New-generation intelligent manufacturing is not simply the addition of artificial intelligence， the industrial Internet， or digital tools onto existing manufacturing systems. Rather， it is centered on the deep integration of artificial intelligence technologies with advanced manufacturing technologies， driving manufacturing systems from “connectivity and visibility” toward closed-loop optimization involving “perception， decision-making， execution， and feedback”. Compared with the earlier concept of “Internet Plus Manufacturing”， “AI Plus Manufacturing” no longer focuses merely on the interconnection of equipment， data， and business processes. Instead， it reconstructs product research and development， production and manufacturing， quality control， supply chain collaboration， and service models， through the coordination of industrial data， model algorithms， manufacturing equipment， process knowledge， and organizational capabilities. Based on existing policy contexts and research foundations， this paper reviews the connotations， key technologies， implementation scenarios， and practical constraints of new-generation intelligent manufacturing， focusing on directions such as smart factories， intelligent products， the industrial Internet， digital twins， trustworthy industrial AI applications， and manufacturing servitization. The study concludes that the key to new-generation intelligent manufacturing lies not in the advancement of individual technologies， but in the formation of sustainable system integration capabilities among manufacturing knowledge， data governance， equipment capabilities， industrial software， standards systems， and organizational processes.

To address the problems of blind expansion， redundant nodes， slow convergence， and tortuous paths that the traditional Rapidly-exploring Random Tree （RRT） algorithm exhibits in manipulator path planning within confined assembly spaces， this study proposes an improved Bidirectional RRT-Artificial Potential Field （BiRRT-APF） hybrid path planning algorithm to achieve efficient， safe， and smooth path generation for a six-degree-of-freedom manipulator in three-dimensional confined spaces. The algorithm establishes a D-H kinematic model for the robot and employs a cylinder-sphere composite bounding box combined with a two-stage screening mechanism to realize efficient collision detection， thereby reducing the detection overhead under large-scale sampling. During the bidirectional random tree expansion， the negative gradient of the artificial potential field is embedded as a local guidance force； a weighted fusion of the random direction and the potential-field direction balances global exploration with local obstacle avoidance， and the algorithm reverts to random exploration when the resultant force approaches zero so as to escape local minima， thereby improving both search efficiency and goal-orientation. An obstacle-distance-driven adaptive step-size strategy is further designed， which accelerates the approach in open regions and automatically contracts the step size near obstacles， thereby enhancing safety when traversing narrow passages. Finally， a post-processing mechanism integrating greedy pruning， equidistant densification， and collision-aware smoothing is incorporated to effectively eliminate redundant detours， improve path continuity and smoothness， and enhance the executability of inverse kinematic solutions. Simulation results show that， across six confined scenarios in both two and three dimensions， the algorithm attains a 100% success rate and obtains the shortest path in every case； the number of sampling nodes is reduced by up to 83.1% compared with the traditional RRT， and the path length in the complex three-dimensional scenario is optimized by 17.1% relative to BiRRT. The algorithm significantly outperforms traditional methods in path quality， convergence efficiency， and obstacle-avoidance reliability， demonstrating promising prospects for engineering applications.

Aiming at the problems associated with directly applying the Cuckoo-Catfish Optimization （CCO） algorithm to path planning， such as the mismatch between continuous variables and discrete path representations， a tendency to get trapped in local feasible channels in complex obstacle environments， and excessive path redundancy， an Improved Cuckoo-Catfish Optimization （ICCO） algorithm is proposed. ICCO couples the search process with grid-path representation through constructing a feasible domain mapping mechanism from continuous variables to discrete path nodes， adopts staged parameter scheduling strategy to balance global exploration and local exploitation， and combines Stochastic Tunnel Jumping （STJ）， Optimal Corridor Scanning （OCS）， and Elite Backward Learning strategy to enhance cross-channel exploration， in-channel path compression， and population diversity. Simulation results show that ICCO achieves good overall performance in average path length， computational efficiency， and solution stability in benchmark tests， ablation studies， and multi-scale grid-map experiments， especially demonstrating strong and stable solution capabilities in medium- and large-scale complex environments.

