The micropolar elastic finite element method has found extensive applications in analyzing materials with complex microstructures. To address the problems of low computational efficiency and shear locking under bending conditions caused by the fully integrated scheme commonly adopted in existing implementations, a universal and efficient reduced-integration first-order hexahedral micropolar finite element was proposed and verified using SAM, a general analysis software for marine structures. The element algorithm, combined with the standard Lagrange interpolation and uniform-strain and curvature formulas, passed the element patch test and ensured the calculation accuracy for distorted elements. Additionally, an artificial stiffness method was introduced to effectively suppress displacement and rotational hourglass instability modes in the reduced-integration micropolar elements. Numerical validation, including force and displacement patch tests for convergence, cantilever-beam tests for shear locking, and modal analysis of vibration responses of sandwich structures used in shipbuilding, demonstrated the proposed element’s effectiveness and advantages over traditional solid elements in finite-element analysis of ship structures.

Efficient and accurate evaluation of shape functions is critical for element stiffness assembly and the global solution in finite element (FE) computations. A general framework was proposed for approximating shape functions was proposed to handle multiple element types and variable node counts. First, learnable type embeddings were used to map discrete element classes to continuous vectors, enabling parameter sharing across element types. Second, geometric information and query-point features were processed by a node encoder; graph convolutions were used to propagated local geometric constraints, while global pooling was applied provided element-level context. Finally, shape function values were produced by a decoder from node- and element-level features, and the model was trained with a loss combining mean square error (MSE) and a physics-inspired sum constraint to enforce shape-function properties. Experiments on synthetic datasets of linear 4-node tetrahedra and quadratic 10-node showed demonstrated that the model achieved a test MSE of approximately 0.001 8 and that shape function values were computed accurately across the test set. Moreover, in a large-scale computing environment, the neural-network inference attained roughly three times the throughput of a conventional interpolation-based implementation. These results suggested a promising route to accelerate or replace classical shape-function evaluations using learning methods.

In continuous production scenarios of process manufacturing, product homogenization directly determines the stability and satisfaction of user experience. Its core value is not only reflected in the consistency of product performance, but also of great significance for improving production efficiency and controlling costs. However, various process parameters in process manufacturing are mutually coupled with complex correlations, which leads to high prediction delay and low accuracy in existing models. To address the above problems, a multi-stage, multi-step prediction model for process-manufacturing quality indicators based on grey relational analysis (GRA) and a Transformer-MC-CF was proposed. Firstly, to mitigate multivariate parameter redundancy in process manufacturing, GRA was used to quantify correlation strengths between input parameters of each process and quality indicators, to identify key influencing parameters, and to eliminate redundant variables with low correlation. This approach reduced the model’s computational complexity and avoided interference from irrelevant information during feature learning. Secondly, a multi-stage, multi-step prediction model based on the Transformer was constructed. To address insufficient local feature fusion caused by mutual coupling of multiple factors, a multi-channel receptive field block (MC-RFB) was designed. By combining multi-scale convolution and dilated convolution, the block simultaneously captured local detailed features and long-term temporal dependencies. Considering the sequential coherence of process manufacturing, a correlation feature fusion (CF) module was proposed; it adopted an adaptive-weighting strategy to fuse latent features of upstream and downstream processes, thereby effectively exploring the indirect inter-process influence relationships. Experimental results showed that the application of GRA effectively improved the prediction performance of the model. Compared with traditional models, the proposed Transformer-MC-CF model outperformed others by more than 4.6% in process-manufacturing quality prediction scenarios, and the average fitting degree for multi-process, multi-step prediction reached 97.5%. This model provided technical support for improving prediction and regulation capabilities of process-manufacturing quality indicators and for achieving homogenized production.

