Pub Date : 2025-08-13DOI: 10.1016/j.cad.2025.103933
Wuheng Gao, Renjie Chen
Delaunay triangulation is a cornerstone of computational geometry, playing a pivotal role in computer-aided engineering (CAE) and finite element methods (FEM) for generating high-quality meshes from scattered data points. While several methods have been developed for parallel Delaunay triangulation using GPUs, these approaches rely on CPU post-processing to produce the final result. In contrast, we introduce the first fully GPU-parallel algorithm for constructing the Delaunay triangulation of a given point set in . Our method is based on a generalization of the Local Delaunay Lemma by Chen and Gotsman, (2013) to , which enables the localization of Delaunay neighbors within a confined region around a given point. We provide a formal proof for this generalized lemma and leverage its advantages to efficiently construct the Delaunay triangulation using a naïvely parallel half-space intersection approach. Each point is processed independently, utilizing only the candidate points identified through the lemma. Additionally, we integrate several acceleration techniques tailored to exploit the hardware capabilities of modern GPUs, further optimizing runtime performance. Extensive experimentation and thorough comparisons demonstrate the efficiency of our method. Notably, our approach outperform the state-of-the-art hybrid GPU-CPU method by a factor of three when the point distribution is close to uniform.
{"title":"Parallel 3D Delaunay Triangulation on the GPU","authors":"Wuheng Gao, Renjie Chen","doi":"10.1016/j.cad.2025.103933","DOIUrl":"10.1016/j.cad.2025.103933","url":null,"abstract":"<div><div>Delaunay triangulation is a cornerstone of computational geometry, playing a pivotal role in computer-aided engineering (CAE) and finite element methods (FEM) for generating high-quality meshes from scattered data points. While several methods have been developed for parallel Delaunay triangulation using GPUs, these approaches rely on CPU post-processing to produce the final result. In contrast, we introduce the first fully GPU-parallel algorithm for constructing the Delaunay triangulation of a given point set in <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>3</mn></mrow></msup></math></span>. Our method is based on a generalization of the Local Delaunay Lemma by Chen and Gotsman, (2013) to <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>3</mn></mrow></msup></math></span>, which enables the localization of Delaunay neighbors within a confined region around a given point. We provide a formal proof for this generalized lemma and leverage its advantages to efficiently construct the Delaunay triangulation using a naïvely parallel half-space intersection approach. Each point is processed independently, utilizing only the candidate points identified through the lemma. Additionally, we integrate several acceleration techniques tailored to exploit the hardware capabilities of modern GPUs, further optimizing runtime performance. Extensive experimentation and thorough comparisons demonstrate the efficiency of our method. Notably, our approach outperform the state-of-the-art hybrid GPU-CPU method by a factor of three when the point distribution is close to uniform.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103933"},"PeriodicalIF":3.1,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-13DOI: 10.1016/j.cad.2025.103938
Kun Liu, Yan Zhang, Yanwen Guo, Jie Guo
Localized neural implicit representations have shown great potential in reconstructing and generating high-quality 3D shapes. However, current works usually decompose shapes in a deterministic manner by uniformly sampling points and encoding these points to latent code. In contrast, we utilize learnable positions and associated latent codes for each of these positions. By adopt the transformer encoder–decoder architecture, we can extract position of interest from a given 3D surface and encode latent feature for each position. The learned positions enable the allocation of more latent vectors to complex areas and fewer in flatter areas, resulting in greater flexibility and efficiency with a limited number of latent vectors. In addition, we show that our proposed representation is compatible with generative models. By decomposing the generation of latent positions and code vectors, we can utilize the diffusion models to generate proposed representation and extract high-quality 3D shapes. Experiments show our method achieve better reconstruction performance compared to existing methods using the same number of latent vectors, and comparable result with SOTA generative models. We show our model can generative novel shapes under various conditions, including category-conditioned, text-conditioned, image-conditioned, and unconditional generation.