Grasping force fluctuation caused by contact impact and pneumatic coupling often reduces the stability of suction-based robotic grasping. To improve the dynamic response of the system， a fuzzy incremental PID force controller optimized by a multi-strategy particle swarm optimization （MSPSO） algorithm is developed. This paper constructs a kinematics model of the manipulator and a force-pneumatic coupling dynamic model of the end-effector， designs a fuzzy incremental PID controller under position-based admittance control framework， uses MSPSO to globally optimize scaling factors and initial PID parameters offline， and retains only the online tuning mechanism of fuzzy incremental PID in real-time grasping to cope with diverse load variations. To make the simulation conditions closer to practical operation， sensor noise， actuator delay， and contact disturbance are considered in the Simulink model. Simulation results show that the proposed controller can effectively suppress grasping force overshoot and improve response speed under different load conditions. Compared with the conventional incremental PID method， the overshoot is reduced from 18.6% to 5.1%， the settling time decreases from 0.72 s to 0.39 s， while the RMSE is reduced to 0.11 N. In repeated optimization experiments， the average ITAE index of MSPSO reaches 0.091， which is approximately 38.5% lower than that of the conventional PSO algorithm. The convergence iterations are also reduced from 35 to 20. The results indicate that the proposed method improves parameter optimization stability and enhances the robustness of grasping force control under load disturbances.

To address the drawbacks of low real-time responsiveness and weak adaptability to dynamic obstacles of conventional robot trajectory planners in dynamic scenarios， an adaptive robot trajectory optimization algorithm combining genetic algorithm and neural network is proposed. The algorithm adopts a two-layer architecture combining offline optimization and online planning. The genetic algorithm is utilized to optimize the connection weights and thresholds of the neural network， so as to obtain a network model with excellent global search capability. During robot movement， optimized trajectory points are output in real time according to the current local environmental information， and a rolling window strategy is employed to update the trajectory to achieve dynamic obstacle avoidance. Simulation and physical experimental results show that the proposed algorithm can generate smooth and collision-free trajectories. The evolutionary curves indicate that the genetic algorithm achieves fast convergence， and the robot can effectively avoid moving obstacles， which verifies the effectiveness and real-time performance of the proposed algorithm.

To address the challenge of insufficient degeneracy accuracy of the two-phase-orthogonal operational modal of annular resonant piezoelectric robots， an annular four-legged piezoelectric robot is taken as the research object， a parametric model is established with ANSYS finite element software， and a method combining modal tracking with sequential optimization is employed to systematically investigate the influence of structural parameters—including the inner diameter of the annular body， body thickness， driving foot length， driving foot radius， and piezoelectric plate thickness—on the modal degeneracy characteristics of the fourth-order operational modal. The results indicate that all structural parameters significantly affect modal degeneracy. Among them， the inner diameter， body thickness， and driving foot length are the dominant regulating parameters， whereas the driving foot radius and piezoelectric plate thickness mainly contribute to fine tuning. Based on parameter sensitivity analysis， a hierarchical sequential optimization strategy was developed， reducing the modal frequency difference between two orthogonal fourth-order operating modes from 1284 Hz to 8.29 Hz， corresponding to an optimization rate of 99.35%. The proposed method achieves high-precision modal degeneracy and provides theoretical guidance for the structural design and modal optimization of annular resonant piezoelectric robots.

To ensure safe operation of equipment， the steel structure on the ceiling of the valve hall must be thoroughly cleaned prior to electrical installation during construction phases of ultra-high voltage converter stations. However， manual cleaning poses significant safety risks and low efficiency. Given the intricate mesh structure of the valve hall's steel framework and the unique challenges of high-altitude cleaning environments， this study proposes a multi-section gripping cleaning robot specifically designed for steel structures. The robot can safely and reliably navigate the complex steel framework. To enhance operational efficiency， this paper investigates path planning strategies. We first develop a digital map model of the valve hall's steel structure and a mathematical model for cleaning duration， simulating the robot's movement patterns on steel beams. Using the Deep Q Network （DQN） algorithm， we sequentially identify optimal nodes while integrating mathematical models to optimize cleaning schedules. This approach achieves comprehensive coverage of the steel structure and yields an optimal cleaning strategy. Results demonstrate that this method significantly improves operational efficiency while ensuring complete cleaning coverage， providing reliable technical support for robots to perform high-altitude cleaning tasks in complex environments.