Rough machining, the initial stage of subtractive manufacturing, rapidly removes the majority of workpiece stock (typically 70%-90% in industry practice) to approximate the final part geometry. Its core objective is high-efficiency material removal, which directly determines overall machining productivity. Conventional roughing toolpath generation algorithms predominantly employ contour-parallel strategies, which ensure satisfactory surface quality but often sacrifice efficiency. To address this given and leveraging the widespread adoption of automatic tool-changing systems in modern CNC (computer numerical control) machines, a novel two-stage roughing optimization algorithm based on collision detection was proposed. The method first utilized GPU-accelerated parallel computing for rapid collision detection to identify feasible machining zones. A large-diameter tool was then deployed to generate direction-parallel (DP) toolpaths in the first stage, enabling aggressive stock removal. Subsequently, the residual stock boundary was precisely updated, and a smaller-diameter tool was engaged to generate contour-parallel (CP) toolpaths for localized precision machining in the second stage. Experimental and simulation results demonstrated that, compared to traditional CP methods, this strategy achieved a 17% improvement in machining efficiency while maintaining surface quality. This work provided new perspectives on resolving the fundamental trade-off between machining quality and processing efficiency during rough machining.

In the context of large-scale customized production, flexible job-shop scheduling faces several critical challenges, including sluggish responses to dynamic demands, limited optimization performance formulti-task integration, and insufficient adaptability touncertain changes and unseen environments. To effectively enhance the optimization quality, generation speed, and adaptive capability of scheduling solutions, this study focused on dynamic flexible job-shop scheduling problems involving multi-resource coordination and multi-task integration in dynamic environments. A multi-agent deep reinforcement learning-based adaptive real-time scheduling method was proposed, along with the development of an adaptive real-time scheduling solver. First, an adaptive real-time scheduling framework based on multi-agent deep reinforcement learning was constructed; an interaction mechanism between the simulation environment and scheduling agents was established, and the distributed decision-making process was formalized as a partially observable Markov decision process. Second, an object-oriented scheduling simulation environment was developed by defining resource and simulation classes (e.g., machines, workpieces, and automated guided vehicles), to comprehensively represent the dynamic characteristics and interactions of typical resources in flexible job shops. Third, scheduling agents supporting machine assignment and job sequencing were designed; a library of deep reinforcement learning algorithms covering value-based, policy-based, and hybrid value-policy approaches was developed to accommodate various scheduling scenarios and complex constraints. Finally, an adaptive solver supporting multi-task and multi-constraint flexible job-shop scheduling was developed, incorporating strategy training and problem-solving modules to achieve adaptive optimization across diverse application settings. Numerical experiments demonstrated that the proposed method exhibited strong generalization and real-time responsiveness in both known and unseen scenarios, effectively handling complex and dynamically changing scheduling problems. In case studies, the developed solver showed significant advantages in response speed while maintaining high optimization performance. The results indicated that the proposed adaptive real-time scheduling method and solver provided an effective solution to multi-resource scheduling challenges and offered solid technical support for the intelligent and autonomous transformation of discrete manufacturing systems.

There is a practical need for reverse modeling technology of ship hulls based on large-scale point clouds in applications such as digital twins and hydrodynamic performance analysis. Currently, few specialized methods addressed large-scale point cloud data for ship hull modeling, and conventional approaches usually relied on neural network models with high training costs or foreign commercial software. In this case, designing a lightweight, one-stop method for hull reverse modeling is valuable: initially, the ship hull was meshed to compress the point cloud data, reducing the data volume by over 90% while preserving the hull’s macro geometric features, thereby significantly decreasing the computational cost for the following algorithms; secondly, contour line detection and fitting algorithms were designed based on the hull’s geometric features, utilizing the least squares method for contour correction to ensure its completeness and smoothness; finally, a smoothing algorithm based on slope anomaly detection was adopted, which efficiently performed automatic smoothing of the hull’s longitudinal and vertical form lines through three steps: slope anomaly detection, preliminary correction, and deep smoothing. The method was validated on two ship models. Compared to the reference models established using commercial CAD software, the principal dimension errors (+0.24% to +2.68%) and displacement volume deviations (-1.05% to -0.88%) were only minor and all within an acceptable range. The entire process was completed within 5 minutes on a conventional computer. The method demonstrated potential for application in related fields such as hull reverse modeling, ship digitalization, and hull form smoothing for 3D generative large models.

