With the growing emphasis on environmental, social, and governance (ESG) investments and initiatives, manufacturers are increasingly seeking innovative ways to enhance energy efficiency and sustainability in their operations, according to the NVIDIA Technical Blog.

Enhancing Operational Efficiency with OpenUSD and AI

Wistron, a leading global supplier of information and communications products, has made significant strides in this area. The company has developed a digital twin platform and AI-enabled simulation tools to optimize the design, performance, and energy efficiency of their run-in test rooms in new NVIDIA DGX and NVIDIA HGX factories. This advancement has the potential to reduce energy consumption by up to 10%.

To achieve this, Wistron’s developers employed the NVIDIA Omniverse platform, leveraging Universal Scene Description (OpenUSD) to build their Digital Twin platform. This platform connects to their building management system and IoT hub, providing real-time data from thousands of physical sensors across their facilities. These sensors monitor critical parameters such as core temperature and air conditioning return temperature.

The integration of OpenUSD has facilitated real-time collaboration for remote teams, streamlining the review of facility layouts and accelerating decision-making processes in facility planning and operations.

Additionally, Wistron’s developers have integrated physics-informed AI models into their Digital Twin platform using the open-source NVIDIA Modulus framework. These models help accelerate simulation work, improve thermal dynamics, and reduce operational risks, ensuring that cooling systems perform optimally even under demanding conditions.

Building Digital Twins of Test Rooms

Wistron has utilized OpenUSD to unify their data pipeline and streamline workflows, enabling the creation of 3D models by their experts. This approach standardizes asset creation and ensures compatibility across various software and simulation tools. John Lu, plant manager at Wistron, highlighted that OpenUSD offers flexible data modeling, allowing the combination of diverse data and outcomes from various 3D modeling and simulation tools.

For example, OpenUSD has facilitated the connection between Wistron’s Digital Twin platform and the Autodesk FlexSim simulation software, enhancing their teams’ ability to simulate, analyze, and experiment with critical manufacturing processes. Developers have built a custom extension to import data and parameters from FlexSim, using an OpenUSD-native Omniverse Connector to integrate this data into their Digital Twin platform.

Accelerating Simulation and Predicting Risk with Physics-Informed AI

Wistron’s simulation experts rely on computational fluid dynamics (CFD) simulations to support the design and management of their run-in test rooms. However, the traditional CFD approaches, running on general-purpose computing architectures, often proved inflexible and resource-intensive.

To address these challenges, Wistron integrated additional capabilities into their Digital Twin platform, including:

• Physics-informed neural networks (PINNs) using NVIDIA Modulus, which significantly speeds up airflow simulations from 15 hours to just 3.6 seconds—a 15,000X improvement.
• An AI-based extension for high-fidelity visualization and analysis of CFD simulations, helping to minimize cooling system loads and reduce operating costs.
• A recommendation system for their automated storage and retrieval system (ASRS) to identify optimal testing locations and autonomously place new supercomputing baseboards in areas with the least thermal risk.

These enhancements enable Wistron to approximate the underlying physics of their thermal systems, providing fast and accurate predictions of temperature distributions and thermal behaviors within their run-in test rooms. Teams can now identify facility hotspots and forecast core temperatures up to 30 minutes in advance.

Read the latest NVIDIA announcement at COMPUTEX to learn how Wistron is adopting NVIDIA technology to build and operate digital twins.

Image source: Shutterstock

. . .

Tags