To address the problems of low sampling efficiency， weak pose adaptability， and susceptibility to local optima in three-dimensional assembly path planning for components in complex obstacle environments and confined spaces， an intelligent assembly path planning method integrating a Vision-Language Model （VLM） with an improved Artificial Potential Field （APF） method is proposed. First， multi-view virtual cameras are employed to capture images of the assembly scene， and VLM is used to identify obstacle distributions and narrow-space features and generate navigation guidance for assembly path planning. Then， the improved APF method adjusts the assembly pose of the component in narrow passages by calculating the repulsive torque between the component and surrounding obstacles. Collision feedback and traversed-path information are further incorporated to establish a closed-loop planning mechanism. Experimental results demonstrate that， compared with the conventional RRT method， the proposed method achieves superior performance in terms of average path length， average planning time， number of pose adjustments， and planning success rate in complex constrained spaces.

Automated storage and retrieval systems are an important component of modern logistics warehousing systems. They enable high-density storage and automated goods retrieval， and have been widely applied in manufacturing， e-commerce， and intelligent logistics. To reduce the multi-round picking operation time of a stacker crane in a single-aisle automated warehouse， this paper addresses the storage-location access sequence optimization problem under capacity constraints and establishes a picking routing optimization model with the objective of minimizing the total order-picking operation time. A hybrid ant colony optimization （HACO） algorithm is proposed to solve the model. The algorithm first constructs chained candidate picking sequences using pheromone concentration and heuristic distances between storage locations， then restores the access sequence into multi-round operation paths according to tote capacity， and further introduces a local improvement strategy based on the round structure. Experiments show that， for medium- and large-scale orders， HACO reduces the average operation time by approximately 26.22%. This indicates that the proposed method effectively reduces cross-region movements and repeated backtracking of the stacker crane， thereby improving the picking efficiency of the automated storage and retrieval system.

Online identification of the winding resistance and rotor flux linkage of a surface-mounted permanent magnet synchronous motor （SPMSM） is an effective approach for monitoring the stator and rotor temperatures. Accordingly， an online identification method for the stator winding resistance and rotor flux linkage of an SPMSM based on the Immune Clonal Algorithm （ICA） is proposed. First， based on the analysis of the steady-state rank-deficient model of the SPMSM， an error analysis method is proposed to construct the penalty function， thereby ensuring the convergence of the ICA during the parameter identification process. Second， the proposed method only requires two sets of stator current， voltage， and speed data—one from steady-state operation of the SPMSM （i_d = 0）， and the other from operation with a narrow， weak magnetic pulse injected （i_d ≠ 0）—to enable the host computer to perform online estimation of the SPMSM’s stator winding resistance and rotor magnetic flux linkage. Finally， the method is validated on an SPMSM， and the experimental results show that the proposed method can accurately track variations in both stator winding resistance and rotor flux linkage.

During the submerged arc welding of circumferential seams on wind turbine towers， the tower rotation causes the welding carriage and the welding torch tip to follow a tangential path relative to the fixed welding station coordinate system. The carriage must therefore travel actively in the opposite tangential direction relative to the tower surface to keep the welding torch tip stable at the target welding station. To clarify the rotation-compensation relationship and suppress speed mismatch caused by friction， load variation and actuator lag， an expanded state observer （ESO） enhanced fuzzy self-tuning PID speed synchronization control method is proposed. A residual endpoint velocity model is established. The carriage reference speed is generated from the measured tower angular velocity and the weld-seam radius， while the ESO estimates and compensates for the total disturbance on the travelling axis. Simulink simulation， Simulink-CoppeliaSim co-simulation and an equivalent scaled experiment are used to verify the engineering control framework. Compared with conventional PID， the proposed method reduces the speed-error RMSE by 46.5% under constant-speed disturbance， reduces the RMS static tip error by 90.2% in co-simulation， and decreases the RMSE， maximum absolute error， IAE， and disturbance recovery time by 32.1%， 13.4%， 88.2% and 12.2%， respectively， in the scaled experiment. The results verify the effectiveness and hardware feasibility of the proposed method.

