Digital twin technology sounds inherently cool and edgy, and it is cool, in many ways.
As a computer representation of a real thing, a digital twin can be the broker to all sorts of cutting-edge applications, such as wind power sold by the hour, remotely managed automated factories and predictive maintenance tools with augmented reality.
Digital twins are everywhere, it seems — if “everywhere” means the websites and shiny brochures of brand-name manufacturers and IT vendors. But the technology is not just a figment of the IT industry’s imagination. In a survey last year, Gartner said digital twins are proliferating and slowly entering the mainstream, and predicted that two-thirds of companies that adopt IoT will have at least one digital twin by 2022.
Digital twin technology is also showing up in real implementations by farm equipment and medical device manufacturers and aerospace companies, among others.
Take Aerion Supersonic, a jet manufacturer based in Reno, Nev. Last year, it began working on a digital twin architecture for the AS2, which it claims will be the first commercial jet capable of flying at supersonic speeds that can exceed 1,000 miles per hour and cut the Los Angeles to New York trip to three hours and 17 minutes — without generating ground noise.
Bissell Smith, Aerion’s CIO and executive vice president of enterprise systems, said digital twins will be needed throughout the AS2’s lifecycle and are worth the significant effort they take to develop and maintain. They drive virtual simulation and operational and fleet management efficiency. Without digital twin technology, for example, problems with an airplane’s design may require grounding the entire fleet. Digital twins of each aircraft will make it possible to isolate issues to a single plane’s unique configuration.
“The better you’ve [digitized a jet’s] ‘secret sauce’ … the more you’ll be able to keep aircraft flying versus on the ground waiting,” he said.
Digital twin technology: Early days
The aerospace industry is not the only sector banking on digital twin technology to confer a competitive edge. Prominent users — and vendors — of digital twins include major names in industrial automation like GE and Rockwell Automation. In the enterprise software world, makers of product lifecycle management (PLM) software lead the way, including Oracle, SAP and Siemens. Digital twins also come from makers of computer-aided design (CAD) and 3D modeling and simulation software, such as Autodesk, Bentley Systems, Dassault Systemes and PTC. Public cloud vendors Amazon Web Services and Microsoft Azure IoT also have offerings. In addition, many users develop their own digital twins (see the infographic, “Sources of digital twins”).
Still, companies interested in using digital twin technology will quickly realize that it presents significant challenges. For starters, its implementation is intertwined with another relatively new technology — IoT.
“There is a very close symbiotic relationship between the two,” said Shawn DeVries, managing director at Kin + Carta, a consulting firm. “The value of the digital twin platform is only as strong as the ability to read the data as near to real time as possible.” IoT sensors are often that data source. Conversely, digital twins are a useful mechanism for organizing and communicating IoT data so it can be analyzed and monetized in applications.
Driving value from digital twin technology and IoT, however, requires mature data management processes and sophisticated systems integration, making digital twin development a complex, multi-year effort — and therefore another challenge companies must reckon with.
So, while existing applications for digital twin technology range widely, from oil and gas companies optimizing oilfield production and electric utilities monitoring their equipment to hospitals keeping tabs on patients after surgery, deployments are limited, according to Benoit Lheureux, a research vice president at Gartner.
“People are still in proofs of concept and experimenting with the technology,” he said.
But Richard Howells, vice president of solution management for SAP digital supply chain, said “the reality is that most people have a digital twin of some sort of their products or their facility or their process. They just don’t call it that.”
“Once you have technology in place that’s an integrated, end-to-end process — from how you design products all the way through to how they operate in a live environment — you start capturing more and more data, digitizing the processes and capturing more information about the assets that you use.”
What is a digital twin, really?
The consensus base definition of a digital twin is that it’s a digital representation of a thing, process or person.
“It becomes a proxy for the state of the thing,” Lheureux said. Digital twins date to the late 1960s when NASA used telemetry and “automated interfaces” to monitor Apollo spacecraft, he said.
Some confusion over the definition might be due to the 3D color images and 2D CAD drawings that some digital twin providers use to market their products. Visualizations that aren’t well integrated with data sources and can’t be queried about the state of the entity they represent aren’t really twins. Simulations also aren’t automatically digital twins for the same reasons, and they require complex algorithms and huge amounts of data to mimic behavior, something most digital twins don’t need.
To Lheureux, what differentiates a digital twin from digital representations that appear similar is that people can query a digital twin to learn something about the thing it represents. “A digital twin is enterprise software that ingests IoT data to produce a finding, to increase situational awareness,” he said.
DeVries said the concept has evolved from his days working in CAD 25-30 years ago when it meant a detailed digital model that could be used in the manufacturing process. What’s changed is the advent of cheap networked sensors that allow digital models to be smarter. “You can see not just what it’s doing now, but what it’s done over time for trend analysis and predictive analysis,” he said.