Conceptual design refers to the process from the proposal of original requirements to the generation of initial schemes during the preliminary design phase. In response to frequent changes in requirements for ship design at this stage, strong dependence on parent models, and the high cost of concurrent design of multiple schemes, a conditional generative adversarial model for hull forms was proposed for scenarios lacking approximate parent models. This model used resistance performance as the conditional label to generate hull-form geometric schemes that met the performance index requirements. Firstly, contour feature curves of the bare hull were constructed using parametric modeling methods; geometric parameters characterizing the hull form were selected, and corresponding three-dimensional meshes were obtained. Secondly, based on the integral theory, the characteristic surface areas and hull-form coefficients were obtained from the three-dimensional mesh, and a consistency dataset of geometric parameters and performance characteristics was constructed via the automated interface of the developed performance-prediction platform. Finally, based on a traditional generative adversarial network (GAN), a multilayer perceptron was employed to encode conditional features represented by resistance, which were then combined with geometric features in the hidden layers of the generator. This approach guided the learning of hull-form sample distributions under varying resistance conditions and generated hull-form geometric parameters that satisfied the specified constraints. The initial realization of a forward-design process for hull forms without reliance on parent-model information and based on overall performance of typical types provided a design basis for the rapid generation and iterative optimization of conceptual schemes under uncertain requirements.

To enable rapid storage of 3D CAD files in the cloud, a cloud CAD file storage method based on command streams was proposed. A command-stream data-storage model was constructed; client operation commands ware classified and represented as command streams, and incrementally transmitted and stored as key-value pairs. A command stream log model was also built, taking commands as the smallest transaction unit, representing user modeling operations as recordable, replayable, and verifiable operation-log sequences, to enable atomic writes of command streams, command replay, version tracking, and state recovery. Meanwhile, a cloud-CAD file storage architecture was established, including an interaction module, a command-stream processing module, and a data-storage module; a two-layer storage mode was adopted to store command streams, geometric and topological data, respectively. Through case testing, the method was demonstrated to enable rapid and accurate transfer 3D CAD files in the cloud, verifying its effectiveness.

The stability of cigarette ventilation rate is affected by multiple factors, such as laser drilling methods and equipment parameters. This makes it difficult for traditional methods to fully capture the complex interactions among these factors and fail to effectively adapt to the needs of modeling long-range dependencies, thus struggling to accurately control the precision of cigarette ventilation rate. To address this challenge, a laser drilling ventilation rate prediction model for the cigarette manufacturing process based on the adaptive multi-scale temporal attention network (AMTA-Net) was proposed. Firstly, to resolve the dimension mismatch between laser drilling process parameters and cigarette categories, a category feature embedding module for laser drilling was designed. By integrating one-hot encoding with a learnable embedding matrix, feature fusion of cigarette category information and drilling process data is achieved. Secondly, to capture the complex nonlinear relationship between hole morphology and cigarette ventilation rate, a multi-scale feature fusion module named ELAN-1D was proposed. This module adopts a two-layer convolutional architecture, realizing deep extraction of temporal features and effective capture of local-global information through convolution kernels with different dilation rates. Finally, to characterize the spatial heterogeneity and sequential correlation of hole morphology features, a sequential attention feature fusion module was constructed. Combining multi-scale sequential convolution with a cross-attention mechanism, this module implemented a “channel-sequence” dual-dimensional cross-logic modeling for the mapping relationship between drilling morphology and ventilation rate. Experimental results demonstrated that the mean squared error (MSE) of the proposed AMTA-Net for filter ventilation rate and total ventilation rate prediction reached as low as 6.799×10^-4 and 6.874×10^-4, respectively. Compared with baseline models, the proposed method reduced the MSE of filter ventilation rate and total ventilation rate by 28.96% and 20.61%, respectively, with the mean absolute error (MAE) decreased by 18.16% and 11.51%. In contrast to traditional models, the MSE and MAE of AMTA-Net were reduced by more than 15.64% and 8.05%, respectively. The method proposed provided robust model support for the precise regulation of laser drilling processes and for achieving cigarette tar reduction and harm mitigation.

