Traditional point cloud registration-based methods enable rapid dimensional inspection of complex geometries but are sensitive to the initial poses and noise. This paper proposes a novel point cloud gravitational field theory (GFT) feature description and registration method to enhance inspection accuracy. A local coordinate system is constructed using principal eigenvectors to achieve coarse alignment. The point mass is defined based on the surface curvature. A multi-dimensional rotation-invariant GFT feature descriptor is designed, including gravitational force, field strength, potential energy and mass density. The correspondence relationship is established according to the GFT feature similarities and a transformation matrix is computed by singular value decomposition for fine registration. The experimental results demonstrate that the proposed method significantly reduces registration errors on complex freeform surfaces such as wind turbine blades, car exteriors and turbine blades, achieving error reductions of 0.4%-27.9%.

Aircraft assembly is characterized by stringent precedence constraints, limited resource availability, spatial restrictions, and a high degree of manual intervention. These factors lead to considerable variability in operator workloads and significantly increase the complexity of scheduling. To address this challenge, this study investigates the Aircraft Pulsating Assembly Line Scheduling Problem (APALSP) under skilled operator allocation, with the objective of minimizing assembly completion time. A mathematical model considering skilled operator allocation is developed, and a Q-Learning improved Particle Swarm Optimization algorithm (QLPSO) is proposed. In the algorithm design, a reverse scheduling strategy is adopted to effectively manage large-scale precedence constraints. Moreover, a reverse sequence encoding method is introduced to generate operation sequences, while a time decoding mechanism is employed to determine completion times. The problem is further reformulated as a Markov Decision Process (MDP) with explicitly defined state and action spaces. Within QLPSO, the Q-learning mechanism adaptively adjusts inertia weights and learning factors, thereby achieving a balance between exploration capability and convergence performance. To validate the effectiveness of the proposed approach, extensive computational experiments are conducted on benchmark instances of different scales, including small, medium, large, and ultra-large cases. The results demonstrate that QLPSO consistently delivers stable and high-quality solutions across all scenarios. In ultra-large-scale instances, it improves the best solution by 25.2% compared with the Genetic Algorithm (GA) and enhances the average solution by 16.9% over the Q-learning algorithm, showing clear advantages over the comparative methods. These findings not only confirm the effectiveness of the proposed algorithm but also provide valuable theoretical references and practical guidance for the intelligent scheduling optimization of aircraft pulsating assembly lines.

To address the issues of strong empirical dependence and low computational efficiency in traditional ball mill cylinder design, this study proposes a lightweight design methodology integrating multiple response surface models with finite element parametric simulation technology. Based on the equivalent density method, stress-strain characteristics are obtained through simplified cylinder structural modeling and finite element static analysis. Parametric finite element simulations are employed to generate sample data. The optimal response surface model is determined by comparing the goodness-of-fit (using evaluation metrics such as the coefficient of determination R2, root mean square error RMSE, etc.) among neural network, Kriging, and genetic aggregation methods. Key structural parameters of the cylinder are identified through sensitivity analysis using data generated from the optimal response surface, enabling the construction of a lightweight mathematical model that is solved using a multi-objective genetic algorithm. Experimental validation on ball mill demonstrates that the proposed method achieves 8.04% reduction in cylinder mass and 12.28% decrease in maximum deformation while maintaining equivalent stress within permissible safety limits.

A large vertical mill is a mineral grinding system. There are various measuring and monitoring operating points and multi-physics coupling, which means that optimizing its operating parameters is crucial for improving its system performance. This paper presents a combined agent model to optimize the system's multi-operation parameters. First, a numerical model of the particle fluid system is created using CFD-DPM, and the performance model of a vertical mill is developed based on factors such as production, specific surface area, and energy consumption. Next, a surrogate model is constructed by combining Kriging, polynomial response surface, and radial basis function. Based on this model, a multi-objective optimization framework is established. Finally, the LGM vertical mill is used to demonstrate the optimization. A comparison analysis reveals that the final product's output has increased by 0.78 %, and its specific surface area has increased by 10.38 %. Furthermore, the energy consumption has decreased by 13.51 %.