{"title":"ANIR: Adaptive Neural Implicit Representation for 3D shape reconstruction and generation","authors":"Kun Liu, Yan Zhang, Yanwen Guo, Jie Guo","doi":"10.1016/j.cad.2025.103938","DOIUrl":"10.1016/j.cad.2025.103938","url":null,"abstract":"<div><div>Localized neural implicit representations have shown great potential in reconstructing and generating high-quality 3D shapes. However, current works usually decompose shapes in a deterministic manner by uniformly sampling points and encoding these points to latent code. In contrast, we utilize learnable positions and associated latent codes for each of these positions. By adopt the transformer encoder–decoder architecture, we can extract position of interest from a given 3D surface and encode latent feature for each position. The learned positions enable the allocation of more latent vectors to complex areas and fewer in flatter areas, resulting in greater flexibility and efficiency with a limited number of latent vectors. In addition, we show that our proposed representation is compatible with generative models. By decomposing the generation of latent positions and code vectors, we can utilize the diffusion models to generate proposed representation and extract high-quality 3D shapes. Experiments show our method achieve better reconstruction performance compared to existing methods using the same number of latent vectors, and comparable result with SOTA generative models. We show our model can generative novel shapes under various conditions, including category-conditioned, text-conditioned, image-conditioned, and unconditional generation.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103938"},"PeriodicalIF":3.1,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144908390","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-13DOI: 10.1016/j.cad.2025.103939
Zijian Mei , Yang Huang , Jingrun Chen
We propose a novel mesh-free topology optimization framework based on the random feature method (RFMTO). It inputs domain coordinates and boundary conditions and minimizes structural compliance under a volume constraint. By coupling a density field neural network with a physics-informed response network, RFMTO eliminates the discretization of design variables and the need for traditional finite element analysis (FEA), enabling direct structural topology optimization. Unlike the physics-informed neural networks used in related work and the FEA in traditional approaches, the RFM solver in this framework retains the advantages of mesh-free methods while efficiently solving the associated partial differential equations, offering high accuracy with reduced computational time, which is essential for topology optimization. Meanwhile, to address the issue of density networks converging to poor local minima in complex cases encountered by existing methods, we incorporate a point-wise density target loss into the loss function to guide the network updates more effectively. We conducted experiments on problems such as linear elasticity and heat sink optimizations. Results on both 2D and 3D benchmark problems demonstrate that RFMTO achieves performance comparable to, or even better than, classical methods such as SIMP, while maintaining similar or improved computational efficiency. Compared to state-of-the-art neural network-based approaches such as DMF-TONN (Direct Mesh-free Topology Optimization Method using Neural Networks), RFMTO can generate smoother and more detailed designs, significantly reduce computation time, and solve problems — such as heat sink optimization — that those methods fail to address. These findings indicate that RFMTO holds strong potential as an efficient and accurate industrial-grade topology optimization tool.
