Everything that is manufactured is first simulated with advanced physics solvers. Real-time digital twins (RTDTs) are the cutting edge of computer-aided engineering (CAE) simulation, because they enable immediate feedback in the engineering design loop. They empower engineers to innovate freely and rapidly explore new designs by experiencing in real time the effects of any change in the simulation. RTDTs are in high demand in aerospace, automotive, electronic design, and indeed the whole of the manufacturing industry.
But developing RTDTs can be challenging when starting from scratch. Digital twins of physical systems (planes, cars, and ships, for example) integrate diverse collections of software tools—a different tool for each part of the simulation—into a unified workflow. At a minimum, advanced physics solvers must be coupled with design tools and realistic visualization. Real-time performance is only possible with near-zero latency and high-performance integration between all the parts of the simulation workflow.
NVIDIA Omniverse Blueprints for physics digital twins help independent software vendors (ISVs) overcome this engineering challenge. These blueprints are reference workflows that include NVIDIA acceleration libraries; physics-AI frameworks; and interactive, physically based visualization. Industry-leading software developers like Ansys, Cadence, Siemens, and others can use the blueprints to develop CAE software tools that are orders of magnitude faster, enabling real-time visualization and analysis of products as they develop. This helps their customers drive down development costs and energy usage while achieving faster time to market.
This post describes how Luminary Cloud, a member of the NVIDIA Inception program for startups, instantiated the Omniverse Blueprint for real-time computer-aided engineering digital twin with their cloud-native, GPU-accelerated solver to realize a real-time virtual wind tunnel. This is one of the first applications of NVIDIA Omniverse Blueprints, and it illustrates how blueprints simplify creating real-time digital twins for computational fluid dynamics (CFD) simulation to advance the designs of cars, airplanes, ships, and many other products.
Building with NVIDIA Blueprints
NVIDIA Blueprints are comprehensive reference workflows built with NVIDIA AI and Omniverse libraries, SDKs, and microservices. Each blueprint includes reference code, deployment tools, customization guides, and a reference architecture, speeding up deployment of AI solutions like AI agents and digital twins, from prototype to production. A key feature of blueprints is that every software component in the blueprint can be modified by a developer or replaced by a third-party solution with similar functionality.
The NVIDIA Omniverse Blueprint for real-time computer-aided engineering digital twins is the first blueprint for CAE workflows. It connects a 3D scene rendered with Omniverse APIs with a simulation AI to create a fully interactive CAE workflow. The AI has been minimally trained for a few automotive geometries to demonstrate how a more capable AI may be created. The tools and frameworks for training such an AI are bundled with the blueprint so developers can easily import proprietary data. This means that the blueprint is a complete software workflow, but it must be integrated with commercial-grade software solutions to realize a fully capable, real-time digital twin. A CFD solver, or an AI surrogate model for CFD, or both, are key parts of a CAE solution built with this blueprint.
Luminary Cloud combined the blueprint with their modern CFD SaaS platform to realize an interactive virtual wind tunnel (Figure 1). Luminary’s cloud-native design with full API integration enables immediate access to powerful CFD simulation tools, making it an ideal fit for this blueprint. The blueprint is provided as a Helm Chart defining a Kubernetes application that integrates all parts of the real-time digital twin.
Training AI for simulation with NVIDIA Modulus
The blueprint uses an AI surrogate model for real-time inference predictions of fluid flow. The surrogate is trained and deployed with the NVIDIA Modulus physics-ML framework. Using Modulus, CAE ISVs can train new surrogate models on ISV application training data. Surrogate models can be trained from scratch, or a foundation model can be used as a starting point to reduce training time. ISVs building on the blueprint can also replace the Modulus-based AI surrogate model with their own AI models.
Luminary Cloud generated the AI training dataset using their compressible flow RANS solver for five base geometries (two sedans, a truck, an SUV, and a sports car) with procedurally generated geometric variations created by modifying parameters such as the ride heights. The geometries also varied their basic features like mirrors, spoilers, and wheel rims. In total, 192 simulations were performed for the different geometry variations as well as inflow speeds, which varied randomly between 33 mph and 11 mph for each configuration.
After performing the CFD simulations with Luminary Cloud, NVIDIA Modulus was used to train an AI on 167 simulation fields. The model was validated on the remaining 25 simulation fields, producing a surrogate model for external aerodynamics. When presented with a surface geometry and wind speed, the surrogate model returns volume and surface flow fields, such as velocity, pressure and wall shear stresses. The results are stored in system memory, not written to disk, so they can be immediately rendered with Omniverse APIs.
Create worlds with Omniverse APIs
The blueprint includes a world state controller built with NVIDIA Omniverse APIs. This software component connects to the simulation AI, maintains the application state, and provides an Omniverse Streaming endpoint to connect the digital twin to the front-end application of the ISV. This enables users to view and interact with the wind tunnel digital twin as part of an ISV’s commercial software suite.
The virtual wind tunnel’s world state controller is built with the Omniverse Kit SDK and combines Omniverse Flow with RTX Scientific. Flow is a Eulerian fluid simulation for smoke and fire, leveraging a sparse voxel grid for an unbounded simulation domain. It supports point clouds as an input, including from Modulus, to produce velocity and pressure NanoVDB. Flow advects smoke using the velocity NanoVDB. NanoVDB is used for high performance data exchange with RTX Scientific and NVIDIA Warp.
RTX Scientific makes use of NVIDIA IndeX to enable interactive visualization and exploration of large volumetric datasets in a collaborative setting. It can scale for a wide range of GPU configurations: from a single GPU, to multiple GPUs, to multinode GPU clusters for scalable real-time visualization and accelerated compute of multivalued volumetric data and embedded geometry data. IndeX provides a range of tools designed for interactive and collaborative analysis. Team members can, for example, change color maps with ease to highlight subtle attributes of the data, view cross-sections across the entire time series, and leverage features like ambient occlusion and shadows to examine key components of the data.
Deploy to the cloud or on-premises
Instantiating the blueprint with Luminary Cloud’s cloud-native CFD solver produces a pair of containers—one for the AI, and one for the world state controller. The containers are configured with a Helm Chart that defines, installs, and maintains the end-to-end application on-premises or in a cloud-native environment. The Omniverse Blueprint can be deployed on all major cloud platforms, including NVIDIA DGX Cloud, Amazon Web Services, Google Cloud Platform, Microsoft Azure, and Oracle Cloud Infrastructure.
Get started
Beyond Luminary Cloud, NVIDIA is also partnering with Rescale to incorporate the blueprint into their physics-AI platform, enabling real-time digital twins for industry software developers. The NVIDIA Omniverse Blueprint for real-time computer-aided engineering digital twin is now available for all software providers to access and adapt to their workflows on-premises or on-cloud. Companies interested in learning more about the Omniverse Blueprint for real-time computer-aided engineering can also sign up for early access.