Maintenance-interval prediction focuses on the proactive scheduling of equipment downtime, arranging maintenance before performance degradation reaches a predefined threshold while aligning with engineering operations. Accurate prediction of such intervals is vital for reliable equipment operation but remains challenging due to difficulties in multi-source data fusion, quantitative degradation characterization, and long-term dependency learning. This study presented a degradation-driven temporal modeling method that dynamically represented performance deterioration during continuous operation and adaptively captured complex dependencies among multi-sensor data. A performance-degradation indicator (PDI) quantified equipment performance decline using time-series measurements. To capture correlations among multi-source features, a sequence-to-sequence prediction model with multi-head attention was constructed and degradation-aware parameters were integrated, which optimized feature weighting and improved long-term trend prediction. The experimental results indicated that the optimal performance of the model improved by nearly 13.5% after integrating PDI. On the TBM (tunnel boring machine) engineering dataset, an RMSE (root mean square error) improvement of approximately 25% was achieved compared to the standard LSTM (long short-term memory), and outperformed other models by nearly 15%, yielding high prediction accuracy. Further evaluation on the C-MAPSS dataset against RNN (recurrent neural network), GNN (graph neural network), and attention-based methods confirmed the approach’s effectiveness, offering a detailed analysis of how varying the number of sensors affected model performance. The method also exhibited strong scalability and could be extended to incorporate environmental-condition awareness, providing technical support for intelligent maintenance decision-making and closed-loop operational control.

To achieve high-precision detection of low-resolution thermal image of insulators, an infrared multi-scale insulator detection method was proposed based on intra-class sparse prior and improved YOLOv8. To address the issue of missed detections due to local density in insulator images, a universal intra-class sparse prior was proposed, combining prior knowledge with training data. This enabled the model to perceive the unique geometric features and morphological information of the insulator, enhancing the accuracy of the target detection model without additional computational cost, and provided a standardized annotation method for insulator data samples. To tackle the difficulty of feature extraction from low-resolution infrared images, a robust feature downsampling module was employed to replace convolutional downsampling, preserving fine-grained detail information and enhancing the robust representation of key feature maps. For the problem of large-scale variations and occlusion in insulator targets, a wise-MPDIoU was utilized, improving the bounding-box loss function and the model’s ability to localize insulators of different sizes. Experimental data demonstrated that, compared to the baseline model, the proposed method achieved improvements of 3.3 and 3.5 percentage points in AP50 and AP50:95 metrics, respectively, providing a new solution for insulator thermal image detection.

Segmentation of skin lesions is an important task in medical image analysis, and is of great significance for the early diagnosis and treatment of skin diseases. However, when processing high-resolution skin images and capturing subtle lesion features, existing models still face challenges such as high computational complexity and insufficient processing of redundant information. To address this end, a lightweight skin lesion image segmentation network based on the Mamba structure was proposed, ResMamba adopted a six-level U-shaped structure, embedding Mamba into the visual state space and introducing it into the codec. The ResVSS module, as the core component of the encoder, reduced the number of parameters by removing a redundant linear layer, and at the same time combined the deep convolution block and learnable scale parameters to scale the residual connection, thereby reducing the complexity of the model while improving the segmentation accuracy. In the hopping connection module, a multi-level multi-scale information fusion module was used to generate spatial and channel attention maps, which effectively fused multi-scale information. Through experimental verification on the public skin dataset ISIC2017 and ISIC2018, the results demonstrated that the ResMamba model achieved good segmentation performance in terms of the number of balance parameters and segmentation performance, thus verifying the effectiveness of the model.

