Computer-Aided Design最新文献

Despite the remarkable advancements in machine learning (ML) techniques for topology optimization, the predicted solutions often overlook the necessary structural connectivity required to meet the load-carrying demands of the resulting designs. Consequently, these predicted solutions exhibit subpar structural performance because disconnected components are unable to bear loads effectively and significantly compromise the manufacturability of the designs.

In this paper, we propose an approach to enhance the topological accuracy of ML-based topology optimization methods by employing a predicted dual connectivity graph. We show that the tangency graph of the Maximal Disjoint Ball Decomposition (MDBD), which accurately captures the topology of the optimal design, can be used in conjunction with a point transformer network to improve the connectivity of the design predicted by Generative Adversarial Networks and Convolutional Neural Networks. Our experiments show that the proposed method can significantly improve the connectivity of the final predicted structures. Specifically, in our experiments the error in the number of disconnected components was reduced by a factor of 4 or more without any loss of accuracy. We demonstrate the flexibility of our approach by presenting examples including various boundary conditions (both seen and unseen), domain resolutions, and initial design domains. Importantly, our method can seamlessly integrate with other existing deep learning-based optimization algorithms, utilize training datasets with models using any valid geometric representations, and naturally extend to three-dimensional applications.

Free-form structures are highly valued for their aesthetic appeal in architecture, but they typically comprise panels of many different shapes, which can pose great challenges for building construction. In this study, we aim to address this issue by proposing a novel clustering–optimization method to reduce the number of different $n$-gonal faces in free-form surface approximations. The method partitions the faces into several groups of similar shapes through clustering and transforms the ones within each group toward congruent forms through optimization. By utilizing this approach, the number of geometrically different panels can be reduced while also satisfying a user-specified error threshold. The potential practical application of this method is demonstrated by redesigning the façade of a real architectural project to achieve cost-effective solutions.

The objective of this study is to address the difficulty of simplifying the geometric model in which a differential problem is formulated, also called defeaturing, while simultaneously ensuring that the accuracy of the solution is maintained under control. This enables faster and more efficient simulations, without sacrificing accuracy in the regions of interest. More precisely, we consider an isogeometric discretisation of an elliptic model problem defined on a two-dimensional simply connected hierarchical B-spline physical domain with a complex boundary. Starting with an oversimplification of the geometry, we build a goal-oriented adaptive strategy that adaptively reintroduces continuous geometrical features in regions where the analysis suggests a large impact on the quantity of interest. This strategy is driven by an a posteriori estimator of the defeaturing error based on first-order shape sensitivity analysis, and it profits from the local refinement properties of hierarchical B-splines. The adaptive algorithm is described together with a procedure to generate (partially) simplified hierarchical B-spline geometrical domains. Numerical experiments are presented to illustrate the proposed strategy and its limitations.

Microstructure reconstruction is an important and emerging aspect of computational materials engineering and multiscale modeling and simulation. Despite extensive research and fast progress in the field, the application of descriptor-based reconstruction remains limited by computational resources. Common methods for increasing the computational feasibility of descriptor-based microstructure reconstruction lie in approximating the microstructure by simple geometrical shapes and by utilizing differentiable descriptors to enable gradient-based optimization. The present work combines these two ideas for structures composed of non-overlapping ellipsoidal inclusions such as magnetorheological elastomers. This requires to express the descriptors as a function of the microstructure parametrization. Deriving these relations leads to analytical solutions that further speed up the reconstruction procedure. Based on these descriptors, microstructure reconstruction is formulated as a multi-stage optimization procedure. The developed algorithm is validated by means of different numerical experiments and advantages and limitations are discussed in detail.

Three-dimensional non-periodic woven composite preforms have sufficient design flexibility that tows can be aligned along principal loading paths even in shaped structural components with detailed local features. While this promises competitive performance, the feasible design space is combinatorially large, far beyond exhaustive search. Seeking a design space that is compact and easily searched yet can span the full potential of 3D weaving, we propose a method for generating candidate designs called the Background Vector Method (BVM) which treats weaving tows as agents in a game competing to match background vectors derived from different design requirements. The BVM generates candidate designs that adapt local architecture to global design goals by adjusting scalar weights. A manufacturing-based parameterization assures fabricability. The scope of possible designs and the speed of the BVM are illustrated by re-creating common periodic 3D weaving patterns and novel complex non-periodic architectures, with a route demonstrated to forming cavities, ducts, and other open volumes. How the BVM might be incorporated within an optimization algorithm is outlined and pathways are shown for systematically enlarging the design space as individual design problems may require.

