Editor’s Note: Dr. Anand Srinivasa Rao is a speaker for ODSC Europe 2022. Be sure to check out his talk, “Digital Twins: Not All Digital Twins are Identical,” there!
As we try to bridge the gap between digital and physical systems, we increasingly hear about “digital twins.” Like many other concepts (e.g., Artificial Intelligence or Metaverse) the term “digital twins” can mean very different things to different people. For some, a digital twin is intimately associated with the Internet of Things (IoT) and is the digital equivalent of a sensor or a physical asset (e.g, an aircraft engine). It allows them to experiment with the digital version that they may not be able to do with the physical system. For example, subject the digital version of the engine to ten times the stress to evaluate its performance which you may not want to do with the real physical engine while the aircraft is flying or even on the ground for experimental purposes. For others, a digital twin is an avatar in the Metaverse that is interacting with other avatars. For some others, a digital twin is a virtual human making decisions in the digital world that can inform decisions and actions in the physical world.
In this article, we discuss the foundational elements of digital twins, including its history, evolution, benefits, and how to help develop the capability around digital twins.
What is a Digital Twin?
A digital twin is defined as follows: “A digital twin is a virtual representation of real-world entities and processes, synchronized at a specified frequency and fidelity”. It is worth diving into a couple of the key terms used in this definition. First, a digital twin is a virtual or digital representation. The digital nature of this representation allows us to manipulate the underlying structure and behavior of this virtual object. Second, a digital twin is a representation of real-world entities. Depending on what one is trying to model the real-world entities can range from sensors to machinery, to people, to organizational entities. Third, a digital twin can also be a representation of a process (e.g., flying a drone or driving an autonomous vehicle). Fourth, the digital and physical representations are synchronized. This allows each of them to be a ‘replica’ of the other or a true ‘twin’. This synchronization is a two-way synchronization where the state of the physical system is reflected in the digital system and the state of the digital system is reflected in the physical system.
A digital twin is sometimes contrasted with digital models and digital shadows. A digital model is a virtual replica of a physical model. Once the digital model is built, it can be studied and improved upon in the digital world and these changes can then be made in the physical world. So, the information flows from the digital to the physical world. In the case of a digital shadow, the real-time information from the physical asset is used to build the digital representation. The information flow is from the physical to the digital world. In a digital twin, information flows from the physical world to the digital world and vice versa.
History of Digital Twins
The concept of ‘digital twin’ is attributed to a 2002 paper on product lifecycle management where the physical and digital spaces were introduced and the flow of information between these two spaces was discussed (See Origins of Digital Twin Concept). The concept was further expanded in 2011 and was referred to as the Information Mirroring Model in the book Virtually Perfect: Driving Innovation and Lean Products through Product Lifecycle Management.
However, the essential aspect of simulating the physical world through a digital model can go back even further. Arguably, agent-based models (or multi-agent systems) and computer simulations are the parents of digital twins. Artificial Intelligence since the days of Alan Turing has been all about mimicking humans and other ‘intelligent’ entities. The concept of an ‘intelligent agent’ as any entity that is embedded in an environment, perceiving the world through sensors, making decisions, and acting in the environment through actuators to achieve certain objectives or goals has been around since the 1990s. Artificial Intelligence is often defined as the study of intelligent agents. One of the first robots, Shakey built by SRI International, is an intelligent agent that moves around in physical spaces by modeling these spaces digitally and then finding the appropriate path. The synchronization between the physical and the digital worlds that the Digital Twin community talks about was present in Shakey the robot. Early definitions of ‘intelligent agents’ also refer to thermostats as one of the simplest forms of agents – no different from today’s digital twins modeling IoT sensors.
The history of computer simulation dates back to the early days of computing history during World War II. The great mathematician Jon Von Neumann and his associate Stanislaw Ulam solved the problem of the behavior of neutrons using computer simulation. Early computer simulations were primarily used in military and aerospace applications. As computing became more ubiquitous computer simulation was applied to a variety of scientific, engineering, and business applications. Agent-based simulations combined the notion of building intelligent agents that ‘mimicked’ the physical entities with the simulation of the physical environment to allow for the testing of these intelligent entities in the digital world.
Benefits of Digital Twins
Digital twins, both in their recent conception as well as the more classical definition of intelligent agents, play five key roles.
- Explain: Digital twins can be used to explain what happened in the past. For example, in the case of sensor-based digital twins one can look for patterns and anomalies in sensor readings.
- Predict: Digital twins can be used to predict future behaviors. For example, given the historical behavior of a specific sensor one can predict future values.
- Explore: The power of the digital twin comes when we can explore alternatives in the digital world and evaluate these alternatives against our desired criteria.
- Change: When we can change the physical world based on the explorations in the digital world, we can derive the full power of digital twins.
- Generate: While we can explore alternative scenarios if we have historical data, we can also generate potential values to test the physical world. This use of digital twins gives a powerful tool to operate with when we have incomplete and uncertain data.
