At a Glance:
- Digital twins provide virtual designs of physical assets, infrastructure or systems.
- Deep learning technologies enhance digital twin capabilities and provide insights on how the physical world is affected.
- Digital twins can be implemented at any scale.
From Bizarro World to the Spiderverse, we’ve always suspected that there is an alternative universe out there. But yesterday’s fiction is today’s reality—and it comes in the form of digital twin technology enhanced with advanced artificial intelligence algorithms.
Digital twins provide virtual designs of cars, factories, buildings, planes, schools, hospitals, governments, or just about any other object or system, enabling insights into real-world behaviors of that object or system under any and every conceivable circumstance. Add to that deep learning technologies that use data gathered from real-world experience to create algorithms that will alter the behavior of the digital twin, and you have a system that provides valuable insights on how a wide variety of real-world environmental, physical or other changes will affect the real-world model the twin is based on.
When paired with deep learning technologies, digital twins answer the question, “What will the effect on my real-world model be if I do X or Y?”
The combination can lead to valuable insights. For example, a car designer can build a digital twin of a vehicle, and use that model to gauge the response of the vehicle under adverse driving conditions—wet roads, gravel or ice. The building blocks of the digital twin include data from sensors, CAD designs and engineering applications, along with real-world data analyzed by advanced deep learning models “trained” to model how it would react given specific circumstances.
Thus, designers could “create” an unprecedentedly heavy rainstorm, and gauge how the digital model handles the challenge—providing valuable insight on how the real-world model will act as well.
The same dual technique of building a digital twin and applying deep learning technology to it has applications in almost everything you can think of, including manufactured products, factory and building design and construction, mining, rocket design, infrastructure, medical care systems and much more.
The system provides insights that will allow for more efficient planning, purchasing, manufacturing and management—saving organizations large amounts of time and money. Digital twin technology that is not enhanced with deep learning won’t provide those kinds of insights.
Deep Learning-Enhanced Digital Twin Technology
It’s been a long road to the development of digital twin-based intelligent systems. First generation digital twins used scripting languages and building blocks to create models resembling the physical world, while second generation digital twins were introduced for industrial applications and other uses around 2012.
We are now in the third generation of digital twin technology, considered an essential aspect of Manufacturing 4.0. Once a digital twin is developed, data relating to the environment it operates in is gathered from time-sensitive sensor, control and other systems. Algorithms enhanced with deep learning technology utilize that data to determine the effect on the twin based on changes in size/scale, temperature fluctuations, chemical reaction or any other change in the twin’s universe.
And the technology can be used to design not just objects, but entire manufacturing plants. In one example, designers develop a digital twin based on the different components and systems in a plant, such as hazardous operations, process, controls, machinery, operator skills and more. The model is built based on input, and subjected to different scenarios based on the deep learning algorithms to cover all possible conditions surrounding construction, operations, output, efficiency and automation.
It’s Not Just for Factories, Either
The value of deep learning-enhanced digital twin technology goes far beyond manufactured products and manufacturing facilities. The technology could be used for processes and systems. For example, a digital twin of a supply chain, along with data on schedules, supplies, sources and delays could allow a company to analyze scenarios involving delays in deliveries, the impact on customers, anticipated price fluctuations of raw materials due to shortages or delivery problems.
A real-world example of this was implemented recently by one of the world’s leading makers of industrial parts made with 3D printing. To build those parts, the company uses powerful laser beams to melt layers of metallic powder, “stacking” each 2D slice at a time. The lasers that do this need to be extremely accurate, with the laser energy precisely measured and directed to ensure proper production.
Using digital twin technology, the manufacturer can use real-world data to accurately predict the heat map and detect anomalies that could occur. With the technology, the manufacturer knows exactly how much energy to apply for each layer—and ensure that a minimum of errors and anomalies crop up.
Building a Deep Learning-Enhanced Digital Twin
Deep learning-enhanced digital twin technology can be implemented on any scale, even for a single component or process; the method is the same as for larger systems. In a vehicle manufacturing plant, for example, the system will indicate if something needs to be done to enhance the performance of vehicle components in a rainstorm. But that single change will affect the entire production cycle, which will need to be altered to account for this one change to a single component.
Experts say it’s best to start small—building a digital twin for just one component and process. Of course, digital twins work on a much larger scale as well, targeting different components on a regular basis to ensure they are functioning as they should, and automating the process throughout the entire system.
From the small, you can move on to the (very) large. If deep learning-enhanced digital twin technology can enhance the development of a single object, manufacturing plants or hospitals, there’s no reason it can’t be used to enhance development of more than one—along with the schools, transportation and energy systems, housing and other aspects of a smart city.
Though fairly new, the technology is quickly becoming key for the development of systems of all kinds. As development makes the technology even more viable and precise, it will be applied in more use cases—enhancing the safety, quality and efficiency of the products and processes we rely on each and every day.
Faustino Gomez is CEO and co-founder at NNAISENSE. Shortly after receiving his Ph.D. in artificial intelligence from the University of Texas at Austin in 2003, he joined the Swiss AI Lab, IDSIA, initially as a post-doctoral researcher and then senior researcher working with Dr. Juergen Schmidhuber. He has published more than 50 papers in the fields of neural networks, evolutionary computation, machine learning and reinforcement learning.