A self-learning finite element extraction system based on reinforcement learning

Pan, Jie; Huang, Jingwei; Wang, Yunli; Cheng, Gengdong; Zeng, Yong

doi:10.1017/s089006042100007x

Cited by 9 publications

(14 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, it is difficult to guarantee the completeness and balance of obtained data, which will comprise the meshing performance significantly. Pan et al (2021) solve training data problems by self-learning from the meshing process based on an RL algorithm, A2C. The introduced A2C method is only designed to provide training samples rather than offer an independent meshing model.…”

Section: Comparison To Nn-based Methodsmentioning

confidence: 99%

“…Yao et al (2005) improved the FreeMesh approach by introducing an artificial neural network (ANN) to learn the element extraction rules from a set of pre-selected samples of good quality quad meshes. Pan et al (2021) built a self-learning automatic quadrilateral mesh generation system by combining a feedforward neural network (FNN) and an RL algorithm (Advantage Actor-Critic, A2C). The RL is used to provide high-quality samples to train an FNN model that serves as the final mesh generator to mesh various geometry domains.…”

Section: Related Workmentioning

confidence: 99%

“…RL is a well-known self-taught learning paradigm for solving sequential decision making problems and has been successfully demonstrated in many fields. By formulating the mesh generation as a Markov decision process (MDP) (Pan et al, 2021), the meshing policy can be learned by the interactions between the agent and the environment without the need for supervised labels. Compared with most existing benchmark problems and complex applications of RL (Machado et al, 2018;Osband et al, 2019), the mesh generation problem poses new challenges for learning an optimal policy: 1) the environments (i.e., the geometries to mesh) have diverse shapes and sizes, which requires the policy to be adaptable and general; 2) the credit assignment for each time step to achieve the final mesh is difficult since the objective is not intuitive; 3) there is a trade-off between the immediate reward achieved by the current action and the future rewards that depend on the state of the environment caused by the current action, because the shape and size of the current generated element will shape the remaining boundary of the geometry to mesh.…”

Section: Related Workmentioning

confidence: 99%

“…Compared with most existing benchmark problems and complex applications of RL (Machado et al, 2018;Osband et al, 2019), the mesh generation problem poses new challenges for learning an optimal policy: 1) the environments (i.e., the geometries to mesh) have diverse shapes and sizes, which requires the policy to be adaptable and general; 2) the credit assignment for each time step to achieve the final mesh is difficult since the objective is not intuitive; 3) there is a trade-off between the immediate reward achieved by the current action and the future rewards that depend on the state of the environment caused by the current action, because the shape and size of the current generated element will shape the remaining boundary of the geometry to mesh. In the previous work (Pan et al, 2021), these challenges are not completely addressed by the RL algorithm, because its purpose is to generate samples for an FNN model rather than complete the mesh. Therefore, this article redesigns the state representation, action formulation, and reward function to fully automatically address those challenges.…”

Section: Related Workmentioning

confidence: 99%

“…In the previous work, the meshing problem is preliminarily formulated as an RL problem (Pan et al, 2021). The mesh generation problem can be seen as a control problem with a sequence of steps to generate mesh elements.…”

Section: Problem Formulationmentioning

confidence: 99%

See 4 more Smart Citations

Reinforcement learning for automatic quadrilateral mesh generation: a soft actor-critic approach

Huang¹,

Cheng²,

Zeng³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

This paper proposes, implements, and evaluates a Reinforcement Learning (RL) based computational framework for automatic mesh generation. Mesh generation, as one of six basic research directions identified in NASA Vision 2030, is an important area in computational geometry and plays a fundamental role in numerical simulations in the area of finite element analysis (FEA) and computational fluid dynamics (CFD). Existing mesh generation methods suffer from high computational complexity, low mesh quality in complex geometries, and speed limitations. By formulating the mesh generation as a Markov decision process (MDP) problem, we are able to use soft actorcritic, a state-of-the-art RL algorithm, to learn the meshing agent's policy from trials automatically, and achieve a fully automatic mesh generation system without human intervention and any extra clean-up operations, which are typically needed in current commercial software. In our experiments and comparison with a number of representative commercial software, our system demonstrates promising performance with respect to generalizability, robustness, and effectiveness.

show abstract

Section: Comparison To Nn-based Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Problem Formulationmentioning

confidence: 99%

See 3 more Smart Citations

Reinforcement learning for automatic quadrilateral mesh generation: a soft actor-critic approach

Huang¹,

Cheng²,

Zeng³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Reinforcement learning for block decomposition of planar CAD models

DiPrete,

Garimella,

Cardona

et al. 2024

Engineering with Computers

View full text Add to dashboard Cite

The problem of hexahedral mesh generation of general CAD models has vexed researchers for over 3 decades and analysts often spend more than 50% of the design-analysis cycle time decomposing complex models into simpler blocks meshable by existing techniques. The decomposed blocks are required for generating good quality meshes (tilings of quadrilaterals or hexahedra) suitable for numerical simulations of physical systems governed by conservation laws. We present a novel AI-assisted method for decomposing (segmenting) planar CAD (computer-aided design) models into well shaped rectangular blocks. Even though the simple examples presented here can also be meshed using many conventional methods, we believe this work is proof-of-principle of a AI-based decomposition method that can eventually be generalized to complex 2D and 3D CAD models. Our method uses reinforcement learning to train an agent to perform a series of optimal cuts on the CAD model that result in a good quality block decomposition. We show that the agent quickly learns an effective strategy for picking the location and direction of the cuts and maximizing its rewards. This paper is the first successful demonstration of an agent autonomously learning how to perform this block decomposition task effectively, thereby holding the promise of a viable method to automate this challenging process for more complex cases.

show abstract

Unstructured surface mesh smoothing method based on deep reinforcement learning

Wang,

Zhang,

Deng

2023

Comput Mech

View full text Add to dashboard Cite

In numerical simulations such as computational fluid dynamics simulations or finite element analyses, mesh quality affects simulation accuracy directly and significantly. Smoothing is one of the most widely adopted methods to improve unstructured mesh quality in mesh generation practices. Compared with the optimization-based smoothing method, heuristic smoothing methods are efficient but yield lower mesh quality. The balance between smoothing efficiency and mesh quality has been pursued in previous studies. In this paper, we propose a new smoothing method that combines the advantages of the heuristic Laplacian method and the optimization-based method based on the deep reinforcement learning method under the Deep Deterministic Policy Gradient framework. Within the framework, the actor artificial neural network predicts the optimal position of each interior free node with its surrounding ring nodes. At the same time, a critic-network is established and takes the mesh quality as input and outputs the reward of the action taken by the actor-network. Training of the networks will maximize the cumulative long-term reward, which ends up maximizing the mesh quality. Training and validation of the proposed method are presented both on 2-dimensional triangular meshes and 3-dimensional surface meshes, which demonstrates the efficiency and mesh quality of the proposed method. Finally, numerical simulations on perturbed meshes and smoothed meshes are carried out and compared which prove the influence of mesh quality on the simulation accuracy.

show abstract

A self-learning finite element extraction system based on reinforcement learning

Cited by 9 publications

References 56 publications

Reinforcement learning for automatic quadrilateral mesh generation: a soft actor-critic approach

Reinforcement learning for automatic quadrilateral mesh generation: a soft actor-critic approach

Reinforcement learning for block decomposition of planar CAD models

Unstructured surface mesh smoothing method based on deep reinforcement learning

Contact Info

Product

Resources

About