Developing and Operating Large-Scale, Connected Digital Twins

AI, simulation, and customization are just some of the key components of enterprise digital twins.

PNY Technologies has submitted this article.

Rendering of BMW’s digital twin factory. (Source: NVIDIA.)

Rendering of BMW’s digital twin factory. (Source: NVIDIA.)

Digital twins will transform enterprises and all industries over the next decade, as they embrace the possibilities of the metaverse and increasingly harness simulation to develop new products and services, enhance customer experiences and relationships, while unlocking speed, agility, and operational efficiencies.

As the 3D internet becomes pervasive and indispensable, designers, planners, and operators within enterprises need new computing architectures, standards like Universal Scene Description (USD), and workflow solutions that enable them to not only digitally design products, processes, and facilities in 3D but also continuously simulate, operate, and optimize them.

Digital twins are key to these metaverse aspirations and organizations of all varieties are already using them for transformative industrial and scientific discovery applications.

We’ve previously discussed how the use of artificial intelligence (AI) and digital twins can change work patterns. Now we’ll explore how organizations can address the needs of creating and operating digital twins for industrial use cases.

What is a Digital Twin?

Digital twins are connected models and simulations of industrial assets, processes, or environments. Accurate and trustworthy virtual testing, experimentation, and optimization is possible only if a digital twin simulation implements specific attributes.

Digital twins operate in real-time and are physically accurate with true-to-reality physics, materials, lighting, rendering, and behavior. To deliver these facilities digital twins provide a representation using a single source of truth for virtual datasets. A digital twin must be a physically accurate replica that strictly and completely obeys the laws of physics. In addition, the digital twin must be perfectly synchronized, and precision timed to the real world.

What Organizations Need to Create and Operate Digital Twins

Building a digital twin is a multi-step journey. Regardless of your industry, use case, or organizational capabilities, the creation of large-scale digital twins can be broken down into these important components:

Must be AI-Enabled and AI-Enabling: AI optimization and AI training of digital twins helps deliver levels of inferencing accuracy to match the physics and synchronization criteria. Utilizing AI-enabled and AI-enabling terminology aids in creating intelligent equipment, embedded with advanced perception, reasoning, and recommendation capabilities so the equipment can engage with our physical world and make recommendations and autonomous decisions based on the laws of physics.

Full Fidelity Visualization and Custom 3D Tools: Creating full fidelity visualization requires the aggregation of full-design-fidelity 3D datasets, materials, lighting, and visualization in real-time. As part of the process, data pipelines and workflows are established, then 3D models and datasets are ingested, aggregated, and normalized using Universal Scene Description (USD) while retaining a likeness to the source data. In addition, 3D models, data, and workflows can be customized and optimized. Materials, lighting, textures, and rendering are added to 3D models to enable full-design-fidelity visualization. The 3D models act as a single source of truth and are continuously shared with, collaborated on, and refined by contributors and key stakeholders in real-time.

Physics Simulation: Physics simulations are applied to 3D models and environments. For example, a physics simulation might involve the simulation of a sensors where results from domain specific solvers are visualized and brought to life. Using AI, physics can be predicted in real-time with high accuracy and fidelity.

Connect and Train Autonomous Systems: As part of the training process, robotics models are imported and synthetic data pipelines are created. The digital twin and synthetic data are leveraged to train, test, and optimize autonomous systems, including robots and perception systems, for scenarios that are impractical or impossible in the real world.

Connect to the Real World: Digital twins are connected to real world systems and live data streams via Internet of Things (IoT) sensors and programmable logic controllers (PLCs).

Meeting Digital Twin Needs

One way some organizations are implementing digital twins is with NVIDIA Omniverse Enterprise, a scalable, end-to-end platform enabling businesses to build and operate metaverse applications.

The Omniverse ecosystem. (Source: PNY.)

The Omniverse ecosystem. (Source: PNY.)

Based on USD, Omniverse Enterprise allows teams to connect and customize complex 3D pipelines and operate large scale, physically accurate virtual worlds. With USD, enterprise teams can aggregate, visualize, simulate, and collaborate on full-design-fidelity datasets from hundreds of 3D applications.

Omniverse Enterprise is extensible and customizable, allowing developers and technical designers to build advanced, AI-enabled tools to connect or accelerate their existing 3D pipelines with Python script.

Based on NVIDIA PhysX, MDL (Materials Definition Library), and RTX, NVIDIA Omniverse Enterprise lets teams scale physically accurate visualization and simulation from workstations to the data center. In addition, Omniverse is fully supported to minimize system downtime and ensure mission-critical projects keep on track with NVIDIA’s enterprise-level support.

How BMW Uses Digital Twins

BMW produces 2.5 million cars every year, and customers tailor 99 percent of those cars before purchase. NVIDIA and BMW collaborated to showcase how digital twins and virtual worlds will drive the future of manufacturing. BMW uses Omniverse Enterprise to simulate many aspects of an entire factory in real time and produce a digital twin or virtual model designed to accurately reflect a physical object. BMW expects to gain as much as 30 percent efficiency improvements using the real-time capabilities and virtual collaboration features of Omniverse.

BMW uses NVIDIA Isaac Sim, a scalable robotics simulation application and synthetic data generation tool, to deploy a fleet of intelligent robots for logistics to improve the material flow in their production. The team can also leverage NVIDIA Isaac for synthetic data generation and domain randomization.

To learn more about implementing digital twins, visit