Graph Neural Network Mesh Refinement for FEM Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optimal mesh generation in finite element method (FEM) simulations is a difficult and time-consuming process, affecting accuracy, computational efficiency, and result interpretation, particularly in complex geometries.
Innovation Solution
A computer-implemented method using graph neural networks (GNNs) to generate a refined mesh by predicting simulation errors and adjusting mesh weights based on ground truth errors, allowing for automated and efficient mesh refinement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fine mesh is used to improve simulation accuracy, then the accuracy of results is improved, but the computational time and resources increase significantly
Solution Approach 1:
The patent applies local quality by refining only specific regions of the mesh where errors are detected, rather than uniformly refining the entire mesh. The error estimation process identifies critical areas (such as regions with high stress concentration or complex geometry), and refinement is applied locally to these areas while maintaining coarser resolution in simpler regions. This selective refinement approach achieves high simulation accuracy in critical areas while significantly reducing overall computational time compared to uniform fine meshing.
Solution Approach 2:
The patent implements feedback through an iterative process where the simulation is performed on a coarse mesh, error estimation is conducted to identify regions needing refinement, the mesh is refined based on the error analysis, and the simulation is repeated. This feedback loop continues until the error between coarse and refined mesh solutions falls below a specified tolerance threshold. The feedback mechanism ensures that computational resources are allocated efficiently by focusing refinement efforts only where necessary to achieve the desired accuracy.
2Manufacturing precision
If manual mesh refinement is performed to achieve optimal mesh quality, then the mesh quality is improved, but the process becomes time-consuming and complex
Solution Approach 1:
The patent applies self-service by implementing an automated error estimation and mesh refinement system that performs refinement without requiring manual intervention. The system automatically computes errors between coarse and fine mesh solutions, identifies regions needing refinement, and generates the refined mesh through algorithmic processes. This automation eliminates the time-consuming manual refinement process while maintaining high mesh quality, allowing the system to self-optimize the mesh based on objective error criteria.
Solution Approach 2:
The patent utilizes parameter changes by dynamically adjusting mesh refinement parameters based on error analysis results. The refinement process modifies parameters such as element size, node distribution, and mesh density in specific regions according to the estimated errors. By changing these parameters adaptively rather than using fixed predetermined patterns, the system achieves optimal mesh quality tailored to the specific problem geometry and physics, significantly reducing the time required compared to manual parameter adjustment.
3Productivity
If a coarse mesh is used to reduce computational cost, then computational efficiency is improved, but the accuracy of stress and strain calculations deteriorates
Solution Approach 1:
The patent applies local quality by maintaining coarse mesh resolution in regions where stress and strain variations are low and computational efficiency is prioritized, while applying fine mesh resolution in critical regions where high accuracy is required. The error estimation process identifies these critical regions based on the physics of the problem (such as areas with high stress gradients or complex boundary conditions), and the mesh is refined locally only in these areas. This selective approach preserves computational efficiency in simple regions while ensuring accurate stress and strain calculations in critical areas.
Solution Approach 2:
The patent implements preliminary action by performing error estimation calculations on the coarse mesh before finalizing the refined mesh. This preliminary error analysis identifies the specific regions and magnitude of refinement needed, allowing the mesh refinement process to be targeted and efficient. By preparing the refinement strategy in advance based on coarse mesh error analysis, the system avoids unnecessary refinement in low-error regions and focuses computational resources only where accuracy is critical, maintaining high productivity while ensuring measurement precision.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method comprising: generating a main graph based on a coarse mesh, the coarse mesh comprising a plurality of mesh elements arranged to fill an area or volume defined by a geometry; generating a vertex graph based on the geometry, wherein the vertex graph comprises vertex graph nodes corresponding to vertices of the area or volume defined by the geometry and vertex graph edges corresponding to links between the vertices in the geometry; generating, using a first graph neural network (GNN), an embedding of the vertex graph; generating, using a second GNN and based on the main graph and the embedding of the vertex graph, a prediction indicative of a refinement of the coarse mesh for generating a refined mesh.