Digital twin use cases in business
The most common types of digital twins deployed are for expensive equipment that is tunable, customizable and critical to a company’s business, DeVries said.
“There seems to be almost a tipping point where the more critical the asset, the greater the chance of having some sort of digital twin associated with it,” he said. “If you’re putting that kind of investment in a piece of equipment, you expect to have some level of visibility to how well that equipment is performing,” he said.
Examples of this phenomenon are the agricultural equipment makers Agco, Caterpillar and John Deere.
Farming equipment has had telemetry for decades, DeVries said. “It’s very similar to what you see in a Tesla these days. You don’t really associate tractors or combines with that type of technology integration, but it very much is the case.”
Besides heavy equipment, the other primary market for digital twins is process automation, DeVries said. ABB is a major player, along with Rockwell Automation, which has a partnership with PTC.
“They have replications of those processes running within their support centers that can basically talk to a maintenance engineer or a plant supervisor,” he said. “Those types of support touchpoints with their customers become a much richer, engaging experience, and they can get to the problem a lot quicker.”
Digital markets for digital twins are springing up, such as the Digital Twin Exchange that IBM opened in May. SAP already had a cloud-based repository for digital twins and other equipment information called the Asset Intelligence Network.
“The easiest way to describe it is as an iTunes for asset information,” said Joe Berti, IBM’s vice president of offering management for AI applications. Customers include an IBM partner who uploaded digital twins for 20 kinds of bridges along with maintenance information and predictive models. A Porsche Boxster’s digital twin contains a bill of materials (BOM), parts list, error codes and metrics such as mean time between failures. IBM splits the revenue with the content producer, which continues to own the intellectual property.
Berti likened the exchange’s digital twins to resume templates for job applicants.
“It depends on the provider of the digital twin, how much information they provide,” he said. “Most manufacturers aren’t going to give away the CAD model for their products. They don’t want people copying their designs. But they’re willing to go with a parts list and the bill of materials. They do that already in their maintenance manuals.”
Companies that own heavy equipment will pay the exchange $1,000 – $5,000 for an operational digital twin, Berti said, because it saves time.
Assembling a digital twin architecture
For large, complicated equipment that is hugely expensive to maintain, has life-or-death safety implications and a useful life measured in half centuries, digital twin technology is both essential and incredibly difficult to pull off. Few industries exemplify that more than aerospace and defense.
Capturing all the information for the digital twin of each jet Aerion Supersonic plans to start manufacturing in 2023 is “no small matter,” Smith said. He detailed the enterprise architecture the company is building, and what’s involved in pulling data from it.
Aerion digital twins start life with the engineering designs in Siemens NX CAD or a partner’s CAD tool. The company also has Siemens PLM software and will soon pick an ERP platform that includes material requirements planning (MRP), the system for calculating and scheduling raw materials and parts. The ERP will run both manufacturing and the supply chain and generate the data the digital twin needs from those processes, including quality control information, rework issues and engineering changes. The ERP will be integrated with a manufacturing execution system (MES) and data lake architectures and analytics set up.
An engineering BOM and a manufacturing BOM will be the main integration mechanisms between PLM and ERP. The BOM information also feeds into MRP, which Aerion will use to buy or make parts.
“All of that’s been captured in what’s called the digital thread that goes into, ultimately, the digital twin,” Smith said. “You have the digital thread all the way up to delivery. That digital thread will be shared with all of our major partners — they’ll be helping to create it. When we get to the digital twin, that will only be between us and the customer.”
Smith said the best way to capture the information is through the transactions handled in the different applications. “Every time an engineering update is made and put in the PLM, PLM is capturing that data. But that’s being done by the hundreds of designers that are working on our aircraft. Every time we [buy parts], we’re going to be releasing that to our supply chain and that supply chain is going to be referencing drawings, and that’s all part of the digital thread.”
People on the manufacturing side do the same as they build their production plans and “redline” the engineering plans for production issues in Siemens Active Workspace. Then mechanical and industrial engineers analyze the drawings and make suggestions, all of which also get captured in PLM.
Bissell SmithCIO, Aerion Supersonic
If, during manufacturing, a deviation from the plan becomes necessary, any changes in the design will also have to be captured. “All that data will be captured in the ERP system and the MES system tied to it,” Smith said. “Our logistics system will be capturing more. And you have to thread all this together.”
“Then it’s up to us — Aerion — to decide, when we’re working on the commercial [aviation] side, what do we want that digital twin to represent?” He said the aerospace industry is “nebulous” about its expectations for digital twin technology and much clearer about how to use it in engineering than in operations. The Air Force, which made digital twins a requirement of its contract, will decide what it wants.