{"title":"Direct mesh-free topology optimization using random feature method","authors":"Zijian Mei , Yang Huang , Jingrun Chen","doi":"10.1016/j.cad.2025.103939","DOIUrl":"10.1016/j.cad.2025.103939","url":null,"abstract":"<div><div>We propose a novel mesh-free topology optimization framework based on the random feature method (RFMTO). It inputs domain coordinates and boundary conditions and minimizes structural compliance under a volume constraint. By coupling a density field neural network with a physics-informed response network, RFMTO eliminates the discretization of design variables and the need for traditional finite element analysis (FEA), enabling direct structural topology optimization. Unlike the physics-informed neural networks used in related work and the FEA in traditional approaches, the RFM solver in this framework retains the advantages of mesh-free methods while efficiently solving the associated partial differential equations, offering high accuracy with reduced computational time, which is essential for topology optimization. Meanwhile, to address the issue of density networks converging to poor local minima in complex cases encountered by existing methods, we incorporate a point-wise density target loss into the loss function to guide the network updates more effectively. We conducted experiments on problems such as linear elasticity and heat sink optimizations. Results on both 2D and 3D benchmark problems demonstrate that RFMTO achieves performance comparable to, or even better than, classical methods such as SIMP, while maintaining similar or improved computational efficiency. Compared to state-of-the-art neural network-based approaches such as DMF-TONN (Direct Mesh-free Topology Optimization Method using Neural Networks), RFMTO can generate smoother and more detailed designs, significantly reduce computation time, and solve problems — such as heat sink optimization — that those methods fail to address. These findings indicate that RFMTO holds strong potential as an efficient and accurate industrial-grade topology optimization tool.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103939"},"PeriodicalIF":3.1,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879823","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-11DOI: 10.1016/j.cad.2025.103934
Yu Chen, Hongwei Lin
Reconstructing models from unorganized point clouds presents a significant challenge, especially when the models consist of multiple components represented by their surface point clouds. Such models often involve point clouds with noise that represent multiple closed surfaces with shared regions, making their automatic identification and separation inherently complex. In this paper, we propose an automatic method that uses the topological understanding provided by persistent homology, along with representative 2-cycles of persistent homology groups, to effectively distinguish and separate each closed surface. Furthermore, we employ Loop subdivision and least squares progressive iterative approximation (LSPIA) techniques to generate high-quality final surfaces and achieve complete model reconstruction. Our method is robust to noise in the point cloud, making it suitable for reconstructing models from such data. Experimental results demonstrate the effectiveness of our approach and highlight its potential for practical applications.
{"title":"Robust Model Reconstruction Based on the Topological Understanding of Point Clouds Using Persistent Homology","authors":"Yu Chen, Hongwei Lin","doi":"10.1016/j.cad.2025.103934","DOIUrl":"10.1016/j.cad.2025.103934","url":null,"abstract":"<div><div>Reconstructing models from unorganized point clouds presents a significant challenge, especially when the models consist of multiple components represented by their surface point clouds. Such models often involve point clouds with noise that represent multiple closed surfaces with shared regions, making their automatic identification and separation inherently complex. In this paper, we propose an automatic method that uses the topological understanding provided by persistent homology, along with representative 2-cycles of persistent homology groups, to effectively distinguish and separate each closed surface. Furthermore, we employ Loop subdivision and least squares progressive iterative approximation (LSPIA) techniques to generate high-quality final surfaces and achieve complete model reconstruction. Our method is robust to noise in the point cloud, making it suitable for reconstructing models from such data. Experimental results demonstrate the effectiveness of our approach and highlight its potential for practical applications.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103934"},"PeriodicalIF":3.1,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144827089","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-11DOI: 10.1016/j.cad.2025.103937
Hui Wang , Xinye Li , Zhi Li , Cheng Wang
High-quality surface designs are increasingly significant in industrial applications, such as architecture and product design, yet they pose challenges in balancing visual appeal and functional requirements. Isogonal nets (I-nets) stand out for their aesthetically pleasing patterns and engineering practicality. However, constructing such nets remains difficult due to their dependence on complex angle constraints or a narrow focus on orthogonal scenarios. We propose a novel representation and construction method for I-nets characterized by similar mid-edge subdivided parallelograms in the quad faces. This approach achieves a simple yet versatile representation that generalizes orthogonal nets and extends to the construction of isogonal 4-webs (I-webs). By focusing on constraining edge ratios, our method enables efficient integration into mesh optimization algorithms. We demonstrate the effectiveness of I-nets and I-webs in freeform shapes through conformal mapping and numerical optimization. Experiments on various surfaces validate our method, showcasing its potential for both theoretical advancements and practical applications.