The cycle time of pulsating aircraft assembly accounts for nearly half of the entire aircraft manufacturing period. Its scheduling process faces challenges such as takt time synchronization of the pulsating line, multi-station collaboration, limited resources, and the complex coupling between worker skill levels and operational efficiency. However, existing studies rarely incorporate explicit quantitative modeling of worker skill-level heterogeneity into pulsating assembly scheduling, and most traditional approaches have limited capability in capturing complex structural job dependencies under large-scale, multi-constraint environments. To address these challenges, this paper proposes a pulsating aircraft assembly scheduling method based on Graph Isomorphism Network (GIN) and Proximal Policy Optimization (PPO) algorithms. First, an assembly scheduling model incorporating skill-level allocation constraints is constructed. The jobs and their dependencies are abstracted into a graph structure, and GIN is employed to extract deep features of job nodes, enabling accurate capture of complex job dependencies and resource demands. Second, within a reinforcement learning framework, the PPO algorithm is used for end-to-end training. By optimizing the scheduling policy, the method improves resource allocation efficiency and shortens the total completion time, while effectively handling complex resource limitations and skill-level constraints. Finally, multiple simulation experiments are conducted to validate the effectiveness of the proposed method, with comparisons made against traditional scheduling methods and modern optimization algorithms. Experimental results show that the proposed GIN-PPO approach significantly outperforms the comparison methods in terms of computational efficiency, solution quality, and convergence speed. It effectively reduces the total completion time of pulsating aircraft assembly jobs considering skill-level allocation, and demonstrates strong robustness and stability across test cases of different scales.

The estimation of failure probability in complex engineering structures suffers from low computational efficiency and challenges in handling with high-dimensional nonlinearity. Traditional methods are inefficient and require large sample sizes, making them difficult to meet the demands of practical engineering applications. This paper proposes a failure probability estimation framework that integrates cross-entropy (CE) optimization with a back propagation neural network (BPNN). The CE method is employed to adaptively optimize the sampling distribution, thereby enhancing the sampling efficiency for rare failure domains. By leveraging BPNN's powerful nonlinear approximation capability, a surrogate model for the limit state function (LSF) is constructed, significantly reducing the number of LSF evaluations. Validation through numerical engineering examples demonstrates that, compared to cross-entropy-based importance sampling (CE-IS) and cross-entropy-based Gaussian mixture sampling (CE-GM) methods, the proposed CE-BPNN approach stably approximates the reference failure probability across varying sample sizes, with lower coefficient of variation (CoV) and mean absolute percentage error (MAPE). The predictive model achieves R-square (R2) values consistently exceeding 0.97. Under identical sample sizes, CE-BPNN exhibits significant accuracy improvement and high stability. The results indicate that the CE-BPNN method offers superior accuracy and efficiency for structural reliability analysis involving high-dimensional nonlinearity and small failure probabilities, providing a promising new approach for reliability assessment of complex engineering structures.

Addressing the limitations of traditional clustering algorithms in unsupervised cross-domain image retrieval, namely, their inability to handle complex data diversity and their tendency to overlook constraints between features and objects-this article proposes a co-clustering method based on multiple correlation measures (CO-MCMs). By integrating multiple correlation measurement mechanisms, this method comprehensively characterizes the multifaceted relationships among objects, features, and object-feature pairs within the data. It constructs a hybrid graph model to more thoroughly uncover the data's intrinsic structure. Experimental results on multitype datasets demonstrate that CO-MCM outperforms existing methods in clustering accuracy, normalized mutual information, and robustness. Furthermore, applying CO-MCM to cross-domain image retrieval enables effective joint learning across domains and invariant feature extraction, significantly enhancing retrieval accuracy. This method demonstrates strong potential for handling complex, heterogeneous data, particularly for intelligent analysis of multisource image data and cross-device content retrieval in Internet of Things (IoT) environments. It provides an effective unsupervised learning solution for edge intelligence and distributed visual perception.

To address the challenge of accurately detecting the S-bend shape of the scraper conveyor in a fully mechanized mining face caused by pullback errors, this paper proposes a virtual-real mapping method for the S-bend based on deep learning and machine vision. First, the overall framework is established. Then, the collaborative sliding process of the hydraulic support and scraper conveyor is analyzed, and a mechanism model for the pullback error is constructed. Next, an improved BiLSTM-LSTM-ReLU model is designed to repair abnormal data from the hydraulic support displacement sensor, yielding ideal data. Subsequently, using the YOLOv8n algorithm and an image coordinate transformation model, the actual position and attitude of the scraper conveyor are detected. The pullback error is calculated, and the ideal data is compensated and corrected. Finally, an experimental platform is constructed, and a digital twin system is developed to complete the closed-loop verification process, encompassing data acquisition, anomaly repair, target recognition, and error feedback. The experimental results demonstrate that, under complex coal seam conditions, repairing abnormal sensor data using the deep learning model improves data smoothness by 72.3 % and trend consistency by 88.7 %. Furthermore, with the integration of visual feedback, the system reduces the final pullback error from the traditional ±3 mm to ±1.44 mm-a reduction of approximately 52 %.