Sandwich panels with excellent mechanical properties are widely used in the aerospace, architecture, and automobile industries. Kirigami-inspired structural designs are receiving increasing attention owing to the shape-induced functions and novel properties imparted by their folds and cuts. In this study, novel graded self-locking kirigami panels based on a tucked-interleaved pattern are developed and analyzed under quasi-static loading. The proposed tucked-interleaved pattern can be assembled to form freely supported self-locking polyhedral structures. The self-locking property is ensured by the interleaved flaps, which create in-plane compression to hold the structure in place. In particular, we analyze the effects of geometric variations in kirigami panels fabricated using a CO₂ laser machining system. The experimental data under quasi-static compression and simulation results both indicate that the proposed kirigami panels have outstanding load-to-weight ratios on the order of 10⁵. It appears that the introduction of a graded design can generate graded stiffness as well as superior specific energy absorption with an appropriate introduction of geometric gradients. These results show that the proposed kirigami panels combining self-locking and programmable non-uniform stiffness have great potential for non-uniform engineering applications.

Hydrogel-based negative hydration expansion (NHE) metamaterials are composite structures composed of responsive hydrogels and polymers, and their properties depend on their unique structures. In this paper, an optimization method based on the combination of the back-propagation neural network (BPNN) and the multi-population genetic algorithm (MPGA) is developed to rapidly design isotropic and anisotropic hydrogel-based metamaterials with specific NHE effects. In this method, several dimensionless design parameters are introduced to describe the structural characteristics of the metamaterial. The initial dataset is constructed based on the finite element method simulation results, and the mapping relationship between the design parameters and the equivalent linear strain is constructed by the BPNN, and the metamaterial with specific effect is efficiently optimized by combining the MPGA. The method is proved to have high accuracy and efficiency, and is applied to design many novel 2D and 3D metamaterials. The 3D metamaterial designed by this method has an ultra-large NHE ratio about 82 %. Compared with the topology optimization method, this method can significantly reduce the amount of computation, and can effectively avoid falling into the local optimum. The results show that the optimization method based on machine learning is an efficient means to design hydrogel-based metamaterials.

Geometry understanding is a core concept of computer-aided design and engineering (CAD/CAE). Deep neural networks have increasingly shown success as a method of processing complex inputs to achieve abstract tasks. This work revisits a generic and relatively simple approach to 3D deep learning – a point-based graph neural network – and develops best-practices and modifications to alleviate traditional drawbacks. It is shown that these methods should not be discounted for CAD tasks; with proper implementation, they can be competitive with more specifically designed approaches. Through an additive study, this work investigates how the boundary representation data can be fully utilised by leveraging the flexibility of point-based graph networks. The final configuration significantly improves on the predictive accuracy of a standard PointNet++ network across multiple CAD model segmentation datasets and achieves state-of-the-art performance on the MFCAD++ machining features dataset. The proposed modifications leave the core neural network unchanged and results also suggest that they can be applied to other point-based approaches.

This paper proposes a multicomponent topology optimization method that considers assemblability. Generally, it is difficult to consider assemblability in topology optimization; however, in this study, we achieve it by introducing a fictitious physical model. To perform multicomponent topology optimization, the extended level set method is used to represent multiple components. First, the assembly constraints are formulated using a fictitious physical model limited to two components. Then, by considering stepwise assembly, the constraint is extended to three or more components. In addition, topology optimization algorithms are constructed using the finite element method. Several numerical examples demonstrate that the proposed method can obtain structures with assemblability and has low initial structure dependence.

Most three-dimensional (3D) computer-aided design (CAD) models of mechanical parts, created during the design stage, have high shape complexity. The shape complexity required of CAD models reduces according to the field of application. Therefore, it is necessary to simplify the shapes of 3D CAD models, depending on their applications. Traditional simplification methods recognize simplification target shape based on a pre-defined algorithm. Such algorithm-based methods have difficulty processing unusual partial shapes not considered in the CAD model. This paper proposes a method that uses a network based on a generative adversarial network (GAN) to simplify the 3D CAD models of mechanical parts. The proposed network recognizes and removes simplification target shapes included in the 3D CAD models of mechanical parts. A 3D CAD model dataset was constructed to train the 3D CAD model simplification network. 3D CAD models are represented in voxel form in the dataset. Next, the constructed training dataset was used to train the proposed network. Finally, a 3D voxel simplification experiment was performed to evaluate the performance of the trained network. The experiment results showed that the network had an average error rate of 3.38% for the total area of the mechanical part and an average error rate of 14.61% for the simplification target area.