Direct prediction
Map materials, geometry and state directly to fields or evolving states for repeated exploration of related problems.
Explore phase-field prediction & speed →MPTT / Mesh-based field prediction
Predict the response inside complex materials.
Interfaces, mesh transfer and scale
Interfaces and internal geometry change how a material carries load. MPTT predicts displacement, stress and evolving states across meshes, supporting direct prediction and FEM-assisted computation.
Complex interfaces. Clear local responses.
Material interfaces redirect load and concentrate stress. Explore the predicted fields as the internal structure changes.
Explore the selected field
Change the mesh. Reuse the trained model.
Reuse the trained model across mesh densities. The void study spans roughly ten thousand to three hundred thousand nodes.
Explore the selected field
Predict directly. Or assist an FEM solver.
One route supports fast queries; the other fits into an established numerical solver. Choose how to use the learned field according to the accuracy required by the task.
An accelerator for FEM
Initialize FEM with a predicted field, then use numerical correction to retain solver accuracy and connect learned prediction to the engineering workflow.
Explore prediction with correction →3D phase fields: predict evolving interfaces.
≈140×
Faster direct evolution prediction
64-step Allen–Cahn prediction: A100 surrogate versus single-core CPU FEM.
Explore the selected field
2D void evolution: bring prediction into FEM.
Let the surrogate and numerical solver work together as the cavity evolves. MPTT predicts the elastic state; FEM corrects it before the next evolution step.
From material meshes to millions of volume points.
Extend prediction from material meshes to large three-dimensional geometries. An Ahmed case reaches about 27.48 million volume points; surface-pressure fields reveal the response across complex shapes.
LowerCpHigher
LowerCpHigher
Explore the selected field