Digital Twins: Components, Use Cases, and Implementation Tips
Digital twins play the same role for complex machines and processes as food tasters for monarchs or stunt doubles for movie stars: They prevent harm that otherwise could be done to precious assets. Having made their way to the virtual world, duplicates save time, money, and effort for numerous businesses — protecting the health and safety of high value resources.
The article covers key questions about digital twins: How do they work? What are they capable of? Who already uses them and for what reason? And — most importantly — what does it take to adopt them?
What is a digital twin?
A digital twin (DT) is a detailed and dynamically updated virtual replica of physical objects or processes, made to monitor performance, test different scenarios, predict issues, and find optimization opportunities. Unlike traditional computer-aided design and engineering (CAD/CAE) models, a DT always has a unique, real-world counterpart, receives live data from it, and changes accordingly to mimic the origin through its lifecycle.
The twinning, however, doesn’t happen out of thin air. This process involves numerous pieces working as a uniform system.
Digital twin system architecture
A digital twin system contains hardware and software components with middleware for data management in between.
Components of the digital twin system.
Hardware components. The key technology driving DTs is the Internet of Things (IoT) sensors, that initiate the exchange of information between assets and their software representation. The hardware part also includes actuators, converting digital signals into mechanical movements, network devices like routers, edge servers, and IoT gateways, etc.
Data management middleware. Its bare-bones element is a centralized repository to accumulate data from different sources. Ideally, the middleware platform also takes care of such tasks as connectivity, data integration, data processing, data quality control, data visualization, data modeling and governance, and more. Examples of such solutions are common IoT platforms and industrial (IIoT) platforms that often come with pre-built tools for digital twinning.
Software components. The crucial part of digital twinning is the analytics engine that turns raw observations into valuable business insights. In many cases, it is powered by machine learning models. Other must-have pieces of a DT puzzle are dashboards for real-time monitoring, design tools for modeling, and simulation software.
Digital thread: a bridge between physical and virtual worlds
Having all the required components in hand, you can interconnect physical systems and their virtual representations into a closed loop known as a digital thread. Within it, the following iterative operations are performed.
Steps within a digital thread.
- Data is collected from a physical object and its environment and sent to the centralized repository.
- Data is analyzed and prepared to be fed to the DT.
- The digital twin uses fresh data to mirror the object’s work in real time, test what will happen if the environment changes, and find bottlenecks. At this step, AI algorithms can be applied to tweak the product design or spot unhealthy trends and prevent costly downtimes.
- Insights from analytics are visualized and presented via the dashboard.
- Stakeholders make actionable, data-driven decisions.
- The physical object parameters, processes, or maintenance schedules are adjusted accordingly.
Then the process is repeated based on the new data.
Digital twins reduce the complexity of the real world to the information necessary for decision-making. This makes the technology welcome across many industries.
DT main applications
First introduced in 2002 by professor Michael Grieves as a new approach to managing product life cycle, the concept of digital twins gained traction in many areas including supply chain management, remote equipment diagnostics, predictive maintenance, and more. They can serve any phase of product development, from designing to post-production monitoring and servicing.
Of course, virtual modeling doesn’t work as well for every business. Its implementation comes with a hefty price tag and may make little to no economic sense for simple products. Virtual models suit complex and large-scale projects and multi-component mechanisms, finding the most successful applications in
- construction of buildings, bridges, drilling platforms, and other large objects;
- industrial environments;
- designing and manufacturing of complex products like cars, jet turbines, airplanes, or new drugs;
- urban planning; and
- the energy sector with its huge equipment for power generation and transmission.
Within these industries, twinning can be performed at different levels — from a separate component to the entire product to production to system of systems.
The basic level of twinning allows engineers to evaluate the durability, resilience, energy efficiency, and other characteristics of the separate parts that constitute a product. They can use simulation software to analyze how the component in question will behave under static or thermal stress and in other real-life scenarios.
Product or asset twinning
The replica of the entire product reveals how individual components work together under various conditions and what can be done to achieve better performance and reliability. Digital twinning also can be used to design new technical solutions — instead of creating multiple prototypes. This shortens development time and allows for faster iterations.
Process and production twinning
Digital twinning is applicable not only to physical assets but to processes as well. In this case, you create complete virtual models of the production steps.
This approach helps answer important questions like: How long will it take to produce a particular product? How much will it cost? Which machine should do what? Which steps can be automated? Is the production of a particular item feasible at all? Additionally, visualizing the entire production process makes it easier to prevent costly downtimes.
A digital twin of the system brings visibility to complex interconnections and interdependencies of products and processes. The twinned system can be as large as a multistory building, electrical grid, or even a whole city, which can be viewed as “a system of systems.” However, investment to build such a replica in many cases may not equal to the hoped-for return. That’s why system twinning is not as widespread as other DT types.
