Manufacturers face increased pressures to shorten production cycles, enhance productivity, and improve quality, all while reducing costs. To address these challenges, they’re investing in industrial digitalization and AI-enabled digital twins to unlock new possibilities from planning to operations.
Developers at Pegatron, an electronics manufacturer based in Taiwan, used NVIDIA AI, NVIDIA Omniverse, and Universal Scene Description (OpenUSD) to build PEGAVERSE, a digital twin platform that transformed the company’s manufacturing operations.
Pegatron’s developers built the flexible digital twin platform for engineers and factory managers to collaboratively plan, simulate, and optimize their production lines, providing real-time insights about facilities, equipment, and maintenance tasks. Teams can now identify and resolve operational issues quickly and effectively.
“With PEGAVERSE we easily integrate all our resources into the virtual world, from design, optimization, maintenance, and operations,” said Andrew Hsiao, associate vice president, for the AI development division at Pegatron.
“It fulfills our need for real-time visibility, predictive maintenance, process optimization, remote operations, quality assurance, risk management, innovation, and cost reduction in our manufacturing operations.”
Figure 1. A comparison between the digital twin and the physical production line
Using virtual factories to optimize operations
Engineers and factory managers leverage PEGAVERSE for many use cases, including predictive maintenance, process optimization, resource planning, remote monitoring, and quality control.
Since beginning development through a successful initial pilot with 15 developers, the development team has expanded to over 100 developers, spanning 3D experts, CFD and FEA specialists, AI engineers, data engineers, and full-stack engineers.
Today, the platform consists of three major components:
PEGAAi, is an MLOps deep learning platform with no-code and low-code capabilities that streamline and automate data collection, generation, labeling, training, deployment, and monitoring.
Large language models (LLMs), enable engineers to develop automation processes tailored to their requirements and unlock flexibility and intelligence in their design and manufacturing processes.
Digital twin technologies help engineers to create physically accurate, large-scale virtual production environments for model training and simulation. Teams use these environments to replay and analyze historical events and predict real-time performance on the production line to guarantee quality.
With the PEGAVERSE platform, Pegatron’s teams developed five virtual factories across Taiwan, India, Indonesia, China, and Vietnam. Teams use these virtual factories for monitoring operations remotely and running simulations of production lines to identify bottlenecks and optimize operations.
With long-standing ambitions to build real-time, physics-based models of their facilities, their teams struggled with connecting their tools and data to achieve their goals. Their teams working on product and factory design and production line simulation rely on critical and specialized tools, including Autodesk Revit, Blender, and Visual Components to get their work done. However, they needed an easy way to unify their data into a digital twin for all stakeholders.
Developing the PEGAVERSE platform on Omniverse, Pegatron developers used OpenUSD, an extensible framework and ecosystem for describing, composing, simulating, and collaborating within 3D worlds. With OpenUSD, their teams integrate different data, tools, and processes within the PEGATRON platform, vastly accelerating the design, simulation, and development of automated equipment and production lines.
The development team sees OpenUSD as a crucial emerging 3D standard, similar to how HTML provided a common language that spurred the growth of the Internet.
“In the age of AI, there is a need for standards that can describe more real-world information, and OpenUSD fills this gap,” said Hsiao. “We see OpenUSD as a new form of the digital asset—it will be the foundation for establishing new processes and management principles to increase efficiency and create higher value for our customers.”
Enhancing digital twin experiences with IoT and generative AI
For access to real-time insights, they integrated generative AI into PEGAVERSE and connected the platform to IoT and sensor technologies inside their facilities using their in-house IoT service called PEGAAIoT. They also developed key features for their teams, like real-time synchronization, remote control, and log replay.
By developing and integrating AI agents, powered by NVIDIA NeMO, engineers can easily interact with their virtual factories in natural language and quickly retrieve critical information required to analyze and optimize operations.
With the NVIDIA Isaac Sim platform, teams can train perception AI models and simulate and fine-tune their fleet of mobile robots and pick-and-place robotic arms.
Using NVIDIA Metropolis, they can rapidly develop and deploy highly accurate automated optical inspection workflows across their production lines.
Figure 2. An illustration of the equipment breakdown-recovery processes using digital twins
With the integration of NVIDIA AI, Omniverse, and OpenUSD into the PEGAVERSE platform, Pegatron’s developers are enabling the company to benefit from AI-enabled digital twins that provide them access to the real-time insights and predictive analytics they need to continuously improve and transform their manufacturing operations.
Read the latest NVIDIA announcement at COMPUTEX to learn how Pegatron is adopting NVIDIA technology to build AI-enabled digital twins.
Additional resources:
