Team Ai
Datasetpublic

HaomingLuo/AgentFEM-Material-Loading-Memory

AgentFEM Material Loading Memory An open, reproducible research dataset for path-dependent material modeling, neural constitutive surrogates and finite-element deployment tests. The repository contains six staged, explicitly separated releases. Start here Goal Recommended entry Understand the current dataset This page and the sealed-test data card Train or benchmark a constitutive model DENIM start guide Reproduce the multiaxial baseline study… See the full description on the dataset page: https://huggingface.co/datasets/HaomingLuo/AgentFEM-Material-Loading-Memory.

sourceHugging Facecc-by-4.0updated 15d agoView on Hugging Face
1likes1.5kdownloads
T2_DATA_MODEL_HANDOFF.md92 linesDownload Raw Back to docs
1# AgentFEM material-memory data and DENIM model handoff2 3## Public products4 5- Dataset: `HaomingLuo/AgentFEM-Material-Loading-Memory`6- Model: `HaomingLuo/AgentFEM-DENIM`7 8The dataset contains 2,660 complete material histories. A sample is always one9ordered trajectory; time points are not counted as independent samples.10 11| Release | Trajectories | Primary use |12|---|---:|---|13| T2 loading-memory v1 | 1,008 | loading-history baseline and proportional cyclic response |14| T2 multiaxial OOD v2 | 1,024 | ID, unseen-path and parameter-OOD comparison |15| DENIM closure v1 | 128 | incomplete internal-state closure |16| DENIM boundary v1 | 500 | path, amplitude, long-history and resolution boundaries |17 18These releases form one evidence chain and retain separate protocols. Pooling19all 2,660 trajectories as exchangeable supervised samples would introduce20conflicting constitutive targets and invalidate the frozen tests.21 22## Model product23 24The recommended checkpoint is `denim-expanded`, version `1.1.0`. It has 91825trainable parameters and retains a small-strain J2 skeleton with two learned26memory channels. The reference material uses three memory channels and hidden27tabulated hardening, so the task measures closure under incomplete material28state and evolution knowledge.29 30Published stress RMSE values are:31 32| Evaluation | DENIM expanded | Incomplete J2 | GRU |33|---|---:|---:|---:|34| Frozen held-out paths | 0.714 MPa | 59.541 MPa | 98.260 MPa |35| New path OOD | 0.743 MPa | 58.177 MPa | 96.959 MPa |36| Amplitude OOD | 2.294 MPa | 69.315 MPa | 95.896 MPa |37| Long history | 10.906 MPa | 54.394 MPa | 90.757 MPa |38 39Long-history state drift is the clearest current boundary. It should remain in40future comparisons rather than being replaced by additional ordinary-path41tests.42 43## Runtime package44 45The model repository includes an `agentfem_bundle/` directory containing:46 47- `model.json`: immutable architecture, state, parameter and applicability contract;48- `weights.safetensors`: runtime weights without pickle deserialization;49- `SHA256SUMS`: authenticated bundle files;50- `README.md`: offline-use notes.51 52The bundle is tied to exact dataset and source-model revisions. AgentFEM owns53the global Newton process, material-state commit/rollback, checkpointing and54result evidence; AgentFEM-learning owns the PyTorch provider and DENIM adapter.55 56## Reproducible evidence57 58The current package has four evidence levels:59 601. dataset integrity and constitutive quality gates;612. material-point equivalence between legacy and safe bundles;623. path, amplitude, long-history and discretization boundary metrics;634. serial and two-rank AgentFEM implicit execution.64 65The plastic automatic-differentiation tangent is currently approximate to66about 0.05–0.80% against fixed-old-state finite differences over the audited67states. The demonstrated global cases converge, but an exact consistent68plastic tangent remains an open numerical improvement.69 70## Material for a university research team71 72The public assets already provide the experimental matrix needed to organize73a computational study:74 75- fixed data splits and model identities;76- black-box, weak-physics, white-box and incomplete-physics comparisons;77- architecture and parameter-count evidence;78- OOD and long-history capability boundaries;79- energy, yield, incompressibility and state diagnostics;80- finite-element deployment evidence and a documented tangent limitation;81- complete generation, training and validation code.82 83The main method proposition available for further theoretical analysis is:84 85> Learn a transferable, incrementally integrable material-memory closure when86> the evolution law and internal-state description are incomplete.87 88Further academic work may analyze identifiability, reduced supervision,89implicit differentiation, error propagation and constitutive stability. The90published claims remain limited to the synthetic fixed-material protocols and91the verified software/runtime versions recorded above.92