{"title":"Discrete isogonal nets with similar parallelograms","authors":"Hui Wang , Xinye Li , Zhi Li , Cheng Wang","doi":"10.1016/j.cad.2025.103937","DOIUrl":"10.1016/j.cad.2025.103937","url":null,"abstract":"<div><div>High-quality surface designs are increasingly significant in industrial applications, such as architecture and product design, yet they pose challenges in balancing visual appeal and functional requirements. Isogonal nets (I-nets) stand out for their aesthetically pleasing patterns and engineering practicality. However, constructing such nets remains difficult due to their dependence on complex angle constraints or a narrow focus on orthogonal scenarios. We propose a novel representation and construction method for I-nets characterized by similar mid-edge subdivided parallelograms in the quad faces. This approach achieves a simple yet versatile representation that generalizes orthogonal nets and extends to the construction of isogonal 4-webs (I-webs). By focusing on constraining edge ratios, our method enables efficient integration into mesh optimization algorithms. We demonstrate the effectiveness of I-nets and I-webs in freeform shapes through conformal mapping and numerical optimization. Experiments on various surfaces validate our method, showcasing its potential for both theoretical advancements and practical applications.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103937"},"PeriodicalIF":3.1,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144860346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-07DOI: 10.1016/j.cad.2025.103932
Zhihe Wu , Yaomin Wang , Zhenzhong Kuang , Jiajun Ding , Min Tan , Xuefei Yin , Yanming Zhu
Point clouds are widely used across various domains, yet their unordered and unstructured nature presents challenges for lightweight models and real-time inference. This paper introduces ViewCloud, a novel multi-view point-cloud-like representation that integrates the advantages of 2D renderings and 3D point clouds while maintaining a compact and efficient structure. Unlike conventional 3D representations, ViewCloud explicitly preserves viewpoint-specific geometric and semantic features, ensuring high information density with minimal redundancy. To construct ViewCloud, we propose an adaptive sampling strategy that extracts contour and interior pixels from multi-view 2D renderings, capturing essential shape characteristics while reducing storage overhead. We further design a ViewCloud-based multi-view feature aggregation Network, incorporating a contrastive learning-based semantic alignment Loss to enhance cross-view consistency and improve 3D recognition. Additionally, we extend ViewCloud to cross-domain retrieval, leveraging it as an intermediate representation to bridge 2D images and 3D point clouds within a shared feature space. Experiments on three benchmark datasets demonstrate that ViewCloud surpasses state-of-the-art methods in 3D recognition and cross-domain retrieval while significantly reducing storage and computational costs. These results establish ViewCloud as a scalable, efficient, and generalizable 3D representation.
{"title":"ViewCloud: A lightweight multi-view point cloud representation for efficient 3D recognition and cross-domain retrieval","authors":"Zhihe Wu , Yaomin Wang , Zhenzhong Kuang , Jiajun Ding , Min Tan , Xuefei Yin , Yanming Zhu","doi":"10.1016/j.cad.2025.103932","DOIUrl":"10.1016/j.cad.2025.103932","url":null,"abstract":"<div><div>Point clouds are widely used across various domains, yet their unordered and unstructured nature presents challenges for lightweight models and real-time inference. This paper introduces ViewCloud, a novel multi-view point-cloud-like representation that integrates the advantages of 2D renderings and 3D point clouds while maintaining a compact and efficient structure. Unlike conventional 3D representations, ViewCloud explicitly preserves viewpoint-specific geometric and semantic features, ensuring high information density with minimal redundancy. To construct ViewCloud, we propose an adaptive sampling strategy that extracts contour and interior pixels from multi-view 2D renderings, capturing essential shape characteristics while reducing storage overhead. We further design a ViewCloud-based multi-view feature aggregation Network, incorporating a contrastive learning-based semantic alignment Loss to enhance cross-view consistency and improve 3D recognition. Additionally, we extend ViewCloud to cross-domain retrieval, leveraging it as an intermediate representation to bridge 2D images and 3D point clouds within a shared feature space. Experiments on three benchmark datasets demonstrate that ViewCloud surpasses state-of-the-art methods in 3D recognition and cross-domain retrieval while significantly reducing storage and computational costs. These results establish ViewCloud as a scalable, efficient, and generalizable 3D representation.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103932"},"PeriodicalIF":3.1,"publicationDate":"2025-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144860792","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
CAD assembly models are typically represented as a collection of components, each of which can share geometric interfaces with others. In the literature, geometric interfaces have been shown to play a fundamental role in assembly model analysis, component characterization, and classification. While these interfaces are not explicitly defined in CAD models, they can be inferred from the relative positioning of components. The resulting geometric interfaces can be categorized as either interference or contact. However, it is often unclear whether these interfaces stem from intentional design choices related to component shape and function, from consistently applied relative positioning, or from unintended errors.