Since the pandemic, the use of digital twins and agent-based simulations seems to have increased significantly as companies have found the many benefits of exploring alternative COVID19 disease progression scenarios and the resulting economic consequences in terms of V-shaped, U-shaped, or W-shaped economic recovery. According to PWC’s 2022 AI Business Survey, nearly 96% of survey respondents said they plan to use AI simulations this year.
Maturity Model of Digital Twins
As companies aim to develop digital twin capabilities it is useful to consider the full spectrum of maturity of digital twins. The simplest form of digital twin one can build is a data twin. A data twin is a digital log of an asset or customer or organizational entity. One can build the data twin using real-world data or synthetic data. Such data twins can also be real-time or non-real-time data twins. For example, a digital replica of a water level sensor used for flood warnings is an example of a digital twin. This digital twin will record the water levels at different points in time based on the design of the sensor and the data transmission mechanism. The levels may be measured every second, minute, hour, etc., or could be based on pre-set values. This data could be stored or transmitted elsewhere in real-time or on a periodic basis. It is not essential that we always have real-world data to build a data twin. We can also synthesize or generate data to represent a specific distribution e.g., rising water levels during a flood.
With the historical log of a data twin, we can help start simulating the physical behavior of an asset. We call this the asset simulation twin. This is the most common type of digital twin used by companies in Industry 4.0 or the world of IoT. The physics of most physical assets, starting from simple sensors to complex systems like aircraft engines, can be specified, understood, and evaluated. Historical data from such digital twins can be used to understand anomalous patterns and predict potential machine failures. For example, based on historical data from temperature and pressure sensors at the tip of oil drilling machines one can predict the failure patterns of these sensors and pro-actively schedule maintenance saving millions of dollars in reduced downtime.
As we move from digital twins of physical assets to decision-making choices of cognitive entities (e.g., customers or organizational entities) we enter the realm of traditional AI or intelligent agents). We call these Choice Simulation Twins. Modeling customer choice functions has been extensively studied in economics to both explain and predict customer preferences and choices. AI has focused on the study of rational agents and how they make decisions in an uncertain and changing environment made up of other agents. For example, we have built customer choice simulation twins to understand, predict, and change the behavior of how customers respond to product promotions. Such choice simulation digital twins provide a rich playground for marketing experts to evaluate different product promotion strategies and their impact on revenues and margins.
Combining machine learning with the choice functions of cognitive agents or the physics of physical systems leads us to more sophisticated Cognitive Learning Twins. Cognitive learning twins can not only explain, predict, and explore alternatives they can also change the way digital twins improve their performance over time. For example, one can build cognitive learning twins that can learn how customers respond to competitive offers and adjust the offers to maximize revenues or margins or any other specified criteria.
When cognitive learning twins can adapt to the changes in the environment and can not only learn from the environment and other digital twins but can also change their learning strategy to adapt to new environmental uncertainties, we have truly Adaptive Autonomous Twins. For example, one can build autonomous vehicles that continuously learn and adapt to changing road conditions. The figure below shows the maturity curve for digital twins.
Building Digital Twin Capability
Several companies want to build digital twin capability to fully explore the emerging areas of IoT, Metaverse, and other new technologies. We recommend companies focus on three critical areas to help achieve this.
First, companies should institutionalize a systems thinking mindset within the entire organization. This requires companies to work outside their traditional boundaries and silos. Companies should go behind the effects or outcomes of their strategies and actions to multiple layers of causation. This will allow companies to address business-level issues with a focus on decisions and/or actions that are under their control and those that are determined by the external environment. This will allow them to design the digital twins in an appropriate fashion.
Second, companies should incorporate digital twins as part of their overall IT stack and allow digital twins to interact and incorporate other analytics and machine learning models. The software and model development lifecycle should incorporate digital twin development and the infrastructure should support the building, calibration, scaling, and deployment of digital twins.
Finally, companies should recruit, retain, and grow digital twin talent. Unlike analytics and machine learning which are generally more appealing to those with a scientific mindset, digital twin development is more appealing to someone with an engineering mindset.
For most organizations, the digital twin journey has just begun, but it promises to provide the necessary return in seamlessly integrating the physical and digital worlds to reduce operational costs and enhance decision making.
About the Author/ODSC Europe 2022 Speaker:
Dr. Anand S. Rao is the Global Artificial Intelligence Leader for PwC. He is also the leader of PwCs AI and Emerging Technology practice. With over 35 years of industry and consulting experience, Anand leads a team of practitioners who advise C-level executives and implement advanced analytics and AI-based solutions on a variety of strategic, operational, and ethical use cases. With his PhD and research career in Artificial Intelligence and his subsequent experience in management consulting he brings business domain knowledge, software engineer expertise, and statistical expertise to generate unique insights into the practice of data science.