Traditional methods for measuring the total elongation at maximum force of reinforcing bars mainly rely on contact extensometers or manual gauge-length measurement， which suffer from complicated operation， equipment damage， and susceptibility to human factors. A non-contact vision-based measurement method is proposed. By synchronously acquiring tensile test videos and load-time data， the image frame corresponding to the maximum force is identified to determine the total elongation at maximum force. First， camera calibration is performed based on the pinhole imaging model， and image distortion is corrected using radial and tangential distortion models. Then， a reference-model-based Dynamic Region of Interest （DROI） extraction method is introduced to achieve stable tracking of the gauge region under large deformation. Finally， the intersections between the marker lines and the centerline in the initial frame and the frame corresponding to the maximum force are detected through geometric pose localization of the reinforcing bar and marker line detection. The variation in gauge-length pixels is then calculated to determine the total elongation at maximum force. Experimental results show that， compared with measurements obtained using a contact extensometer， the proposed method achieves a mean absolute error of 0.218% and a mean relative error of 1.130%. The proposed method provides high measurement accuracy and offers an effective technical solution for the automatic measurement of the total elongation at maximum force of reinforcing bars.

To reduce the cost of manual annotation for flexible printed circuit （FPC） defect datasets and to alleviate the prompt dependence and background interference of general-purpose segmentation foundation models in industrial images， an FPC defect pre-annotation method based on SAM2 feature fusion and regional constraints is proposed. A single manually annotated sample is used as the reference. Multi-scale defect prototypes are extracted from the SAM2 image encoder， and similarity responses between the target image and the reference defect are calculated. A temporary point-prompt decoding step is used to obtain a probability map for similarity fusion， after which foreground and background prompts are generated automatically. Connected component filtering， morphological constraints， candidate box number control， and class-adaptive post-processing are then introduced to generate candidate boxes for manual review or downstream detector training. Experiments on 700 test images from seven FPC defect categories show that， at an IoU threshold of 0.5， the proposed class-adaptive strategy improves macro F1 from 0.0548 to 0.2664， micro F1 from 0.0631 to 0.2763， and average best IoU from 0.1738 to 0.2536， while reducing the average processing time from 0.7645 s to 0.6253 s per image. The method performs better on certain texture and structural defects， such as gold-surface scratches， stains， and stiffener offset， but remains insufficient for low-contrast or irregular defects. Therefore， it is positioned as a candidate generation method for FPC defect pre-annotation rather than a complete replacement for manual annotation.

In the case of composite faults of rolling bearings， multi-source impacts are coupled with each other， and the fault energy is dispersed across different frequency bands. Traditional fixed-frequency-band demodulation struggles to adaptively determine the optimal frequency band， which leads to the easy masking of weak fault features by strong fault components or noise， resulting in limited diagnostic accuracy. Therefore， an adaptive diagnostic method for composite faults is proposed. A multi-scale sequential statistical filter with increasing window length is used to detect and fuse the signal amplitude spectrum， obtaining adaptive frequency band partition； based on the boundaries， an empirical wavelet filter bank is constructed to obtain the modal components of different resonance frequency bands； a multi-point kurtosis index is introduced to select the optimal modal containing rich periodic impact information for square envelope spectrum analysis， and extract fault features. Experimental results show that the multi-scale sequential statistical filter can effectively improve the stability of boundary detection and reduce the number of false boundaries； the multi-point kurtosis based on the fault frequency can accurately select the optimal demodulation frequency band. The experimental results show that multi-scale sequential statistical filtering helps improve the stability of boundary detection and reduces false boundaries； the multi-point kurtosis oriented to fault frequency can be used to screen the demodulation frequency band. The proposed method extracted the characteristic frequencies and their harmonics of inner and outer ring faults from the two sets of composite fault data in this paper， demonstrating a good effect in extracting fault features.