In industrial practice, the design of complex products often involves models sourced from public catalogs for third-party components. These catalog models frequently include shape simplifications, which can lead to unintended intersections or clearances with surrounding components — deviations that do not exist in the final physical product. This study aims to provide a comprehensive analysis and formalization of geometric interfaces, based on the complementary roles of CAD assembly modules and digital component catalogs, both widely used in industry as foundational resources for generating assembly models. The results are directly applicable to industrial CAD assembly models and can serve as a reference for CAD developers seeking to improve and extend assembly processing, as well as for researchers conducting assembly analysis.
This work introduces a formalization of geometric interfaces, including contacts, interferences, and interface envelopes, which are essential for defining component mounting requirements. An analysis of geometric interface perturbations caused by repetition operators is performed, leading to the concept of an interface envelope to model specific interface repetitions. The nominal assembly representation, presented as a reference model, facilitates the formalization of interface consistency, supporting more robust reasoning processes.
{"title":"A structured analysis of CAD assembly model interfaces for their enhanced computerized processing","authors":"Jean-Claude Léon , Flavien Boussuge , Franca Giannini , Marina Monti , Katia Lupinetti , Brigida Bonino , Jean-Philippe Pernot , Roberto Raffaeli","doi":"10.1016/j.cad.2025.103911","DOIUrl":"10.1016/j.cad.2025.103911","url":null,"abstract":"<div><div>CAD assembly models are typically represented as a collection of components, each of which can share geometric interfaces with others. In the literature, geometric interfaces have been shown to play a fundamental role in assembly model analysis, component characterization, and classification. While these interfaces are not explicitly defined in CAD models, they can be inferred from the relative positioning of components. The resulting geometric interfaces can be categorized as either interference or contact. However, it is often unclear whether these interfaces stem from intentional design choices related to component shape and function, from consistently applied relative positioning, or from unintended errors.</div><div>In industrial practice, the design of complex products often involves models sourced from public catalogs for third-party components. These catalog models frequently include shape simplifications, which can lead to unintended intersections or clearances with surrounding components — deviations that do not exist in the final physical product. This study aims to provide a comprehensive analysis and formalization of geometric interfaces, based on the complementary roles of CAD assembly modules and digital component catalogs, both widely used in industry as foundational resources for generating assembly models. The results are directly applicable to industrial CAD assembly models and can serve as a reference for CAD developers seeking to improve and extend assembly processing, as well as for researchers conducting assembly analysis.</div><div>This work introduces a formalization of geometric interfaces, including contacts, interferences, and interface envelopes, which are essential for defining component mounting requirements. An analysis of geometric interface perturbations caused by repetition operators is performed, leading to the concept of an interface envelope to model specific interface repetitions. The nominal assembly representation, presented as a reference model, facilitates the formalization of interface consistency, supporting more robust reasoning processes.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103911"},"PeriodicalIF":3.1,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144896241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-06DOI: 10.1016/j.cad.2025.103943
Junfeng Gao, Zihao Yang, Yuan Liang, Yongcun Zhang, Kangjie Liu
Inspired by natural shell-infill systems with spatially adaptive coating thicknesses (e.g., human femur bones), this paper proposes a mixed-variable topology optimization method for collaboratively designing the base topology and the adaptive coating thickness distribution of shell-infill structures. The optimization framework consists of two coupled levels. At the first level, a discrete-variable topology optimization method is employed to generate a base structure (shell and infill) with uniform coating thickness, effectively eliminating intermediate density elements to ensure a clear material interface for coating identification. At the second level, the coating size optimization is realized through density-based topology optimization combined with a novel holeless coating constraint based on a virtual temperature field. Meanwhile, to ensure manufacturability, a minimum coating thickness constraint is introduced. A density field mapping strategy further couples the two optimization levels, enabling iterative updates of both the base topology and coating thickness distribution. Three numerical examples demonstrate the effectiveness of the proposed method. The shell-infill structure with adaptive coating thickness achieves over 10 % mass reduction. Additionally, the constraints successfully eliminate unmanufacturable holes while preserving thickness continuity. Moreover, a large-scale 3D case validates the capability of the method for handling complex three-dimensional coating problems. The results highlight the potential of the method in designing bio-inspired, high-performance shell-infill structures.
