration network communication security under this method is higher.

Aiming at the problems of low sample utilization rate， slow training convergence speed， and poor path planning performance in deep reinforcement learning， an improved path planning algorithm based on Dueling DQN is proposed， and directional rewards and filtering strategies are introduced. According to the angle between the line connecting the current state and the next moment state and the line connecting the current state and the target point， the reward function is redesigned to alleviate the problem of sparse rewards. During training， the actions resulting in collisions with obstacles are put into the blacklist， so that the action can be filtered in the next round of action selection， and the exploration speed of the algorithm is improved. The experimental results show that the improved algorithm can effectively improve the efficiency of path planning， and the exploration efficiency of the agent in complex environment is increased by about 95%， so that the agent can reach the target point with fewer exploration steps and less time.

Aiming at the problems of insufficient coordinated control and poor adaptability due to the single-arm design of the traditional pork cutting robot， a dual-arm segmentation supple control method based on six-dimensional force sensors is proposed， in which the right arm decomposes the cutting task into the orthogonal subspace of force and position control by constructing a master-slave force-position hybrid control framework， adjusting the cutting force and motion trajectory in real time to ensure the smooth cutting， while the left arm tracks the change of the master arm position and attitude in real time based on the Jacobi matrix mapping and acceleration constraint transfer， to ensure the balance and stability of the two arms when they move together. The simulation results show that the control method has good robustness and flexibility， and is able to accomplish the high-precision pork cutting task.

To address the problem of the crank-slider mechanism to overturn due to excessive reciprocating frequency of the sliding frame in vibrating membrane bioreactors， the instantaneous center of velocity （ICV） line model was derived through analytical solution research on the crank-slider mechanism. Calculations based on this model reveal that when the crank-connecting rod length ratios are 1⁄5， 1⁄4， and 1⁄3， respectively， the instantaneous center line exhibits the following variation pattern： with the crank length fixed， a longer connecting rod causes the overall ICV line to shift to the right. Except at the midpoint， the connecting rod of ICV line approaches a nearly horizontal straight line. Building upon the ICV line model， the variation patterns of the sliding frame's velocity and acceleration were further derived for the overturning limit frequency of 1.9 Hz and the safe frequency of 1.8 Hz within a 10 minutes interval. Using ADAMS Software to verify the effectiveness of the ICV line model for the crank-slider mechanism， providing crucial design theory references for predicting the stable operation of vibration membrane module mechanisms. Extending the ICV line model， the ICV line of the four-bar linkage mechanism was derived and applied to the design of the human knee joint. The ICV lines at two extreme positions （leg upright and bent） were obtained， which are similar to the J-shaped ICV trajectory of the human knee joint. The proposed ICV line model exhibits excellent alignment with the ideal instantaneous center trajectory of human joints， providing a reference for optimizing designs with human-machine compatibility. The ICV line mode established in this paper holds promising application prospects in the field of biomechanics.

In order to solve the problems of low powder utilization rate caused by poor powder feeding convergence and clogging of powder outlet caused by high-intensity laser cladding in current laser coaxial powder feeding nozzle， the finite element analysis method and orthogonal experimental design method based on the theory of gas-solid two-phase flow movement are adopted in this project. Under the conditions of the determined nozzle structure and environmental parameters， the orthogonal experiments are designed according to the parameters such as the powder concentration in the central axis， the distance between the convergence points， the powder flow concentration distribution per unit distance， and the powder spot diameter， which measure the powder flow convergence ability. The influence laws of the angle between the upper wall of the channel and the axis， the angle between the channel walls， and the outlet width of the particle path， on the gas powder flow distribution are obtained.

A dynamic resource scheduling method based on enhanced double Q-learning （EDQL） is proposed to address the issues of insufficient efficiency in dynamic resource scheduling and difficulty in ensuring differentiated service quality when multiple services coexist in the cloud radio access network （C-RAN） of the fifth generation mobile communication （5G） system. Firstly， establish a virtualized slicing architecture for ultra reliable low latency communication （uRLLC）， enhanced mobile broadband （eMBB）， and massive machine type communication （mMTC）， and achieve adaptive correction of resource weights through a multi priority preemption mechanism and dynamic queue adjustment factors. On this basis， design an EDQL algorithm that integrates competitive network architecture and dynamic reward scaling， and combine Markov decision processes to jointly model the network load， channel state， and queue delay. The experimental results show that the proposed method reduces the forced termination probability of uRLLC services by 82.3%， improves the eMBB service completion rate by 41.2%， increases system resource utilization by 28.5%， and reduces the average queuing delay of mMTC by 76.9%， compared to traditional polling scheduling， static priority， and heuristic rule algorithms. This study provides a new paradigm for 5G multi service resource scheduling through the deep integration of virtualized slicing and reinforcement learning.

The machining quality of electric spindles critically determines precision， efficiency， and stability in precision manufacturing. However， the machining process faces challenges due to diverse product types， multiple operating conditions and scarce target-condition data， making consistent quality of electric spindle difficult to guarantee. To address this， this paper proposes a transfer-learning-based method for multi-operating-condition quality prediction. The method first extracts spindle time-series signals and employs the Synthetic Minority Over-sampling Technique to balance historical and target-condition data distributions. Subsequently， constructs a two-stage regression model， TrAdaboost.R2， and leverages knowledge transfer to predict spindle quality under target conditions. Finally， the proposed method is validated with electric spindle data， demonstrating its superior prediction performance. This approach provides an effective framework for the precise quality prediction of electric spindles across varying operating conditions.

