Kit-CAE: A Bridge Connecting CAE Results to Omniverse

Overview

CAE analysis results have traditionally been viewable only through dedicated viewers such as ParaView or Ansys post-processing tools.

However, NVIDIA provides Kit-CAE to bring these results into Omniverse, enabling real-time visualization, collaboration, and AI integration.

https://developer.nvidia.com/blog/how-to-run-ai-powered-cae-simulations/

https://github.com/NVIDIA-Omniverse/kit-cae/tree/main

Why It’s Needed

• Analysis data are complex and heavy, but visualization is usually limited to single-user/local analysis.
• In industrial applications, design–analysis–simulation–operation must be tightly integrated.
• Traditional tools alone cannot easily support real-time collaboration or AI/digital-twin connectivity.

Purpose

• Directly import and visualize CAE result data (.cgns / .vtk (.vti) / .npz) in Omniverse.
• Standardize results into USD format for reuse and collaboration within the same scenes as Isaac Sim or USD Composer.

• AI Pipeline Integration: Physics NeMo (training) → NIM (API inference) → instant reflection in the scene.
• Real-Time Digital Twin: Instant prediction and visualization based on robot or equipment state changes.
• Large-Scale Data Optimization: High-quality, real-time rendering powered by IndeX/NanoVDB.

1) What Is This Tool For?

• Load and visualize CAE analysis results (.cgns / .vtk / .npz) directly in Omniverse.
• Perform basic post-processing such as External Faces, Streamlines, Volume/Slice.
• Integrate seamlessly with USD-based applications such as Isaac Sim and USD Composer.

Dynamic Streamline / Flow Visualization in Kit-CAE

• Loads existing CAE result data (velocity, pressure fields, etc.) and renders them visually in real time.
• “Real-time” here means not recomputing physics, but generating streamlines or particle animations on a given vector field (e.g., flow velocity).
• Thus, it serves not as a CAE solver, but as a post-processing and interactive visualization tool.

GPU Acceleration

• Streamlines, voxelization, and flow visualization are computation-heavy; GPU-accelerated libraries are essential.
• Kit-CAE is implemented as an Omniverse Kit extension and optionally integrates NVIDIA’s Warp library:
  • Example: Warp-based Streamlines (faster but supports limited element types)
  • Example: GaussianWarp Voxelization (GPU-parallel voxelization using a Gaussian kernel)

Role of Warp

• Warp is NVIDIA’s CUDA-based GPU physics-computation framework for Omniverse.
• Performs numerical operations such as streamline integration or point-cloud → volume voxelization on GPU in parallel instead of CPU.
• Hence, Kit-CAE provides Warp-optimized versions for Streamlines, Flow, and Voxelization algorithms.

2) Differences from ParaView

• USD-based collaboration: multiple teams can access the same scene simultaneously.
• Digital-twin integration: overlay analysis results on robot or equipment simulations.
• AI pipeline connection: Physics NeMo → NIM API → Kit-CAE visualization.
• Python automation: supports scripted/repetitive workflows.

ParaView = Post-processing viewer

Kit-CAE = Extension platform that bridges the Omniverse ecosystem

3) Usage Patterns

• CAE Data – Simulation results from tools such as Ansys, Star-CCM+, or OpenFOAM.
• Curator – Converts data into AI-training-ready formats.
• Physics NeMo – Trains AI surrogate models from converted data.
• NIM – Deploys trained models as APIs for external access.
• Omniverse – Uses Kit-CAE to fetch API results for real-time 3D visualization within digital-twin environments.

A. Comparison & Analysis Mode (for Researchers / Analysts)

• Export analysis result files (.cgns, .vtk, .npz) from existing CAE tools (Ansys, Star-CCM+, OpenFOAM etc.).
• Import and visualize them in Kit-CAE.
• Retrieve inference results from AI surrogate models deployed via NIM API calls.
• Compare traditional analysis and AI prediction results side-by-side in the same Omniverse scene.

Context

• Researchers and developers can intuitively compare actual analysis data against AI predictions.
• Enables rapid design evaluation and model validation without repetitive experiments.

B. Real-Time Digital-Twin Mode (for Operations / Simulation)

• rain a Physics NeMo surrogate model on existing analysis data and deploy it through NIM API.
• The AI model immediately returns physical analysis outputs (flow fields, stress, temperature distribution, etc.).
• In apps like Isaac Sim, real-time state data (velocity, position, temperature, etc.) from robots or equipment are sent to the NIM API.
• Kit-CAE visualizes these results on the USD scene, providing instant feedback.

Context

• Manufacturing digital twin – Monitor thermal stress and deformation in real time during operation.
• Logistics simulation – Predict and display airflow or temperature changes instantly as AMRs move.
• Robotics & AI training – Reinforcement learning within simulation including a physical feedback loop.

Flow NanoVDB

concept_car.npz

The Kit-CAE examples (e.g., Streamlines, Volume Rendering, etc.) visualize existing CAE analysis data in real time, producing an animation that appears as dynamically moving fluid.

Dynamic Visualization (Without AI Computation)

• Streamline and Flow visualizations are based on existing analysis data (e.g., velocity vector fields) and update paths in real time.
• For instance, when you move the seed sphere, streamlines update instantly to make the fluid flow appear animated.

However, this is not a simulation run (AI or physics computation) but a purely visual update.

Difference from AI-Based Dynamic Prediction