Unleashing the Power of Generative AI: Transforming Digital Twins into Intelligent Twins
Digital twins are not inherently generative in nature, but the evolution of generative AI is poised to revolutionize them with unprecedented intelligence and predictive capabilities, thus transforming them into "intelligent twins”.
Digital twin is a virtual representation of a physical entity – an object, behavior, process, system, or product. It is essentially an animate ecosystem that is trained to mirror and sync with the physical entity. A digital twin is created by collecting real-time data from sensors embedded in physical entities and using that data to create a digital replica. This digital replica can then be used for various purposes such as monitoring, analysis, simulation, and optimization.
Select any image to see a larger version.
Mobile users: To view the images, select the "Full" version at the bottom of the page.
A digital twin ecosystem is an inherently dynamic, interconnected network comprised of software, generative and non-generative algorithms, batch and streaming data, and of course, business logic. Digital twins often converge AI/ML, simulation, forecasting, optimization, and visual and streaming analytics to stress test a physical entity in the digital world to prescribe actions that optimize the entity in the physical world – to improve the lives of individuals, the lives of populations, the environment, cities, organizations, systems, products, and more.
A simple example of a Factory Digital Twin is illustrated in a SAS post authored by Biller, B. (2022) titled Digital twin development: Why simulations are critical.
Generative AI as a Digital Twin
Generative AI is set to revolutionize digital twins by transforming them into “intelligent twins”.
Envision a friendly looking doctor wearing a white lab coat with a stethoscope around her neck. She has a warm and approachable smile, exuding confidence and professionalism. Her eyes convey empathy and understanding, and she appears focused yet compassionate. This is an AI-generated version of your doctor.
AI-generated version of doctor is created using https://app.simplified.com/graphic-design/tools/text-to-image/results
Now, imagine a scenario where you can post a question to this virtual persona who embodies expertise, care, and a dedication to helping others and could serve as a virtual advisor. This AI-powered digital twin would respond to your queries based on the knowledge it has acquired from the imperfections and behavioral traits of your doctor. It goes beyond simply analyzing real data. Instead, it taps into the essence of your doctor's persona and assisting in decision-making, providing expert insights, and even offering personalized treatment recommendations.
This concept of generative AI digital twins is still in its early stages, but it has already captured the attention of researchers and industry experts.
Generative Ecosystem of Digital Twins
Digital twins can be generative in nature. The simulation aspect of a digital twin is often emphasized, but the role of generative AI is severely undervalued.
The interconnected networks in a digital twin are comprised of numerous deep learning algorithms that are inherently generative. For example, generative adversarial networks (GANs) are increasingly being adopted in the field.
A digital twin uses real-time data to generate or create simulations that mirror the behavior and characteristics of real-world entities. Besides, a digital twin uses generative outputs as automated inputs to the ecosystem. For example, it generates synthetic data as an output from the ecosystem that is, in turn, used as a feedback loop and contributes back as an input into the ecosystem to optimize the ecosystem.
A digital twin generates new data and insights about an entity’s behavior in different scenarios and conditions to proactively guide real-world strategy and to improve real-world outcomes – not as a static replica, but as an animate virtual representation of an entity.
Similarly, it generates new observations to continuously learn and adapt to changes in the real-world entity, leading to more accurate predictions and better decision making about the real-world entity.
What is more, a digital twin generates large amounts of synthetic data in a controlled ecosystem when real-world data is limited or too expensive to collect. This real-time synthetic data creates simulations to mirror and emulate real-world objects, systems, or processes.
Methodological Perspective of Digital Twin Ecosystem
Frank Emmert-Streib presented a methodological perspective of a digital twin system (shown below in a light blue rectangle) with an interface to a physical entity and external data. Its structural architecture highlights the various contributions that AI can offer for optimizing the outcomes of a digital twin. These contributions will be represented from 1 to 6.