Semantic segmentation, a core task for autonomous driving perception, faces challenges under low-light and high-speed scenarios due to the limitations of conventional cameras. Event cameras, with their microsecond temporal resolution and high dynamic range, effectively mitigate motion blur and extreme lighting conditions. However, their asynchronous sparse event data lacks texture and color information, while uneven event distributions caused by relative motion between background and objects pose significant difficulties for semantic feature extraction. To address these issues, a multi-frequency hypergraph fusion method for event-based semantic segmentation was proposed. First, the approach decomposed event frames into multi-scale spatiotemporal features through a frequency separation module, distinguishing high-frequency motion edges from low-frequency structural information. A dynamic hypergraph construction algorithm then mapped these multi-frequency features into hypergraph nodes, utilizing hypergraph convolution to capture long-range dependencies across frequencies. Finally, an attention mechanism adaptively fused multi-frequency features to enhance inter-class discriminability. Experiments on Carla-Semantic and DDD17-Semantic datasets demonstrated that this method achieved 88.21% MPA and 82.68% mIoU, outperforming existing methods and validating the effectiveness of the multi-frequency hypergraph model for event-based semantic understanding. This research provided a novel solution for robust environment perception with event cameras, particularly suited to challenging autonomous driving scenarios involving low-light conditions and rapid motion.

With the popularity of coffee culture and growing consumer demand, the maturity of coffee fruits has become a key determinant of quality and market value. However, irrational harvesting leads to uneven quality and impacts economic benefits. Through advanced ripeness detection techniques, harvesting accuracy can be improved to provide data-based decision support for farmers; however, the existing methods still have technical challenges in terms of robustness in complex backgrounds and high-density small-target detection. Therefore, a coffee tree fruit ripeness prediction model based on image-chunking interaction was proposed, which achieved the complementary fusion of local and global feature information by introducing a spatial-blocking interaction attention mechanism (SBIAM), so that the model can focus on the fruit region as well as effectively inhibit the background interference, enhancing the model's ability to pay attention to key features. In addition, a normalized Wasserstein distance (NWD) loss function was introduced to solve the problems such as the prediction-position deviation common in coffee-fruit classification, thereby improving the accuracy and robustness of coffee-fruit ripeness detection in complex scenes. Experimental results demonstrated that the proposed improved model not only enhanced the detection accuracy, but also achieved a good balance between performance and efficiency.

The collection of UAV (unmanned aerial vehicle) video datasets is costly and faces issues such as limited quantity, low quality, and scenario constraints. To address these challenges, an unsupervised UAV-object-tracking model based on temporal cycle consistency and dynamic memory enhancement was proposed. First, salient-object detection was introduced for unlabeled object discovery. By combining salient object detection with unsupervised optical flow techniques and incorporating dynamic programming based on image entropy, the quality of pseudo-labels was improved. Second, a weight is defined for each frame in the video, and these weights are utilized for single-frame training to fully leverage the information from all frames. Finally, inspired by long short-term memory (LSTM) networks, the memory queue was transformed into a dynamic memory queue, along with a self-attention branch designed to control its updates. Target-features changes over long spans were learned without increasing the queue length. The proposed method achieved 68% accuracy on UAV datasets, outperforming other unsupervised trackers and matching typical supervised-tracker performance. On general scene datasets, it attained 77?% accuracy, comparable to other unsupervised trackers. Experimental results on both UAV and general scene datasets demonstrated that the proposed method achieved excellent performance in scenarios involving rapid motion and large-scale variations.

The distribution gap between training and test sets poses generalization challenges for deep-learning models. Systematic analysis identified two key urgent problems: insufficient optimization of knowledge transfer from training data to test data, and the impact of uneven categories in the dataset. To address these challenges, a novel test-time adaptation algorithm, fine-tuning with trusted pseudo-labels (FTP), was proposed. By optimizing the sample selection process, the test samples with low entropy value were selected to construct a fine-tuned dataset, and the model was fine-tuned by combining with the original training set, significantly enhancing the generalization performance of the image classification model on the test set. Extensive experiments on MNIST, FashionMNIST, and CIFAR10 datasets showed that the image classification model combined with FTP generally achieved a performance improvement on the test set, with an accuracy increase of up to about 3%, and a corresponding increase in F1 score, outperforming the current commonly used test adaptation methods such as TENT, COTTA, EATA, and OSTTA. In addition, the gradient-based visual analysis confirmed that the FTP-fine-tuned model preserved good interpretability while maintaining high prediction accuracy, offering reliable guarantee for practical applications.

