In the rapidly evolving field of global technology device manufacturing, Pegatron Corp. initially exploited the AI to gain an advantage. Now, it’s set to create digital twins to further streamline its efficiency.
Whether or not they know the name, most people have probably used smartphones, tablets, Wi-Fi routers or other products made by Taiwan-based Pegatron at nearly a dozen factories spread across in seven countries. Last year, it manufactured more than 10 million laptops.
Andrew Hsiao, Associate Vice President of Pegatron’s Software R&D Division, is leading the company’s evolution toward machine learning and the 3D Internet known as the Metaverse.
Build an AI platform
“We’ve been collecting factory data since 2012 to find patterns and insights that improve operations,” said Hsiao, a veteran technical director who has been with the company for 14 years since it spun off from ASUS. , one of the largest PC manufacturers in the world. .
In 2016, Pegatron COO Denese Yao launched a task force to apply new technologies to improve operations. Hsiao’s team of AI experts collaborated with factory workers to find AI use cases. One of their first pilots used deep learning to detect anomalies in products as they came down the line.
He got solid results using modified versions of neural network models like ResNet, so they pressed the gas.
Today, Pegatron uses Cambrian, an AI platform it built for automated inspection, deployed in most of its factories. It supports hundreds of AI models, trained and running in production on NVIDIA GPUs.
Fewer defects, more consistency
The new platform detects up to 60% more defects with 30% less variation than human inspectors, and plant workers love it.
“Manual inspection is boring and repetitive work, so it’s no surprise employees don’t like it,” he said. “Now we see employees motivated to learn more about new technology, enabling people to do more value-added work.”
The system can also improve throughput as factories adjust workflows at assembly and packaging stations to account for faster inspection lines.
Models deployed 50 times faster
Pegatron’s system uses NVIDIA A100 Tensor Core GPUs to deploy AI models up to 50 times faster than when training them on workstations, reducing weeks of work to hours.
“With our unified DGX-based platform, we have our data lake, datasets, and training in one place, so we can deploy a model with the click of a button,” Hsiao said.
By utilizing the multi-instance GPU capability of the A100 GPUs, Pegatron has reduced the wait time for developers to access an accelerator from nearly an hour to 30 seconds. “It allows us to dynamically schedule tasks like AI inference and lightweight model training,” he said.
As part of its AI inference work, the system analyzes more than 10 million frames per day using NVIDIA A40 and other GPUs.
Triton, NGC Simplifying AI Jobs
Pegatron uses NVIDIA Triton Inference Server, an open-source software that helps deploy, run, and scale AI models on all types of processors and frames. It works hand-in-hand with NVIDIA TensorRT, software that simplifies neural networks to reduce latency.
“Triton and TensorRT make it easy to service multiple clients and convert jobs to the most cost-effective levels of accuracy,” he said.
Hsiao’s team optimizes pre-trained AI models that it uploads to embedded Kubernetes containers from the NVIDIA NGC hub for GPU-optimized software.
“NGC is very useful because we get the deep learning frameworks and all the other software components we need with the click of a button, things that used to take us a long time to put together,” he said.
Next step: digital twins
Reaching a New Milestone in Smarter Manufacturing, Pegatron Pilots NVIDIA Omniverse, a Digital Twin Development Platform
It has two use cases so far. First, test Omniverse Replicator to generate synthetic data on what products following the line of inspection might look like under different lighting conditions or orientations. This information will make its perceptual models more intelligent.
Second, it creates digital twins of inspection machines. This allows remote workers to manage them remotely, gain insight into predictive maintenance, and simulate software updates before deploying them to a physical machine.
“Today, when a system goes down, we can only check logs that might be incomplete, but with Omniverse, we can replay what happened to figure out how to fix it, or run simulations to predict how it will behave in the future”, he said.
In addition, industrial engineers who care about throughput, automation engineers responsible for downtime, and equipment engineers who manage maintenance can work together on the same virtual system at the same time, even when connecting from different countries.
Vision of a Virtual Factory
If all goes well, Pegatron could have Omniverse on its inspection machines before the end of the year.
Meanwhile, Hsiao is looking for partners who can help create virtual versions of an entire production line in Omniverse. In the longer term, his vision is to create a digital twin of an entire plant.
“In my opinion, the biggest impact will come from building a full virtual factory so that we can try things like new ways of moving products through the factory,” he said. “When you just build it without prior simulation, your mistakes are very costly.”