{"title":"Mixed-variable topology optimization for shell-infill structures with adaptive coating thickness","authors":"Junfeng Gao, Zihao Yang, Yuan Liang, Yongcun Zhang, Kangjie Liu","doi":"10.1016/j.cad.2025.103943","DOIUrl":"10.1016/j.cad.2025.103943","url":null,"abstract":"<div><div>Inspired by natural shell-infill systems with spatially adaptive coating thicknesses (e.g., human femur bones), this paper proposes a mixed-variable topology optimization method for collaboratively designing the base topology and the adaptive coating thickness distribution of shell-infill structures. The optimization framework consists of two coupled levels. At the first level, a discrete-variable topology optimization method is employed to generate a base structure (shell and infill) with uniform coating thickness, effectively eliminating intermediate density elements to ensure a clear material interface for coating identification. At the second level, the coating size optimization is realized through density-based topology optimization combined with a novel holeless coating constraint based on a virtual temperature field. Meanwhile, to ensure manufacturability, a minimum coating thickness constraint is introduced. A density field mapping strategy further couples the two optimization levels, enabling iterative updates of both the base topology and coating thickness distribution. Three numerical examples demonstrate the effectiveness of the proposed method. The shell-infill structure with adaptive coating thickness achieves over 10 % mass reduction. Additionally, the constraints successfully eliminate unmanufacturable holes while preserving thickness continuity. Moreover, a large-scale 3D case validates the capability of the method for handling complex three-dimensional coating problems. The results highlight the potential of the method in designing bio-inspired, high-performance shell-infill structures.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103943"},"PeriodicalIF":3.1,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144810677","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-06DOI: 10.1016/j.cad.2025.103944
Min Wu , Yinghui Wang , Liangyi Huang , Jinlong Yang , Wei Li , Jiaxing Shen , Xiaojuan Ning
Normal estimation for point clouds is fundamental to 3D geometric processing and applications. Despite recent advances by deep learning-based methods, effectively representing geometric structures in regions with sharp features and complex geometries remains challenging. This limitation primarily arises from the use of general architectures (e.g., CNNs, PointNet) or conventional graph convolutions, which limits the ability to capture fine geometric details in local point cloud patches. Moreover, the persistent issue of scale ambiguity in selecting optimal neighborhoods further hinders precise encoding of local structures. To address these challenges, we propose EPR-Net, a novel framework that enhances local patch representation learning for normal estimation in point clouds. Specifically, we introduce the GraphFormer module, which builds on the PoolFormer architecture to improve feature learning and incorporates graph convolution with adaptive kernels to capture geometric details across different semantic regions, thereby enabling more discriminative feature encodings. Additionally, we design the pyramid dynamic graph update (PDGU) strategy, which guides multi-scale feature aggregation through geometric weights to alleviate the scale ambiguity in neighborhood selection. PDGU also dynamically updates the local k-nearest neighbor (kNN) graph to expand the receptive field, thereby enhancing the ability of the model to extract long-range semantic information from point cloud patches. Extensive experiments are conducted on both synthetic and real-world datasets, and the qualitative and quantitative evaluations demonstrate the superiority of our method in point cloud normal estimation.