Heavy icebreakers are specifically designed to operate continuously in polar regions. During navigation, the hull structure withstands repeated loading and unloading effects caused by continuous collisions with sea ice. To better understand the performance of the EH500 high-strength steel, which is widely used in heavy icebreakers, experimental tests were performed under cyclic loads with varying amplitudes to simulate demanding conditions. The outcome reveals that EH500 steel exhibits a remarkable cyclic softening feature, with a stress difference of up to 300 MPa between its maximum and minimum yield strengths throughout the entire cyclic process. However, the differences between tensile and compressive yield strengths remain minimal across cycles, indicating that the Bauschinger effect is negligible. Additionally, as the amplitude of cyclic loading increases, the fatigue life of EH500 steel decreases following a power-law relationship. This research provides a foundation for improving the structural safety design of heavy icebreakers operating in extreme ice environments.

Traditional robot control methods often encounter limitations such as lengthy development cycles and insufficient flexibility when addressing dynamic production environments and complex task requirements. To overcome these challenges, this paper constructs an integrated robot embodied control (EC) system that organically combines digital twins (DT), machine vision, and deep reinforcement learning (DRL). The method follows a closed-loop perception-decision-action framework. First, machine vision senses the environment in real time and precisely maps the 3D pose of the target object to the DT space. Second, DRL is conducted in the DT environment for training and strategy optimization. Finally, continuous state synchronization between the physical robot and the DT enables cross-environment policy transfer and online optimization. Taking the robotic arm pressing an emergency stop button as a representative task scenario, experimental results show that the system achieves a task success rate of 88% in the DT environment and 73% in the real physical environment, which was further improved to 76% through fine-tuning. In an extended lamp switch task, the success rate reached 79%, further verifying the generality and cross-environment adaptability of the framework. Overall, this integrated system significantly enhances the intelligence and operational efficiency of robotic systems, demonstrating its potential for achieving programming-free autonomous control in complex industrial environments.

The quantification of model parameter uncertainty is of great importance for the safety of engineering structures. However, traditional particle filter methods face challenges in avoiding particle degradation and impoverishment when estimating and predicting model parameters. Thus, this paper uses three improved particle filters combined with a Gaussian mixture model to evaluate the uncertainty of the model parameters. Fatigue crack growth model is established based on the Paris equation, and the performance of the filter is validated through two numerical examples with engineering backgrounds. Numerical example 1 analyzes a pressure vessel with a central crack, comparing the performance of different filters in estimating crack length, stress intensity factor, and posterior parameter distribution. By comparing the performance of PFGM, IBIS, and SMC under 107 loading cycles, it was found that SMC, compared to the other two methods, effectively reduces the estimation error of the posterior parameter distribution, with the error not exceeding 3%. Additionally, the time consumption of the resampling process was reduced by 12.7%. Numerical example 2 studied the central crack model of a Q235 steel plate, combining the K-L expansion method to quantify the material parameter random field. The results show that SMC outperforms the other filtering methods in terms of dynamic adaptability, prediction accuracy (RMSE reduced by 23.5%), and computational efficiency. In summary, SMC, through its adaptive resampling method, significantly reduces computation time and improves prediction accuracy in parameter estimation. It effectively quantifies the uncertainty of structural model parameters, enhancing the precision and stability of uncertainty assessments, thus providing a reliable tool for uncertainty quantification in large-scale structural health monitoring.

