Extend atomic knowledge
Learned potentials and energy models carry atomic information into larger-scale computation. NN-MGSF with DFD shows how layer interactions shape collective defect evolution.
NN-MGSF / DFD →Materials Physics AI Platform
Physics AI. Connected to materials R&D.
Bring high-fidelity computing, material knowledge and learned prediction into a workflow for material design, virtual testing and faster simulation.
Many scales. One material story.
Equivalent material parameters describe a response. Atomic interactions, defects and microstructure explain how that response changes. We connect these scales to bring composition and processing into material design.
01 / Generate data
Start with the physics. Build the data.
From atomic energies to evolving microstructures, numerical methods reveal the behavior that a model needs to learn. GPU parallelism supports repeated simulations and parameter exploration.
Material systemInitial state & geometryConditions & boundaries
DFT / MDConfigurations, energies & forces
DFD / phase fieldDefects, interfaces & trajectories
FEMDisplacement, stress & response
Parallel computation & parameter sweepsGPU-accelerated numerical computing
Physics data & material states
Turn material knowledge into useful data.
Choose the right scale: DFT and MD for atomic behavior, DFD and phase field for defects and structure, FEM for mechanical response.
Use computation to explore more material states and build the foundation for reusable prediction.
Explore parallel numerical computation →02 / Train models
Learn from data. Learn from physics.
Build models around the prediction you need. Use simulation and experimental data, physical laws, or both to learn material energies, fields and evolving states.
Material parameters, geometry, initial state & boundaries
Simulation & experimental dataEnergy / equations / boundary constraints
Energy model / field & evolution operator
Make evolving states predictable
Learn reusable responses as material distributions, interfaces and initial states change. Choose the model around the structure and behavior you want to explore.
PDCO / MPTT / evolution surrogates →03 / Validate
Bring speed and confidence into the workflow.
Compare predictions with high-fidelity results and physical behavior, then assess the benefit in the computation that matters to your team.
Compare material responses
Follow changes in geometry, material state and loading.
Keep the material physics
Connect local fields, interfaces and evolving states.
Measure the R&D benefit
Put prediction quality and total computation side by side.
04 / Predict & solve
Predict directly. Accelerate the solver.
Use a surrogate to explore material responses directly, or bring its predicted fields into FEM. Numerical correction connects learned prediction with established engineering computation.
Compare more material options.
Query fields and changing states across related material problems, making repeated exploration easier.
Explore direct void-evolution prediction →Give FEM a better starting point.
Start from a predicted field and retain the solver’s correction process and accuracy requirements.
Explore MPTT + FEM correction →- SurrogatePredict an initial guess
- FEMNumerically correct the field
- Material responseContinue the computation
Bring AI into an existing solver workflow.
MPTT predicts the elastic field, FEM corrects it, and the evolving void continues to change. Learning and numerical physics work together in the same loop.
Explore MPTT with FEM →
Keep building the material knowledge behind the next prediction.
Let prediction answer material design questions.
Connect the response you want to the composition, defects and processing choices you can change. Build material-specific workflows around strength, deformation, damage and fatigue objectives.
Composition & processing→Defect & microstructure evolution→Properties & candidate comparison
What should your next material do?
Start with the material, the performance target and the decision you need to make.