“If they buy aircraft from us, they’re going to expect that infrastructure,” Smith said.
Then there is the information needed for certification, such as results of virtual tests of aircraft models for factors such as noise, weight and drag.
“Everywhere we’ve got any kind of simulated testing out on the system, we’re going to try to capture all that through the Siemens architecture,” Smith said.
The digital twin becomes each aircraft’s unique signature. “We try and get as close as possible, but there are always errors that enter the process — a mechanic drilled a hole too large that has to be filled. There’s a deviation waiver that goes with that, and approval,” he said.
Challenges of digital twin technology
Digital twin users face the same hurdles as Aerion, if not always on so massive a scale.
“Deploying those sensors, the communications, designing the algorithms, implementing the cloud platform, integrating the digital twins with a back-end system to automate your response — that is a lot of technology. It’s not trivial,” Lheureux said. What’s more, the job takes a range of skills in integration, security, analytics and application development. Much of the technology is unproven, so people lack confidence in it.
Industry standards from entities like the Industrial Internet Consortium could help, Lheureux said, but most manufacturers have many brands in their shops, and much of the value comes from standalone digital twins.
The “blue-sky” vision is for interoperability to occur between the hierarchical levels of all the digital twins a company has — say, between the digital twin for a piece of equipment — what Gartner calls a discrete digital twin — and the digital twin for the whole factory, a type Gartner labels as composite. That data can then feed into the highest-level category, an organizational digital twin.
“We will get there,” Lheureux said.
For now, interoperability comes from APIs, which most digital twins have. A composite twin can use an API to get data from a discrete twin. The challenge comes in normalizing the data to ensure it meets the needs of the receiving twin. One composite digital twin of a manufacturing process might reside in the AWS public cloud, while the discrete digital twins for equipment run on Microsoft Azure IoT.
“That sounds like a mess, right? And it is a mess,” Lheureux said.
Sometimes the data will be available in a discrete digital twin and the user will call APIs to access it, he said, but sometimes it won’t be. “You’ll go after other sensors, you’ll put out new sensors, you’ll collect data from different sources. You’ll do whatever it takes to get whatever data is needed to achieve a particular outcome,” he said.
Benoit LheureuxResearch vice president, Gartner
The usual solution is for owner-operators of digital twins like GE to only develop them for a handful of the most troublesome pieces of equipment. “To create the composite, you’re going to be pulling data from many sources anyway,” he said. “Anybody who will be successful at deploying digital twins at scale must be a master of integration.”
Berti said an industry standard for equipment information is needed. Certain pieces are already covered by bodies such as the ISO. “As people see the need to electronically import information and exchange it, standards typically follow, but they typically follow over the next three to five years,” he said.
The IT infrastructure needs of the digital twins themselves are minimal, according to Berti, because most of the data is already being stored and managed in enterprise systems such as CAD and ERP as part of maintenance management.
Getting the right type and amount of data requires the equipment manufacturer to foster a close partnership between its business side, which understands customer use cases, and the engineering side, which knows the hardware integration issues, DeVries said.
“There’s almost always a business or executive sponsor who understands, and gets out there in the field and interviews the growers and other people who work with this equipment, whether it be a mining operation, construction, etc., and tries to declutter a lot of those interfaces to make sure they’re getting the right information at the right time,” he said.
But the challenges aren’t primarily technical, DeVries said. “It’s a cultural and educational shift in the frontline workers, the maintenance engineers and plant supervisors. Like anything else, they need to see the benefits of a solution to support that more advanced digital mindset.”
Do you even need digital twins?
Digital twin technology sounds a lot like the data management and analytics that have long been required to make IoT sensors practical. So why bother with a digital twin?
“A lot of people do think that a lot of this is just really event-driven analytics, and I don’t disagree with that,” Lheureux said, noting it’s similar to the event stream processing that has been around for decades.
Nevertheless, digital twins will still be important in realizing the automation potential of IoT. “Everything we do in business is event-driven,” he said. When a product breaks, the consumer contacts a call center to request a spare part. If a production-line robot breaks down, a service technician notes the problem and logs it in a help desk system.
“Right now, we respond to those events at the ‘last mile’ of integration with humans,” Lheureux said. “That last-mile gap between things in the world and our IT systems is what we’re closing when we implement IoT and digital twins.”
He said managers who are responsible for their organization’s enterprise applications must prepare for a fundamental shift from applications in which workflows are controlled by humans managing business transactions to ones where IoT-connected things call more of the shots. “It’s going to be a thing that’s sending data to a digital twin and the digital twin will execute some new workflow that will trigger the backend systems.”
“It’s generational. It’ll be bit by bit, but it’ll be a fundamental shift by death by a thousand cuts for the enterprise application leader.”