A digital twin is created corresponding to a dynamic system that is implemented via a computer simulation capturing essential features of the physical entity. The physical entity to be modeled can come from a wide range of applications including engineering, manufacturing, healthcare, urban development, and more. Usually, this involves external data.
Source: Emmert-Streib, F. What Is the Role of AI for Digital Twins? AI 2023, 4, 721-728.
The additional external data is used to estimate the parameters of the digital twin. This is an optimization process of model creation, the first AI contribution. Importantly, there is another optimization step involving AI that is different from the first one.
The second optimization step ensures the synchronization of the digital twin and its physical counterpart during its operation. This leads to the optimization of the updating mechanism of the digital twin called model updating.
The third AI involvement is generative modeling. For example, a GAN can be used to generate data with characteristics learned from large-scale data to replace or complement a traditional simulation model. That means AI does not only help optimize a simulation model, but it can constitute the simulation model itself.
The fourth and fifth AI associations are principally the data analysis steps. However, the sources of the data for such an analysis are entirely different.
External data is thoroughly analyzed to identify patterns and relationships. This data analysis step is called descriptive analytics. Descriptive analytics seeks to describe an event, phenomenon, or outcome. It helps understand what has happened in the past and provides businesses the perfect base to track trends and can also be used to compare the performance with others within the same industry.
A lot of data is generated by the digital twin. The data generated by the digital twin and the data that is coming from descriptive analytics using the external data are combinedly used to predict future outcomes like component failure of a particular component in a manufacturing unit. This process is called predictive analytics. Evidently, the source of data has a crucial influence on the interpretation of an analysis. Hence, descriptive analytics and predictive analytics provide complementary means, and the technique we discuss next aims at their integration.
The sixth AI application, called decision making, is used for summarizing all individual results achieved up to this point to enable decision makers to base their choices on factual data and insights rather than intuition or assumptions without testing the physical entity. This step integrates everything together and produces a quantitative or qualitative summary that can be seen as the ultimate output of a digital twin system.
How Digital Twin Is Different from Simulation
Data simulation is not a new idea, and it has been a fundamental tool for statistical modeling. Digital twin technology is similar to simulations as both utilize digital models to replicate various processes of a system. However, there are some key differences between the two.
The most notable is that a simulation model is a brute-force computational technique that relies on repeating a computation on many different static random samples to estimate a statistical quantity where modelers must set all parameters manually, whereas digital twin technology uses real-time data to mimic processes by integrating all relevant data into a digital system, effectively mirroring the entirety of the product, service, or process.
It is important to understand that a simulation model is not equipped with an updating mechanism, whereas a digital twin is a mathematical model with an updating mechanism. Digital twins are designed around a two-way flow of information that first occurs when object sensors provide relevant data to the system processor and then happens again when insights created by the processor are shared back with the original source object.
A simulation replicates what could happen to a product, service, or process. Any changes to a simulation are limited to the imagination of the modeler. But a digital twin replicates what is happening to an actual outcome in the real world as it offers active interaction. Thus, the modeler can see if it is working as intended and then determine any improvements based on actual use.
The difference between digital twin and simulation is largely a matter of scale. A simulation studies one process, but a digital twin can itself run any number of useful simulations to study multiple processes. A digital twin is far more flexible, allowing modelers to test the entity through all stages in many different settings. As long as you have the data, the digital twin will replicate the entire process. Digital twin simulations are more versatile and offer a deeper understanding of processes when compared to simulations.
Concluding Remarks
Generative AI represents a paradigm shift in the realm of digital twins, unlocking new possibilities for intelligent decision-making, predictive analysis, and scenario forecasting. By harnessing the power of generative models, digital twins are evolving into intelligent twins that not only replicate physical entities but also simulate, predict, and optimize their behavior in real-time. As this transformative technology continues to mature, it holds the promise of revolutionizing diverse industries and shaping the future of smart, interconnected systems.
Find more articles from SAS Global Enablement and Learning here.
