Create and Manage Datasets
Datasets are collections of one or more simulation results used for training or testing a model. The following results files are supported for PhysicsAI:
Table 1. List of Supported Files
| Supported Solvers | File Format |
|---|---|
| Abaqus | .odb |
| AcuSolve | .ensight, .h3d |
| ANSYS | .rst, .rth |
| CCM+ | .ensight |
| Custom/user | .ensight, .unv, .h3d |
| Fluent | .ensight |
| LS-DYNA | .d3plot, .d3eigv |
| Marc | .t16 |
| Nastran | .op2, .h5, .xdb |
| PAM-CRASH | .dsy |
| OptiStruct | .h3d, .op2 |
| Radioss | .h3d, .anim |
| ultraFluidX | .ensight |
Understanding your setup options
When creating datasets, Physics AI can work with a broad set of inputs (geometric and non-geometric such as thickness, materials, boundary conditions, loads) and outputs (full 3D field results, scalar and vector KPIs, and curves).
Your dataset setup depends on:
- What you want to predict (field results, KPIs, or curves)
- What varies between your simulation cases (geometry only, or geometry plus non-geometric parameters)
The following sections explain which files are needed for each scenario.
Base input - geometry (required)
Source: Result files (h3d, .odb, etc.) OR Input decks (.fem, .inp, etc.)
Physics AI automatically extracts mesh geometry including node coordinates, mesh connectivity, and part labels.
Outputs (What you want to predict)
- Field results (Contours) → from result files
- KPIs (Scalar values) → from *.json file
- Curves (1D arrays) → from *.json file
Additional inputs (Non-geometric parameters - Optional, only if these vary between data)
- Global Hook (*_inp.json): Model-level inputs applied to whole model
- Nodal Hook (*.csv): Node-specific inputs applied per node
Setup table
Use this table to determine which files you need:
| What You Want to Predict (Outputs) | What Varies (Input) | Files Needed |
|---|---|---|
| Field results only | Geometry only | Result file |
| Field results + KPIs | Geometry only | Result file + *.json |
| KPIs or curves only | Geometry only | Input deck OR Result file + *.json |
| Field results | Geometry + model-level inputs (Global hooks) | Results file + *_inp.json |
| Field results | Geometry + node-specific inputs (Nodal hooks) | Results file + *.csv |
| Field results + KPIs | Geometry + model-level + node-specific inputs | Result file + *_inp.json +.csv + *.json |
File naming rule
All files for one simulation case must share the same base name.
Example for case
"model_001":
- model_001.h3d = Result file (contains geometry + field results)
- model_001.fem = Input deck (contains geometry)
- model_001.json = KPIs or curves you want to predict (outputs)
- model_001_inp.json = Global Hook inputs - additional inputs applied to whole model
- model_001_nodal.csv = Nodal Hook inputs - additional inputs applied node-specific
It is important to carefully curate the datasets to avoid noise being introduced into the model due to the presence of outliers. This can be done after a dataset has been created.
Learn more about the data compatibility of PhysicsAI.