The blind image super-resolution (SR) problem is concerned with recovering high-resolution (HR) images from low-resolution (LR) images with unknown degradation patterns. Currently, most existing methods primarily use explicit modeling to estimate blur kernels for characterizing image degradation processes. However, real-world image degradation is complex and diverse, making explicit modeling unable to fully cover multiple degradation types. Although implicit modeling is more effective in handling complex degradations, its model structures are often complicated with huge parameter sizes, leading to high computational costs and poor model stability. To address these issues, a lightweight blind SR reconstruction method named BDSSR was proposed, achieving efficient reconstruction through an implicit learning mechanism. The core framework of BDSSR consisted of a degradation factor eliminator (DFE) and a feature-fusion SR (FFSR) network. The DFE separated images with complex degradations into a clear LR image containing only bicubic down-sampling and non-bicubic degradation features such as noise and blur. Specifically, the clear LR image was provided as high-quality input for the SR process, reducing noise and blur interferences; the separated degradation features were fused into the SR network through feature-modulation coefficients to adaptively adjust the network weights, guiding the model to focus on the fine-grained reconstruction of high-frequency details. The FFSR further employed a multi-scale convolution strategy to enhance the capture capability of image content through efficient fusion of cross-scale features, thereby generating rich and realistic details and enabling robust modeling of complex degradations within a lightweight architecture. Experimental results demonstrated that BDSSR exhibited superior performance on multiple standard datasets. Taking the Urban100 dataset as an example, at ×2 and ×4 magnification factors, BDSSR improved the PSNR values by 0.97 dB and 0.47 dB, respectively, compared to DASR, with SSIM values increased by 0.012 2 and 0.015 8. Additionally, its parameter count was only 1.7 M, approximately 3/10 of that of DASR. This method provided a new theoretical perspective, and broad application prospects in practical scenarios were demonstrated, contributing novel ideas and tools to the development of blind super-resolution technology.

Multiple reference images were used to guide video colorization, which provides an efficient means of user-intent guidance and can better handle scene changes in videos. However, challenges remain in the allocation of color information from the reference images, in ensuring that the colorized result faithfully matches the user’s reference images, and in maintaining color naturalness and temporal consistency. To address these challenges, a video colorization method based on multiple reference images was proposed. First, a reference image feature extraction and recommendation module was designed. Convolutional neural networks were employed to extract features from multiple reference images and to calculate their semantic similarity to the grayscale video frames, upon which color information was recommended for the grayscale frames based on this similarity. Next, a temporal color module was introduced, in which a constrained attention mechanism used color information from the previous frame to guide the colorization of the current frame, ensuring natural color transitions and temporal consistency. Then, a color fusion network fused the recommended color from reference images with temporal color features, resolving conflicts between colors among multiple sources and generating a consistent color representation. Finally, a decoder module decoded the fused color information into the final color video frames. Experimental results demonstrated that the proposed method performed well on several public datasets, especially in handling complex scene transitions. The generated videos significantly improved visual quality, color transition smoothness, and overall consistency, demonstrating its great potential for application in video colorization.

Steel arch towers, as the primary load-bearing structures of steel arch cable-stayed bridges, require early detection and assessment of their surface defects—such as corrosion, spalling, and cracks—to ensure structural safety. However, traditional manual inspection methods are inefficient, highly subjective, and unable to access concealed areas at high altitudes. To address these challenges, an intelligent detection method based on an improved YOLOv8n-Seg deep-learning framework integrated with the OSRA attention mechanism was proposed. High-resolution internal images of steel arch towers were collected using a self-developed rail-guided inspection robot system. A comprehensive dataset containing 5 846 original images was constructed from both collected and open-source data, and data augmentation techniques—including random cropping, mirroring, and brightness adjustment—expanded the dataset to 23 378 images. At the algorithmic level, the OSRA attention module was innovatively embedded into the feature fusion layer of the YOLOv8n-Seg network. By leveraging an overlapping patching strategy and a local refinement mechanism, the model’s ability to capture irregular boundaries and small-scale defect features was significantly enhanced. Experimental results demonstrated that the optimized YOLOv8-OSRA model achieved notable performance improvements on an independent test set: corrosion detection mAP@0.5 reached 90.9% (+2.6%), crack identification accuracy reached 87.0% (+1.1%), and spalling detection accuracy reached 81.9% (+2.1%). Ablation experiments further confirmed that the OSRA module, while maintaining computational efficiency (increasing GFLOPs by only 0.8%), outperformed conventional attention mechanisms such as SE and CBAM. The findings provided a lightweight and deployable solution for steel arch tower defect detection, and the proposed multi-scale feature enhancement approach offered valuable insights for detecting surface defects in complex steel structures.