To address the reduced diagnostic robustness caused by multi-condition response coupling and sensor-channel degradation in helicopter rotor blades， a reliability-mask-enhanced physical-topology residual graph-attention multi-layer perceptron is proposed. Based on a physical data chain from aerodynamic loads and sectional forces to strain acquisition， a 16-channel dataset comprising six health states is constructed， and public datasets are used to constrain load trends and fault spectral patterns. Channel validity and quality factors are incorporated into neighborhood attention， while physical-topology aggregation and a residual branch jointly provide cross-sensor compensation and preserve local fault features. Experimental results show that the proposed model achieves accuracies of 95.77% and 93.88% under complete-channel and mixed-degradation tests， respectively， and requires approximately 8.6 ms for on-board inference per window on the ZYNQ7010. The results demonstrate stable recognition performance under channel loss and noise/drift， satisfying the accuracy and real-time requirements of edge-based helicopter rotor blade health monitoring.

This paper optimizes the metering accuracy of a certain type general-aviation refueling truck during intelligent preset operations. The study established a water-hammer-flow coupling simulation model based on the Method of Characteristics （MOC） to address the overshoot prediction deviation. A quantitative two-stage valve-closing design method was developed using the target flow decay trajectory as the control variable. The study analyzed the impact of different strategies on overshoot stability and pipeline safety. Results showed that a 70% rapid flow closing in the first 2 seconds reduced the valve-closing overshoot by 19.5% and significantly weakened its sensitivity to fuel tank level fluctuations and valve aging. Pressure oscillations were effectively mitigated and metering stability was maintained when the first-stage rapid closing ratio was set within the range of 60%-70%. By optimizing control parameters， the proposed strategy achieved a relative metering error of less than 0.15% during 100 L small-batch refueling， without requiring any physical modification of existing hardware. The resulting performance satisfied the civil-aviation accuracy requirement of Class 0.2 metering.

The development of military aero-engines is characterized by high technology intensity， long lifecycles， and substantial investment. To address the limitation of existing research， which predominantly focuses on manufacturability analysis while lacking systematic consideration of producibility throughout the entire development process， this paper proposes an integrated methodology for producibility design and comprehensive evaluation tailored for military aero-engines. In the design phase， the methodology permeates the entire development lifecycle， balancing manufacturability and assemblability requirements during the early design stage to implement producibility measures. In the evaluation phase， a producibility assessment index system is established， encompassing key elements across the dimensions of design， production， schedule， and cost. The methodology integrates subjective weights derived from expert scoring with objective weights determined by the CRITIC method to obtain comprehensive weights， and introduces the normal cloud model to conduct an uncertainty-based comprehensive evaluation. This approach provides quantitative decision-making support for the iterative optimization of design schemes. Analytical results demonstrate that the proposed method effectively identifies producibility weaknesses in design， offering significant reference value for enhancing the development efficiency and engineering maturity of military aero-engines.

The study utilizes pulsed laser for paint cleaning on 16 Mn automotive skin substrates. An orthogonal experimental design was established with four parameters—laser power， scanning speed， repetition frequency， and spot overlap rate—as influencing factors， with surface cleaning quality as the response variable. The Taguchi method was applied to analyze the mean and S/N ratio of experimental results based on different paint thicknesses. Analysis of variance was used to determine the significance of each parameter's influence on the results. The order of influence of the four parameters on the surface cleaning quality was determined as follows： laser power > spot overlap rate > frequency > scanning speed. The optimized laser paint cleaning parameters were identified as while the painting coating thickness by 50 µm~90 µm： laser power 210 W， scanning speed 7000 mm/s， repetition frequency 50 kHz， and spot overlap rate 0.4. Experimental validation of the optimized parameters confirmed that they effectively and consistently complete the paint cleaning process without causing deformation or microstructural changes in the substrate material. Additionally， the adhesion test results of the repainted film after cleaning met the required standards.

A nonlinear mechanical characteristic analysis and structural selection method is developed for spherical disc springs used in the thrust bearing spring pack of automated equipment. The spherical disc spring is transformed into an equivalent conical disc spring under geometric constraints， and the equivalent coefficient is calibrated by finite element load-displacement results， forming a computational workflow for parametric simulation and automated sizing. The theoretical and finite element curves show consistent trends under different height-thickness ratios， with an average error below 10%. Parametric analysis indicates that spring thickness， height and outer radius strongly affect peak load and the low-stiffness interval. After geometry selection for the 70-74 kN target load， the stable working interval increases by more than 30%， which provides a basis for digital design and monitoring threshold setting of thrust bearing spring packs.