Mesh-gen-train / README.md
Iris314's picture
Update README.md
72b8d22 verified
metadata
license: mit
task_categories:
  - tabular-classification
  - tabular-regression
language:
  - code
tags:
  - finite-element-analysis
  - meshing
  - structural-engineering
  - cae
  - simulation
  - geometry
  - stress-analysis
  - adaptive-mesh-refinement
  - engineering-education
  - ansys
  - python
pretty_name: MeshRefine-FEA Corner Bracket Dataset
size_categories:
  - 10k<n<100k

MeshRefine-FEA Corner Bracket Dataset

Overview

This dataset contains simulation-driven mesh refinement information for right-angle corner bracket geometries under various loading conditions. All simulation data was generated using the ANSYS Mechanical Python API, applying up to five localized mesh refinement iterations per case based on stress concentration analysis.

The dataset supports research in:

  • Feature-aware mesh refinement
  • Convergence prediction
  • Intelligent simulation-preprocessing assistants
  • Lightweight agentic decision-making for FEA

Dataset Structure

feature_level_augmented.csv

Each record corresponds to a single CAD face and includes:

  • Geometric category: {hole, plane, fillet}
  • Spatial information: center (cx, cy, cz), normal (nx, ny, nz)
  • Area of the face
  • Load relation: {load, fixed, free}
  • Distance from load application region
  • Final local mesh size
  • Binary refinement label

Primary use: feature-level refinement recommendation & local sizing


part_level.csv

Each record represents a full simulation case and includes:

  • Total and per-type face counts
  • Load type, direction, and magnitude
  • Final global mesh size
  • Convergence status (after ≤5 iterations)

Primary use: global mesh sizing & convergence prediction


Data Generation Workflow

  • CAD Source: Public STEP models from McMaster-Carr (corner brackets)
  • Initial mesh based on bounding box + thickness metrics
  • Stress-driven refinement:
    • Local size reduction: 0.75–0.85× per iteration
    • Convergence rule: <15% change in max von Mises stress
  • Marked as singularity if not converged after 5 iterations

All simulations were scripted using ansys.mechanical.core to ensure consistency and automation.


Target ML Tasks

This dataset enables supervised learning for 4 predictive objectives:

Task Level Type
Region refinement decision Feature Binary classification
Local mesh sizing Feature Regression
Convergence prediction Part Binary classification
Global mesh sizing Part Regression

These tasks reflect the common failure points for early-career FEA users.


Dataset Statistics

Property Value
# of CAD parts 25
# of simulation cases 7,500
Max refinement iterations 5
Domain Industrial brackets

Example Use Cases

  • Predict stress-driven refinement regions
  • Prevent non-convergent simulation setup
  • Automatically adjust mesh density for faster solvers
  • Provide explainable refinement guidance for students

License

  • Based on publicly accessible mechanical CAD
  • Free for research and educational use
  • Please cite this dataset in derivative work

Acknowledgements

Created to support intelligent mesh refinement for beginner FEA simulation tools.


Citation

@dataset{meshrefine_fea_2025,
title={MeshRefine-FEA Corner Bracket Dataset},
author={X. Tang et al.},
year={2025},
note={Finite Element mesh refinement dataset for feature-aware learning}
}