Automated assembly lines, while improving production efficiency, are also accompanied by energy consumption issues. To address the balancing and energy consumption optimization problem of non-robotic automated assembly lines with inherent rigidity, a mathematical model is established. The model aims to minimize the smoothness index, production energy consumption, and total production cost. It considers operation constraints, device parallelism, and device power. An improved multi-objective harmony search algorithm based on reinforcement learning is proposed to solve this problem. The algorithm uses a three-layer encoding scheme with a random key based on process constraints. To address the dependency of the harmony search algorithm on the initial harmony memory, an elite pool generated through fast non-dominated sorting is used to initialize the harmony memory. Dual metrics of population uniformity and diversity are introduced to evaluate solution quality. Based on these metrics, a variable neighborhood search algorithm combined with the Q-Learning algorithm in reinforcement learning is proposed, enabling the adaptive selection of search strategies and improving the efficiency of the population's local search. An improved elitist strategy combined with the Metropolis criterion is proposed, which probabilistically retains non-dominant individuals to enhance population diversity. The feasibility of the proposed algorithm is validated through examples of different scales. The effectiveness of the algorithm is further verified by an intelligent meter automated assembly line engineering case. Experimental results show that the proposed algorithm demonstrates excellent performance in the automated assembly line problem, providing multiple assembly line configuration options for business decision-makers.

Topology optimization can automatically generate the optimized topology under various boundary conditions. However, optimized results usually lack compatibility with computer-aided design (CAD) systems, necessitating time-consuming and laborious manual reconfiguration, thereby significantly reducing the efficiency of the designing process. This paper presents a novel automatic construction to generate an editable and closed CAD model based on the parametric surface for those tending toward beam-like structures. The process can be divided into three parts: pre-processing, solution, and post-processing. In the pre-processing stage, a complete and easy to implement multi-level hybrid topological data structure is proposed for automated reconstruction. In the solution stage, the rotation minimizing frames are calculated to generate the lofting surfaces with minimal twists. Meanwhile, the proposed principal plane projection addresses the robust reconstruction of the multiple freeform cross-sections. The reconstructed model is performed with Boolean union operations with non-designed domains in the post-processing stage. Five case studies are provided to validate the effectiveness of this new approach to obtain the editable CAD model. The derived CAD model is pivotal in subsequent applications, such as assembly constraints, simulation analysis, and parametric design.

As the market is gradually subdividing, the customer demands in the market of equipment operation and maintenance (O&M) services are increasingly diverse. Specifically, different customers may propose different service level requirements for the critical spare parts, which greatly increases the decision complexity of the optimal production scheduling faced by service manufacturers. To this end, this paper takes a parallel machine environment as an operating scenario and studies a critical spare parts production scheduling problem with multi-category mixed orders. Specifically, the problem considers three order categories: in-warranty orders, general orders outside the warranty period but having some tolerance for order delays, and temporary orders outside the warranty period but having no tolerance for order delays. First, a mathematical model of the studied problem is developed with the objectives of minimizing the total completion time and maximizing the total net revenue. Then, a memetic algorithm incorporating Q-learning (QMA) is proposed to obtain the near-optimal Pareto frontiers. In QMA, an initialization method and a local search operator based on the problem feature are designed, and a Q-learning-guided adaptive mutation operator is constructed to enhance the searching performance. Finally, numerous test cases with different scales are constructed for testing. Through comparison with other well-known algorithms, the results demonstrate the effectiveness of QMA for the proposed problem.

There are numerous quantities and types of electrical loads, and their electrical characteristics have similarities and differences. To adapt to the development trend of refined management and scheduling on the load side, it is necessary to explore the electricity consumption patterns of loads and classify them. However, the classification performance is affected by data redundancy, the complexity of feature selection, and the diversity of power consumption behavior. To adapt to the development trend of refined management and scheduling on the load side, it is imperative to classify loads based on their electrical characteristics. Firstly, based on a statistical analysis of load-side electricity consumption data, the monthly electricity consumption of each load throughout the year is extracted to reflect the continuous electricity consumption characteristics of each load. By calculating the annual load rate, maximum load utilization hours, and rated capacity of each load and then using a Gaussian Mixture Model (GMM) for clustering analysis, the discrete electricity consumption characteristics of each load are obtained. Then, based on the K-prototypes clustering model, a load classification method is proposed based on continuous and discrete hybrid electricity characteristics. By setting the weight between continuous and discrete electrical characteristics, the optimal number of categories can be determined through the elbow method. Finally, using 86 industrial electricity-consuming enterprises in a region of Northwest China as experimental subjects, the results demonstrate that the method proposed in this study outperforms the K-means, GMM, and Gower.