Below, we’ll give real-life examples of how digital twins work in different industries across all levels.
Aircraft industry: pinpointing the right time for engine maintenance
Up to 70 percent of airplanes in the world fly on engines, produced by General Electric (GE). This fact makes the corporation partially responsible for the safety of millions of passengers. To forecast the degradation of the aircraft’s heart over time, GE created a digital twin for its engine GE90 that powers long-range Boeing 777.
GE created a digital twin for fan blades of its most popular GE90 engine that surpassed 100 million flight hours. Source: Aerospace Manufacturing and Design.
The twin represents not the entire mechanism, but its composite fan blades that are prone to spallation or thepeeling off of the fragments of material due to the impact of rough conditions. This is especially valid for the regions like the Middle East, where engines are exposed to such an additional damaging factor as sand. The DT helps pinpoint the right time for maintenance before any issues arise.
Automotive industry: running Tesla car replicas for remote diagnostics
Each new car produced by Tesla has its own digital twin. Sensors embedded in a vehicle constantly stream data about the environment and performance to the virtual copy that lives in the cloud. AI algorithms analyze these feeds to identify whether the car works as expected. If not, the problems are fixed by sending over-the-air software updates.
In this way, Tesla adapts the vehicle’s configurations to different climate conditions, virtually improves its performance, and provides remote diagnostics, minimizing the need for visiting service centers.
Tire manufacturing: reducing the wear of wheels
Bridgestone, the world’s top tire and rubber manufacturer, is taking advantage of digital twins on a regular basis to understand how speed, road conditions, driving style, and other factors affect the performance and lifespan of their products. Armed with these insights, the company helps fleets select the best options for their specific needs and advise on what can be done to prevent breakages and extend the life of the wheels.
The industry leader also uses digital twinning to design and test new types of tires. According to Bridgestone estimations, this approach cuts development time by 50 percent.
Power generation: predicting the performance of gas turbines
The Europe’s largest industrial manufacturing company and a digital twinning pioneer, Siemens developed a virtual avatar of its gas turbine and compressor business purchased from Rolls-Royce. The digital twin called ATOM (Agent-Based Turbine Operations and Maintenance) represents the production and servicing of their turbine fleet, spanning the supply chain operations.
ATOM digital twin reflects complex interactions across the entire lifecycle of the gas turbine. Source: Anylogic.
ATOM digests live data from multiple sources to thoroughly model cobwebs of engine parameters, performance metrics, maintenance operations, and logistics steps across the entire turbine lifecycle. By running different what-if scenarios and visualizing their results, it helps stakeholders make better investment decisions.
Supply chain simulation: bringing visibility to logistics
In September 2021, Google presented a new service that enables companies to build digital twins of their physical supply chains. The solution focuses on organizations in the retail sector, healthcare, manufacturing, and automotive industry. It aggregates information from multiple sources into one place and helps customers get a complete and clear view of their logistics.
Google claims that the DT paves the way to much faster data analysis: tasks that previously took up to two hours now take just a few minutes. With its fresh offering, the company is snapping at the heels of IBM, Amazon, and Microsoft, all of which launched supply chain and other digital twin options a little bit earlier.
Urban planning: creating profiles of buildings to reduce energy consumption
Automatic Building Energy Modeling or AutoBEM for short enables the generating of digital twins for any building in the US. The project took the developers
From the Department of Energy’s Oak Ridge National Laboratory five years and became available in 2021.
AutoBEM relies on public information like satellite imagery, street views, light detection and ranging (LIDAR), prototype buildings, and standard building codes to generate energy profiles of structures. A twin reflects all critical external and internal characteristics including a building’s height, size, and type, a number of windows and floors, building envelope materials, roof type, heating, ventilation, and cooling systems.
Advanced algorithms behind the twin predict which technologies are to be implemented to save energy. This includes modern water heaters, smart thermostats, solar panels, and more. Supposedly, AutoBEM will be widely used in urban planning and maintenance as there is much concern about energy consumption in cities across the US.
How to approach digital twinning
Theoretically, you can build a digital twin for almost everything. In practice, it’s far from feasible to create a replica that covers every single aspect of a product or manufacturing process. If you’ve already jumped to the conclusion that your business will benefit from DTs or at least want to test the idea, choose a single component or operation that is most vulnerable or crucial for your business.
Once you understand what you are going to twin in the first place, the next steps may be the following.
Choose the type of digital twin: physics-based vs data-driven vs hybrid models
Generally, there are two types of DTs — physics-based twins and data-based twins. The former rely on physical laws and expert knowledge. They can be built from CAD files and used to simulate the work of comparatively simple objects with predictable behavior — like a piece of machinery on the production line.
