---
license: cc-by-4.0
task_categories:
- tabular-regression
tags:
- engineering
- fatigue-life
- fea
- finite-element-analysis
- surrogate-model
- node-wise-prediction
- 3d-mesh
size_categories:
- 100M<n<1B
language:
- en
pretty_name: FatigueNet
---

# FatigueNet Dataset

## Dataset Description

- **Homepage:** [GitHub Repository](https://github.com/AnK-Accelerated-Komputing/FatigueNet)
- **Repository:** [Hugging Face Dataset](https://huggingface.co/datasets/accelerated-komputing/FatigueNet)
- **Paper:** FATIGUENET: DEEP LEARNING SURROGATE MODEL FOR NODE-WISE FATIGUE LIFE PREDICTION IN 3D STEEL SHAFTS
- **Point of Contact:** sachin@ank-world.com, amit@ank-world.com

### Dataset Summary

The FatigueNet dataset is a comprehensive 3D finite element analysis (FEA) dataset designed for node-wise fatigue life prediction across unstructured 3D meshes of steel shafts. It provides 1,200 unique, automated FEA simulations covering multi-axial load states on complex stepped and dog-bone geometries. 

The primary objective of this dataset is to facilitate the development and benchmarking of machine learning models (such as MLPs, Graph Neural Networks, and Neural Operators like Transolver) for direct spatial field prediction of fatigue life, bypassing computationally expensive traditional FEA fatigue simulations. The dataset spans both high-cycle fatigue (HCF) and low-cycle fatigue (LCF) regimes under fully reversed loading (R = -1).

## Dataset Structure

### Data Instances

The dataset is divided into six compressed HDF5 (`.h5`) files based on the geometry type and loading condition:

- `dogbone_shaft_tensile.h5`
- `dogbone_shaft_torsion.h5`
- `dogbone_shaft_torsion_tensile_combined.h5`
- `step_shaft_tensile.h5`
- `step_shaft_torsion.h5`
- `step_shaft_tensile_torsion_combined.h5`

Each file contains 200 independent simulation cases (totaling 1,200 cases across all files). 

### Data Fields

Inside the `.h5` files, data are stored as key-value groups (e.g., `Group_0`, `Group_1`, ..., `Group_199`). Each group represents one simulation case and contains the following:

- **Nodes:** Node coordinates stored in Cartesian form `(x, y, z)`.
- **Cells:** Tetrahedral connectivity (4-node elements).
- **Geometric Descriptors (Features):**
  - Local curvature
  - Surface normals `(Nx, Ny, Nz)`
  - Surface indicator `(Is_Surface)`
  - Distance-to-step (specific to stepped shaft geometries)
- **Targets:** Node-wise fatigue life targets (number of cycles to failure).

## Dataset Creation

### Source Data

All geometries were generated in ANSYS® SpaceClaim using an automated script linked to Workbench. The design space was sampled using Latin Hypercube Sampling (LHS) to provide efficient coverage of multidimensional parameter ranges.

- **Stepped Shafts:** Varied large diameter (30-42 mm), small diameter (16-26 mm), and fillet radius (1.2-5 mm).
- **Dog-Bone Shafts:** Varied grip diameter (12-24 mm), gauge diameter (8-12 mm), and transition radius (16-72 mm).

### Simulations

Finite-element simulations were performed in ANSYS R2024. The material was modeled as homogeneous isotropic structural steel (Young’s modulus E = 210 GPa, Poisson’s ratio ν = 0.3, density 7850 kg/m³). We simulated:
- Pure axial tension
- Pure torsion
- Combined tension–torsion

## Citation

If you use this dataset in your work, please consider citing the following publication:

```bibtex
@inproceedings{FATIGUENET2026,
  author    = {Sachin Saud and Bipsan Nepal and Bipin Shrestha and Rachit Rijal and Susil Chhetri and Tulsi Narayan Shrestha and Amit Regmi and Akio Tanaka},
  title     = {{FATIGUENET: DEEP LEARNING SURROGATE MODEL FOR NODE-WISE FATIGUE LIFE PREDICTION IN 3D STEEL SHAFTS}},
  booktitle = {Proceedings of the ASME 2026 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC-CIE2026)},
  year      = {2026},
  address   = {Houston, TX, USA},
  paperid   = {DETC2026-193892},
  publisher = {American Society of Mechanical Engineers (ASME)}
}
```