HaomingLuo/AgentFEM-Layered-Thermoelastic-Bending-2D

AgentFEM Layered Thermoelastic Bending 2D

This dataset contains 512 verified two-dimensional finite-element simulations of perfectly bonded bilayer and trilayer strips under a uniform positive temperature increment. It supports scalar surrogate modeling, mesh-based field learning, scientific-ML benchmarking and reproducible CAE workflow studies.

Dataset summary

• 256 bilayers and 256 trilayers.
• 256 forward and 256 reversed stack orders.
• Deterministic, family-stratified splits: 384 train, 64 validation and 64 test.
• Eight HDF5 shards with 64 independent cases per shard.
• P2 displacement degrees of freedom, vertex displacement, material IDs, and unsmoothed material-aware DG0 thermoelastic stress and von Mises stress.
• SI units throughout.

The physics assumptions are isotropic linear elasticity, plane stress, small strain, perfect bonding and a uniform temperature increment. The reported stress is C:(epsilon-alpha*DeltaT*I).

Verification

All 512 cases passed combined absolute/relative checks against an independent axial A-B-D layered-beam reference for central curvature and layer-average axial stress. The maximum curvature relative error for non-negligible reference values was 3.570%, the median was 0.598%, and the maximum layer-average stress relative error was 3.404%. Four worst-case family/orientation combinations passed coarse/base/fine mesh audits; the maximum base-to-fine curvature change was 0.002522%. A homogeneous free-expansion test produced 0.127611 Pa maximum residual DG0 Mises stress. Full evidence is in artifacts/t1_layered_thermoelastic_v1/QUALITY_REPORT.md.

The frozen dataset was generated with AgentFEM commit 159bfba491e3b72d7ed3a8ce86222c7890772e78. After the thermoelastic API upgrade, the project was independently revalidated against AgentFEM commit 8fe642a: 149 relevant framework, workflow, dataset and cross-version tests passed. Two representative cases were rebuilt with the new native thermal eigenstrain API and agreed with the frozen fields and scalar responses.

Files and loading

The canonical sample index is data/t1_layered_thermoelastic_v1/index.jsonl; shard paths and SHA-256 digests are recorded in data/t1_layered_thermoelastic_v1/manifest.json.

from src.load_t1_layered_thermoelastic import load_sample

sample = load_sample("data/t1_layered_thermoelastic_v1", 0)
print(sample["parameters"])
print(sample["fields"]["vertex_displacement_m"].shape)

The lightweight reader requires only NumPy and h5py. Reproduction code and the frozen configuration are included under src/ and configs/.

Limitations

All material parameters are synthetic. The dataset is not experimentally calibrated and does not represent delamination, contact, plasticity, cure shrinkage, transient heat conduction or three-dimensional free-edge effects. The analytical reference validates central curvature and layer-average axial stress, not local end stresses. DG0 peak stress is mesh- and representation-dependent.

License

Dataset contents are released under CC BY 4.0. Included source code is released under Apache-2.0; see DATA_LICENSE.md and CODE_LICENSE.