The key downside is that updating such twins takes hours rather than minutes or seconds. So, the approach makes sense in areas where you don’t need to make immediate decisions.
Contrasted with the physics-based type, data-based twins don’t require deep engineering expertise. Instead of understanding the physical principles behind the system, they use machine learning algorithms (typically, neural networks) to find hidden relationships between input and output.
The data-based method offers more accurate and quicker results and is applicable to products or processes with complex interactions and a large number of impact factors involved. On the other hand, to produce valid results it needs a vast amount of information not limited to live streams from sensors.
Algorithms have to be trained on historical data generated by the asset itself, accumulated from enterprise systems like ERP, and extracted from CAD drawings, bills of material, Excel files, and other documents.
Today, various combinations of two methods — or so-called hybrid twins — are often used to take advantage of both worlds.
Narrow down the focus
Even if we’re speaking of the same component or product, different models are to be created for different tasks. For example, the General Electric Digital department defines four categories of models that can be integrated into the DT designed for the needs of power plants.
Lifing models track equipment condition and forecast how it will age providing for its operations and exposure to damaging factors. This helps optimize maintenance outages and extends the asset’s working lifespan.
Anomaly models are responsible for early fault detection to reduce unplanned downtimes.
Thermal models simulate parameters related to thermal efficiency to better manage degradations.
Transient models predict the plant’s speed, reliability, and emissions under different conditions. The gained insights are used to achieve the best operational flexibility with consideration for equipment and site limitations.
Anyway, at this step, you need to define what exactly your model will do: monitor conditions, prevent faults, simulate behavior under different parameters, help in product design, you name it. Focus on one task, test results, and only then augment your digital twin with other capabilities.
Consider investments to be made
Let’s presume that your company already has sensors and CAD or CAE software to create a basic representation of your assets. At the next phase, you will need to invest in
- additional hardware — for example, edge computing devices to process data on the periphery, closer to IoT sensors;
- services of a data management or IoT platform or other middleware to ingest and process data from disparate systems and store it in one place;
- simulation software;
- analytics solutions;
- domain experts to run physics-based simulations; and/or
- data scientists if you opt for data-driven or hybrid methods.
It’s worth noting that large cloud providers and leaders in digital twinning offer services that cover many aspects of digital twinning.
Explore ready-to-use solutions
Here’s a short overview of the DT products from industry leaders that will save you time and effort.
IBM Digital Twin Exchange works as a virtual shop where organizations may search, purchase, and download digital twins or data related to them from different manufacturers. The assortment includes 3D CAD models, bill of material (BOM) lists, engineering manuals, etc. They can be teamed with IBM Maximo asset management solution to predict asset performance and schedule maintenance operations based on fresh data.
Azure Digital Twins is a platform as a service (PaaS) for visualizing physical environments with all connected devices, locations, and occupants involved. The relationships between these objects are represented with spatial graphs. The service is paired with Azure IoT Hub that collects data from IoT sensors and other Azure services.
GE Digital Twins software allows companies to rapidly create digital twins and get value from them, using “blueprints” from their catalog. The three core areas covered by GE are manufacturing assets, grid networks, and production processes. DT software is powered by machine learning and integrated with GE’s IIoT platform called Predix.
Digital twins are at the core of GE’s Predix platform.
Oracle IoT Digital Twin Framework enables you to create both physics-based and data-driven or predictive digital twins. The former compares observed and desired parameters to detect existing problems. The latter run machine learning algorithms to forecast future issues and prevent or prepare for them.
Keep in mind, though, that DT tools, elements, and blueprints work well when bought from one vendor. Otherwise, compatibility and integration issues are more than possible. As with other emerging and evolving technologies, digital twinning lacks generally accepted standards, leading to poor interoperability between systems.
Remember that DT is not off-the-shelf technology
In any case, each DT is as unique as a product or process it represents. While ready-to-use infrastructures, platforms, and models can facilitate the development, but they won’t do all the work. You will still need experts in data, machine learning, cloud technologies, and, of course, engineers, capable of integrating different parts of hardware and software puzzles.
What to expect?
Despite all promises and even proven examples of success, digital twins still don’t see wide adoption. To some extent, the complexity of their creation is to blame. Another reason that we already mentioned is a scarcity of industry standards that restrains communication and data exchange across platforms and apps from different vendors.
Hopefully, this will change soon: More and more tech leaders including Microsoft, GE Digital, and Dell have become part of the Digital Twin Consortium to facilitate the development, adoption, and interoperability of DTs. This goal is hard to achieve without clear technical guidance and agreed-upon frameworks.