I wrote this article for Age of Robots and I like to share it with you on this blog since many of the ideas discussed derive from the work being done in the IEEE FDC Symbiotic Autonomous System Initiative. By the way, this will be one of the key topic discussed at the SAS workshop on October 30th 2018, colocated with TTM Conference. Don’t miss it!
Digital twins can represent objects and entities as varied as a turbine, a robot, a whole ship, a cow, a human being or a city, and everything else in between. More recently they have started to be used to represent intangible entities like services, processes and knowledge.
Digital twins are already used in design, planning, manufacturing, operation, simulation and forecasting. They are also used in agriculture, transportation, health care and entertainment. Applications will continue to grow through the next decade, hence it is not surprising that they named among the ten most strategic emerging concepts for the coming years by Gartner, or that MPL Systems expects 25% of asset-intensive companies to be using them by 2020 and supporting technology spending of $10.96 billion in 2022.
When considering digital twins, the key word is “represents”. Basically, a digital twin mimics in bits an object’s atoms and their structural/functional relations. It does not necessarily represent all of them (something conceptually impossible, as you cannot represent a single atomic electron cloud with unlimited precision), but what matters is that the representation is accurate enough to support the goals that have been identified and that are being pursued. For example, if you want to check the proper working of an engine you need to represent all aspects that are functional to that goal (e.g., you may disregard the color used to paint parts of that engine). However, if you are mirroring a car then the color of the paint is important because retouching a car after an accident requires knowing the original paint color.
Note that a digital twin can also, and usually it does, contain more data than its real counterpart. As an example, an engine’s digital twin is likely to contain the list of suppliers of the various components of the engine as well as the identity of the robots and workers that assembled it. A digital twin is also a historical repository of its counterpart. Thus, in the case of an engine, it may include extensive data on maintenance events and operations, for example, the minute-to-minute monitoring of airplane engine data including rotation speed, oil usage, pressure, temperature, and so forth.
All these data sets can be used for real-time analyses and simulation. They can also be used collectively to identify patterns and meanings. Take the example of General Electric (GE), which creates a digital twin for each of the turbines they produce. Once these turbines are assembled on a windmill to generate electricity or deployed on an aircraft to generate thrust, the turbines report operation status back to GE in quasi real time. This information is then compared with data generated by each unique digital twin for consistency. Any deviation activates an application to analyse the discrepancy and take action if needed, such as ordering the turbine flying on the plane to reduce power and decreasing the rotation speed to safeguard the integrity of the engine. Of course, this affects the other digital twin engine on the aircraft—in this case making sure that balancing measures are implemented by increasing the thrust of the other engine and repositioning the wings’ moving parts to maintain equilibrium. At the same time, the applications will look for an emerging pattern related to the situation (present and past) of other digital twins and will store data on any mismatches.
There are similar scenarios with our own digital twins. For instance, I have a bunch of data on Facebook that identifies my friends, information on my travel logged on Instagram, and a Twitter account that shows my reactions to events. Additionally, I have “sensors” on my body (a smartphone) that tracks my daily activity and provides further data. More data may come from my health records, if I am willing to share them, and in the future I could even have the data from my sequenced genome. Applications can continuously analyse my digital twin and detect emerging patterns that may require attention. This is particularly true in the health care domain, where Bill Ruh, CEO of GE Digital, in a recent presentation stated, “I believe we will end up with health care being the ultimate digital twin.”
Similar to a digital twin of an object like a turbine, my digital twin is more than a representation of myself within a specific domain of interest. It contains the copy of my past, very possibly keeping memories of something I forgot long ago. There is a need, therefore, to distinguish between the instantaneous digital twin, which represents me at a specific moment in a specific context, and my global digital twin that remembers what I had for dinner a year ago and what pill I took to ease my digestion.
A digital twin can be used to monitor its real twin and to simulate the effect of some actions (e.g., increasing the rotation speed of a turbine or changing a person’s diet). It can be used to derive relevant information from other digital twins, such as detecting a malfunction that could affect other turbines or determining the side effects of a particular remedy. Statistical information and pattern data can be used to monitor changes in a particular activity, for example, turbines on a specific assembly line showing a power decrease in certain conditions or several persons taking two different kind of pills being subjected to undesirable side effects.
Digital twins can also become “impersonators”, where they can act out in cyberspace the part of the object in the real space. Hence, they can be used when designing a new object to study the interactions that may happen, as well as to solicit a digital twin to learn from those interactions, and then to transfer what has been learnt to the physical object. This may be particularly useful in robotics where the digital twin of a robot can be solicited by other digital twins, including ones representing a specific situation, and can try different approaches to find the most effective one. This experience (or knowledge) can then be shared with/downloaded to the real robot, providing it with an experience that it could not have had in the real space, perhaps because the situation could not be replicated at will or because the “real” experience might damage the robot.
There are many situations that are easier and cheaper to replicate in cyberspace with no likelihood of collateral damage and many examples of applications getting smarter by challenging themselves and learning from the experience. In the future, digital twins may become an essential component in the evolution of machines and the growing symbiosis between humans and machines. This is a major topic currently being studied at the IEEE FDC Symbiotic Autonomous Systems Initiative that will be also addressed later this year at a workshop to be held in San Diego on 30 October in conjunction with the 2018 IEEE Technology Time Machine Conference, San Diego, October 31–November 1.
Over the next decade, many objects will be created with their own digital twin and will live in symbiosis with them. In various situations, as I’ve described, robot intelligence will emerge through interactions with a digital twin in cyberspace, and over time the evolution and continuous learning of robots and other objects will undoubtedly be fostered by digital twins.
As with any new technology, digital twins are prompting ethical, legal and societal questions. Imagine a company with hundreds of human and robot workers, each one with a digital twin! Suppose a robot breaks down and needs to be replaced—wouldn’t it be normal to associate the new robot to the previous robot digital twin so that it immediately inherits the previous robot’s experience? Of course, no discussion about that.
Now consider a human worker who decides to retire, or just change jobs. What about her digital twin? Will it remain the property of the company and as such be used by the company to train a new worker? What might happen when it becomes feasible to replace a worker with a robot? Can the company associate the human digital twin to the robot, hence transferring the experience previously accumulated by the human worker to the robot? Is it possible in the future that companies will hire humans just for creating a digital twin that along with a robot will make them redundant?
As a robot digital twin learns by interacting with a human digital twin, is it likely to become smarter and smarter, accelerating the process of human displacement in factories and, more generally, in the labour market? These are just a few of the questions that are popping up as we walk the unexplored trails heading towards a future that probably is just around the corner.