A novel neural network approach for airfoil mesh quality evaluation

Chen, Xinhai; Gong, Chunye; Liu, Jie; Pang, Yufei; Deng, Liang; Chi, Lihua; Li, Kenli

doi:10.1016/j.jpdc.2022.03.006

Cited by 9 publications

(3 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It should be noted that the common node is the same vertex number in different element types, in order to prevent the error of data information transmission at the common intersection of the mesh in different subregions. The difference between the hybrid element and the traditional finite element is that the basis function of the hybrid element is the basis function combination of multiple different types of elements, while the traditional finite element is the linear combination of the same type of basis function [30,31]. Therefore, the hybrid finite element method is an extension of the traditional finite element method and has stronger flexibility.…”

Section: Theory Of Hybrid Element Solving Elastic Equationmentioning

confidence: 99%

Thermal-Mechanical Coupling Model Based on the Hybrid Finite Element Method for Solving Bipolar the Plate Deformation of Hydrogen Fuel Cells

2022

View full text Add to dashboard Cite

New energy is the focus of attention all over the world, and research into new energy can inject new vitality into the industrial system. Hydrogen fuel cells are not only environmentally friendly, but also rich in reserves that can be used as a strategic resource for the entire country. The difficulty lies in the safe design of application equipment and the batch generation and storage of hydrogen. In addition, fuel cells have the disadvantage of a slow start-up. Based on the above problems, this paper proposes a hybrid-element method to solve the thermal-mechanical coupling model of fuel cell plate, which can effectively solve the thermal stress change, temperature field distribution and displacement change of the battery plate when working. Firstly, the hybrid-element algorithm is given for 2D plate deformation. Then, the deformation application of a 3D fuel cell plate is given. The 2D numerical results show that the hybrid finite element method (FEM) is more flexible for realizing the flexible combination of sub-mesh and finite element basis functions, and has a better mesh quality compared to the traditional constant strain triangular element (CST) adaptive FEM and quadrilateral isoparametric element (Q4) adaptive FEM. This method achieves a balance between numerical accuracy and solving efficiency for the multi-porous elastic plate. In addition, a deformation control formula is given which can display the displacement deformation and stress merge to same graph, since it is convenient to quickly compare the regions where the displacement and stress extremum appear. In short, the hybrid finite element method proposed in this paper has good mesh evaluation results, and when the number of discrete elements is equivalent, the hybrid element converges faster and the solution efficiency is higher. This paper also provides a good numerical theory and simulation reference for industrial mechanics and new energy applications.

show abstract

Section: Theory Of Hybrid Element Solving Elastic Equationmentioning

confidence: 99%

Thermal-Mechanical Coupling Model Based on the Hybrid Finite Element Method for Solving Bipolar the Plate Deformation of Hydrogen Fuel Cells

2022

View full text Add to dashboard Cite

show abstract

“…Chen et al [ 9 , 30 , 31 ] first introduced neural networks to the mesh quality evaluation task. They proposed an automatic quality indicator for 2D NACA0012 airfoil meshes using CNNs.…”

Section: Related Workmentioning

confidence: 99%

A Neural Network-Based Mesh Quality Indicator for Three-Dimensional Cylinder Modelling

Chen

Wang

Liu

et al. 2022

Entropy

Self Cite

View full text Add to dashboard Cite

Evaluating mesh quality prior to performing the computational fluid dynamics (CFD) simulation is an essential step to ensure the acceptable accuracy of cylinder modelling. However, traditional mesh quality indicators are often insufficient since they only check geometric information on individual distorted elements. To yield more accurate results, the current evaluation process usually requires careful manual re-evaluation for quality properties such as mesh distribution and local refinement, which heavily increase the meshing overhead. In this paper, we introduce an efficient quality indicator for varisized cylinder meshes, consisting of a mesh pre-processing method and a neural network-based indicator, Mesh-Net. We also publish a cylinder mesh benchmark dataset. The proposed indicator is trained to study the role of CFD meshes on the accuracy of numerical simulations. It considers both the effect of element geometry (e.g., orthogonality) and quality properties (e.g., smoothness and distribution). Thereafter, the well-trained indicator is used as a black-box to predict the overall quality of the input mesh automatically. Experimental results demonstrate that the proposed indicator is accurate and can be applied in the mesh quality evaluation process without manual interactions.

show abstract

“…However, existing CL methods primarily focus on continuous data streams with annotations, neglecting the fact that unlabelled data streams are prevalent in real-world applications. Meanwhile, Self-Supervised Learning (SSL) has recently achieved great success in visual representation learning [5,7,6,15,8], which is built upon the assumption of a vast amount of unlabelled and unbiased data. Nevertheless, this assumption does not always hold in a real-world scenario, e.g.…”

Section: Introductionmentioning

confidence: 99%

Adaptive Self-Supervised Continual Learning

Wu,

Wang,

Liu

2023

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

Continual Learning (CL) studies the problem of developing a robust model that can learn new tasks while retaining previously learned knowledge. However, the current CL methods exclusively focus on data with annotations, disregarding that unlabelled data is the mainstream in real-world applications. To close this research gap, this study concentrates on continual self-supervised learning, which is plagued by challenges of memory over-fitting and class imbalance. Besides, these challenges are exacerbated throughout incremental training. Aimed at addressing these challenges from both loss and data perspectives, we introduce a framework, Adaptive Self-supervised Continual Learning (ASCL). Specifically, we devise an Adaptive Sharpness-Aware Minimization (ASAM) module responsible for identifying flatter local minima in the loss landscape with a smaller memory over-fitting risk. Additionally, we design an Adaptive Memory Enhancement (AME) module responsible for rebalancing self-supervised loss with new and old tasks from a data perspective. Finally, the adaptive mechanisms in AME and ASAM modules dynamically adjust the loss landscape sharpness and memory enhancement strength with the feedback of intermediate training results. The results of our extensive experiments demonstrate the state-of-the-art performance of our methods in continual self-supervised learning scenarios across multiple datasets.

show abstract

A novel neural network approach for airfoil mesh quality evaluation

Cited by 9 publications

References 28 publications

Thermal-Mechanical Coupling Model Based on the Hybrid Finite Element Method for Solving Bipolar the Plate Deformation of Hydrogen Fuel Cells

Thermal-Mechanical Coupling Model Based on the Hybrid Finite Element Method for Solving Bipolar the Plate Deformation of Hydrogen Fuel Cells

A Neural Network-Based Mesh Quality Indicator for Three-Dimensional Cylinder Modelling

Adaptive Self-Supervised Continual Learning

Contact Info

Product

Resources

About