Simulation 5.0 with AI Agents and High Speed: Welcome to the Omniverse

Before a Krones filling line goes into operation, every process step is tailored to the customer’s individual requirements. Precision is essential for the smooth processes, maximum throughput, and minimal losses Krones guarantees its customers. This promise is delivered through detailed physics‑based simulations that examine every aspect of the customer’s production process.

To ensure, for example, that bottles are filled with exactly the required amount of product or that contents do not spill during transport, these operations are calculated and analyzed by Krones using flow simulations. Once the ideal configuration has been identified in the digital twin, it is transferred by the process engineers to the real system.

The one “drop of bitterness”: because the models are highly detailed and therefore computationally demanding, and because many isolated, proprietary, sequential simulations were involved, results often required around four hours of compute time. Only then could they be linked with further factors and used to derive and implement optimization measures.

The digital twin was also used for the development and optimization of Krones’s new Ingeniq line concept, presented at drinktec 2025. | © Krones

Integrated Instead of Sequential: Simulation in NVIDIA Omniverse

With increasing competitive pressure and rising costs, such waiting times were increasingly scrutinized, especially as new technologies promise entirely new capabilities. One of these is NVIDIA Omniverse. More platform than tool, NVIDIA Omniverse creates a continuous, open, and interconnected engineering environment that brings together simulation, AI, visualization, automation, and real data sources in a single digital space.

Krones embraced this approach. Together with a partner ecosystem, they developed within two months an Omniverse‑based solution that

• compresses simulation times from four hours to just minutes,

• enables near real‑time system optimization, and

• uses a new GPU‑based simulation method for Ansys Fluent developed by CADFEM to precisely map filling and sloshing processes. AI agents are used that learn and independently make data‑driven optimization decisions.

Technologies from NVIDIA, Synopsys, and Microsoft

The technological foundation is the combination of NVIDIA Omniverse with Ansys Fluent, the CFD simulation software from Synopsys, and the Microsoft Azure cloud.

NVIDIA Omniverse, as a real‑time platform for collaborative engineering, is part of the Industrial Metaverse. In a consistent OpenUSD scene, it integrates CAD, physical field simulation, and IoT data, enabling interdisciplinary collaboration and physics‑based decision‑making. With its modular architecture of connectors, extensions, apps, and cloud services, it forms a flexible foundation for GPU‑accelerated simulation and AI‑supported automation.

Together with PyAnsys, the Python‑based access to all Ansys solvers from Synopsys, end to end workflows based on Ansys Fluent are created: fully automated, consistently reproducible, and seamlessly integrated from simulation to USD‑based visualization.

In addition to GPU‑accelerated simulation, the required compute power is provided via the secure and scalable Microsoft Azure cloud infrastructure.

Expertise From SoftServe and CADFEM

Beyond the key technologies used, the service providers SoftServe and CADFEM are part of Krones’s partner ecosystem. While IT specialist SoftServe integrates all components into a comprehensive digital twin, CADFEM focuses on developing a simulation workflow based on flow simulations for filling and sloshing processes. CADFEM developed a GPU‑based method for Ansys Fluent specifically for Krones, laying the foundation for extremely fast and AI‑supported CFD applications in the Omniverse environment.

CADFEM has been a close partner of Ansys for more than 40 years, now part of Synopsys. For nearly as long, CADFEM has collaborated with Krones in the field of simulation, with this project representing the latest milestone. CADFEM’s CFD experts for flow simulation with Ansys Fluent were involved, achieving previously unknown levels of expressiveness and speed in NVIDIA Omniverse.

CFD experts from Krones’s simulation partner, CADFEM, analyzed and optimized filling and sloshing effects of various bottle and liquid configurations in Ansys Fluent. | © Krones

Filling and Sloshing Simulations With Ansys Fluent by CADFEM

Not only is every bottle different, but the beverages also vary in consistency, density, and viscosity. This results in a wide range of machine configurations that must be represented in the computational setup. Managing all these variables to cover the full spectrum of customer and system‑specific aspects was the task of CADFEM’s CFD experts as they prepared the baseline simulations for the subsequent automation. Another key aspect was the final visualization and full integration into Omniverse to ensure seamless and appealing representation of the calculated field data, such as free surfaces.

Conclusion

Omniverse, as an overarching platform that unites all data, models, AI mechanisms, and workflows into a coherent, interactive, automated engineering ecosystem, is the link between simulation, AI, and real operations. It enables a new, highly detailed generation of digital twins that map not only products and systems but also environmental and operational parameters.

Krones uses Omniverse and GPU‑accelerated CFD simulations to digitally model filling processes and line layouts. Sloshing effects can be visualized in detail and evaluated in realistic line environments. Together with the partner ecosystem, these approaches have been combined into a fully digital system twin. It connects planning, operation, and service, and enables greater process reliability. This allows Krones to guarantee its customers system performance for different requirements even before commissioning.

The project at Krones demonstrates how CADFEM integrates these capabilities with established CAE workflows and makes them usable for engineering.