In the field of computer graphics and vision, 3D scene graph generation (SGG) has gained widespread attention in recent years. While existing research has improved the accuracy of coarse-grained classification and single-relation labels, performance in fine-grained classification and multi-label scenarios remains inadequate, limiting real-world applications. To address this, an innovative framework was proposed to fully utilizes contextual information to achieve fine-grained entity classification, multi-relation labeling, and enhanced accuracy. Our method comprised two core modules: the graph feature extraction (GFE) module and the graph context inference (GCI) module. The GFE module was used to extract entity and interaction semantic features from input data to ensure the extraction of key information. The GCI module introduced structural features from both traditional graphs and hypergraphs, analyzed relationships between entities, identified relational proximity within neighborhoods, and merged entities with similar interaction patterns to learn their interactions. The geometric hypergraph structure was dynamically generated based on scene layouts, providing structured organizational information. Experimental evaluations on the 3DSSG dataset, by integrating the organizational capabilities of both traditional graphs and hypergraphs for node and relationship clustering, the proposed work effectively improved fine-grained classification and multi-relation label recognition in 3D SGG tasks.

Multiphysics-field coupling analysis is an important tool for the design and optimization of complex engineering products. However, in practical applications, the incremental replacement of CAE software and algorithms faces many difficulties, such as the strong closed nature of commercial software, the limited functionality of open-source software, and the complexity of integrating self-developed algorithms. To address these problems, through in-depth research on the multi-physics coupling scheme and based on a loosely coupled scheme of the incremental integration architecture and an unstructured grid mapping algorithm, a fully automatic incremental replacement multi-physics coupling platform was developed relying on open-source and self-developed CAE software. The platform adopted a modular design and supported heterogeneous parallel computing of CPU/GPU. A multi-level framework of Field (data), Module, Node, and SceneGraph enabled flexible management and customization of the simulation process. Based on the plug-in system, it realized the rapid integration of custom algorithms. By implementing the extended development of data transfer algorithms, fluid analysis software, and structural analysis software in fluid-structure coupling analysis and computation on the platform, it was demonstrated that this platform possessed a more convenient advantage in the incremental replacement of multi-physics field analysis and could meet the actual engineering simulation requirements.

In order to improve the emotional experience of products and the efficiency of design, and aiming at the operational problems in the process of implementing the design affordance and emotionalization, this paper proposes an approach to the design of products based on feature mapping for the fusion of emotion and affordance. Firstly, the relevant concepts of the affordance mapping are analyzed, and the affordance items to be studied are determined by using the universal affordance template, which provides the basis for the generation of emotional afford. The Kano and rough number methods are applied to determine the affordance problem items affecting the emotional experience. Secondly, the association rules of “affordance-interaction design” and “interaction design principle-standard engineering parameter” are generated based on the Apriori algorithm, and the path of affordance mapping is constructed, then the affordance mapping ontology is retrieved, and the TRlZ conflict resolution theory is applied to the solution of the affordance problem items, and the overall design scheme is perfected based on theance structure matrix. Finally, based on the three-level theory of emotion and the theory of affordance, the evaluation index system for product design is established, and the weight of index is calculated by the combined weighting method of AHP-DEMATEL-CRITIC, and the product based on the feature mapping of emotion and affordance is designed evaluated. On this basis, the process of product design based on feature mapping is formed, and this process is applied to the improvement design of home oxygen machine, which verifies its and provides a new perspective for the design method of products based on the fusion of emotion and affordance.