TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

Digital Twins Are Modeling the World

If your enterprise is considering adopting digital twins, the best advice may be to start simply.

• By Barry Devlin
• March 30, 2018

The concept of digital twins recently made it into Gartner's 2018 list of Top 10 Strategic Technology Trends, paired with more obviously mainstream concepts such as artificial intelligence (AI), analytics, and the Internet of Things (IoT). Where are these twins going? What should you do about them?

For Further Reading: Protect Your Network: Use Caution with IoT Data Sources 3 Keys to Maximizing Machine Learning in Your Enterprise How Predictive Analytics Will Change the Supply Chain of Tomorrow

Defining Doubles

What is a digital twin? Ian Skerrett writes on Medium: "In simple words, a digital twin is the virtual representation of a physical asset." A software-based, virtual representation of something in the physical world is traditionally simply called a model. Going back to high school physics, the motion of a pendulum can be described by a mathematical equation, which can be translated into a BASIC program, et voilà, you have a model of a pendulum.

Is it a digital twin? That depends on how completely the model describes the physical reality of the pendulum. Our basic (and BASIC) program might only capture its motion in a vacuum -- the simplest case. Our program could be extended in stages to capture the initial push that sets the pendulum in motion, the decay of its swing in air, then in the case when a wind is blowing, and if the string were to break, and so on.

As the model represents more of the object's behaviors, covering more of its life cycle, it becomes a digital twin of the real-world object. At a certain point in this evolution, we might argue that high school students no longer need to experiment with a real pendulum. All necessary learning can be derived from playing with its digital twin.

Enterprise Applications

Applying the same logic to industry, a software model of a truck's engine, for example, can first be derived from the laws of physics. This is useful in the early stages of design. However, a digital twin also carries information gathered from the behavior of the engine in real situations -- when it is new or well-worn; when operated in different conditions of temperature, humidity, or pollution; when it's driven hard or carefully. This is more useful, allowing engineers to predict performance of both existing engines and proposed changes in newer designs.

A truck's engine is, of course, only one part of a truck. Its behavior and performance over its lifetime is inextricably linked to all the other components of the truck, from the drive system to the fuel supply, from brakes to wheel bearings. This is the leading edge of digital twins today, as described to me by Dave McCarthy of Bsquare. That company is working with truck maker PACCAR to create such comprehensive models to improve overall truck reliability, maximize uptime, and enable predictive maintenance.

Useful as this may be, it hardly explains why digital twins appear as one of 10 top strategic technology trends. For that, we need to look further into the future, where we are not just modeling a single truck, but including all its environs and any situation it may encounter.

The Future of Digital Twins

IBM describes its digital twin vision this way: "The digital twin is a virtual doppelganger of ... a complex ecosystem of connected things, such as an autonomous car in the middle of rush-hour traffic in Los Angeles. It's not just a 3D model -- it's a living model in 3D that sees the car as part of a complex technology ecosystem of electronics, navigation, communication and entertainment, collision avoidance, climate control, and so on."

What is missing from that model is the human element. Should the digital twin also represent the potential behaviors of drivers of older, not-yet-autonomous vehicles as well as cyclists and pedestrians dodging the rush-hour traffic? Gartner's strategic technology trends document envisages the extension of the concept to human medicine and to digital twins of smart cities "infused with AI-based capabilities to enable advanced simulation, operation, and analysis."

Of course, significant gaps remain between these visions of the future and today's messy reality. Data from IoT devices is notoriously noisy and incomplete. Most interesting behaviors comprise data from multiple devices (even within a single machine). So far, there are no standards for data sharing and merging between devices from different vendors. The reality is that we are still in the early stages of digital twin evolution in the IoT world.

Nonetheless, the importance of the digital twin concept should not be underestimated. It isn't a new concept. It is, in fact, some 15 years old. The change today is in the explosion of intelligence and connectivity of physical devices throughout the world, presenting both the opportunity and the need to model these complex systems accurately and extensively, even before unleashing them in the wild.

For enterprises beginning their journey on the Internet of Things and considering adopting digital twins, the advice is to start simply. Work with data from discrete sensors first. Master the challenges of dirty data. Extend to devices consisting of multiple sensors and actuators. Learn from the experience of major engineering companies that have already trodden this path with older, proprietary technology.

The Taoist philosopher, Lao Tzu, said that a journey of one thousand miles begins with one step. In the case of digital twins, there are two steps: the first is in the model and the second in the physical world.