Why digital twins require an incremental process

Digital twins show promise for streamlining collaborative development in factories, construction, smart cities, and product development. But enterprises need to move incrementally to work out the data integration issues in their existing digital tools. 

One pioneer is Pegatron, a leading Taiwanese manufacturer of smartphones, tablets, and other equipment made in about a dozen factories. Andrew Hsiao, Associate Vice-President at Pegatron, argues that digital twins are an important part of their evolution from more traditional product lifecycle management (PLM), design tools, and quality assurance processes to a more agile manufacturing process. 

Hsiao has been leading work to integrate workflows across various NVIDIA tools for hosting digital twins and streamlining machine vision apps for quality control and managing 3D models using the Universal Scene Description (USD) format. For example, Pegatron currently uses Solidworks for mechanical design and CAD modeling. It converts Solidworks models to USD and then use various tools to integrate and present the aggregated models connected across various NVIDIA tools. 

New workflows

Pegatron is currently planning a new factory in Taiwan that is pushing the new digital twin based process through its paces. It helps individuals across different teams, such as facility personnel, automation engineers, and industrial engineers, communicate and collaborate. This makes it easier for everyone to see how designs are affected by different simulated scenarios and understand the potential impacts. 

Additionally, they are using digital twins to plan layouts for a new warehouse system for an overseas factory. This helps simulate and assess different scenarios more accurately, such as determining the appropriate number of automated guided vehicles, charging locations, and routing.

A third use case is improving Pegatron’s computer vision workflows. Pegatron developed its own internal AI platform and is now using the NVIDIA machine vision platform to streamline and enhance the performance of deep learning models within its internal platform. 

Many challenges

One challenging aspect lies in accurately representing the factory as it is built. Despite improvements in tools for capturing existing factories, it also requires a lot of work. Hsiao explains: 

In the past, we have tried using reality capture tools to assist, but it often requires a significant amount of post-processing work, and the cost is relatively high. Now, we are also experimenting with different tools to construct various objects. For example, Revit can be used to construct buildings, while Visual Components is used to construct automation equipment.

Another challenge lies in coordinating digital twin efforts across different locations, roles, and teams. Like many new technologies, it is not as simple as installing a new digital twin server and plugging in all the existing apps. Hsiao says: 

Currently, we are using a task force approach and inviting personnel from different teams to participate in a pilot project together. Our plan is to first build a platform that can support a larger user base, enabling people from different regions to work together. Next, we will gradually integrate the tools and processes used by different teams and provide education and training on critical workflows in collaboration with these teams.

Silos and spaghetti

Pegatron has built its own internal private data center to host the NVIDIA equipment underpinning its digital twin efforts. One issue is that the private data center primarily runs Linux, while most collaborative design tools run on Windows, so they must prepare workstations. 

Despite the challenges in getting things up and running, they are already starting to see some benefits in a more unified approach to managing design data across tools. Hsiao says:

In the past, due to the lack of common formats and collaborative platforms, the management of digital models was more like 'silos and spaghetti.'

The Nvidia tools improve the search and management of Pegatron’s 3D data using USD file formats, which can improve these issues. However, data can sometimes be lost, or conversion processes get hung when moving back and forth. So, Hsiao’s team has been gradually developing systemic approaches, such as using scripts to convert the existing digital models into USD.

The USD file format was initially conceived to streamline 3D rendering in the movie industry. This presents issues for industrial design tools which use different file format structures optimized for simulating mechanical and electrical properties rather than just how they look. For example, Solidworks uses the STEP format, which sometimes causes the conversion process to hang. Hsiao’s team is also turning to various third-party tools like Blender for converting other file formats, such as OBJ and MTL files, into USD. 

Building new bridges

In the long run, Hsiao hopes different vendors improve support for the USD format for digital models. For now, connectors serve as the best bridge for collaborating between various tools. 

Today most of their design tools run on Windows, so developers need to have their own development environment, which increases the cost of usage. New approaches for connecting things up into a private cloud could help reduce the costs of provisioning and updating tools. Hsiao notes:

If it is possible to move all these to a data center, the development efficiency and multi-device communication would greatly improve.

My take

Comprehensive digital twins for creating models of physical things and the processes that build them show tremendous promise. But getting all the tools to talk to each other is incredibly challenging. There are over fifty different 3D file formats optimized for various design, simulation, production and quality control processes. 

Today conversion tools can translate across these different file formats, but data can get lost in translation. Live connectors that synchronize changes across these file formats will be required to streamline workflows across various tools. All major PLM vendors are building their own digital twin platforms that can do this for their own ecosystem of tools.