Engineering with Computers最新文献

Mesh quality directly affects the accuracy and efficiency of numerical simulation. Mesh quality evaluation aims to evaluate the suitability of the mesh generated in CAE pre-processing for numerical simulation. Recent work has introduced deep neural networks for mesh quality evaluation. However, these methods treat the mesh quality evaluation task as a multi-classification problem, resulting in serious correlations among different quality categories, which makes it difficult to learn the boundaries of different categories. In this paper, we propose a topology-guided graph neural network, MTGNet, which treats the mesh quality evaluation task as a multi-label task. Specifically, we first decomposed the categories in traditional multi-classification problems and obtained three completely orthogonal mesh quality labels, namely orthogonality, smoothness and, distribution. Then, MTGNet introduces a topology-guided feature representation for structured mesh data, which can generate multiple blocks of element-based graphs through the mesh topology. In order to better fuse features in different blocks, MTGNet also introduces an attention-based block graph pooling (ABGPool) method. Experimental results on the NACA-Market dataset demonstrate MTGNet shows superior or at least comparable performance to the state-of-the-art (SOTA) approaches.

A thin shell formulation is developed for the approximation by a meshfree Reproducing Kernel Particle Method (RKPM). The formulation is derived from a degenerated shell approach where the structure is treated as a 3D solid subjected to kinematic constraints of the Kirchhoff–Love (KL) shell theory. To address the challenge of surface geometry representation in a meshfree method, a local parameterization using principal component analysis (PCA) is employed. Taylor-series expansion adapted to the shell formulation is developed to address the accuracy and stability issues of nodal quadrature. Several approaches that address membrane locking are also considered. The effectiveness of the proposed RKPM KL shell formulation is demonstrated using an extensive set of linear-elastic and finite-deformation inelastic test cases.

This study proposes a shape optimisation framework for unsteady electromagnetic scattering problems on the basis of the time-domain boundary integral equation method, focusing on the perfectly electric conductors (PECs). The boundary-only formulation is ideal for treating a shape optimisation problem in an exterior domain. However, the electromagnetic shape optimisation in concern has been unrealised with the boundary integral approach regardless of the fact that the boundary-type shape derivative has been known in the literature. The first contribution of the present study is to derive a novel expression of the shape derivative in terms of the surface current densities of the primary and adjoint problems, by considering that the surface current density is handled by usual integral equations methods. The second contribution is to clarify the integral representations and equations of the adjoint electromagnetic fields in terms of the reversal time. These theoretical achievements possess a high affinity with the standard spatial discretising approach (i.e. RWG basis) whenever the temporal basis is sufficiently smooth. The numerical experiments confirmed the reliability of the proposed shape optimisation methodology and indicated the capability to deal with scientific and engineering applications.

It is vital to control the vibration of cellular composites under harmonic excitation in engineering. Due to numerous design variables and expensive frequency domain integration operation, the majority of multiscale topology optimization methods for frequency response minimization of cellular composites tend to be conservative, where a small number of types of microstructures are considered. This paper proposes an efficient multiscale topology optimization method to minimize the frequency response of cellular composites over specified frequency intervals. This method utilizes multiclass graded lattice unit cells (LUCs) as design candidates, offering great design space to improve the dynamic performance of cellular composites. At microscale, the proposed method leverages Kriging metamodels to replace the the homogenization method in each iteration step, thus accelerating the performance estimation of multiclass graded LUCs. At macroscale, the second-order Krylov subspace with moment-matching Gram-Schmidt orthonormalization (SOMMG) method is introduced to expedite the frequency response analysis of cellular composites. Two types of design variables are employed to construct the Kriging metamodel assisted Uniform Multiphase Materials Interpolation (KUMMI) model, facilitating the concurrent updating of LUCs’ classes and relative densities. Several numerical examples are presented to validate the effectiveness and efficiency of the proposed method in minimizing the frequency response of cellular composites.

Efficient and robust anisotropic mesh adaptation is crucial for Computational Fluid Dynamics (CFD) simulations. The CFD Vision 2030 Study highlights the pressing need for this technology, particularly for simulations targeting supercomputers. This work applies a fine-grained speculative approach to anisotropic mesh operations. Our implementation exhibits more than 90% parallel efficiency on a multi-core node. Additionally, we evaluate our method within an adaptive pipeline for a spectrum of publicly available test-cases that includes both analytically derived and error-based fields. For all test-cases, our results are in accordance with published results in the literature. Support for CAD-based data is introduced, and its effectiveness is demonstrated on one of NASA’s High-Lift prediction workshop cases.

Machine learning is employed for solving physical systems governed by general nonlinear partial differential equations (PDEs). However, complex multi-physics systems such as acoustic-structure coupling are often described by a series of PDEs that incorporate variable physical quantities, which are referred to as parametric systems. There are lack of strategies for solving parametric systems governed by PDEs that involve explicit and implicit quantities. In this paper, a deep learning-based Multi Physics-Informed PointNet (MPIPN) is proposed for solving parametric acoustic-structure systems. First, the MPIPN introduces an enhanced point-cloud architecture that encompasses explicit physical quantities and geometric features of computational domains. Then, the MPIPN extracts local and global features of the reconstructed point-cloud as parts of solving criteria of parametric systems, respectively. Besides, implicit physical quantities are embedded by encoding techniques as another part of solving criteria. Finally, all solving criteria that characterize parametric systems are amalgamated to form distinctive sequences as the input of the MPIPN, whose outputs are solutions of systems. The proposed framework is trained by adaptive physics-informed loss functions for corresponding computational domains. The framework is generalized to deal with new parametric conditions of systems. The effectiveness of the MPIPN is validated by applying it to solve steady parametric acoustic-structure coupling systems governed by the Helmholtz equations. An ablation experiment has been implemented to demonstrate the efficacy of physics-informed impact with a minority of supervised data. The proposed method yields reasonable precision across all computational domains under constant parametric conditions and changeable combinations of parametric conditions for acoustic-structure systems.

Many natural materials exhibit highly complex, nonlinear, anisotropic, and heterogeneous mechanical properties. Recently, it has been demonstrated that data-driven strain energy functions possess the flexibility to capture the behavior of these complex materials with high accuracy while satisfying physics-based constraints. However, most of these approaches disregard the uncertainty in the estimates and the spatial heterogeneity of these materials. In this work, we leverage recent advances in generative models to address these issues. We use as building block neural ordinary equations (NODE) that—by construction—create polyconvex strain energy functions, a key property of realistic hyperelastic material models. We combine this approach with probabilistic diffusion models to generate new samples of strain energy functions. This technique allows us to sample a vector of Gaussian white noise and translate it to NODE parameters thereby representing plausible strain energy functions. We extend our approach to spatially correlated diffusion resulting in heterogeneous material properties for arbitrary geometries. We extensively test our method with synthetic and experimental data on biological tissues and run finite element simulations with various degrees of spatial heterogeneity. We believe this approach is a major step forward including uncertainty in predictive, data-driven models of hyperelasticity.