This study aims to propose an improved motion model for tobacco particles in the drum dryer. It also explores a method to estimate the residence time and other indicators when the inherent parameter values and operating conditions are known. By considering multiple motion forms of tobacco in the dryer and estimating the real-time positions of tobacco particles， we can accurately estimate the residence time in the drum. This estimation is significant for controlling the moisture and temperature of tobacco. It provides valuable information and improvement directions for enhancing tobacco processing quality. Numerical simulation results indicate that， under the condition where inherent parameters cannot be adjusted， the residence time and contact heating time are more sensitive to the rotation speed of the dryer compared to the hot air velocity. Therefore， when optimizing the drying process， controlling the rotation speed of the dryer as an effective operating variable can achieve precise control of tobacco moisture content.

Path planning is an important part of the process when an unmanned aerial vehicle （UAV） executes tasks， and it is also a crucial guarantee for the successful completion of tasks by the UAV. In this paper， the three-dimensional environment is modeled to construct a cost function for UAV path planning. Based on the analysis of the Elk Herd Optimizer （EHO） algorithm， an improved algorithm is proposed to solve the UAV path planning problem. The improved algorithm （Improved Elk Herd Optimizer， IEHO） initializes the population distribution through the Logistic chaotic mapping to enhance the diversity of the population. An S-shaped transfer function is utilized to adaptively and dynamically adjust the ratio of male elks， so as to balance the abilities of exploration and exploitation. Additionally， Gaussian mutation is introduced to perturb the position update of young elks， aiming to improve the search efficiency. Furthermore， the Lévy flight strategy is integrated to prevent the population from falling into a local optimum. Meanwhile， the numerical solution of the obtained path planning is smoothed by using the Bézier curve. The IEHO algorithm is compared with the EHO algorithm and five other algorithms through the cec2017 test functions， which verifies the effectiveness of the proposed algorithm. Finally， tests were conducted under two different environmental models. The results demonstrate that IEHO outperforms EHO， DBO， and COA in UAV path planning， exhibiting superior convergence speed and path optimization capabilities.

Due to the multi-degree-of-freedom， strong coupling， and nonlinear characteristics of wheeled humanoid robots， their motion control， especially whole-body motion control， is highly challenging. To address this issue， a whole-body motion control scheme based on robot kinematics and Model Predictive Control （MPC） is proposed. A simplified kinematic model of the wheeled humanoid robot is established， and a quadratic cost function related to the end-effector's pose is defined. The Relaxed Barrier Functions （RBF） are employed to set the constraints for robots to follow during operation， such as joint position limits， control input constraints， collision avoidance constraints， and tip-over prevention constraints. A simulation experiment for end-effector pose planning is designed， and the results demonstrate that the proposed control scheme not only enables the robot to achieve whole-body motion for end-effector control but also satisfies various constraints.

The realization of industrialized intelligent construction for steel structures is contingent upon two prerequisites： first， the development of a fully assembled steel structure system that is inherently efficient for repeated disassembly； second， the development of automated assembly robots to address the issues of low efficiency， low precision， and poor quality associated with on-site manual installation. This paper proposes a solution involving an automated assembly robot for the installation of torsion-shear high-strength bolts in Core-tube type steel column joint. Focusing on the assembly process of M16 torsion-shear high-strength bolts， this study emphasizes the mechanism design and structural reliability analysis of the robot end-effector. A hierarchical control system for bolt-hole assembly， based on machine vision， is designed. Simulation experiments demonstrate that the proposed robotic mechanism satisfies the assembly process requirements and significantly enhances the efficiency， precision， quality， and safety of installing torsion-shear high-strength bolts in core-tube column connections.

Difficult-to-cut materials are widely used in the aerospace and aviation industries. These materials have characteristics such as high cutting difficulty， high material cost， and difficult calibration experiments. In the finite element simulation modeling process of difficult-to-cut materials， the setting of material mechanical properties and tool chip friction performance， will significantly affect the prediction accuracy of the simulation model. How to achieve efficient acquisition of mechanical property parameters of difficult-to-cut materials， is of great significance to study the rapid and accurate uncertainty calibration strategy of simulation models. Taking the milling process of Ti₂AlNb intermetallic compound as an example， a finite element simulation model uncertainty calibration method for difficult-to-cut materials under the Bayesian framework is proposed. Firstly， an uncertainty analysis of the model is conducted， and a Bayesian based model uncertainties quantification method is proposed. The uncertainty coefficients are solved using the Markov chain Monte Carlo method. Secondly， finite element modeling， simulation experiment design， and simulation dataset construction for milling process are carried out based on finite element simulation software. A surrogate modeling method based on Gaussian process regression and supporting vector regression is proposed. Finally， Ti₂AlNb milling experimental design is carried out， and the working condition dataset is constructed to quantify the JC constitutive parameters and tool chip friction coefficient within the finite element model. The experimental results show that the uncertainty-calibrated finite element simulation model has significantly improved the prediction accuracy， and the relative error in predicting the cutting force of three-dimensional has decreased from 21.47% before calibration to 12.17%.

An improved starfish search algorithm was proposed to optimize workshop facility layouts， so to address the existing problem of high handling costs and time. Latin hypercube sampling was used to diversify the initial population. Levy flight and Brownian motion strategies are introduced to enhance the diversity of solutions and local exploration capabilities， achieving a balance between global and local search during the iteration process. Algorithmic complexity analysis indicates no increase in computational burden. Performance comparisons against 5 algorithms （SFOA， GA， PSO， GWO， and WOA） with 12 benchmark functions from CEC2017 demonstrates that the improved starfish search algorithm yields superior mean and standard deviation results over the other 5 algorithms in most cases. It reduces transportation costs by 44.47%， transportation time by 46.33%， and the objective function value by 44.81%. This method significantly enhances layout optimization efficiency and solution accuracy， making it practical and feasible for real-world applications.