The term “digital twins”—digital representations of physical objects, persons, or processes that can help organizations simulate real situations and outcomes in order to make better decisions—has surged into prominence, evoking images of meticulously crafted virtual replicas, spanning from infrastructural marvels to high-tech machinery. Promises to revolutionize industries with their potential for unparalleled design optimization, means of training, or problem prediction before they occur have been made.
The idea of digital twins isn’t entirely new. Who can forget the case of Apollo 13 (“Houston, we have a problem”)—probably the most classic example of a digital twin—when NASA engineers relied heavily on simulations and models to understand the spacecraft’s condition, troubleshoot problems, and devise solutions. Recent advancements in AI, coupled with stunning visualizations, have given this concept new life. The allure lies in the ability to convert real-world projects into mesmerizing digital replicas, enabling us to anticipate potential issues, be it weather disruptions, security events, design efficiencies, or sustainable construction. But do digital twins hold water in the real world?
The Methodology
To answer this question, I did my research, delving into articles, TEDx Talks, podcasts, and explanatory videos. I found that digital twins are already making waves, promising rosy scenarios of efficiency gains across manufacturing, construction, transportation, medicine, retail, smart cities, and more. These digital doppelgängers are hailed for their promise of new levels of efficiency and the ability to predict the future with AI’s assistance.
Some claim that now we can design and model structures in virtual reality, simulating various forces and impacts. But isn’t that what simulations have been doing for years? In my quest for clarity, I stumbled upon a crucial distinction: Digital twins require a multitude of input models and real-time feedback on an already functioning object, setting them apart from conventional simulations. Yet it left me questioning their viability and accuracy in real-world scenarios.
Current Digital Twins Adoption
In reality, digital twins find their true calling in industries characterized by highly specialized teams, advanced sensors providing vast data streams, expensive equipment, and high-stakes consequences for errors. Sectors like Formula 1, aerospace, military ballistic missiles, the U.S. Air Force, and skyscraper construction/maintenance are prime examples where digital twins shine.
For example, Boeing uses digital twins to model the life cycles of its airplanes, allowing it to track maintenance needs and predict potential part failures to minimize aircraft downtime. Airbus employs digital twins to analyze data from its aircraft engines, which enables the company to predict potential maintenance issues, allowing for proactive and cost-effective maintenance.
In the automotive sector, Tesla creates a digital twin of every vehicle it sells. Sensors from the cars continuously stream data into each car’s simulation in the factory, where AI interprets the data and determines if a car needs maintenance. This massive data asset empowers the company to optimize performance and predict maintenance needs. General Motors took a different approach, creating a digital twin of its entire factory. By modeling the interactions between robots, conveyors, and other equipment, it can continuously improve the facility’s layout and output.
Other pioneering applications are found in fields like wind energy, where Siemens Gamesa is working with NVIDIA to generate AI-powered digital twins of its turbines. Amsterdam’s Schiphol Airport maintains a digital twin to simulate passenger flows and energy consumption to optimize terminal operations. Carnival Cruise Line employs a virtual replica of its ships for individual, group, and ship-wide analytics looking into boarding, luggage location, housekeeping, and emergency management. Bergen, a city in Norway, employed a digital twin process in designing a new light rail extension and claims that costly errors were reduced by 25%.
The bottom line is that digital twins hold genuine promise to aid design, testing, and operations in applications where ample sensor data is available and the cost of failure is extremely high. But we’re far from the utopian vision of organizations optimizing their every move through virtual replicas symbiotically linked to their physical counterparts.
Real-Life Anecdotal Evidence
Seeking further insight, I looked for real-life evidence from individuals. One user’s perspective resonated strongly: “A digital twin is only as good as the model it is based on, necessitating dedicated resources to maintain and update the model, the sensors, and the infrastructure continuously.” This implies that beyond the annual licensing costs, companies need to allocate significant funds for model maintenance.
Another user, who is a chemical engineer, concurred, commenting that digital twins are “like trying to fill a bottomless pit with money.” The data inputs from sensors are rarely robust enough and the computational models themselves lack the precision necessary for the promised accurate forecasting in real time. This engineer explained that every digital twin model he has seen uses flawed equilibrium assumptions, while real-world processes are rarely in equilibrium. Dynamic models fare even worse. For any company considering a digital twin initiative, he suggested asking the software vendor for a list of demonstrably successful implementations, which he believes would be sparse on most occasions.
However, amidst all the criticisms, I did manage to find some real-life success stories. For instance, in process automation, digital twins have proven to be invaluable for software factory acceptance testing and operator training. Another pharmaceutical plant control engineer stated that digital twins are excellent for testing software changes in a simulated environment. For example, they can simulate different process scenarios to ensure the logic handles every case before deploying changes on the actual equipment. This is invaluable for factory acceptance testing.
Market Size and Key Drivers
The hype surrounding digital twins has driven market estimates to astronomical figures. McKinsey & Company predicts digital twin investments exceeding $48 billion by 2026, Autodesk suggests $89 billion by 2028, and MarketsandMarkets projects a whopping $110.1 billion by the same year. North America is poised to dominate this market, thanks to its abundance of tech companies, capital, and investments in digitization. Predictive maintenance is set to account for the largest portion of digital twins’ market growth. However, in reality, I couldn’t find tangible examples where this technology will actually be able to predict a breakdown better than a good old facilities manager who knows his building inside out.
Curious about the labor market’s response, I conducted a quick job search related to digital twins. To my surprise, the search yielded more than 100 jobs, which were mostly in IT infrastructure, construction, and aerospace. The required experience for the jobs entailed engineering/physics degrees; computer science and software development; knowing coding languages, computer-aided design, Agile, building information modeling, Revit, C++, enterprise data warehouse, SAP, and Python; and similar knowledge/qualifications.
The buzz surrounding digital twins is undeniable, and they have made inroads into various industries. However, their effectiveness varies depending on the sector and the investment in accurate data models and maintenance. As we navigate this evolving landscape, one thing remains clear: Digital twins are not a one-size-fits-all solution, and their true potential lies in specific niches where they can genuinely make a difference. For now, be wary of overzealous marketing claims and focus digital twin initiatives on high-value use cases where virtual modeling can make an immediate impact. The revolutionary effects prophesied by proponents are coming, but until then, digital twins will remain more fiction than fact for most organizations outside of narrow, data-rich environments.