{"title":"EPR-Net: Enhanced patch representation network for point cloud normal estimation","authors":"Min Wu , Yinghui Wang , Liangyi Huang , Jinlong Yang , Wei Li , Jiaxing Shen , Xiaojuan Ning","doi":"10.1016/j.cad.2025.103944","DOIUrl":"10.1016/j.cad.2025.103944","url":null,"abstract":"<div><div>Normal estimation for point clouds is fundamental to 3D geometric processing and applications. Despite recent advances by deep learning-based methods, effectively representing geometric structures in regions with sharp features and complex geometries remains challenging. This limitation primarily arises from the use of general architectures (e.g., CNNs, PointNet) or conventional graph convolutions, which limits the ability to capture fine geometric details in local point cloud patches. Moreover, the persistent issue of scale ambiguity in selecting optimal neighborhoods further hinders precise encoding of local structures. To address these challenges, we propose EPR-Net, a novel framework that enhances local patch representation learning for normal estimation in point clouds. Specifically, we introduce the GraphFormer module, which builds on the PoolFormer architecture to improve feature learning and incorporates graph convolution with adaptive kernels to capture geometric details across different semantic regions, thereby enabling more discriminative feature encodings. Additionally, we design the pyramid dynamic graph update (PDGU) strategy, which guides multi-scale feature aggregation through geometric weights to alleviate the scale ambiguity in neighborhood selection. PDGU also dynamically updates the local k-nearest neighbor (kNN) graph to expand the receptive field, thereby enhancing the ability of the model to extract long-range semantic information from point cloud patches. Extensive experiments are conducted on both synthetic and real-world datasets, and the qualitative and quantitative evaluations demonstrate the superiority of our method in point cloud normal estimation.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103944"},"PeriodicalIF":3.1,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144842237","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-06DOI: 10.1016/j.cad.2025.103928
Jun Min , Xin Li , Li-yong Shen
This paper extends the generalized NURBS representation to support local refinement via truncation hierarchical mechanism. The new representation is called truncated hierarchical GNURBS (TH-GNURBS), which provides adaptive refinement on arbitrary topological unstructured quadrilateral control mesh with non-uniform knot intervals. To construct TH-GNURBS, this paper builds the hierarchical structure and applies the truncation mechanism for highly localized refinement. During the construction, we modify TH-GNURBS basis functions to maintains the continuity around the extraordinary points (EPs). The TH-GNURBS basis functions satisfy partition of unity, everywhere except at the local region surrounding EPs. Finally, we provide a fitting algorithm to approximate an arbitrary triangle mesh with TH-GNURBS. Experimental results show that higher fitting accuracy with fewer control points via adaptive spline surface fitting.
{"title":"Truncated hierarchical GNURBS for adaptive spline surface fitting","authors":"Jun Min , Xin Li , Li-yong Shen","doi":"10.1016/j.cad.2025.103928","DOIUrl":"10.1016/j.cad.2025.103928","url":null,"abstract":"<div><div>This paper extends the generalized NURBS representation to support local refinement via truncation hierarchical mechanism. The new representation is called truncated hierarchical GNURBS (TH-GNURBS), which provides adaptive refinement on arbitrary topological unstructured quadrilateral control mesh with non-uniform knot intervals. To construct TH-GNURBS, this paper builds the hierarchical structure and applies the truncation mechanism for highly localized refinement. During the construction, we modify TH-GNURBS basis functions to maintains the continuity around the extraordinary points (EPs). The TH-GNURBS basis functions satisfy partition of unity, <span><math><msup><mrow><mi>C</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> everywhere except <span><math><msup><mrow><mi>G</mi></mrow><mrow><mn>1</mn></mrow></msup></math></span> at the local region surrounding EPs. Finally, we provide a fitting algorithm to approximate an arbitrary triangle mesh with TH-GNURBS. Experimental results show that higher fitting accuracy with fewer control points via adaptive spline surface fitting.</div></div>","PeriodicalId":50632,"journal":{"name":"Computer-Aided Design","volume":"189 ","pages":"Article 103928"},"PeriodicalIF":3.1,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144827088","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号